Specimens of B. vivipara at IS, BS, and VS were collected from Datong, Xining, Qinghai (101.6937° E, 36.9328° N, 2731 m). The growth forms of these developmental stages are illustrated in Fig. 1. The inflorescences, bulbils, and viviparous plantlets were flash-frozen in liquid nitrogen for use as experimental materials.

The total RNA of the experimental materials was extracted using Trizo^® (Invitrogen-Thermo Fisher Scientific, Carlsbad, CA, USA). The RNA underwent DNase I treatment. Each sample contained 1.5 µg of RNA. RNA purity was assessed using a NanoPhotometer^® spectrophotometer. RNA concentrations were determined using the Assay Kit in a Qubit^® 2.0 Fluorometer (LifeTechnologies, Carlsbad, CA, USA). RNA integrity was evaluated using the RNA Nano 6000 Assay Kit of the Agilent Bioanalyzer 2100 system (Agilent Technologies, Carlsbad, CA, USA), with RNA Integrated Number (RIN) values ≥8. RNA that met the quality standards was enriched for mRNA using Oligo(dT)-attached magnetic beads for cDNA library construction.

The library was prepared using the NEBNext^® Ultra^TM RNA Library Prep Kit for Illumina^® (NEB, Ipswich, MA, USA). Library fragments underwent purification using the AMPure XP system (Beckman Coulter, Beverly, CA, USA). Assessment of library quality was conducted using the Agilent Bioanalyzer 2100 system. Raw reads were obtained through Illumina HiSeqTM 2500 sequencing; clean reads were generated by eliminating reads containing adapters, poly-N sequences, and low-quality reads from the raw data. Concurrently, Q20, Q30, GC-content, and sequence duplication levels of the clean data were calculated. All subsequent analyses were based on high-quality, clean data. The transcriptome was assembled by splicing clean reads using Trinity [16]. For each gene, the longest transcript was designated as the Unigene for further analysis.

The Unigenes were queried against public databases for functional annotation based on sequence similarity. The quantity and percentage of annotated Unigenes were evaluated across various public databases, including Nt (E-value ≤ 1E − 5), Pfam (E-value ≤ 0.01), EuKaryotic Orthologous Groups (KOG)/Clusters of Orthologous Groups (COG) (E-value ≤ 1E − 3), Swiss-Prot (E-value ≤ 1E − 5), Kyoto Encyclopedia of Genes and Genomes (KEGG) (E-value ≤ 1E − 10), and Gene Ontology (GO) (E-value ≤ 1E − 6).

Gene expression levels of the samples were quantified using RSEM, with a threshold of FPKM > 0.3 employed to determine gene expression [17]. The FPKM method is currently the most widely adopted approach for gene expression quantification [18].

The differential expression analysis between two samples was conducted using the DESeq (2010) R package. The p-value was adjusted using q-value [19]. A threshold of p-adj < 0.05 and an absolute value of log2 (fold change) > 1 was established to determine significantly differentially expressed genes (DEGs).

The total RNA was extracted from flesh samples using the plant RNA extraction kit (TaKaRa, Dalian, China). Following DNase Ⅰ treatment, cDNA synthesis was performed using the 1st Strand cDNA Synthesis Kit (TaKaRa, Dalian, China). For validation, twenty DEGs involved in viviparous plantlet development were selected for qRT-PCR analysis. Primers were designed using Primer 5.0 software and are presented in Table A1. Primer amplification efficiency was evaluated through PCR amplification, 3% agarose gel electrophoresis, sequencing, and qRT-PCR amplification curve analysis. The qRT-PCR analysis was conducted using TB Green^® Premix Ex Taq^TM II (TaKaRa, Dalian, China) on a CFX ConnectTM Real-Time System (BIO-RAD, Hercules, CA, USA). Each 20 μL reaction mixture contained 10 μL of TB Green Premix Ex Taq II, 0.8 μL of each primer (10 μM), 6 μL of 60-fold diluted cDNA, and 2.4 μL of water (TaKaRa, Dalian, China). The amplification program consisted of one cycle at 95°C for 30 s, followed by 40 cycles of 95°C for 5 s and 60°C for 20 s. The relative expression levels of the selected genes were normalized to the expression of B. vivipara β-actin and analyzed using the 2^−ΔΔCT Method [20,21]. The qRT-PCR analysis for each gene was performed with six independent biological replicates and three technical repeats per biological replicate.

The sequencing process yielded 452 million raw reads (Error < 0.01, GC > 47.3%) from the nine libraries. Of these, 439 million high-quality reads (Q > 20) were selected for further analysis. Notably, 61.65%–64.4% of the clean reads aligned with the de novo transcriptome sequence (Table 1). The assembly process generated 265,021 Unigenes (N50 ≥ 1312, N90 ≥ 444) from 338,770 transcripts (N50 ≥ 1206, N90 ≥ 335) across various developmental stages.

Pearson’s correlation coefficient was employed to evaluate the correlation between evaluation indexes of samples [22]. The correlation coefficients of gene expression levels demonstrated strong repeatability among replicated samples within groups, particularly between IS-2 and IS-3, BS-2 and BS-3, and VS-1 and VS-2 (Fig. 2). Consequently, development-related DEGs were analyzed using IS-2, IS-3, BS-2, BS-3, VS-1, and VS-2.

Of the 265,021 Unigenes, 20,782 were matched across all public databases, while 154,813 were matched in at least one public database (Table 2). Additionally, 99,344 Unigenes were annotated by GO assignments and categorized into 56 groups, encompassing Cellular Component (28.92%), Biological Process (47.91%), and Molecular Function (23.17%) (Fig. 3). The KOG function classification sorted 52,662 Unigenes into 26 sub-categories (Fig. 4). Furthermore, 56,240 Unigenes were classified into 19 sub-categories using the KEGG classification (Fig. 4).

The differential gene expression across three developmental stages was analyzed, and DEGs were identified through pairwise comparisons of the nine libraries (Fig. 5). The analysis revealed 15,216 up-regulated DEGs in the VS vs. BS comparison, 13,225 in the BS vs. IS comparison, and 13,458 in the VS vs. IS comparison (Fig. 5b,d–f). Conversely, the analysis identified 13,303 down-regulated DEGs in the VS vs. BS comparison, 14,512 in the BS vs. IS comparison, and 13,212 in the VS vs. IS comparison (Fig. 5c–f). In aggregate, 49,731 genes exhibited differential expression across the three developmental stages (Fig. 5a). These findings suggest a substantial involvement of DEGs in the development of viviparous.

We employed GO assignments to categorize the functions of DEGs in pairwise comparisons of cDNA libraries. The GO categories of DEGs exhibited significant differences in the comparisons of VS vs. BS, BS vs. IS, and VS vs. IS. In the BS vs. IS comparison, down-regulated DEGs were enriched in ‘microtubule-related functions’, ‘copper ion binding’, ‘tetrapyrrole binding’, and ‘serine-type carboxypeptidase activity’, while up-regulated DEGs were enriched in ‘oxidation-reduction process’, ‘hormone metabolic process’, ‘tetrapyrrole binding’, ‘response to water’, ‘metabolic process’, ‘response to abiotic stimulus’, and ‘iron ion binding’. In the VS vs. IS comparison, down-regulated DEGs were enriched in ‘translation-related functions’, ‘biosynthetic process’, ‘tetrapyrrole binding’, ‘copper ion binding’, ‘oxidation-reduction process’, ‘cytoplasm related functions’, and ‘metabolic process’, while up-regulated DEGs were enriched in ‘oxidation-reduction process’, ‘macromolecule modification’, ‘metabolic process’, ‘response to auxin’, ‘transmembrane transport’, ‘response to biotic stimulus’, ‘ion binding’, ‘tetrapyrrole binding’, and ‘glutamine biosynthetic process’. In the VS vs. BS comparison, down-regulated DEGs were enriched in ‘translation-related functions’, ‘steroid related’, ‘oxidation-reduction process’, ‘metabolic process’, ‘cytoplasmic-related functions’, ‘structural molecule activity’, and ‘starch binding’, while up-regulated DEGs were enriched in ‘oxidation-reduction process’, ‘macromolecule modification’, ‘metabolic process’, ‘kinase activity’, ‘photosynthesis’, ‘ion binding’, ‘adenyl ribonucleotide binding’, ‘tetrapyrrole binding’, ‘polysaccharide binding’, ‘defense response’, and ‘transport’. Further analysis of the overrepresented GO functions within each cluster revealed significant enrichment of genes associated with ‘metabolic process’ and ‘biosynthetic process’. Certain GO terms were enriched in particular clusters, including ‘heme binding’, ‘tetrapyrrole binding’, ‘glutamine biosynthetic process’, and ‘steroid metabolic process’, which were enriched specifically in the comparisons of VS vs. BS, BS vs. IS, and VS vs. IS. To further explore the biological pathways associated with the DEGs, we conducted a KEGG analysis. This analysis revealed that the DEGs were involved in pathways such as ‘phenylpropanoid biosynthesis’, ‘plant hormone signal transduction’, ‘protein processing’, ‘starch and sucrose metabolism’, and ‘plant-pathogen interaction’.

Previous studies have identified DEGs involved in plant hormones, phytochromes, enzymes, and transcription factors as being associated with plant development [23–26]. In this study, we identified putative homologs of these genes in B. vivipara. Nineteen classes of DEGs related to the development of B. vivipara were identified in comparisons of VS vs. BS, BS vs. IS, and VS vs. IS (Fig. 6). These DEGs exhibited distinct expression patterns during different developmental periods. In the IS vs. BS comparison, 48 DEGs were identified and grouped into 15 categories, including enzymes associated with secondary metabolites, embryogenesis and cell division, plant hormones, proteins, and transcription factors that regulate plant growth (Fig. 6a). The comparison between BS and VS yielded 61 DEGs classified into 16 categories (Fig. 6b). In the VS vs. IS comparison, 56 DEGs were identified and grouped into 15 categories (Fig. 6c). Furthermore, variations in expression types and levels of DEGs were observed across different developmental stages (Fig. 6a–c). Cluster-22162. (42282, 208449, 133388, 110919, 46118, 214884, 32166) showed differences only in the IS and BS comparison. Cluster-22162. (183713, 14777, 106603, 106607, 86809, 110915, 218977, 120903, 122741, 127615), Cluster-22404.0, and Cluster-6515.0 revealed differences only in the BS and VS comparison. Cluster-22162. (90417, 76686, 149199, 199887, 110908, 66968, 46184, 200698, 149633, 204391, 160297, 160155) exhibited differences only in the IS and VS comparison.

To identify the DEGs associated with viviparous plantlet development, we selected 20 candidate genes for verification using qRT-PCR. Statistical analysis of the results revealed that 17 of these genes exhibited significantly different expressions (p ≤ 0.05; Fig. 7). Moreover, 12 genes demonstrated significant correlations between the qRT-PCR data and RNA sequencing (RNA-seq) results, indicating good reproducibility between the transcript abundance measured by RNA-seq and the expression profile obtained from RT-qPCR. These genes include Cinnamate 4-hydroxylase (C4H) (Cluster-22162.149884), CMK (Cluster-22162.117067), DXS (Cluster-22162.126257), DELLA (Cluster-22162.133295), plantlets. Phospholipase D (PLD) (Cluster-22162.160155), somatic embryogenesis receptor-like kinase (SERK) (Cluster-22162.108908), DXS (Cluster-22162.126243), Sterol C24-methyltransferase (SMT) (Cluster-22162.143764), UDP-glucuronic acid decarboxylase (UXS) (Cluster-22162.206500), Delta24-sterol reductase (DWF1) (Cluster-22162.127615), Abscisic acid (ABA) Insensitive 5 (ABI5) (Cluster-22162.183709) and Ascorbate peroxidase (APX) (Cluster-22162.214879) (Figs. 6 and 7).