Reference-guided transcriptomics resolves CoHMGR family composition and expression diversification during postharvest warm conditioning of Camellia oleifera seed kernels

Abstract

Background

Camellia oleifera is an important woody oil crop whose seed kernels contain squalene and other bioactive oil constituents. Although short-term postharvest conditioning is commonly used in oil-tea processing, the molecular behavior of the 3-hydroxy-3-methylglutaryl-CoA reductase (HMGR) system during this early postharvest window remains poorly characterized. Here, mature C. oleifera seed kernels were subjected to controlled warm conditioning at 35 °C and 95% relative humidity for 12–24 h to characterize HMGR-related biochemical responses, gene-family features, and transcriptomic changes associated with squalene accumulation.

Results

Warm conditioning was associated with a rapid increase in squalene content, rising from 177.58 ± 8.2 µg/g FW at T-0 to 314.30 ± 9.7 µg/g FW at T-12 and remaining elevated at T-24. An ELISA-based estimate of HMGR activity showed a similar early increase. Transcriptome-guided curation resolved five HMGR-related loci, including four canonical CoHMGR genes with conserved class I HMGR-domain support and one truncated HMGR-like locus, LOC_00050219. Integrated phylogenetic, structural, domain-architecture, localization-prediction, and collinearity evidence supported this curated four-member canonical CoHMGR set. Reference-guided RNA-seq and RT–qPCR revealed divergent expression trajectories among canonical members, with HMGR2 showing the strongest warm-conditioning-associated response. NetPhos analysis further prioritized candidate phosphorylation sites in a representative CoHMGR protein, and global transcriptome profiling placed the CoHMGR response within a staged program involving stress signaling, protein homeostasis, and carbon/lipid metabolic reallocation.

Conclusions

This study establishes a reference-guided framework for resolving CoHMGR family composition and interpreting HMGR-associated responses during postharvest warm conditioning in C. oleifera seed kernels. The findings identify prioritized CoHMGR isoforms, candidate phosphorylation sites, and transcriptomic contexts for future protein-level validation and functional studies of postharvest squalene regulation. Importantly, the current evidence supports candidate prioritization rather than direct mechanistic proof, and causal links among candidate kinases or phosphatases, HMGR protein regulation, and squalene biosynthesis remain to be experimentally established.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12864-026-13038-6.

Introduction

Camellia oleifera (C. oleifera) is an economically important woody oil crop in China and a major source of high-quality edible oil [1, 2]. Its seed oil contains a high proportion of unsaturated fatty acids as well as minor bioactive constituents, including squalene, sterols, tocopherols, and polyphenols [3, 4]. In commercial production, freshly harvested fruits typically undergo a short postharvest conditioning period before shelling, drying, and oil extraction, a step that can measurably affect oil-related quality traits. Recent studies on postharvest piling and related conditioning treatments indicate that this early postharvest phase is biologically active rather than merely preservative, highlighting the need to define the molecular responses that occur in C. oleifera kernels during this processing window [5, 6].

Within this processing-relevant window, 3-hydroxy-3-methylglutaryl-CoA reductase (HMGR) represents a biologically relevant target because it catalyzes a major regulatory step in the cytosolic mevalonate (MVA) pathway, which supplies precursors for squalene and downstream triterpenoid- and sterol-associated metabolism [7, 8]. In plants, HMGR is typically encoded by a small multigene family rather than a single-copy locus, and the corresponding proteins are membrane-associated enzymes that are predominantly targeted to the endoplasmic reticulum [9]. Recent work in C. oleifera seed kernels further suggests that CoHMGR family members are not functionally equivalent, with CoHMGR2 emerging as a prominent isoform during a critical phase of squalene biosynthesis [10]. These observations make precise family-level definition particularly important in C. oleifera, because structurally reduced or HMGR-like predictions could confound interpretation if all related sequences were treated as canonical family members. Accordingly, we first re-evaluated CoHMGR family composition using a transcriptome-guided annotation framework, thereby defining a reliable canonical gene set for subsequent analyses of isoform-level expression divergence and candidate regulatory features during postharvest warm conditioning.

Despite these advances, the molecular basis of HMGR-associated responses during the early postharvest conditioning window remains incompletely understood. Existing studies in C. oleifera have established the importance of HMGR-linked squalene biosynthesis during seed development [10, 11], but they do not explain how the HMGR system behaves after harvest, when kernels enter a processing-relevant transition rather than a developmental maturation phase. It also remains unclear whether canonical CoHMGR members respond as a coordinated group during early postharvest warm conditioning or instead partition into distinct baseline-abundance and inducibility patterns across the treatment window. A further unresolved issue concerns potential phosphoregulation. Although plant HMGR is subject to multilevel regulation and phosphatase-associated control in Arabidopsis [12], residue-level candidate sites that could provide direct experimental entry points have not been defined for the postharvest conditioning context in C. oleifera.

To address these questions, we established a controlled postharvest warm-conditioning regime in mature C. oleifera seed kernels and integrated biochemical measurements with transcriptome-guided family curation, reference-guided transcriptomics, RT–qPCR validation, and sequence-level phosphosite prioritization. This design enabled us to re-evaluate canonical CoHMGR membership within a chromosome-scale C. oleifera genomic coordinate framework, resolve expression diversification among canonical members during the early postharvest interval, and define residue-level candidates for downstream validation. By integrating these analytical layers within a processing-relevant framework, this study links early conditioning-associated biochemical responses with family-level HMGR interpretation while maintaining clear boundaries among transcript-level association, sequence-based prediction, and direct mechanistic inference.

Materials and methods

Plant material, postharvest warm conditioning, and sample preparation

Fully mature fruits of Camellia oleifera cv. Cenruan No. 3 were harvested from the same experimental plantation in Guangxi, China, following Shuangjiang, the first-frost solar term that typically occurs from late October to early November, when the fruits had reached full maturity. Freshly harvested fruits were assigned to three groups: an untreated control sampled at 0 h (T-0) and two warm-conditioning treatments maintained at 35 °C for 12 h (T-12) or 24 h (T-24). The 35 °C regime was selected based on our previous postharvest optimization study and was used here as a defined, processing-relevant treatment to investigate early molecular and biochemical events in seed kernels, with particular emphasis on HMGR-associated responses and squalene accumulation [13].

Warm conditioning was conducted in darkness in an artificial climate chamber (RDN-1000 C, 1000 L; Ningbo Yanghui Instrument Co., Ltd., Ningbo, Zhejiang, China) set to 35 °C and 95% relative humidity. At each sampling point, the pericarp and seed coat were removed, and seed kernels were collected for analysis. Each treatment included three biological replicates (n = 3), and each replicate consisted of kernels pooled from 200 fruits, corresponding to approximately 400 kernels. This pooling strategy provided sufficient and compositionally matched material for parallel RNA-seq, RT–qPCR, ELISA-based estimation of HMGR activity, and squalene quantification, and was intended to capture treatment-level responses at the batch scale rather than variation among individual fruits. After collection, kernels were immediately frozen in liquid nitrogen and stored at − 80 °C until analysis. The experimental workflow and sampling timeline are summarized in Fig. 1.

Fig. 1.

Experimental workflow and sampling timeline for postharvest warm conditioning of C. oleifera fruits. Fresh fruits were conditioned at 35 °C and 95% RH, and seed kernels were collected at T-0, T-12, and T-24 for transcriptomic and biochemical analyses

RNA extraction, library preparation, sequencing, and RT–qPCR

Total RNA was extracted from seed-kernel samples using the Total RNA Extractor (TRIzol) kit (B511311; Sangon Biotech, Shanghai, China) according to the manufacturer’s instructions. Residual genomic DNA was removed by treatment with RNase-free DNase I. RNA integrity was evaluated by 1.0% agarose gel electrophoresis, and RNA quality and concentration were assessed using a NanoPhotometer^® spectrophotometer (IMPLEN, CA, USA) and a Qubit^® 2.0 Fluorometer (Invitrogen, USA). Qualified RNA samples were submitted to Sangon Biotech (Shanghai, China) for library construction and sequencing.

For each sample, 1 µg of total RNA was used to construct sequencing libraries with the VAHTS™ mRNA-seq V2 Library Prep Kit for Illumina^® according to the manufacturer’s protocol. Index sequences were added to distinguish individual samples. Briefly, mRNA was enriched using poly(T) oligo-attached magnetic beads and fragmented at elevated temperature in 5× First Strand Synthesis Reaction Buffer. First-strand cDNA was synthesized using random hexamer primers and M-MuLV Reverse Transcriptase (RNase H−), followed by second-strand cDNA synthesis using DNA polymerase I and RNase H. After end repair and 3′ adenylation, sequencing adapters were ligated to the cDNA fragments. Fragments of approximately 150–200 bp were selected using the AMPure XP system (Beckman Coulter, Beverly, USA). The size-selected, adapter-ligated cDNA was treated with 3 µL USER Enzyme (NEB, USA) at 37 °C for 15 min and then at 95 °C for 5 min before PCR amplification. PCR amplification was performed using Phusion High-Fidelity DNA polymerase, Universal PCR primers, and Index (X) Primer. PCR products were purified with the AMPure XP system, and library quality was assessed using an Agilent Bioanalyzer 2100 system (Agilent Technologies, CA, USA). Qualified libraries were quantified, pooled, and subjected to paired-end sequencing on an Illumina NovaSeq platform (Illumina, San Diego, CA, USA).

For transcript validation, selected genes were analyzed by RT–qPCR. Total RNA (1,500 ng per sample) was reverse-transcribed into cDNA using the following program: 65 °C for 5 min, 25 °C for 10 min, 50 °C for 30 min, and 85 °C for 5 min. The resulting cDNA was stored at − 20 °C and diluted 10-fold before amplification. RT–qPCR was performed on an ABI 7500 Real-Time PCR System using 2× SGExcel FastSYBR Mixture. Each 20 µL reaction contained 10 µL 2× mix, 0.4 µL of each primer (10 µM), 2 µL diluted cDNA, and 7.2 µL nuclease-free water. The amplification program consisted of an initial denaturation at 95 °C for 3 min, followed by 45 cycles of 95 °C for 15 s and 60 °C for 45 s, with melt-curve analysis performed at the end of each run. Relative transcript abundance was normalized to SAND and calculated using the 2^−ΔΔCt method, with T-0 used as the calibrator. Technical replicates were included for all samples.

Data assessment and quality control

Raw sequencing reads were initially assessed using FastQC v0.11.2. Adapter sequences and low-quality bases were removed using Trimmomatic v0.36. The filtering procedure included adapter trimming, removal of low-quality bases from both the 5′ and 3′ ends (Phred score < 20), sliding-window trimming with a 5-bp window and a minimum average Phred score of 20, and removal of reads shorter than 35 nt together with their paired mates. The resulting clean reads were used for all downstream analyses.

Reference-guided read alignment and transcript assembly

Clean reads were aligned to the chromosome-scale C. oleifera reference assembly GCA_022316695.1 (ASM2231669v1) using HISAT2 v2.1.0. The reference genome sequence was downloaded from NCBI, and the corresponding baseline gene-model annotation was obtained from the CON genome dataset released by the Research Institute of Subtropical Forestry, Chinese Academy of Forestry. A HISAT2 index was built from the reference genome sequence before read alignment. Alignment statistics were summarized using RSeQC v2.6.1. Read-distribution uniformity and genomic feature assignment were evaluated using Qualimap v2.2.1, and gene coverage and chromosome-level read-density distributions were assessed using BEDTools v2.26.0. Reference-guided transcript assembly and expression estimation were performed using StringTie v1.3.3b. The assembled transcripts were compared with the baseline gene models using GffCompare v0.10.1 to identify novel transcript regions and to support downstream gene-model curation.

Expression quantification and differential expression analysis

Transcript abundance was estimated using StringTie v1.3.3b, and expression values were reported as transcripts per million (TPM) to facilitate cross-sample comparison and visualization. Principal component analysis (PCA) was performed using the gene-level expression matrix to evaluate sample-level transcriptomic structure. For differential-expression analysis, gene-level count data were analyzed using DESeq2 v1.12.4 for the three pairwise comparisons: T-12 vs. T-0, T-24 vs. T-0, and T-12 vs. T-24. Genes were considered differentially expressed when the Benjamini–Hochberg-adjusted P value was ≤ 0.001 and the absolute fold change was ≥ 2. For log-scale visualization, expression values of zero were replaced with 0.01. Genes with normalized expression values < 1 in both groups of a given pairwise comparison were excluded from downstream differential-expression analysis. Expression differences were visualized using scatter plots, MA plots, and volcano plots.

Functional enrichment analysis of differentially expressed genes

Functional enrichment analyses of differentially expressed genes (DEGs) were performed for Gene Ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. For GO enrichment analysis, DEGs were assigned to GO terms, and enrichment was tested against the reference gene background using a hypergeometric test. For KEGG pathway analysis, DEGs were mapped to metabolic and signaling pathways in the KEGG database, and pathway enrichment was assessed using the same statistical framework. Multiple testing correction was performed using the false discovery rate (FDR) approach. GO terms and KEGG pathways with an FDR-adjusted q value < 0.05 were considered significantly enriched.

CoHMGR family curation, phylogenetic analysis, collinearity analysis, and protein-feature prediction

HMGR-related candidate sequences were curated from the transcriptome-guided annotation output by integrating gene structure, coding capacity, predicted protein length, and conserved-domain support. Conserved domains were annotated using the NCBI Conserved Domain Database (CDD). Protein sequences from C. oleifera, Camellia sinensis, and Arabidopsis thaliana were used for cross-species phylogenetic reconstruction. Intragenomic collinearity among CoHMGR-containing genomic regions was evaluated using MCScanX. Transmembrane topology was predicted using DeepTMHMM, subcellular localization and membrane-association type were predicted using DeepLoc 2.1, and N-terminal targeting-peptide classes were assessed using TargetP 2.0 in Plant mode. Candidate phosphorylation sites in the representative CoHMGR protein were predicted using NetPhos. Sequence-derived hydrophobic segments and online localization predictions were summarized for architectural interpretation only and were not used as decisive criteria for final CoHMGR classification.

ELISA-based estimation of HMGR activity and GC–MS quantification of squalene

An ELISA-format commercial plant HMGR kit was used to obtain an apparent estimate of HMGR activity in seed-kernel extracts. Crude extracts were prepared from seed-kernel tissues on ice, clarified by centrifugation, and loaded into 96-well plates together with blanks and serially diluted standards. For each sample well, 40 µL of sample diluent and 10 µL of extract were added, resulting in a final 5-fold dilution. The assay was then performed according to the manufacturer’s instructions. Signals were developed with TMB substrate and measured at 450 nm. Activity values (IU/L) were calculated from the standard curve, corrected for the total dilution factor, and, where indicated, converted to IU/g FW using the extraction volume and tissue mass. All biological replicates were assayed in technical duplicate.

Squalene was quantified by GC–MS using a TRACE GC 1300 gas chromatograph coupled to an ISQ LT mass spectrometer (Thermo Fisher Scientific) equipped with a TG-5MS capillary column (30 m × 0.25 mm i.d., 0.25 μm film thickness). Finely ground seed-kernel tissue was extracted with dichloromethane (DCM) by ultrasonication (5 mL, 30 min), followed by centrifugation. The residue was re-extracted with DCM (5 mL, 10 min), and the two supernatants were combined and brought to a final volume of 10 mL with DCM before analysis. GC conditions were as follows: helium carrier gas at 0.8 mL min⁻¹, split ratio of 40:1, injection volume of 2 µL, and injector temperature of 300 °C. The oven was held at 80 °C for 2 min, ramped to 300 °C at 20 °C min⁻¹, and held at 300 °C for 15 min. MS conditions were electron-impact (EI) ionization at 70 eV, with an ion source temperature of 220 °C and a transfer-line temperature of 280 °C. Data were acquired in selected-ion monitoring (SIM) mode using m/z 81, 217, 368, and 386. Quantification was performed using an external calibration curve prepared from authentic squalene standards, and squalene content was reported as µg/g FW of seed tissue.

Statistical analysis

Unless otherwise stated, experimental measurements were performed with three independent biological replicates. Data from biochemical assays and RT–qPCR analyses are presented as mean ± standard deviation (SD). Differences among the three treatment stages (T-0, T-12, and T-24) were evaluated using one-way analysis of variance (ANOVA), followed by Tukey’s honestly significant difference (HSD) test for multiple comparisons when appropriate. When the assumptions of normality or homogeneity of variance were not met, the Kruskal–Wallis test was used as a non-parametric alternative. Statistical significance was defined as P < 0.05. Statistical analyses of experimental data were performed using SPSS Statistics 26.0 (IBM, Armonk, NY, USA).

Sequence-based curation and prediction outputs were treated as descriptive bioinformatic evidence rather than inferential statistical datasets. Conserved-domain annotation, phylogenetic reconstruction, gene-structure assessment, MCScanX-based collinearity analysis, and online predictions of transmembrane topology, subcellular localization, membrane-association type, and N-terminal targeting peptides were summarized according to their output categories, coordinates, or support status. Specifically, DeepTMHMM was used to infer transmembrane topology and transmembrane-helix coordinates; DeepLoc 2.1 was used to predict subcellular localization and membrane-association type; TargetP 2.0 was used to assess classical N-terminal targeting-peptide classes; and Phobius or SignalP 6.0 was used, where applicable, to help distinguish signal-peptide-like features from transmembrane regions. These predictions were interpreted qualitatively as supportive evidence and were not subjected to statistical testing or treated as experimental proof of protein localization.

Results and discussion

Postharvest warm conditioning is associated with increased squalene accumulation and an elevated ELISA-based estimate of HMGR activity in C. oleifera seed kernels

Postharvest warm conditioning was associated with pronounced early increases in both squalene accumulation and the ELISA-based estimate of HMGR activity in C. oleifera seed kernels (Fig. 2). Squalene content increased from 177.58 ± 8.2 µg/g FW at T-0 to 314.30 ± 9.7 µg/g FW at T-12 and remained elevated at 309.10 ± 7.5 µg/g FW at T-24. The ELISA-based estimate of HMGR activity showed a similar temporal pattern, increasing from 10.70 ± 0.52 IU/g FW at T-0 to 13.74 ± 0.48 IU/g FW at T-12 and 14.03 ± 0.55 IU/g FW at T-24. For both measurements, values at T-12 and T-24 were significantly higher than those at T-0. Together, these results indicate that the major biochemical shift occurred within the first 12 h of warm conditioning and was then largely maintained through 24 h.

Fig. 2.

Squalene accumulation and ELISA-based HMGR activity estimate in C. oleifera seed kernels during postharvest warm conditioning. A Squalene content and (B) ELISA-based HMGR activity estimate at T-0, T-12, and T-24 under 35 °C and 95% RH in darkness. Data are shown as mean ± SD (n = 3 biological replicates). Asterisks indicate significant differences from T-0 (P < 0.05)

This temporal pattern is consistent with previous evidence that C. oleifera seeds remain physiologically active after harvest and that postharvest handling conditions can alter oil-related quality traits rather than merely preserve a static preharvest state [5, 6]. It is also consistent with recent work in developing oil-tea seed kernels linking CoHMGR-associated expression dynamics to squalene biosynthesis [10, 11]. In this context, the coupled increase in squalene accumulation and the ELISA-based estimate of HMGR activity is consistent with an early HMGR-associated biochemical response in postharvest kernels. At the same time, the limited difference between T-12 and T-24 suggests early establishment followed by relative stabilization, rather than a continuously intensifying response across the full 24-h window. Because these measurements were obtained at the bulk-kernel level, however, they do not by themselves identify the responsible CoHMGR isoform or establish whether phosphorylation directly contributes to the observed biochemical shift. The gene-family and expression analyses below were therefore used to define the molecular context of this response, while avoiding direct attribution of the bulk-kernel phenotype to any single CoHMGR isoform.

Transcriptome-guided curation identifies four canonical CoHMGR genes and one truncated HMGR-like locus

To define the CoHMGR candidate set for subsequent molecular interpretation, we curated HMGR-related gene models from the transcriptome-guided annotation output. This analysis identified five HMGR-related loci in C. oleifera (Table 1). Four loci—LOC_00008077, LOC_00033676, LOC_00002567, and LOC_00050218—encoded predicted proteins of 429–573 aa, and each showed clear support for the class I HMG-CoA reductase domain (cd00643) in CDD-based analysis [14]. By contrast, LOC_00050219 was represented by a single-exon model with a 375-bp coding sequence predicted to encode a markedly shorter 124-aa polypeptide, and no detectable canonical HMGR-domain support was recovered for this locus. Considered together, reduced coding capacity, simplified gene structure, and absence of conserved-domain support distinguished LOC_00050219 from the other four candidates more clearly than any single feature alone. Accordingly, LOC_00008077, LOC_00033676, LOC_00002567, and LOC_00050218 were retained as canonical CoHMGR genes, whereas LOC_00050219 was classified as a truncated HMGR-like locus rather than a canonical family member. This conservative curation established the working CoHMGR set used in the subsequent phylogenetic, structural, collinearity, and expression analyses.

Table 1.

Genomic features and curation-based assignment of HMGR-related loci in C. oleifera

Gene ID	Simplified label used in figures	Chromosome	Genomic coordinates	Strand	Exon number	CDS length (bp)	Protein length (aa)	Conserved-domain support	Final assignment
LOC_00008077	HMGR1	CM039309.1	190,106,755–190,111,474	+	4	1644	547	Present; cd00643 (HMG-CoA_reductase_class I)	Canonical CoHMGR gene
LOC_00033676	HMGR2	CM039307.1	588,732–593,042	−	5	1341	446	Present; cd00643 (HMG-CoA_reductase_class I)	Canonical CoHMGR gene
LOC_00002567	HMGR3	CM039311.1	124,659,229–124,665,648	−	4	1722	573	Present; cd00643 (HMG-CoA_reductase_class I)	Canonical CoHMGR gene
LOC_00050218	HMGR4	CM039314.1	157,943,839–157,950,656	−	4	1290	429	Present; cd00643 (HMG-CoA_reductase_class I)	Canonical CoHMGR gene
LOC_00050219	HMGR-like	CM039314.1	157,950,733–157,951,107	−	1	375	124	Not detected	Truncated HMGR-like locus

Locus features were compiled from the transcriptome-guided annotation output, including chromosomal position, strand orientation, exon number, coding-sequence length, and predicted protein length. Domain support was evaluated at the protein level. Final assignments were made by integrating structural completeness and conserved-domain evidence. Simplified labels HMGR1–HMGR4 are used in the expression and validation figures to denote the four canonical CoHMGR genes; LOC_00050219 is labeled as HMGR-like because it lacks detectable canonical HMGR-domain support

Integrated phylogenetic and architectural analysis of HMGR-related loci in C. oleifera

To place the curated C. oleifera HMGR-related loci in an evolutionary context, we reconstructed a midpoint-rooted phylogeny using HMGR-related proteins from C. oleifera, Camellia sinensis, and Arabidopsis thaliana (Fig. 3A; Table S1). This cross-species framework was informative because plant HMGRs are typically represented by small gene families rather than single-copy loci, and A. thaliana itself contains two differentially expressed HMGR genes [7]. Within this framework, the four full-length C. oleifera loci clustered within the Camellia-associated portion of the tree together with C. sinensis homologs, rather than appearing as isolated or aberrant sequences. This topology supported their inclusion in the working CoHMGR set for subsequent evaluation. By contrast, LOC_00050219 required a more conservative interpretation. Although LOC_00050219 could be incorporated into the HMGR-related alignment and tree reconstruction, phylogenetic placement alone was insufficient for canonical assignment. As summarized in Fig. 3B, this locus showed reduced structural completeness relative to the four full-length loci. Together, the phylogenetic placement and structural summary supported the four full-length loci as canonical CoHMGR candidates, whereas LOC_00050219 was best interpreted as a truncated HMGR-like locus. This interpretation is consistent with recent seed-focused work in C. oleifera indicating that CoHMGR family members associated with squalene biosynthesis are not functionally equivalent [10].

Fig. 3.

Phylogenetic placement and structural architecture of HMGR-related loci in C. oleifera. A Midpoint-rooted phylogeny of HMGR-related proteins from C. oleifera, Camellia sinensis, and Arabidopsis thaliana. The five C. oleifera loci are highlighted, and branch-support values are shown at selected nodes. B Transcriptome-guided gene models and corresponding protein-domain architectures. Exons, coding regions, and CDD-supported class I HMGR domains are indicated as shown in the legend. LOC_00050219 is shown as a truncated HMGR-like locus lacking detectable canonical HMGR-domain support

Structural and domain-level evidence further refined the phylogenetic interpretation. The four full-length loci—LOC_00008077, LOC_00033676, LOC_00002567, and LOC_00050218—shared multi-exon coding architectures and encoded substantially longer proteins than LOC_00050219 (Fig. 3B; Fig. S1). CDD-based annotation detected the class I HMG-CoA reductase domain (cd00643) in all four full-length proteins, whereas no corresponding canonical HMGR-domain support was recovered for the 124-aa LOC_00050219 product (Fig. S2; Table S2) [14]. This distinction is consistent with the expected architecture of canonical plant HMGRs, which are generally membrane-associated enzymes containing conserved catalytic regions [7, 8]. Thus, gene structure, coding capacity, and conserved-domain evidence provided the primary basis for defining LOC_00008077, LOC_00033676, LOC_00002567, and LOC_00050218 as canonical CoHMGR genes and for excluding LOC_00050219 from the canonical set. Sequence-based topology and localization predictions provided an additional supportive layer but were not used as decisive classification criteria. DeepTMHMM predicted N-terminal transmembrane helices for LOC_00008077 and LOC_00002567, and DeepLoc 2.1 assigned LOC_00008077, LOC_00002567, and LOC_00033676 to endoplasmic-reticulum/transmembrane-compatible categories, whereas LOC_00050218 was predicted to be cytoplasmic/soluble (Table S2). TargetP 2.0 did not identify a classical N-terminal targeting-peptide class for any of the five proteins. Notably, although LOC_00050219 showed membrane-associated prediction features, its short coding sequence and lack of canonical HMGR-domain support precluded its assignment as a canonical CoHMGR gene. Overall, conserved-domain support and coding completeness were used as the primary criteria for family curation, whereas sequence-based localization predictions were treated as supportive bioinformatic evidence rather than experimental proof of protein localization [7–10].

Genomic distribution and intragenomic collinearity are consistent with the curated four-member CoHMGR set

We next examined the genomic distribution and intragenomic collinearity patterns of the curated CoHMGR loci. In the transcriptome-guided annotation output, the four canonical CoHMGR genes were located in four distinct chromosomal regions, whereas LOC_00050219 occurred in the same local genomic window as LOC_00050218 on CM039314.1 (Fig. 4). This distribution indicates that the four canonical loci are embedded in separate genomic contexts across the C. oleifera genome rather than representing a single local cluster expansion. MCScanX-based intragenomic collinearity analysis further refined this interpretation. Five collinear links were detected among genomic regions containing the four canonical CoHMGR loci, including links involving LOC_00033676, LOC_00008077, LOC_00002567, and LOC_00050218 (Table S3). These links indicate that the canonical CoHMGR-containing regions are connected through conserved gene-order relationships across related chromosomal segments, consistent with a curated four-member CoHMGR set rather than an arbitrary collection of HMGR-like predictions [15]. By contrast, no MCScanX collinearity support was detected for LOC_00050219 in the current output, and no tandem relationship involving LOC_00050219 and the canonical CoHMGR loci was recovered. This distinction is notable because LOC_00050219 lies close to LOC_00050218 in the same local genomic window but lacks the collinearity evidence observed for the canonical loci. Overall, the genomic distribution and intragenomic collinearity evidence complemented the phylogenetic and structural analyses presented above. The four canonical loci were distributed across distinct genomic regions but connected by MCScanX-supported collinear relationships, whereas LOC_00050219 lacked detectable collinearity or tandem-duplication support.

Fig. 4.

Local genomic distribution and intragenomic collinearity of CoHMGR-related loci in C. oleifera. Gray bars show local genomic regions, blue arrows indicate the four canonical CoHMGR genes, and the red arrow indicates the truncated HMGR-like locus, LOC_00050219. Curved lines represent MCScanX-detected collinear links among canonical CoHMGR-containing regions. Chromosome identifiers and local coordinate ranges are shown at the left

Reference-guided transcriptomics reveals divergent CoHMGR expression trajectories during postharvest warm conditioning

Reference-guided RNA-seq analysis provided a coherent sample-level context for interpreting CoHMGR expression during postharvest warm conditioning. Principal component analysis based on the gene-level TPM matrix separated the three sampling stages, with PC1 and PC2 explaining 32.77% and 18.17% of the total variance, respectively (Fig. 5C). The sample-to-sample distance heatmap further showed close clustering among biological replicates within each treatment stage (Fig. 5D), and the mapping summary indicated comparable alignment profiles across libraries (Fig. 5E). Together, these sample-level analyses provided a consistent transcriptomic background for downstream gene-level interpretation.

Fig. 5.

Reference-guided transcriptome structure and expression diversification of canonical CoHMGR genes during postharvest warm conditioning. A Expression trajectories of the four canonical CoHMGR genes at T-0, T-12, and T-24, shown as log2(TPM + 1). Points represent biological replicates, lines indicate stage means, and error bars denote SD. B Heatmap of stage-mean CoHMGR expression shown as row Z-scores. C PCA based on the gene-level TPM matrix; PC1 and PC2 explain 32.77% and 18.17% of the variance, respectively. D Sample-to-sample distance heatmap based on the gene-level TPM matrix. E Mapping summary showing uniquely mapped, multiple-mapped, and unmapped read proportions. HMGR1–HMGR4 correspond to the canonical CoHMGR loci listed in Table 1

Within this analytical framework, all four canonical CoHMGR genes were transcriptionally detected, but their expression profiles were clearly divergent across the warm-conditioning series (Fig. 5A, B). HMGR1 showed the highest transcript abundance at T-0, decreased at T-12, and increased again at T-24, indicating that high baseline expression did not correspond to a simple monotonic response to conditioning. HMGR2 displayed the strongest treatment-associated induction, increasing from T-0 to T-12 and remaining elevated at T-24. By contrast, HMGR3 showed a progressive decline across the series, whereas HMGR4 remained a comparatively low-abundance member but increased at the later stage. Thus, the canonical CoHMGR genes were not transcriptionally activated as a uniform block; instead, they partitioned into distinct temporal expression modes.

This expression partitioning is consistent with the broader biology of plant HMGR families, in which paralogous members often differ in expression pattern and biological contribution rather than functioning as interchangeable copies [7]. Recent work in C. oleifera seed kernels also supports non-equivalent CoHMGR behavior during squalene biosynthesis, reinforcing the need to interpret this family as a differentiated gene set rather than as a single generic HMGR signal [10]. In the present warm-conditioning context, HMGR2 showed the expression trajectory most closely aligned with the early increases in squalene accumulation and the ELISA-based estimate of HMGR activity described above, whereas HMGR1 appeared to represent a high-abundance but non-monotonic transcript. These relationships provide a rational basis for candidate prioritization, but they remain associative and do not by themselves establish causal control over the biochemical phenotype. Overall, the reference-guided transcriptomic results extend the preceding phylogenetic, structural, domain-architecture, and collinearity evidence by showing that the curated canonical CoHMGR genes are also differentiated at the expression level during the postharvest warm-conditioning window.

RT–qPCR provides directional support for CoHMGR expression divergence and cross-platform concordance during postharvest warm conditioning

RT–qPCR analysis provided an independent validation layer for the expression divergence observed among representative canonical CoHMGR genes (Fig. 6). Relative transcript abundance was normalized to SAND and calculated using the 2^−ΔΔCt method, with T-0 used as the calibrator [16]. Among the three assayed CoHMGR genes, HMGR2 showed the strongest early increase at T-12 and remained above baseline at T-24, whereas HMGR1 increased more progressively and reached its highest level at T-24. HMGR3 exhibited a comparatively modest increase across the same interval (Fig. 6A, B). These trajectories provide directional support for temporally differentiated CoHMGR expression and are consistent with the reference-guided transcriptomic patterns described above. The cross-platform comparison further supported the directional stability of the principal expression trends. Across the representative target–contrast pairs, RNA-seq-derived and RT–qPCR-derived log2 fold-change estimates showed strong positive concordance (Spearman’s ρ = 0.92, P = 2.04 × 10^−4, n = 10; Fig. 6C). This agreement indicates that the main directional responses were preserved across platforms, even though strict numerical equivalence between RNA-seq and RT–qPCR is not expected. Additional RT–qPCR profiling of selected contextual targets showed that the HMGR response occurred alongside treatment-responsive MAPK- and phosphatase-related candidates (Fig. 6D), providing a broader signaling context for the CoHMGR-focused observations without implying direct regulatory control.

Fig. 6.

RT–qPCR validation of CoHMGR expression divergence and cross-platform concordance. A RT–qPCR profiles of HMGR1, HMGR2, and HMGR3 at T-0, T-12, and T-24. Relative expression was normalized to SAND, calculated using the 2^−ΔΔCt method with T-0 as the calibrator, and shown as log2-transformed values. Points represent replicate-level measurements, lines indicate stage means, and error bars denote SD. B Heatmap of stage-mean RT–qPCR expression shown as row Z-scores. C Concordance between RNA-seq- and RT–qPCR-derived log2 fold changes for representative target–contrast pairs. The dashed line indicates the fitted trend, and Spearman’s correlation coefficient is shown. D RT–qPCR log2 fold-change profiles of selected HMGR-, MAPK-, and phosphatase-related targets relative to T-0

These validation results should nevertheless be interpreted within the methodological limits of the assay. Because SAND Ct values showed dispersion across samples, the resulting RT–qPCR estimates are best regarded as directional, reference-level evidence rather than absolute quantitative measurements. This interpretation is consistent with MIQE-oriented expectations, in which transparent reporting of normalization behavior is essential for evaluating RT–qPCR-based inference [17–19]. Taken together, the RT–qPCR results support the interpretation that expression divergence is a consistent feature of CoHMGR behavior during postharvest warm conditioning, while maintaining an appropriately cautious distinction among cross-platform concordance, candidate prioritization, and causal regulation.

Predicted phosphosite candidates define testable post-translational hypotheses for a representative CoHMGR protein

To explore potential post-translational regulatory features of CoHMGR, we analyzed a representative 501-aa CoHMGR protein using NetPhos. The prediction identified multiple putative serine, threonine, and tyrosine phosphorylation candidates, among which S47, S90, S116, T208, T238, and T246 were prioritized for interpretation in the present study (Fig. 7). These candidates were distributed across the protein sequence rather than concentrated at a single dominant residue, indicating that the current analysis defines a set of experimentally tractable phosphosite hypotheses rather than a uniquely resolved regulatory switch. Because NetPhos is a sequence-based predictor, these sites should be interpreted as candidates for follow-up validation rather than as evidence of in vivo phosphorylation.

Fig. 7.

Predicted phosphorylation-site candidates in a representative CoHMGR protein. NetPhos analysis of the representative 501-aa CoHMGR sequence identified multiple putative serine, threonine, and tyrosine phosphorylation candidates. Vertical lines indicate the top retained NetPhos score at each positive site, and symbols distinguish S, T, and Y residues. Circled labels indicate the prioritized candidate sites S47, S90, S116, T208, T238, and T246. The horizontal bar represents the full protein sequence from the N terminus to the C terminus

This conservative interpretation is important in the context of plant HMGR biology. Plant HMGR proteins are membrane-associated enzymes primarily linked to the endoplasmic reticulum and retain a conserved catalytic region, although their N-terminal and membrane-associated regions may vary among family members [8, 9]. In Arabidopsis, PP2A has been implicated in the regulation of HMGR transcript abundance, protein accumulation, and enzyme activity, supporting the biological plausibility of phosphatase-associated control of plant HMGR [12]. Recent work in C. oleifera seed kernels has also highlighted non-equivalent CoHMGR family behavior during squalene biosynthesis, further supporting targeted follow-up of HMGR regulation in this species [10].

Accordingly, Fig. 7 should be viewed as a prioritization map rather than direct evidence of site-specific modification under postharvest warm conditioning. The prediction does not identify the responsible kinase or phosphatase, establish phosphorylation occupancy in vivo, or demonstrate a direct connection between any individual residue and the ELISA-based estimate of HMGR activity or squalene accumulation. Its primary value is to define a focused set of candidate residues for phosphoproteomic validation, targeted proteomics, site-directed mutagenesis, and protein-level functional assays.

Reference-guided transcriptomics places the CoHMGR response within a staged warm-conditioning program

The broader reference-guided transcriptome indicated that postharvest warm conditioning elicited a staged transcriptional program rather than a simple before–after transition. Across the three pairwise contrasts, 5,189, 5,434, and 2,613 differentially expressed genes (DEGs) were identified in T-12 vs T-0, T-24 vs T-0, and T-12 vs T-24, respectively (Fig. 8A). The two contrasts against T-0 captured the largest transcriptomic divergence, and their overlap indicated that a substantial fraction of the response was already established by 12 h and remained evident at 24 h. At the same time, the T-12 vs T-24 comparison retained a sizeable DEG set, indicating that the transcriptome continued to adjust after the initial shift. The representative DEG heatmap further illustrated this stage-resolved structure at the expression-pattern level (Fig. S3A). Together, these results indicate that warm conditioning induced an early transcriptional reprogramming phase followed by continued temporal refinement, rather than a single static response. This interpretation is consistent with RNA-seq analytical practice, in which sample-level structure, differential-expression architecture, and functional interpretation are treated as complementary layers of transcriptomic inference [20–22]. Functional enrichment analysis further showed that the staged response converged on recurring biological themes rather than unrelated comparison-specific outputs (Fig. 8B). The most reproducible axes involved stress signaling, protein homeostasis, and carbon/lipid metabolic reallocation. These enrichments are interpreted here as transcriptomic context rather than evidence for a single dominant regulatory node. This restraint is important because postharvest warm conditioning is expected to involve coordinated changes in signaling, protein folding and turnover, redox adjustment, and metabolic redistribution. The module-level enrichment detail in Fig. S3B further supported this interpretation by showing that representative temporal modules were associated with related but non-identical functional themes. In this regard, the enrichment structure observed here is broadly consistent with C. oleifera studies reporting hormone signaling, phenylpropanoid or secondary metabolism, and broader metabolic remodeling under cold-stress and yield-associated contexts [21, 22]. It is also compatible with heat-stress studies in tea, in which heat-shock proteins, chaperones, heat-shock factors, and thermotolerance-associated protective programs have been emphasized [23]. More broadly, plant temperature-response reviews indicate that heat-associated transcriptional responses frequently involve protective protein homeostasis, stress signaling, and metabolic adjustment, supporting the interpretation of these enrichments as part of a coordinated temperature-response program [24]. Temporal module analysis placed the CoHMGR-focused results within a dynamically partitioned transcriptomic framework (Fig. 8C). The reference-guided DEG set resolved into four representative modules—early-induced, transient-induced, late-induced, and progressively repressed—containing 2,809, 1,612, 1,081, and 2,672 genes, respectively. This organization indicates that the warm-conditioning response was coordinated but not temporally uniform. The selected candidate-gene heatmap further showed that HMGR-related, stress-associated, proteostasis-related, MAPK-related, and phosphatase-related candidates occupied distinct temporal neighborhoods within this module framework (Fig. S3C). Accordingly, the module analysis is used here to provide biological context rather than to infer direct regulatory causality. Read together with the CoHMGR expression profiles, RT–qPCR validation, and phosphosite-prioritization analysis described above, the reference-guided transcriptomic background supports a conservative model in which CoHMGR-associated changes occur within a staged reorganization of signaling and metabolism during postharvest warm conditioning. However, these transcriptomic patterns alone do not establish direct in vivo regulatory relationships among candidate kinases, phosphatases, the ELISA-based estimate of HMGR activity, and squalene accumulation.

Fig. 8.

Reference-guided transcriptomic context for the CoHMGR-focused response during postharvest warm conditioning. A Differential-expression summary for T-12 vs T-0, T-24 vs T-0, and T-12 vs T-24, including up- and down-regulated DEG counts and DEG intersections. B KEGG enrichment summary grouped by recurring functional themes; point position, size, and adjacent values indicate the number of enriched pathways, significant genes, and mean −log10(q), respectively. C Temporal expression modules derived from the reference-guided DEG set. Gray lines indicate individual genes, colored lines indicate module means, and the bar plot shows module sizes

Conclusions

This study establishes a reference-guided framework for resolving CoHMGR family composition and interpreting HMGR-associated responses during a processing-relevant postharvest warm-conditioning window in C. oleifera seed kernels. The combined biochemical and transcriptomic evidence shows that harvested kernels remain metabolically responsive during short-term warm conditioning, with increased squalene accumulation occurring alongside an elevated ELISA-based estimate of HMGR activity. By integrating gene-model curation, phylogenetic placement, structural and conserved-domain evidence, protein-feature prediction, and intragenomic collinearity, we refined the CoHMGR family into a curated four-member canonical gene set and distinguished it from a truncated HMGR-like locus. Reference-guided RNA-seq and RT–qPCR further revealed isoform-level expression divergence, identifying HMGR2 as the most responsive candidate during the early warm-conditioning phase.Beyond family annotation, these findings provide a prioritized molecular framework for future studies of postharvest squalene regulation in oil-tea seed kernels. The candidate CoHMGR isoforms, predicted phosphorylation sites, and staged transcriptomic context reported here can guide subsequent phosphoproteomic validation, site-directed mutagenesis, and protein-level functional assays. Importantly, the present evidence supports candidate prioritization rather than direct mechanistic proof. Direct links among candidate kinases or phosphatases, HMGR protein regulation, and squalene biosynthesis remain to be experimentally established. Overall, this work strengthens family-level interpretation of CoHMGR biology and provides a basis for future studies of postharvest regulation of bioactive lipid accumulation in C. oleifera seed kernels.

Authors' contributions

Jianwen Wu: conceptualization, funding acquisition, writing–original draft. Rong Qin: data curation, software, validation, writing–original draft. Yingying Chen: formal analysis, writing–original draft, project administration. Jihua Guan: investigation, methodology, writing–review and editing. All authors have read and approved the submitted version of the manuscript.

Funding

This research was funded by the Central Guiding Local Science and Technology Development Fund (Grant no. Guike ZY23055025), the Guangxi Natural Science Foundation (Grant no. 2025GXNSFAA069040) and the Guangxi Forestry Science and Technology Promotion Demonstration Project (Grant no. 2023GXLK22).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.