Screening and Validation of Stable Reference Genes for Real-Time Quantitative PCR in Indocalamus tessellatus (Munro) P. C. Keng Under Multiple Tissues and Abiotic Stresses

Abstract

Indocalamus tessellatus (Munro) P. C. Keng is a bamboo species with significant economic and ecological value, and exhibits considerable resistance to abiotic stresses. However, systematic evaluation of reference genes for gene expression analysis in this species is lacking. Analysis of multi-tissue transcriptomes yielded 3801 relatively stable genes; from these, we selected eleven new candidates along with nine widely adopted reference genes. We then evaluated these candidates under four conditions: drought (15% PEG-6000), salt (200 mM NaCl), waterlogging (root submergence in water), and a multi-tissue panel (leaf, leaf sheath, culm, shoot, and root). Under stress, early and sustained time points were sampled to capture dynamic transcriptional responses. Expression stability was assessed using geNorm, NormFinder, BestKeeper, and ΔCt, and results were integrated with RefFinder to generate comprehensive stability rankings for each condition. The most stable reference genes were condition-dependent: MD10B and PP2A under drought, eIF1A and Ite23725 under salt stress, PP2A and eIF4A under waterlogging, and 60S and UBP1 across different tissues. Notably, commonly used genes such as UBI and Actin7 were less stable. Peroxidase (POD) was used as a validation marker because it is a known stress-responsive gene, providing a sensitive readout of normalization accuracy. Validation confirmed that selecting suitable reference genes is essential for dependable expression quantification. These findings provide a robust set of reference genes for qRT-PCR studies in I. tessellatus, supporting future molecular and functional research in bamboo.

1. Introduction

Indocalamus species are bamboos of significant economic and ecological value, widely distributed across southern China [1]. Several species provide leaves traditionally used for food packaging, especially for wrapping zongzi, and these leaves are rich in bioactive compounds such as flavonoids, polysaccharides, and volatile constituents with potential applications in food preservation and medicine [2,3,4]. For example, essential oils from Indocalamus latifolius (Keng) McClure leaves were shown to suppress Aspergillus flavus, a major food contaminant [5]. In natural ecosystems, Indocalamus contribute to soil and water conservation, support biodiversity, and improve habitat quality [6,7].

Environmental constraints strongly influence bamboo growth and productivity. In bamboo stands, drought restricts shoot emergence and lowers yield [8]. Prolonged waterlogging imposes hypoxic stress, leading to wilting and leaf abscission and thereby accelerating senescence [9]. Moreover, bamboo growth is highly sensitive to soil salinity, particularly in managed plantations [10]. Despite these challenges, the genus Indocalamus exhibits considerable tolerance to multiple abiotic stresses and holds promise for ecological restoration. For instance, I. latifolius exhibited relatively high tolerance to waterlogging compared with numerous bamboo species [11], and Indocalamus decorus Q. H. Dai withstands combined heat and drought stress by regulating the ascorbate-glutathione cycle, thereby mitigating oxidative damage and maintaining cellular redox balance [12]. Members of the genus have demonstrated potential for the phytoremediation of lead-contaminated soil [13] and enhanced tolerance to freezing injury [14].

Over the past decade, bamboo molecular biology has advanced substantially, aided by high-throughput sequencing and multi-omics, which helped overcome earlier bottlenecks posed by genome complexity and limited materials [15,16]. At least 13 bamboo genome assemblies are now available, including high-quality, fully sequenced genomes for Phyllostachys edulis (Carrière) J. Houz., Dendrocalamus sinicus L. C. Chia & J. L. Sun, and Bambusa emeiensis L. C. Chia & H. L. Fung [17,18,19]. These resources have enabled the elucidation of genes involved in culm elongation, hormone signaling, cell-wall dynamics, and abiotic-stress responses [16]. Despite the widespread use of RNA-seq for discovery, quantitative real-time PCR (qRT-PCR) remains indispensable for targeted validation and fine-scale expression profiling due to its sensitivity, accuracy, and cost-effectiveness [20,21]. Accordingly, accurate qRT-PCR depends on normalization to reference genes with stable expression across tissues, developmental stages, and environmental conditions [22].

In bamboos, several studies have evaluated candidate reference genes—typically including Actin, Tubulin, EF1α, UBQ, and TIP41—mainly in P. edulis and D. brandisii [23,24]. These studies demonstrate that housekeeping genes often vary in stability across organs, stages, and stresses, which can introduce multiplicative errors in 2^−ΔΔCt and bias fold-change estimates. Multi-algorithm workflows (geNorm, NormFinder, BestKeeper, and ΔCt) are therefore recommended to derive robust consensus rankings [25,26]. However, a systematic, condition-resolved reference gene panel for Indocalamus has not been reported.

Here, we present the first systematic identification and validation of reference genes for qRT-PCR in Indocalamus, focusing on Indocalamus tessellatus (Munro) P. C. Keng. Newly generated genome and transcriptome resources for I. tessellatus were utilized to systematically identify and evaluate stable reference genes under multiple abiotic stress conditions. The assessment spanned drought, salinity, and waterlogging, with multi-timepoint and multi-organ sampling to capture condition-specific stability. Furthermore, we evaluated stability using geNorm, NormFinder, BestKeeper, and the ΔCt method, and validated the optimal genes by profiling a stress-responsive peroxidase gene (ItPOD).

2. Materials and Methods

2.1. Plant Materials

Healthy and vigorous seedlings of I. tessellatus were collected in March 2024 from field-grown plants in Suichang County, Zhejiang Province, China. To ensure a uniform genetic background, all seedlings were derived from a single bamboo rhizome. Healthy culms with uniform growth were selected, divided, and transplanted into the greenhouse of Lishui University (119.91° E, 28.47° N). Environmental conditions in the greenhouse were maintained at 70 ± 10% relative humidity, natural photoperiod, and an average temperature of 23 ± 1.0 °C. Seedlings were watered and fertilized regularly to maintain optimal growth, and after two months of acclimation, and vigorous individuals were used for subsequent experiments.

2.2. Stress Treatments

Drought, salt, and waterlogging were selected as representative stresses due to their ecological relevance in subtropical forests and their importance for land restoration and stress biology. To standardize physiological status, all seedlings underwent a 2-day water-withholding pre-treatment prior to stress application. Three biological replicates were included per treatment and time point, and the second fully expanded leaves were harvested at 0, 5, 12, 24, and 48 h after stress initiation. For organ-specific analysis, leaves, leaf sheaths, stems, roots, and young shoots were sampled from the same plants. All samples were immediately frozen in liquid nitrogen and stored at –80 °C, with at least three biological replicates included for each treatment and time point.

Salt stress was initiated at time 0 by irrigating each pot once with 200 mL of 200 mM NaCl [27,28], ensuring thorough wetting with only minimal drainage into the saucer to avoid excessive leaching so that most of the applied salt remained in the substrate. No further irrigation was provided during the 48 h sampling window, and control pots were irrigated with 200 mL deionized water under otherwise identical conditions. Drought (osmotic) stress was imposed at time 0 by irrigating each pot once with 200 mL of 25% (w/v) PEG 6000 solution to reduce substrate water potential [28,29], again ensuring thorough wetting with only minimal drainage into the saucer so that the applied solute remained in the substrate. No additional irrigation was provided during treatment, and control pots were irrigated with 200 mL deionized water. Waterlogging was applied at time 0 by submerging the roots completely in deionized water while keeping the shoots above the water surface; water was maintained at room temperature, and the level was adjusted as needed to keep the root zone submerged, with controls maintained under otherwise identical conditions without root submergence.

2.3. Total RNA Extraction and cDNA Reverse Transcription

Approximately 100 mg of plant tissue per sample was ground to a fine powder in liquid nitrogen with an RNase-free mortar and pestle. Total RNA was subsequently extracted using the Plant RNA Kit (Omega Bio-tek, Norcross, GA, USA) according to the manufacturer’s instructions. RNA integrity was assessed by 1% agarose gel electrophoresis; samples showing clear 28S and 18S rRNA bands without visible degradation or contamination were considered high quality. RNA concentration and purity were measured with a NanoDrop microvolume spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). Only RNA with an A260/280 ratio of 1.8–2.2, and an A260/230 ratio above 2.0 was used for subsequent experiments.

First-strand cDNA was synthesized from 1 μg of total RNA per reaction using the PrimeScript™ FAST RT Reagent Kit with gDNA Eraser (Takara, Shiga, Japan), which effectively removed genomic DNA according to the manufacturer’s protocol. The resulting cDNA was appropriately diluted for subsequent analyses and stored at −20 °C until use.

2.4. Identification of Reference Genes from Transcriptome Data

In the present work, we reanalyzed previously generated RNA-Seq datasets as secondary data; these datasets were originally produced by our group for the I. tessellatus genome project. Unpublished RNA-Seq data from six tissues were used, with biological replicates as follows: roots n = 3, stems n = 2, mature leaves n = 3, young leaves n = 2, leaf sheaths n = 2, and shoots n = 2. Libraries were prepared as strand-specific and sequenced as paired-end 2 × 150 bp on the DNBSEQ-T7 platform (MGI Tech, Shenzhen, China), yielding approximately 101.02 Gb in total. Sequencing datasets are deposited in the NCBI SRA (SRR32130617-SRR32130622 and SRR32130625-SRR32130632). Gene-level expression estimates were normalized as transcripts per million (TPM) to allow comparison across samples.

For each gene, TPM values were log₂-transformed across all samples. The mean log₂(TPM) was calculated, and genes with an average log₂(TPM) below 5 were excluded to remove low-abundance transcripts that might not be reliably detected or quantified by RT-qPCR. For the remaining genes, the mean, standard deviation (SD), and coefficient of variation (CV; SD divided by the mean on log2-transformed TPM) were calculated using R (version 4.3.2). Candidate reference genes were required to have SD < 1 and CV ≤ 0.2, indicating relatively stable expression [30,31].

2.5. KEGG and GO Enrichment Analyses

To elucidate the biological functions of the candidate reference genes, Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed. Annotation results for all transcriptome genes and the candidate reference genes were uploaded to the OmicShare online platform (https://www.omicshare.com/tools/, accessed on 15 April 2025), and enrichment analyses were carried out using the Bioinformatics Cloud Toolkit (OmicShare, Guangzhou, China).

2.6. Primer Design and Specificity Validation

We selected the eleven novel candidate reference genes with the lowest SD and CV of log₂ (TPM) values, together with nine commonly used reference genes, for stability analysis (

Table 1. Overview of 20 candidate reference genes in Indocalamus tessellatus and their homologous genes in rice (Oryza sativa) and Arabidopsis thaliana.

Category	Gene	ID	Description	Rice Ortholog	Arabidopsis Ortholog
TRGs *	Actin7	Ite22542	Actin	Os11t0163100	AT5G09810
	Tubulin	Ite01366	Alpha-tubulin	Os03t0219300	AT5G19780
	UBI	Ite18473	Ubiquitin-like protein	Os08t0295100	AT5G25270
	eIF1A	Ite39018	Translation initiation factor 1A	Os06t0342200	AT2G04520
	eIF4A	Ite16714	Translation initiation factor 4A	Os03t0566800	AT3G19760
	PP2A	Ite19575	Protein phosphatase 2A	Os01t0691700	AT1G50370
	SAMDC	Ite30893	S-adenosylmethionine decarboxylase	Os02t0611200	AT3G25570
	60S	Ite41075	60S ribosomal protein L36	Os05t0459900	AT5G02450
	CYP	Ite17392	Cyclophilin	Os06t0670500	AT1G53720
NRGs *	BS	Ite29857	bet1-like SNARE	Os02t0820700	AT3G58170
	Ite23725	Ite23725	unknown protein	Os08t0374200	AT4G10970
	ARF	Ite03155	ADP-ribosylation factor	Os05t0489600	AT5G14670
	UBP1	Ite37126	oligouridylate-binding protein 1	Os11t0620100	AT1G17370
	PEX13	Ite18045	peroxisomal membrane protein 13	Os07t0152800	AT3G07560
	SKA	Ite10401	shaggy-related protein kinase alpha	Os01t0252100	AT5G26751
	PNN	Ite00796	pinin	Os03t0701900	AT1G15200
	RPN8	Ite09253	Regulatory Particle Non-ATPase 8	Os04t0661900	AT5G05780
	MD10B	Ite31918	Mediator complex subunit 10B	Os09t0528300	AT1G26665
	SUGP1	Ite07289	SURP and G-patch domain protein 1	Os09t0281600	AT3G52120
	HNRPQ	Ite14890	Heterogeneous nuclear ribonucleoprotein Q	Os11t0250000	AT4G00830

* TRGs: traditional reference genes; NRGs: new reference genes.

). Sequences of the commonly used reference genes were identified by BLAST (TBtools version 1.120) against the reference genome of I. tessellatus (GenBank assembly accession: GCA_048537245.1). Primers were designed using NCBI Primer-BLAST (https://www.ncbi.nlm.nih.gov/tools/primer-blast/, accessed on 26 April 2025), with the following parameters: constraining primer length to 18–22 bp, GC to 40–60%, Tm to 58–62 °C with ΔTm ≤ 2 °C between primer pairs, and an amplicon size 80–250 bp. Because a complete reference genome of I. tessellatus is not available in Primer-BLAST, primer specificity was assessed against closely related bamboo genomes and by in silico screening of our I. tessellatus CD sequences. Off-targets were required to contain at least two mismatches with ≥1 mismatch within the last five 3′-terminal bases. Potential secondary structures and primer–dimers were examined in IDT OligoAnalyzer (https://www.idtdna.com/pages/tools/oligoanalyzer, accessed on 26 April 2025), accepting hairpin/heterodimer ΔG > −9 kcal·mol⁻¹. Whenever possible, primers spanned exon–exon junctions. All primer sequences are listed in Table S1. Primer specificity was validated by standard PCR, and amplifications were visualized on 2% agarose gel electrophoresis to confirm single, specific bands.

2.7. Real-Time Quantitative PCR Analysis

RT-qPCR assays were carried out with TB Green^® Premix Ex Taq™ II (Takara, Shiga, Japan) on a Roche LightCycler^® 96 real-time PCR instrument (Roche Diagnostics GmbH, Mannheim, Germany), according to the manufacturer’s instructions. Each 20 μL reaction mixture comprised 0.6 μL of forward primer, 0.6 μL of reverse primer, 10 μL of 2× TB Green Premix Ex Taq II, 7.8 μL of ddH₂O, and 1 μL of diluted cDNA solution. Cycling conditions: 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. At the end of the amplification, melt curve analysis was performed as follows: 95 °C for 1 s, 65 °C for 15 s, followed by a gradual increase to 95 °C while continuously monitoring fluorescence. Each reaction was carried out in three technical replicates to ensure reliability.

Prior to quantitative analyses, standard curves were generated for each primer pair using pooled cDNA adjusted to a concentration of 1200 ng/μL and subjected to five-fold serial dilutions (undiluted, 1:5, 1:25, 1:125, and 1:625), with each dilution assayed in triplicate to obtain Ct values. After amplification, melt curve analysis confirmed the specificity of the reactions. Linear regression of Ct versus log₁₀ dilution provided the slope and correlation coefficient (R²), and amplification efficiency (E) were calculated using the formula:

$$E=({10}^{-{\displaystyle \frac{1}{slope}}}-1)\times 100\%$$

Primer pairs were accepted for quantification when they produced single-peak melt curves, R² ≥ 0.97, and slopes between −3.60 and −3.22 corresponding to E = 90%–110%, with no amplification in controls; under these conditions, the 1:5 diluted cDNA was selected as the working template to keep Cq values within the dynamic range of the standard curves.

2.8. Data Analysis and Statistics

PCR standard curves (linear equations) and amplification efficiencies were calculated in Microsoft Excel. Raw Cq values were imported into GraphPad Prism (version 9.5) to visualize the distribution of Ct values for the twenty candidate reference genes. We evaluated reference gene stability using four algorithms, geNorm [32], NormFinder [33], BestKeeper [34], and the ΔCt method [35], all run separately within RefFinder [36] (http://blooge.cn/RefFinder/, accessed on 20 June 2025) to generate individual stability values and rankings. RefFinder subsequently combines these rankings by weighting each gene accordingly and computing the geometric mean to yield an overall comprehensive ranking. According to the geNorm algorithm, the optimal number of reference genes can be determined by analyzing the pairwise variation (V_n/n+1) of normalization factors. A V_n/n+1 value less than or equal to 0.15 indicates that n reference genes are sufficient for accurate normalization. If the value exceeds 0.15, additional reference genes should be included until the value falls below this threshold. In this study, the pairwise variation values were calculated using the R (version 4.5.1) packages ReadqPCR and NormqPCR [37]. Visualizations of geNorm and NormFinder outputs were generated in GraphPad Prism (version 9.5) and TBtools (version 1.120) [38], respectively.

2.9. Reference Genes Validation Using Gene ItPOD

To validate the stability of reference genes, the expression of the ItPOD gene, which encodes a peroxidase involved in plant stress responses, was analyzed. Expression levels were examined under waterlogging, salt, and drought stress conditions, as well as across different tissues. The relative expression of ItPOD was calculated using the two most stable and the least stable reference genes identified by the RefFinder stability ranking. Each qPCR reaction was performed in three technical replicates to ensure reproducibility. Relative expression levels were determined using the 2^−ΔΔCt method [39]. For each time point, group genes were compared by one-way ANOVA (two-tailed, α = 0.05), followed by Tukey’s HSD for post hoc pairwise comparisons. Analyses and visualizations were performed in R (version 4.5.1).

3. Results

3.1. Transcriptome-Based Screening and Functional Enrichment

In this study, we utilized RNA-Seq data from six different tissues of I. tessellatus, including roots, stems, mature leaves, young leaves, leaf sheaths, and shoots. The transcript abundance was normalized to TPM for all samples. To identify suitable candidate reference genes, genes with average log₂(TPM) values less than 5 across all tissues were first excluded, thereby removing low-abundance transcripts. For the remaining genes, we calculated the SD and CV of log₂(TPM) values across all samples in order to evaluate expression stability. Only genes with SD less than 1 and CV no greater than 0.2 were retained. Using these criteria, a total of 3801 genes were identified as candidate reference genes that exhibited both high and stable expression across all examined tissues (Table S2).

GO and KEGG enrichment analyses were conducted to investigate the biological functions and pathways associated with these candidate genes. GO results revealed that the majority of annotated genes were assigned to the biological process (BP) category, followed by the cellular component (CC) and molecular function (MF) categories. In the BP category, most selected genes were involved in cellular processes, metabolic processes, single-organism processes, and responses to stimulus. Within the CC category, these genes were mainly enriched in cell, cell part, organelle, organelle part, and membrane. For MF, binding and catalytic activity were the predominant terms (Figure 1A and Table S3). These CC and MF patterns point to ubiquitous subcellular structures and fundamental molecular activities that typically support general cellular maintenance rather than specialized roles, which is consistent with cross-tissue expression stability.

KEGG pathway analysis further demonstrated that these 3801 stably expressed genes are widely distributed across essential metabolic and cellular pathways (Figure 1B and Table S4). The most significantly enriched KEGG pathways included ribosome, spliceosome, protein processing in endoplasmic reticulum, ubiquitin-mediated proteolysis, mRNA surveillance pathway, and oxidative phosphorylation. Genes involved in these pathways are largely associated with maintaining basal transcription, translation, protein turnover, RNA quality control, and energy production, and they are in constitutive demand across tissues. Classical pathways such as carbon metabolism, proteasome function, and endocytosis, were also prominently represented, aligning with the typical characteristics of reference genes.

Overall, the depletion of tissue-specific pathways and the enrichment of core biosynthetic and quality-control processes are consistent with our low SD and CV criteria. This consistency offers a mechanistic basis for cross-tissue invariance and supports their suitability for RT-qPCR normalization.

3.2. Verification of Amplification Specificity and Efficiency for Reference Genes

From the transcriptome analysis, eleven genes exhibiting the lowest SD and CV values were selected as novel candidate reference genes for further validation. These genes included BS, Ite23725 (unknown function), ARF, UBP1, PEX13, SKA, PNN, RPN8, MD10B, SUGP1, and HNRPQ. In addition, nine commonly used traditional reference genes were selected for comparison: Actin 7, Tubulin, UBI, eIF1A, eIF4A, PP2A, SAMDC, 60S, and CYP (Table 1).

Primer pairs for all candidate reference genes were designed using NCBI Primer-BLAST (Table S1). To verify primer specificity and product size, conventional PCR was conducted, and the resulting products were assessed by agarose gels. As shown in Figure S1, all PCR products were within the expected size range (80–250 bp), and each lane displayed a single, clear band, confirming high primer specificity. Subsequently, qRT-PCR was conducted using serially diluted mixed cDNA templates to generate standard curves. The amplification efficiencies ranged from 90% to 105%, and the correlation coefficients (R²) ranged from 0.979 to 0.999 (Table S1), demonstrating high amplification efficiency and strong linearity. Melting curve analysis further confirmed primer specificity, with a single peak detected for each of the 20 genes within the 80–90 °C range (Figure S2).

Together, these results demonstrate that all selected primer sets are highly specific, produce amplicons of the expected size, and have suitable PCR efficiency, making them reliable for use in subsequent RT-qPCR analyses.

3.3. Expression Patterns of Candidate Reference Genes

The threshold cycle (Ct) reflects transcript abundance: lower Ct values indicate higher expression. In this work, we analyzed 20 candidate reference genes across various tissues of I. tessellatus and under multiple abiotic stresses (drought, salt, and waterlogging). The Ct values ranged from 19.91 to 34.97 (Figure 2), reflecting substantial variation in expression. Among the genes, ARF exhibited the highest expression level, with a mean Ct value of 21.01 and a median of 20.78. This was followed by 60S (median 21.97), UBP1 (median 21.99), and PEX13 (median 22.54). In contrast, Tubulin displayed the lowest expression levels, with a mean Ct of 26.90 and a median of 26.56. Regarding expression stability, PP2A, UBI, and Ite23725 showed the narrowest interquartile ranges, indicating relatively consistent expression across samples; however, UBI did exhibit some variation under certain conditions. Additionally, SAMDC, CYP, and PNN demonstrated minimal differences between their maximum and minimum Ct values, further supporting their stable expression.

3.4. geNorm Analysis

The geNorm algorithm was used to assess the expression stability of candidate reference genes based on the M value (Figure 3), where lower M values indicate higher expression stability. An M value below 1.5 is generally considered acceptable for reference gene selection. Under drought stress, PP2A and PEX13 exhibited the highest stability (M = 0.15), whereas UBI showed the lowest stability (M = 0.93). In salt-stressed samples, MD10B and Ite23725 were the most stable (M = 0.22), while BS ranked as the least stable (M = 0.78). During waterlogging stress, eIF4A and PP2A (M = 0.29) demonstrated the highest stability, in contrast to Ite23725 (M = 0.83) and MD10B (M = 0.79), which exhibited lower stability. With an M value of 0.38, eIF4A and 60S emerged as the most stable genes across different tissues, whereas Actin7 displayed the greatest variability (M = 1.14).

Interestingly, the identity of the most stable reference genes was not consistent between the different experimental conditions. For instance, PP2A consistently ranked among the most stable under both drought and waterlogging stress, whereas Ite23725 and MD10B were highly stable only under salt stress. In contrast, some genes such as Actin7, BS, and UBI repeatedly appeared among the least stable across multiple conditions.

To determine the optimal number of reference genes required for reliable normalization, we calculated pairwise variation values (V_n/n+1) using geNorm (Figure 4). A common threshold of 0.15 was used to evaluate whether including an additional reference gene would significantly improve normalization accuracy. For drought stress, salt stress, waterlogging stress, and different organs, the V_2/3 values were substantially below 0.15. This indicates that using two reference genes is sufficient for accurate qRT-PCR normalization in all tested conditions. Inclusion of more reference genes did not provide further substantial benefit, so the use of two reference genes is recommended for subsequent gene expression analysis across these experimental settings.

3.5. NormFinder Analysis

NormFinder was used to evaluate the expression stability of candidate reference genes under various experimental conditions. In this method, genes with lower stability values are considered to have more consistent expression. As shown in Figure 5, under drought stress, PP2A and MD10B exhibited the lowest stability values and were thus identified as the most stable reference genes. Under salt stress, SAMDC and eIF1A showed the highest stability. Similarly, during waterlogging, PP2A and Tubulin emerged as the top stability performers. When analyzing gene stability across different organs, UBP1 and 60S consistently emerged as the top-performing reference genes. In contrast, UBI was generally unstable across multiple conditions. Notably, PP2A demonstrated stable expression in nearly all settings, supporting its suitability as a reference gene for normalization across diverse experimental conditions.

3.6. BestKeeper Analysis

BestKeeper analysis was used to evaluate the expression stability of candidate reference genes under different experimental conditions by comparing the SD and CV values. Genes with the lowest SD and CV values are considered the most stable. As shown in Table S5, under drought stress, MD10B was identified as the most stable reference gene, followed by PP2A and eIF4A. For salt stress conditions, PP2A exhibited the highest stability, with PEX13 and RPN8 ranking closely behind. Under waterlogging stress, eIF4A demonstrated the most stable expression, while PP2A and CYP were also highly stable. In the analysis of different organs, PNN and eIF1A ranked as the top two stable reference genes.

In contrast, UBI consistently showed excessively high SD and CV values, ranking as the least stable gene under drought, salt, and waterlogging stress. For different organs, Actin7 displayed the highest SD and CV values, indicating it ranked last in reference gene stability. These results suggest that MD10B, PP2A, eIF4A, and PNN are recommended for qRT-PCR normalization under specific conditions, while UBI and Actin7 should be avoided due to their poor stability.

3.7. ΔCt Analysis

As shown in Table S6, stability was evaluated using the ΔCt method. For each experimental condition or tissue type, the standard deviation of ΔCt values between pairs of candidate reference genes was calculated across all samples. Genes with smaller variation in ΔCt values are considered to be more stably expressed. Under drought stress, MD10B and 60S exhibited the highest stability, with the lowest average STDEV values. For salt stress, eIF1A and SAMDC showed the highest stability among the candidates. During waterlogging stress, PP2A and Tubulin displayed the most stable expression. In the analysis of different organs, 60S and UBP1 were identified as the most stable genes. On the other hand, UBI showed extremely poor stability under drought stress, with an average STDEV of 4.54, the highest among all tested genes. This result indicates that UBI was the least stable reference gene under this condition. BS, Ite23725, and Actin7 were the least stable under salt stress, waterlogging, and across different organs, respectively.

3.8. Comprehensive Analysis of RefFinder

The four commonly used methods, namely, geNorm, NormFinder, BestKeeper, and delta Ct, differ to some extent in their mathematical principles, resulting in variation in the stability rankings of candidate reference genes (Table S7). Although certain genes consistently ranked among the most stable under specific conditions, the discrepancies among algorithms emphasize the importance of employing multiple analytical methods to identify suitable reference genes. According to the results from multiple algorithms, PP2A emerged as a highly suitable and stable reference gene under both drought and waterlogging stresses. In contrast, under salt stress, the top-ranked genes varied by method, geNorm favored MD10B and Ite23725, NormFinder favored SAMDC and eIF1A, and BestKeeper favored PP2A and PEX13, indicating less agreement and the necessity for precise validation under this condition. For different tissues, 60S and UBP1 were consistently placed among the top performers by multiple algorithms.

To derive a consensus ranking, RefFinder integrated the method-specific results using the geometric mean of ranks across conditions (

Table 2. Stability ranking of candidate genes assessed using RefFinder software.

Ranking	Drought		Salt		Waterlogging		Different Organs
	Genes	Stability Score	Genes	Stability Score	Genes	Stability Score	Genes	Stability Score
1	MD10B	1.78	eIF1A	2.89	PP2A	1.32	60S	2.40
2	PP2A	1.86	Ite23725	3.31	eIF4A	2.65	UBP1	3.35
3	PEX13	3.03	SAMDC	3.46	CYP	3.13	BS	3.94
4	60S	4.12	MD10B	3.66	Tubulin	3.72	eIF4A	6.00
5	Ite23725	4.16	PP2A	3.76	Actin7	4.56	SKA	6.34
6	ARF	5.96	60S	5.05	HNRPQ	5.70	UBI	6.36
7	UBP1	7.26	PEX13	6.48	ARF	6.32	Ite23725	6.62
8	HNRPQ	7.75	UBP1	6.96	SAMDC	6.91	PP2A	6.82
9	eIF4A	9.69	ARF	7.17	PEX13	9.49	ARF	7.33
10	SKA	10.12	RPN8	8.66	SKA	10.59	PNN	7.48
11	RPN8	10.29	SUGP1	9.53	BS	10.92	MD10B	9.32
12	PNN	10.72	Actin7	10.79	SUGP1	11.89	eIF1A	10.10
13	eIF1A	11.17	UBI	11.13	UBP1	12.24	CYP	10.44
14	Actin7	13.07	CYP	12.62	UBI	13.48	PEX13	11.03
15	SAMDC	14.46	SKA	14.00	60S	14.49	SAMDC	11.98
16	BS	15.23	PNN	14.42	eIF1A	14.98	HNRPQ	13.24
17	CYP	15.66	Tubulin	16.36	PNN	15.17	SUGP1	13.94
18	SUGP1	17.20	HNRPQ	17.73	RPN8	16.49	Tubulin	14.49
19	Tubulin	19.00	eIF4A	18.74	MD10B	19.00	RPN8	15.54
20	UBI	20.00	BS	20.00	Ite23725	20.00	Actin7	20.00

). Under drought stress, MD10B and PP2A consistently displayed the highest overall stability among the tested genes, with geometric mean rank values of 1.78 and 1.86, respectively. For salt stress, eIF1A (2.89) and Ite23725 (3.31) were identified as the most stable genes, while BS (20.00), eIF4A (18.74), and HNRPQ (17.73) were among the least stable. Under waterlogging stress, PP2A (1.32) and eIF4A (2.65) ranked highest, indicating strong stability in these conditions. When comparing different organs, 60S (2.40) and UBP1 (3.35) were found to be the most reliable reference genes. Collectively, these findings suggest that MD10B, PP2A, eIF1A, eIF4A, and 60S are suitable as internal controls for accurate normalization for gene-expression analyses across a range of conditions in I. tessellatus, whereas UBI and Actin7 should be used with caution due to their poor stability.

3.9. ItPOD-Driven Assessment of Normalization Controls

To further corroborate these findings, the two top-ranked and the lowest-stability reference genes for each treatment or tissue (as identified by RefFinder) were selected for normalization in expression analysis. POD (Peroxidase), a widely recognized stress response gene, was used as the target to evaluate the reliability of the selected reference genes in calibrating expression patterns. Under drought stress, normalization with the stable reference genes MD10B, PP2A, or their combination produced consistent expression patterns, showing moderate increases at 5 h and 12 h followed by declines at later stages. There were no significant differences among these stable genes, with similar statistical groupings across time points. In contrast, normalization with the unstable reference gene UBI caused dramatic overestimation of expression, especially at 5 h, where values exceeded 30,000-fold (Figure 6A). Similarly, under salt stress, using eIF1A, Ite23725, or their combination as reference genes resulted in significantly different ItPOD expression levels at 5 h and 48 h compared to using the least stable gene, BS (Figure 6B). Under waterlogging stress, normalization with Ite23725 led to significantly lower ItPOD expression at 5 h, 12 h, 24 h, and 48 h, in contrast to normalization with PP2A, eIF1A, or their combination, which showed consistent and higher expression patterns (Figure 6C). For expression analysis across different organs, using the low-stability reference control Actin7 for normalization yield lower ItPOD expression levels in the leaf sheath, bamboo shoot, leaf, and root compared to stable reference controls. Moreover, Actin7 altered the expression pattern, showing higher expression in the leaf sheath than in the bamboo shoot, which was opposite to the trend observed with stable reference genes (Figure 6D).

4. Discussion

4.1. Transcriptome-Guided Candidate Identification and Functional Enrichment

Stable reference genes are crucial for accurate RT-qPCR normalization [40]. Although housekeeping genes such as Actin, Tubulin and GAPDH are frequently used for normalization, their expression stability often varies across species, tissues, and experimental conditions [25,26]. To overcome this, we combined transcriptome-based screening with conventional candidates in I. tessellatus, selecting genes with low expression variability across tissues and confirming that such transcriptomic metrics correlate with RT-qPCR Ct values. Enrichment analyses showed that candidate reference genes participate in core metabolic and cellular pathways (Figure 1), consistent with their expected constitutive roles. Similar transcriptome-guided selection strategies have been successfully applied in other plants, such as Panax ginseng C. A. Mey. [41], Medicago sativa L. [42] and Commelina communis L. [43], providing a systematic basis for normalization.

4.2. Cross-Method Stability Assessment and Integrated Ranking of Reference Genes

We evaluated nine traditional housekeeping genes and eleven RNA-seq–derived candidates across drought, salt, and waterlogging stresses, as well as across multiple tissues, using geNorm (Figure 3), NormFinder (Figure 5), BestKeeper (Table 2), and the ΔCt method (Table S5). Consistent with findings in Dendrocalamus brandisii (Munro) Kurz. [24], Betula luminifera H. J. P. Winkl. [44], and other species [45,46], the rankings did not fully agree across methods: under drought, PP2A ranked first in geNorm and NormFinder, whereas MD10B ranked highest in BestKeeper and ΔCt; under salt, the top choices diverged further (Table S7). These differences are expected because each algorithm quantifies stability in a distinct way, rather than one being universally superior [36,47]. For example, geNorm prioritizes the consistency of pairwise expression ratios among candidate genes across samples but can be misled when candidates are co-regulated under a treatment [32]. In contrast, NormFinder minimizes between-group shifts relative to within-group variability [22,48]. BestKeeper emphasizes the dispersion of raw Ct values (e.g., SD and CV) and the correlation of each gene with a composite index, thereby favoring genes that are stable in absolute Ct terms rather than in pairwise ratios [34]. To mitigate method-specific bias and obtain a consensus, we used RefFinder to integrate rankings via the geometric mean, an approach that is increasingly adopted in reference gene validation studies [23,32,46]. The composite analysis highlighted MD10B, eIF1A, PP2A, and 60S as consistently among the most stable across stresses and tissues, with eIF4A and Ite23725 also showing relatively high stability. This integrated strategy aligns with MIQE recommendations for multi-gene, verifiable normalization [49].

4.3. Performance and Functional Basis of Key Reference Genes

PP2A is a conserved serine/threonine phosphatase whose heterotrimeric holoenzyme underpins diverse regulatory functions [50], and it serves broadly as a reference control due to typically stable transcript levels across tissues and conditions in species such as Nicotiana benthamiana Domin [51], Brassica napus L. [52] and Setaria viridis (L.) P. Beauv. [53]. Consistent with this, PP2A showed excellent stability in I. tessellatus under drought and waterlogging (Table 2). MD10B, a Mediator complex subunit that links sequence-specific transcription factors to RNA polymerase II [54,55], likewise ranked among the most stable candidates (Table 2; Figure 3B). PP2A and MD10B likely stay stable under stress because their functions are buffered by regulation at the holoenzyme/protein-complex level [54,56], keeping core subunit transcripts relatively constant while still allowing rapid functional adjustment, making them suitable for normalization. Although Actin7 and UBI are commonly used reference genes [52], their transcripts were less stable in I. tessellatus. This likely reflects active reprogramming of cytoskeletal dynamics and proteostasis across tissues under abiotic stress. Actin7 responds to stress-driven changes in cell architecture and trafficking (e.g., cell wall remodeling, aerenchyma formation, altered root growth), causing condition- and tissue-specific expression shifts [57]. UBI tracks protein turnover demands that rise with misfolded-protein clearance, selective autophagy, and repair, with E3 ligases and the ubiquitin–proteasome system modulating pathway components context-dependently [58].

4.4. Optimal Number of Reference Genes

A key question in RT-qPCR studies is how many reference genes are required for robust normalization. A single reference gene is discouraged because it is susceptible to condition-, tissue-, and batch-specific effects that can introduce systematic bias into normalization [32,33,49]. We therefore determined the number of references in a data-driven manner: geNorm pairwise variation V_n/n+1 was computed and complemented with stability metrics (M-values, CV), accounting for potential between-group shifts. Across all abiotic stress and tissue conditions, it remained below our predefined threshold (0.15; Figure 4), indicating minimal benefit from adding a third gene. Thus, two reference genes provided accurate and robust normalization in our study, consistent with geNorm’s context-dependent criterion. Nonetheless, in datasets with stronger between-group or batch effects, three or more reference genes may be warranted, in line with MIQE [49].

4.5. ItPOD Validation Confirms Reference Gene Impact

The functional impact of appropriate reference gene selection was further underscored by our validation using the peroxidase gene (ItPOD), a stress-responsive marker. Peroxidases (PODs) are key antioxidant enzymes that scavenge reactive oxygen species (ROS) and play crucial roles in plant defense against multiple abiotic and biotic challenges, including drought, waterlogging, and salt [59]. The expression of POD genes is often rapidly induced in response to oxidative stress, making them reliable indicators of plant stress responses [60,61,62]. Expression profiles of ItPOD normalized with stably expressed reference genes (e.g., PP2A, MD10B, eIF1A, and 60S) were consistent and biologically interpretable across stress treatments and tissues. In contrast, normalization with unstable reference genes (UBI, Actin7 and BS) resulted in either dramatic over- or underestimation of gene expression levels and, in some cases, reversed the expected tissue-specific expression patterns (Figure 6). This confirms previous cautions that inappropriate reference gene choice can lead to potentially misleading biological conclusions [40].

5. Conclusions

In summary, by integrating transcriptomic screening and comprehensive experimental validation, we identified context-appropriate reference genes for RT-qPCR normalization in I. tessellatus. Specifically, MD10B was recommended for drought, eIF1A for salt, PP2A for waterlogging, and 60S for different tissues, with the combined use of two stable reference genes advised for precise normalization under various conditions. Validation with the stress marker ItPOD confirmed that appropriate reference gene choice critically impacts expression estimates. Within the tested tissues and stress regimes, our context-matched reference genes provide a practical foundation for accurate expression analysis and subsequent functional studies in I. tessellatus. Looking ahead, adopting these normalizers across species and expanding validation to additional stresses and developmental stages will establish a cross-bamboo normalization atlas, improving data interoperability and powering meta-analyses to uncover conserved regulatory mechanisms.