Active prey attraction and capture
Field observations and herbarium specimen analyses show that gnats belonging to several genera (e.g., Polypedilum, Chironomus, Cricotopus, Dicrotendipes, and Paratanytarsus) of the family Chironomidae were the main insect visitors found trapped by glandular hairs of Saxifraga candelabrum (all insects identified by entomologists; Supplementary Data 1). In contrast, species from Calliphoridae, Muscidae, Syrphidae, Sarcophagidae, Colletidae, and Halictidae were the main uncaptured flower-visiting (potential pollinator) insects (Supplementary Data 1, Fig. S1). Field observations demonstrated that the number of insects trapped on S. candelabrum individuals correlated positively with both the number (Spearman’s rank correlation ρ = 0.38, n = 30, p = 0.038) and length (Spearman’s rank correlation ρ = 0.33, n = 92, p = 0.001) of inflorescence branches (Fig. 4). Juvenile plants with only a few branches exhibited minimal or no insect trapping. In contrast, mature individuals harbored an average of 71 trapped insects, predominantly localized on the glandular hairs of inflorescence branches, with significantly fewer occurrences observed on stems. This observed correlation suggested that the increased number and length of inflorescence branches may enhance insect attraction and provide a larger surface area bearing glandular hairs, potentially increasing the probability of insect capture.
In addition, examination of all available flowering herbarium specimens of S. candelabrum revealed that insect prey were present on glandular hairs in the vast majority of specimens (43 out of 45), spanning both the geographical range and the entire temporal span represented in the collections (Fig. 4, Supplementary Data 2). This finding indicates that insect trapping by glandular hairs is a recurrent and widespread feature of the species, rather than a sporadic or recent phenomenon.
To test whether floral visual or olfactory attractiveness was responsible for the active attraction of insects, we conducted a field experiment in which we enclosed the flowering individuals of S. candelabrum either with a glass enclosure (odor-blocked) or with a white net enclosure (visual-blocked). More flower-visiting insects were observed in the visual-blocked and the untreated groups than in both the odor-blocked and the net-only control groups (Fig. 4b). The visual-blocked group showed higher visitation than the odor-blocked group (two-tailed Welch’s t-test: t(8.38) = 2.76, p = 0.024, Fig. 4b), indicating an olfactory response in the active attraction of insects in S. candelabrum.
We further analyzed the floral volatile compounds of S. candelabrum using gas chromatography-mass spectrometry (GC-MS), which revealed a diverse suite of floral volatiles (Supplementary Data 3). Several compounds we identified, including β-myrcene and eucalyptol, have been reported to be involved in insect perception or attraction, suggesting that the floral scent of S. candelabrum contains volatiles with the potential to attract insects21,22. Using inflorescences to capture insects may result in a conflict with pollination in S. candelabrum, as Saxifraga also depends on insects for pollination (Fig. S1)23. In contrast to many carnivorous plants that spatially separate flowers and traps to reduce such conflict24,25, here we found another strategy to balance carnivory and pollination: selectively capturing non-pollinators. The effective pollinators of S. candelabrum were large-bodied insects belonging to the families Calliphoridae, Halictidae, Muscidae, Syrphidae, and Tachinidae, which were rarely observed trapped by the glandular hairs of the plant. In contrast, the glandular hairs predominantly trapped smaller insect visitors, primarily species of Chironomidae, which were largely ineffective as pollinators of S. candelabrum (Fig. S1, Supplementary Data 1).
Prey digestion and nutrition absorption
As the most widely investigated and readily detectable enzyme in studies of carnivory, phosphatase was considered a key digestive enzyme in carnivorous plants that mobilizes nutrients from prey26. Most carnivorous plants display phosphatase activity in glandular cells organized into glandular hairs26. Following treatment with the ELF97 phosphatase substrate, yellow-green fluorescence indicative of phosphatase activity was detected in the glandular hairs of the carnivorous plants Pinguicula primuliflora and S. candelabrum. This enzymatic activity indicated that S. candelabrum has the capacity to directly digest trapped insects, rather than relying on alternative decomposition processes. In contrast, S. filicaulis, a closely related and often sympatric species of S. candelabrum in Shangri-La, Yunnan, China, showed no detectable fluorescence under identical conditions, consistent with the observation that its glandular hairs rarely captured insects (Fig. 5).
In our isotopic investigation, we examined S. candelabrum and the distantly related carnivorous flypaper trap plant Drosera peltata (Droseraceae) along with two non-carnivorous sympatric species: Salvia przewalskii (Lamiaceae) and Saxifraga diversifolia. All samples used for interspecific comparisons were collected from the same locality (Wufeng Mountain, Shangri-La), although minor microhabitat differences were present within the site, as reflected by variation in natural δ¹⁵N values of local soils (Supplementary Data 4), with negative control species exhibiting higher baseline δ¹⁵N values than S. candelabrum and D. peltata (Fig. 6a). Application of 15N-labeled Drosophila to S. diversifolia, S. candelabrum, D. peltata and S. przewalskii revealed no significant change of δ15N in the non-carnivores S. przewalskii (two-tailed Welch’s t-test: t(7.64) = 0.03, p = 0.978) and S. diversifolia (two-tailed Welch’s t-test: t(3.56) = 0.20, p = 0.853) (Fig. 6, Supplementary Data 4). In contrast, δ15N showed a significant increase in D. peltata (two-tailed Welch’s t-test: t(4.00) = 4.75, p = 0.009) and S. candelabrum (two-tailed Welch’s t-test: t(10.10) = 5.86, p = 0.0001) (Fig. 6, Supplementary Data 4). The absence of isotopic enrichment in the non-carnivorous controls indicated that the observed signal in S. candelabrum cannot be explained by passive absorption or surface contamination alone.
We further compared the δ15N enrichment in different organs of S. candelabrum and found the highest δ15N enrichment in flowers, followed by cauline leaves, with the lowest δ15N enrichment in basal leaves (Fig. 6, Supplementary Data 4). The higher content of 15N accumulated in reproductive organs than vegetative organs could not be fully explained by their proximity to glandular trichomes that absorb 15N, given that cauline leaves, which were similarly close to the nitrogen sources, exhibited lower 15N levels. This contrast indicated that plants actively allocate more nitrogen to their reproductive organs, a strategy that optimizes resource use to maximize reproductive output of this monocarpic plant.
To rule out 15N-labeled nutrient uptake from soil, we measured δ15N in soil around treated individuals. Soil δ15N remained low across all treatments (Fig. 6), confirming that the δ¹⁵N enrichment in S. candelabrum was not from the soil. The decline in δ15N enrichment from 2 weeks to 20 days likely reflects the complete absorption of the limited labeled prey (10 fruit flies) within the first 2 weeks (Fig. 6), after which unlabeled soil nitrogen continued to be incorporated, diluting the isotopic signal.
Comparative genomics and carnivory
To investigate the genomics of carnivorous plants, we generated a chromosome-scale annotated genome for S. candelabrum (Figs. S2–S4; Supplementary Data 5–15). Our broad comparative analysis included: (1) available genomes of other Saxifragales (Supplementary Data 5; all non-carnivorous); (2) representative rosid lineages (e.g., Brassicales, Celastrales, Cucurbitales, Fabales, Fagales, Malpighiales, Myrtales, Rosales, Sapindales, Vitales: see Fig. S3); and (3) carnivorous species with flypaper traps (Drosera spatulata, Droseraceae; Roridula gorgonias, Roridulaceae), as well as other trap types, including pitcher traps (Cephalotus follicularis, Cephalotaceae; Nepenthes gracilis, Nepenthaceae), and snap traps (Aldrovanda vesiculosa, Droseraceae). We investigated gene family expansions, molecular convergence across independently evolved carnivorous lineages, signatures of positive selection, and assessed gene family downsizing.
Comparative genomic analysis of S. candelabrum and two distantly related sticky trap species (D. spatulata, Droseraceae; R. gorgonias, Roridulaceae) revealed 20 expanded gene families shared by all three sticky flypaper trap carnivorous lineages (Fig. 3; Supplementary Data 16). These shared genes were functionally enriched for key carnivorous traits, including prey attraction (GO:0009836, fruit ripening, climacteric)27, digestion (GO:0016788, hydrolase activity)3,9, and nutrient absorption (GO:0006817, phosphate ion transport)12.
No common gene expansions were found across all six examined carnivorous species representing a diversity of trap types. Further pairwise comparisons between S. candelabrum and each carnivorous species identified additional gene family expansions associated with distinct stages of the hunting cycle, from prey attraction and capture to digestion and nutrient absorption (Fig. S5; Supplementary Data 17–21), demonstrating putatively convergent evolutionary pathways in the genomic underpinnings of plant carnivory28,29,30,31.
Comparative genomic analysis of three distantly related carnivorous species with sticky traps (S. candelabrum, Saxifragaceae; D. spatulata, Droseraceae; R. gorgonias, Roridulaceae), revealed 53 genes exhibiting convergent evolutionary rate acceleration (Supplementary Data 22). Functional annotation demonstrated significant enrichment in key aspects of carnivory: (1) mucilage production and secretion (GO:0048193, Golgi vesicle transport)31; (2) amino acid biosynthesis (GO:0009086, methionine biosynthetic processes; GO:0009088, threonine biosynthetic process; GO:0009097, isoleucine biosynthetic processes)32; and (3) protein transport (GO:0043328, protein transport to vacuole)12.
Expanding our comparative analysis to six carnivorous species representing diverse trap types, we identified 21 genes showing convergent acceleration across all lineages (Supplementary Data 23–28), including functionally important candidates involved in amino acid metabolism (aspartate-semialdehyde dehydrogenase: ASADH), prey digestion (ATP-dependent Clp protease ATP-binding subunit ClpX-like: CLPX), and nutrient processing (golgi SNAP receptor complex member 1: GOSR1).
Our selection analyses revealed strong evidence for adaptive evolution in carnivorous lineages. Specifically, we identified 16 genes under positive selection across all three sticky trap species (S. candelabrum, D. spatulata, and R. gorgonias; Supplementary Data 29), including genes putatively involved in digestion (proteasome subunit beta type-7: PSMB7) and signal transduction (germinal center kinase 1: GCK1). However, no genes showed a signature of positive selection across all six examined carnivorous species (Supplementary Data 30–34).
Notably, we found 59 convergent amino acid substitutions in the three flypaper trap species (Supplementary Data 35), functionally enriched for leaf morphogenesis (auxin response factor 2: ARF2), phosphatase activity (trehalose 6-phosphate synthase/phosphatase: T6PS), and nutrition absorption (nuclear transport factor 2-like: NTF2-like, nuclear pore complex protein Nup98: NUP98, ATP-binding cassette, subfamily C1: ABCC1).
Expanding our analysis to six species with diverse trap types revealed 58 additional genes with convergent substitutions (Supplementary Data 36–41), including digestive enzyme genes (CTD phosphatase-like 4: CPL4, a homolog of the papain family cysteine proteases: CPs, one homolog of the glycosyl hydrolase family 10: GH_10). These findings reinforce the suggestion that convergent evolution of digestive enzymes in carnivorous plants likely reflects strong selective pressures associated with alternative nutritional strategies29,33,34,35.
Among the three sticky-trap species (S. candelabrum, D. spatulata, and R. gorgonias), we identified two convergently contracted gene families involved in cellular plasticity (GO:0030042, actin filament depolymerization), and hormone signaling (GO:0009873, ethylene-activated signaling pathway; Supplementary Data 42–45). No gene families were observed to contract across all six examined carnivorous species. Our exploration of shared gene losses in carnivorous plants revealed putative convergence of function in three key areas (Supplementary Data 46): root development (including RCP2 and RID3), kinetochore formation (H2B-related genes and MDC1-like)36,37,38, and photosynthesis (CA and NAD(P)H-quinone oxidoreductase)39. While these patterns imply potential associations, further functional studies are needed to test whether these gene losses causally underlie carnivorous adaptations.
Observation of insect attraction and capture
In Saxifraga candelabrum, insect capture occurs predominantly during the flowering (reproductive) stage, whereas only occasional insect trapping is observed during the vegetative (rosette) stage. Accordingly, insect visitors representing both uncaptured flower visitors and prey organisms trapped by glandular hairs of S. candelabrum were collected from flowering individuals in the field. All voucher specimens were deposited in the National Animal Collection Resource Center (NACRC) at the Institute of Zoology, Chinese Academy of Sciences, for taxonomic identification and permanent curation (Supplementary Data 1). To quantify prey capture in living plants and test the relationship between the number of trapped insects and trapping surface, we recorded the number and length of inflorescence branches, as well as the number of trapped insects on each branch, for 32 full-blooming individuals of S. candelabrum from Wufeng Mountain of Yunnan province in China. The relationships between the number of trapped insects and inflorescence branch number and length were assessed using Spearman’s rank correlation. Linear models were used solely to visualize trends, with shaded areas indicating 95% confidence intervals.
Additionally, to assess the consistency of insect trapping across the species’ geographical and temporal distribution, all available flowering specimens of S. candelabrum from the Herbarium of Kunming Institute of Botany, Chinese Academy of Sciences (KUN) were examined, and the number of trapped insects (prey) on glandular hairs was recorded (45 specimens spanning 1888–2016; Supplementary Data 2).
To test whether floral visual or olfactory cues attract insect visitors to S. candelabrum (Fig. S1), we conducted an experiment on Wufeng Mountain with the following treatments: (1) Odor-blocked: flowering individuals of S. candelabrum were enclosed within a transparent glass enclosure (visual attraction but no odor, with the open end of the glass enclosure sealed using transparent plastic film to minimize floral odor escape); (2) Visual-blocked: flowering individuals of S. candelabrum were covered with a white breathable net enclosure (permitting odor attraction); (3) Net-only control: white breathable net enclosure without S. candelabrum in it (testing the possibility of color attraction by the white netting); (4) Untreated: flowering individuals of S. candelabrum without any treatment. Each treatment included 6–16 flowering individuals of S. candelabrum in full bloom. Each individual was observed for 10 minutes, and the number of insect visitors was recorded during this period. Differences in the number of insect visitors among treatments were tested using two-tailed Welch’s t-tests.
Floral volatiles collection and component analysis
The volatile compounds of Saxifraga candelabrum flowers were collected using the dynamic headspace sampling method39. Floral volatile samples were collected from three independent flowering individuals of Saxifraga candelabrum (SC1–SC3), each with three replicates (−1 to −3). Flowers of S. candelabrum were enclosed in odorless, transparent polytetrafluoroethylene (PTFE) collection bags. The air from a ventilation pump was purified through a charcoal filter, primarily to remove moisture and dust from the air. The purified air then entered the collection bag through its inlet, promoting the volatilization of floral odors inside the bag. Finally, the volatile compounds were extracted from the bag’s outlet and transferred into a glass tube packed with an adsorbent. After collection, the adsorbent tube was wrapped in aluminum foil and placed in a sealed bag, then stored in a deep-freezer box and transported to the laboratory. Three non-flowering S. candelabrum individuals were used as controls to distinguish floral volatiles from environmental and non-floral sources.
Adsorbent tubes were eluted with 2 mL of n-hexane, and 0.5 mL of the eluted solution containing floral volatiles was collected in a 2 mL amber vial. The solution was concentrated to 100 μL by nitrogen blowdown39. Finally, the sample was stored at −18 °C in a freezer until it was removed for instrumental analysis. The GC-MS technique was used to analyze the floral volatiles39, with the NIST11 library serving as the spectral library for volatile identification, and only compounds with a match score ≥ 80 were retained (Supplementary Data 3). Analyses of the floral volatiles were performed with an Agilent 7890 gas chromatograph/5975 mass selective detector with a DB-5MS capillary column with a length of 30 m. Helium was used as the carrier gas at a constant flow rate of 1 mL/min. The oven conditions included an initial temperature of 60 °C for 1 min, the temperature increased at a rate of 10 °C/min to 130 °C, and then 6 °C/min to 250 °C for 10 min. The inlet temperature was kept constant at 250 °C, and the MS transfer line was set at 290 °C. MS acquisition parameters included scanning from m/z 50–600 in the electron impact (EI) mode for routine analysis. Data processing and peak identification were conducted using MSD ChemStation (Agilent Technologies, version E.02.02.1431).
Determination of enzyme activity
Inflorescence stems of field-collected Saxifraga candelabrum were kept intact and immersed in a solution of ELF97 phosphatase substrate (Invitrogen/Molecular Probes of Thermo Fisher Scientific)40. Samples were incubated in a substrate solution of 250 μM ELF97 phosphatase substrate in Milli-Q water (diluted 1:20) at room temperature for 15 min. Materials were then examined for fluorescence under an epifluorescence microscope (ZEISS Discovery V20 microscope equipped with an EBQ 100 isolated mercury lamp; excitation, 450–490 nm; dichromatic mirror, DM 510 nm; emission filter, 510 nm) at the Microscopy and Image Analysis Facility (Kunming Institute of Botany, Chinese Academy of Sciences). Fluorescence was documented using a Canon EOS Rebel T5 digital camera. Unstained samples of S. candelabrum were used to check for possible autofluorescence. To provide biological controls, inflorescence stems of Pinguicula primuliflora C.E. Wood & R.K. Godfrey (carnivorous) and Saxifraga filicaulis Wall. & Ser. (closely related, noncarnivorous) were processed in parallel.
Experiments for nutrient uptake
More than 1000 fruit flies (Drosophila melanogaster) were fed with 15N-labeled amino acids for approximately two months, allowing multiple generations to develop and ensuring stable isotopic enrichment. The labeled diet was prepared by adding 100 mg algal amino acid mixture (Sigma-Aldrich, 98 atom%¹⁵N) to 100 g standard Drosophila medium. Another 500 fruit flies were fed with normal standard Drosophila medium as a control19. Adult Drosophila melanogaster (n = 12) were randomly selected and weighed individually using an analytical balance, and the average fresh body mass was 0.403 ± 0.011 mg (mean ± SD). Fruit flies were then frozen at −20 °C before the experiment. Field experiments were conducted on 60 individuals of Saxifraga candelabrum across three localities in Shangri-La (Wufeng Mountain, Napahai, Songzanlin-Lamasery) of Yunnan province in China. Laboratory experiments were also conducted using 20 individuals to eliminate potential environmental influences. Fruit flies were placed on the upper stems of S. candelabrum, Saxifraga diversifolia (negative control), calyces of Salvia przewalskii (negative control), and leaves of Drosera peltata (positive control). S. diversifolia was selected as a negative control as it is a close relative of S. candelabrum and, based on our observations, shows no evidence of insect capture, despite possessing glandular hairs. Salvia przewalskii was chosen as a control because it often co-occurs with S. candelabrum in Shangri-La, and has sticky glandular hairs restricted to its calyx, but no evidence of carnivory. For interspecific comparisons, S. candelabrum, S. diversifolia, S. przewalskii, and D. peltata were all sampled from the same locality (Wufeng Mountain, Shangri-La).
For the first three species noted above (S. candelabrum, S. diversifolia, S. przewalskii), fruit flies were gently placed on the glandular surfaces. To prevent the insects from falling off, small strips of parafilm were wrapped around the stem near the insects, following the method described by Lin et al.19 Five to ten individuals of each species were treated with ten unlabeled fruit flies for two weeks as controls. Five to ten individuals of each species were treated with ten 15N-labeled fruit flies for each treatment, and tissues were harvested two weeks after treatment for isotopic analysis. In addition, for S. candelabrum, additional samples were collected 20 and 28 days after treatment to examine temporal changes in prey-derived nitrogen uptake.
Stems or leaves of D. peltata, S. przewalskii, S. candelabrum, and S. diversifolia were collected and analyzed for nutrient absorption19. For S. candelabrum, the δ15N of flowers, cauline leaves, and basal leaves (different plant organs from the same individuals under each treatment) were also measured. We also measured the δ15N of soil around the treated S. candelabrum to evaluate the potential contributions from labeled fruit flies falling into soil and being absorbed by roots. To avoid potential overestimation caused by residual labeled material remaining on glandular surfaces, the upper stems of Saxifraga, calyces of Salvia, and leaves of D. peltata directly exposed to labeled fruit flies were excluded from δ15N measurements.
All plant samples were dried at 105 °C for 30 min, and then at 65 °C for 48 h. After that, all samples were ground into powder and sent to the Stable Isotope Facility (Kunming Biological Diversity Regional Center of Instruments, Chinese Academy of Sciences) for isotopic analysis. δ15N values were measured using a Delta V Advantage mass spectrometer (Thermo Scientific) and expressed in units of per mil (‰) where δ = [(Rsample /Rstandard) − 1]×1000, R = 15N:14N, with nitrogen in the air as the standard. The precision was typically ±0.3‰ for δ15N. Differences in δ¹⁵N values among species and treatments were assessed using two-sided Welch’s t-tests.
Genome sequencing, assembly, and assessment
Fresh basal leaves of Saxifraga candelabrum were collected from Wufeng Mountain, Yunnan province, China [27.820°N, 99.723°E; voucher specimen (Deng 14228) was deposited in the Herbarium of Kunming Institute of Botany, Chinese Academy of Sciences (KUN)]. Genomic DNA was extracted using standard CTAB methods41. The DNA quality was monitored on 1% agarose gels, while DNA purity and concentration were checked using the NanoPhotometer® spectrophotometer (Implen, CA, USA) and the Qubit 2.0 Fluorimeter (Life Technologies, CA, USA), respectively. The genome size of the sequenced S. candelabrum individual was estimated to be ~2.24 Gb using a k-mer approach (Fig. S2A)42,43.
Karyotype and chromosome number determination of S. candelabrum were performed using root tips. Field-collected seeds were germinated on sterilized filter paper moistened with 12.5 mg/L gibberellic acid (GA) in distilled water, incubated at 20 °C with ample light. Root tips (0.3–0.5 cm) were harvested from non-detected seedlings. Root tips were pretreated with 0.002 mol/L 8-hydroxyquinoline at room temperature in darkness for 3 h to arrest metaphase, and then fixed in Carnoy’s solution (3:1 ethanol: acetic acid) at 4 °C for ≥24 h. Fixed samples were hydrolyzed in 1 M HCl at 60 °C for 13 min, rinsed in distilled water, and stained with 20–25% acetocarmine for ≥24 h. Stained root tips were squashed on slides under a coverslip and observed using an Olympus BX-51 microscope at 100× oil immersion. Well-spread metaphase cells were photographed with Images Advanced 3.0 software to confirm the karyotype (Fig. S2B; 2n = 2x = 16).
Two paired-end libraries were constructed with insert sizes of 350 bp and sequenced on the Illumina NovaSeq X Plus platform (Illumina, San Diego, CA, USA). For the long-read sequencing, third-generation sequencing 20-kb libraries were constructed and sequenced using the PacBio HiFi platform (Pacific Biosciences, Menlo Park, CA, USA) according to the manufacturer’s protocol. For the Hi-C technology, fresh young leaves of S. candelabrum (Deng14228) were preserved in liquid nitrogen for library construction. This process involved chromatin extraction, digestion, ligation, fragmentation, and followed by sequencing on an Illumina NovaSeq X Plus platform. Raw reads from Illumina sequencing were filtered using fastp v0.21.044, while quality control of third-generation data was performed using SMRT Link v11.0 and CCS v6.0.0 software45,46. For Hi–C data quality control, reads were filtered using fastp v0.21.0, and HICUP v0.8.0 was performed for further filtering47. Accordingly, a total of 117.32 Gb (~58×) clean PacBio long reads were utilized for genome assembly with Hifiasm v0.16.1 (Supplementary Data 5–7)48. The preliminary de novo assembly with PacBio long reads contained 770 contigs with a total length of 2.09 Gb (contig N50 = 10.21 Mb). To anchor these contigs to chromosomes, 372.75 Gb (~186×) Hi–C clean reads were aligned to the draft reference genome using ALLHiC v0.9.8 to assist in genome assembly through clustering, ordering, and orienting scaffolds onto chromosomes49.
Visualization of the genome maps was achieved using Juicebox v1.11.0850. According to the strength of chromosome interaction, 98.56% of the initial de novo assembly contig sequences were then anchored to a model of 8 pseudochromosomes based on ALLHiC v0.9.8, being consistent with the chromosome number result (Fig. S2B-D). With the integration of Hi-C data, the outassembled genome size was finalized at 2.086 Gb, with scaffold N50 reaching 232.01 Mb. Finally, Illumina paired-end reads were mapped to the assembled genome to evaluate the accuracy of the genome assembly using the Burrows-Wheeler Aligner (BWA) v0.7.12-r103951. Illumina paired-end reads mapping rate exceeded 99.77%, indicating a nearly complete assembled genome (Supplementary Data 8–10). The completeness of the genome assembly was assessed using Benchmarking Universal Single-Copy Orthologs (BUSCO) v4.0.552, which showed that 95.4% of 1539 genes were found in the S. candelabrum genome (Supplementary Data 11).
Gene repeats and genome annotation
Both de novo prediction and homologous alignment were used to detect repeated sequences. RepeatModeler v1.0.11 was used to build the repeat sequence database53, LTR_FINDER v1.0.754, and LTR_retriever v2.8 were used for de novo estimation55. RepeatMasker and RepeatProteinMask programs were then used to conduct homologous sequence alignment by searching the RepBase database56. For the ncRNA, transfer RNAs (tRNAs) were predicted using tRNAscan-SE v1.2357. The rRNAs and their subunits were predicted using the rRNA database. MicroRNA, small nuclear RNA, and small nucleolar RNA were identified using INFERNAL v1.1.2 to search the Rfam database58. As a result, repetitive sequences accounted for 84.20% of the whole-genome size (Supplementary Data 12), which was highly similar to the results of the k-mer distribution analysis (83.54%). Long terminal repeat retrotransposons (LTR-RTs) constituted the largest portion of repetitive sequences (76.17%).
Three independent strategies, including ab initio prediction, homology search, and reference-guided transcriptome assembly, were used for gene prediction. For de novo prediction, AUGUSTUS v3.3.2 and GlimmerHMM v3.04 were utilized for ab initio gene prediction59,60. For RNA-seq-based gene prediction, RNA was extracted using an RNAprep Pure Plant Kit (Tiangen Biotech, Beijing, China) from the roots, stems, flowers, and leaves of Saxifraga candelabrum, and RNA was used to produce cDNA for transcript sequencing on an Illumina NovaSeq X Plus platform (Illumina, San Diego, CA, USA). For homology searches, Blast v2.7.1 was used to align the homologous peptides from four species (i.e., Arabidopsis thaliana, Sedum album, Aldrovanda vesiculosa, and Drosera spatulata) and Exonerate v2.4.0 was employed for coding region prediction and the gene structure information61,62. The predicted results were integrated using the MAKER v2.31.10 and were ensured by combining evidence63. After gene prediction, gene function information, motifs, and domains were assigned by comparisons with public databases with an E-value cut-off of 1e − 5, including Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG)/EuKaryotic Orthologous Groups (KOG), UniProt, and NCBI nonredundant protein (NR) databases. Ultimately, 42,143 (88.63%) from the 47,550 annotated genes showed functional evidence from combined evidences (Supplementary Data 13–14).
Gene family identification and phylogenetic analysis
A total of 31 species were investigated in a broad phylogenomic analysis to construct orthogroups using OrthoFinder v2.4 (Fig. S3; Supplementary Data 5)64. We included Saxifraga candelabrum and six representative species from Saxifragales for which published genomic data are available (Crassulaceae: Kalanchoe fedtschenkoi, Kalanchoe marnieriana, Sedum album, and Rhodiola crenulata; Altingiaceae: Liquidambar formosana; Saxifragaceae: Tiarella polyphylla)65,66,67,68,69. At a broader scale we used 15 representative species from various rosids clades (Brassicales: Arabidopsis thaliana; Celastrales: Tripterygium wilfordii; Cucurbitales: Cucumis sativus; Fabales: Trifolium pratense, Vigna radiata; Fagales: Juglans regia, Morella rubra; Malpighiales: Linum usitatissimum; Populus trichocarpa; Myrtales: Corymbia citriodora; Sapindales: Citrus maxima; Rosales: Morus notabilis, Prunus avium, Prunus persica; Vitales: Vitis vinifera)70,71,72,73,74,75,76,77,78,79,80,81,82,83,84, five available representative carnivorous species (Caryophyllales: Aldrovanda vesiculosa, Drosera spatulata, Nepenthes gracilis; Oxalidales: Cephalotus follicularis; Ericales: Roridula gorgonias)85,86,87,88, two non-carnivorous species in Caryophyllales (Amaranthus hypochondriacus and Spinacia oleracea)89,90 and two outgroup species representing the monocots and basal angiosperms, respectively (Poales: Zea mays; Nymphaeales: Nymphaea colorata)91,92.
For the identification of gene families among these 31 species, we used OrthoFinder v2.464; protein sequences were extracted from the available genome data (Supplementary Data 5). Alternative splicing was filtered out, and only the longest transcript was retained using a Python script for gene family analysis of each species. The similarities between protein sequences were calculated based on an all-vs-all BLAST with an E-value threshold of 1e − 3. In total, 64,092 orthogroups were identified among all of the 31 species (Fig. S4, Supplementary Data 15).
Four-fold synonymous (degenerative) third-codon transversion (4DTv) and synonymous substitution rates (Ks) methods were performed to search for putative WGD events93. LTR_FINDER v1.0.7 was used to search for LTR sequences54, and LTR_retriever v2.8 was used to filter out redundant LTR sequences to obtain accurate LTR sequences55. The flanking sequences on both sides of each LTR were extracted and compared to MAFFT v7.30 (parameters: –localpair), and EMBOSS v6.6.0 was used to estimate the distance (k) between the two sides in terms of the Kimura model94,95. The t = k/(2 × r) was used to estimate the divergent time with the molecular clock (Fig. S5; r = 7 × 10−9).
Convergent gene family evolution
Gene family expansion and contraction were investigated with CAFÉ v4.2103 using a birth and death process to model gene gain and loss based on the phylogenetic tree from OrthoFinder v2.4 of the 31 species64. Gene Ontology and KEGG enrichment analyses were further performed to identify significantly expanded or contracted terms using the ‘clusterProfiler’ package in R v4.396. To further determine whether there were orthologs experiencing convergent shifts in selective pressure, especially to assess possible acceleration along branches to the carnivorous species, the ‘RERconverge’ package in R v4.3 was employed to estimate the relative evolutionary rate (RER) in carnivorous species97.
Briefly, RERs are calculated as relative branch lengths by normalizing branch lengths for focal branches (i.e., carnivorous branches) across all genes. The program compares rates of change in focal foreground branches (i.e., carnivorous branches) and the remaining branches (i.e., the non-carnivorous branches), and identifies genes with a significant correlation between relative evolutionary rate and a phenotypic trait (i.e., carnivorous species). The RERs analyses only allow one gene copy per species in each gene family, but the resulting number of single-copy genes is very low, as determined by OrthoFinder v2.4, due to the distant phylogenetic relationships among some of the 31 species. Thus, we further reduced the number of orthogroups detected by OrthoFinder v2.4 with multiple gene copies per species to one copy per species. We focused our analysis on orthogroups containing at least one carnivorous species representative. Protein sequences from these filtered orthogroups were then aligned using MAFFT (v7.30)95. The gene trees were further constructed based on each orthogroup using RAxML v8.2.12, and only one gene copy per species was extracted according to the branch length and phylogenetic position. Finally, 3161 one-to-one genes were obtained for subsequent analyses, and protein sequences were aligned using PRANK v17042798. The corresponding coding sequences were further extracted based on the gene identification of the proteome sequences. For RERs analyses, the branch lengths of each gene tree of 3161 one-to-one genes were estimated using the ‘phangorn’ package, and RERs were further calculated for each branch of the corresponding gene tree by the ‘RERconverge’ package in R v4.3. The carnivorous species were set as foreground branches (i.e., carnivorous branches) to test for a significant association between RERs and phenotype (i.e., carnivorous species) across all branches of each tree with a p-value less than 0.05.
To further characterize the patterns of molecular evolution associated with carnivory, we estimated the ratio of non-synonymous to synonymous rate (dN/dS) for each carnivorous and non-carnivorous plant for each gene using a branch-site test aBSREL in HyPhy v2.5 with six convergent carnivorous plants set as foreground branches99. The positively selected genes were recorded when they occurred in at least three independent carnivorous species from the aBSREL results. To determine if convergent amino acid substitutions were present at specific gene sites, we identified convergent substitution sites using FADE (FUBAR Approach to Directional Evolution) in HyPhy v2.5 for each gene. FADE identifies sites experiencing directional selection towards specific amino acids in foreground relative to background branches.
For the above analyses, we first compared Saxifraga candelabrum to two other sticky trap species (Drosera spatulata, Droseraceae; Roridula gorgonias, Roridulaceae) to explore the molecular convergence in all three sticky flypaper trap carnivorous lineages. We further performed comparative genomic analyses across all six examined carnivorous species, including not only sticky flypaper traps (D. spatulata, Droseraceae; R. gorgonias, Roridulaceae; S. candelabrum, Saxifragaceae), but also pitcher traps (Cephalotus follicularis, Cephalotaceae; Nepenthes gracilis, Nepenthaceae), and snap traps (Aldrovanda vesiculosa, Droseraceae). Finally, pairwise comparisons between S. candelabrum and each carnivorous species were conducted to identify possible additional trap-specific molecular convergence associated with carnivory.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.