Introduction

Reverse-transcription quantitative polymerase chain reaction (RT-qPCR) is a widely used technique for gene expression analysis that requires accurate quantification of mRNA transcripts1. Gene expression analysis has fundamental applications for investigating gene function and regulation, and elucidating broader gene networks involved in essential cellular and physiological processes. RT-qPCR is a more recent alternative to northern blot hybridization that provides greater specificity and sensitivity, enabling detection of a broader range of mRNA transcript levels2,3.

A requirement for accurate gene expression quantification is normalization to reference genes (also known as housekeeping genes or endogenous control genes) that are stably expressed across experimental conditions (e.g., control versus treatment) and parameters (e.g., age, sex, tissue)4,5,6. However, many reference genes initially adopted for RT-qPCR are vestiges of northern blot hybridization erroneously assumed stable due to their cellular housekeeping activities. The high sensitivity of RT-qPCR has since shown that many reference genes (e.g., those belonging to TUBULIN, UBIQUITIN, and GADPH gene families) can be more variable in expression than acceptable for accurate normalization under certain experimental contexts3,4,7,8. Therefore, it has become standard practice to thoroughly evaluate the expression stability of potential reference genes under given experimental conditions using a combination of statistical approaches to identify a minimum of two reference genes suitable for normalization. Statistical methods, such as ∆Ct, geNorm, normFinder, and BestKeeper, have been developed specifically to identify such reference genes with minimal variation across experimental treatments4,5,9.

Upland cotton (Gossypium hirsutum) is an economically important textile crop that has been cultivated for thousands of years for its fiber10,11. Due to its global importance and susceptibility to many insect pests, cotton was one of the first commercialized transgenic Bt crops12. However, field-evolved resistance to the insecticidal Cry proteins and newer VIP3a toxins expressed in Bt crops is an emergent global concern13,14,15,16,17. As such, alternative biotechnological approaches to insect pest management will require a better understanding of the genetics underpinning cotton-herbivore interactions. Virus-induced gene silencing (VIGS) is currently an undervalued reverse genetics tool for studying cotton-herbivore interactions often used in functional genomics studies to transiently silence genes in whole plants18,19,20. Gene silencing using VIGS is achieved by infiltrating plants with recombinant viral vectors (e.g., tobacco rattle virus) harboring coding sequence fragments of target gene(s)21. Expression of the viral vector(s) lead to systemic infection, eliciting the plants native antiviral pathway and downstream degradation of the target gene’s mRNA transcripts in the cytoplasm22. RT-qPCR is routinely used to validate successful knockdown of the target gene.

While VIGS has been extensively used in cotton (see20,23,24,25,26 for examples), a thorough evaluation of reference gene stability during systemic infection by VIGS viruses has not been previously reported, especially when insect herbivory stress is under investigation. Here, we tested the expression stability of six candidate reference genes: actin-7 (GhACT7), ubiquitin extension protein (GhUBQ7), polyubiquitin 14 (GhUBQ14), transmembrane 9 superfamily member 5-like (GhTMN5), trichome birefringence-like 6 (GhTBL6), and serine/threonine protein phosphatase 2a1 (GhPP2A1). Their expression stability was evaluated in wild-type and VIGS-infiltrated plants that were either uninfested or infested with cotton aphids (Aphis gossypii), measured at two time points, using a fully factorial experimental design. In a validation study, we demonstrate the importance of proper stable reference gene selection by comparing our worst performing and best performing reference genes for expression normalization of the ∆8–∆7 sterol isomerase gene GhHYDRA1 required for phytosterol biosynthesis27. This study highlights the importance of proper reference gene selection for cotton VIGS studies and facilitates further investigations evaluating genes that mediate cotton-herbivore interactions.

Materials and methods

Plant material and growth conditions

A high-yielding conventional cotton variety ‘Tamcot 73’ with seeds sourced from the Cotton Improvement Laboratory at Texas A&M University were used for all experiments in this study. Cotton seeds were germinated in 10 cm flowerpots containing Miracle-Gro Potting Mix and maintained at 23 °C with a 14:10 L:D photoperiod in Percival Incubators under humidity domes.

Cotton aphid colony maintenance

Laboratory cotton aphid colonies originally sourced from cotton fields near College Station, Texas, USA, were maintained on seedlings of the same variety used in the experiments, ‘Tamcot 73’. Colonies were kept in plexiglass cages under a 12:12 L:D photoperiod at 26-28ºC and ~ 50–60% relative humidity.

A. tumefaciens infiltrations for VIGS

Tobacco rattle virus (TRV) RNA2 (pYL156; Addgene #148969) and RNA1 (pYL192; Addgene #148968) vectors were transformed into the A. tumefaciens strain GV3101. Standard cotyledon Agro-infiltrations using TRV vectors were performed following the protocol reported by21. GV3101 glycerol stocks harboring respective TRV vectors were first plated to LB agar media containing the antibiotics kanamycin (50 µg/mL) and gentamicin (25 µg/mL) and grown at 28 °C for 2 days. Single colonies were inoculated into 5 mL liquid LB containing both antibiotics and shaken overnight at 50 rpm at 28ºC. The 5 mL liquid cultures were then diluted 1:10 in 50 mL liquid LB supplemented with antibiotics and a final concentration of 10 mM 2-(4 morpholino)-ethane sulfonic acid (MES) and 20 µM acetosyringone. The 50 mL liquid cultures were shaken overnight at 50 rpm at 28 °C. Bacterial pellets were harvested when the overnight cultures reached an OD600 ~ 0.8–1.2 and resuspended in induction buffer (10 mM MES, 10 mM MgCl2, and 200 µM acetosyringone) to an OD600 1.5. The resuspended bacteria in induction buffer were maintained at room temperature for 3 h. To prepare plants for cotyledon infiltrations, the bacteria were first mixed together at a 1:1 RNA1:RNA2 v/v ratio. A 25G needle was then used to puncture superficial wounds on the abaxial side of each cotyledon from 7–10-day-old seedlings and flooded with the TRV mixture using a needleless syringe until the cotyledon was fully saturated. Infiltrated seedlings were covered with humidity domes and incubated at room temperature overnight in a low-light setting. Plants were returned to the previously described conditions the following day for the remainder of the study. A total of 56 plants were VIGS-infiltrated with one RNA2 construct targeting a non-endogenous control gene (GFP; n = 28) and a second targeting a gene of interest (n = 28) described in a later section. Ten (10) additional plants were infiltrated with TRV targeting the essential chloroplast development gene, cloroplastos alterados 1 (GhCLA1), as a positive control with resulting albinism serving as a visual marker (Supplementary Fig. S1) for effective systemic gene silencing but were not used in the aphid herbivory experiment.

Aphid infestations for herbivory stress

Half (n = 12) of the wild-type ‘Tamcot 73’ plants (hereafter referred to as uninfiltrated plants) and half (n = 28) of all VIGS-infiltrated plants were infested 21 days post-infiltration with 5 adult reproducing cotton aphids and wrapped with microperforated plastic bags to keep the aphids on the plant. Uninfested plants (n = 12 wildtype; n = 28 VIGS-infiltrated) were similarly wrapped with microperforated plastic to control for any effect (e.g., evapotranspiration rate) the plastic covering may have on gene expression. In total, experimental sample sizes were n = 6 and n = 7 biological replicates for uninfiltrated and VIGS-infiltrated plants, respectively, for each infestation status (uninfested or infested) at each time point (14- and 21 days post-infestation).

Tissue sampling, RNA isolation, and quantification

Leaf tissues were collected at 14- and 21-days post-infestation to test gene expression stability over time and across multiple reproducing aphid generations. Tissues from both the 2nd and 4th true leaves of cotton plants were harvested to control for within-plant variation, which could be due to differences in tissue age, heterogenous establishment of TRV, and heterogenous distributions of aphids feeding on the plants. Total RNA was isolated from cotton leaf tissues using the Spectrum Total RNA Extraction Kit (Sigma Aldrich, STRN250) following the manufacturer’s instructions. RNA concentrations and purity were initially determined using spectrophotometry. Low purity RNA samples were further purified following a standard EtOH precipitation protocol (https://projects.iq.harvard.edu/files/hlalab/files/ethanol-precipitation-of-rna_hla.pdf). Final RNA concentrations were measured using an Invitrogen Qubit 4 fluorometer (Thermo Fisher Scientific).

Initial choice and primer design for candidate reference genes

Several different reference genes have been reported for expression normalization in cotton, including in VIGS studies and those investigating candidate reference gene stability in other experimental contexts (see3,25,28,29,30 for examples). Comparative transcriptomics has also been used to identify novel candidate reference genes not commonly used in cotton, some of which are specifically recommended for transgenic cotton31. It is important to note that in all published cases we found, only one reference gene was used for expression normalization whereas using at least two is currently considered best practice32. Based on these studies, we screened a modest panel of six candidate reference genes GhUBQ7 (DQ116441.1), GhUBQ14 (XM_016855771.2), GhTMN5 (XM_016895405.2), GhTBL6 (XM_016880182.2), and GhPP2A1 (XM_041080556.1) to identify the most stably expressed reference genes under VIGS and herbivory stress over time. Multiple qPCR primer sets for each candidate gene were designed using Integrated DNA Technologies online PrimerQuest tool. Primer sets were first tested for amplification specificity and minimal primer dimerization using gel electrophoresis (Supplementary Fig. S2). Full details on gene accession numbers, primer sequences, and amplicon sequences can be found in Supplementary Table S1.

cDNA synthesis and RT-qPCR optimization

A stepwise approach following guidelines provided by the Berrick laboratory (https://barricklab.org/twiki/bin/view/Lab) at the University of Texas Austin was taken to optimize the qPCR reaction conditions for each qPCR primer set. Prior to cDNA synthesis, qPCR reaction efficiencies (Supplementary Table S1) were first measured across sixfold dilution series for each reference gene primer set using the respective conventional PCR amplicons as the template in the reaction. cDNA synthesis was performed using the iScript™ gDNA Clear cDNA Synthesis Kit (Bio-Rad, 1725035) following the manufacturer’s instructions with 500 ng input RNA in 20µL reactions. Optimal cDNA input for qPCR reactions was determined using a composite cDNA library from all sample cDNA libraries combined in equal 1 µL volumes. The composite library was diluted across several sixfold dilution series and tested for cDNA inputs resulting in cycle threshold (Ct) values ranging from ~ 18 to 32 for all RT-qPCR reactions. Final RT-qPCR reactions were performed using 1:10 diluted cDNA (10 ng total input cDNA). All qPCR reactions in this study were conducted in 10 µL volumes using the SsoAdvanced™ Universal SYBR® Green Supermix (Bio-Rad, 172575) following the manufacturer’s instructions. Reactions were run in technical triplicates on a BioRad CFX384 Touch Real-Time PCR Detection System with the following cycle program: initial denaturation at 95 °C for 3 m followed by 40 cycles of denaturation at 95 °C (15 s) and anneal/extend at 60 °C (30 s). Biological replicates from all experimental conditions were distributed across the 384-well plates to obviate potential plate effects on qPCR cycles. Melt curves were performed by heating qPCR reactions to 95 °C (0.5 °C/cycle) to confirm amplification specificity (Supplementary Fig. S3).

RT-qPCR quality control and candidate reference gene stability analyses

RT-qPCR relative fluorescence units and melt curves were initially scrutinized using the Bio-Rad CFX Maestro Software 1.1 (v.4.1.2433.1219) to ensure amplicon specificity. Technical triplicates were filtered for standard deviations (SD) < 0.5 of Ct values whereby some reactions were reduced to technical duplicates when a given technical replicate deviated from the Ct values of the other two technical replicates having a SD < 0.5.

All data analyses and visualizations were performed in R (v.4.4)33. Expression stability analyses included ∆Ct calculations and geNorm5, NormFinder9, and BestKeeper4, algorithms implemented using R packages RQdeltaCT (1.3.2)34 and ctrlGene (v.1.0.1)35. ∆Ct values were determined by normalizing the Ct values for each given reference gene using all other candidate reference genes as reference for normalization. Linear mixed effect models (R package lme4 v.1.1–35.4) were conducted to test for the effects, if present, of experimental conditions on ∆Ct values36. Infiltration status, infestation status, and days post-infestation were used as fixed effects. The individual plant (plant ID) was used as a random effect as up to two leaves were collected from each plant. The inclusion of plant ID as a random effect for all models was assessed using Likelihood Ratio Tests on models for Maximum Likelihood. Benjamini-Hochberg (BH) false discovery rate was performed for each fixed effect in the mixed effect models for multiple testing corrections. The R package RankAggreg (v0.6.6) was used to perform a weighted rank aggregation analysis using its brute force algorithm (k = 6; distance = “Spearman”) of ranked results from ∆Ct standard deviations, geNorm and BestKeeper37.

Validating selected reference genes for relative expression in VIGS herbivory studies

To validate results from the reference gene expression stability analyses, we performed an independent relative expression analysis of GhHYDRA1 using the top two performing and single worse performing candidate reference genes identified in the present study. Relative expression was conducted using E-∆∆Ct where E was the reaction efficiency factor for GhHYDRA1 relative to the geometric mean of reaction efficiencies of the reference genes, which is recommended to account for measured differences in reaction efficiencies > 5%38. For relative expression analyses, we fit linear mixed effects models (~ infestation.status*time + infiltration.status + (1|plant ID)) to determine the significance and magnitude of the effect aphid infestations and VIGS infiltrations had on relative GhHYDRA1 expression outcomes over time. A mixed analysis of variance (ANOVA; Satterthwaite’s method) was conducted to determine the statistical significance of experimental conditions on relative GhHYDRA1 expression outcomes. The estimated marginal means were used for a Bonferroni-adjusted post-hoc test on the mixed effects model using the R package emmeans (v.1.10.2). Distributions of model residuals were inspected visually and tested for normality using Shapiro–Wilk normality tests (base R shapiro.test). Levene’s tests were performed to check for equal variance (homogeneity) of residuals across the levels of the fixed effects in our linear mixed-effects models.

Results

Ct values, standard deviations, and ∆Ct analyses of candidate reference genes

RT-qPCRs were performed for each of the six candidate reference genes GhACT7, GhUBQ7, GhUBQ14, GhPP2A1, GhTMN5, and GhTBL6, across the experimental condition variables of (1) infiltration status (uninfiltrated or VIGS-infiltrated), (2) infestation status (uninfested or infested), and (3) infestation duration (14- or 21 days post-infestation) using a fully factorial design. Cycle threshold (Ct) values and standard deviations (SDs) evaluated for the entire experiment revealed a wide range of expression levels and expression variability (Fig. 1a). Complete summary statistics for Ct values can be found in Supplemental Table S2. GhACT7 had the lowest Ct mean (21.7) indicating it was the most highly expressed candidate reference gene in this study. Intermediate expression levels were found for GhUBQ7 (24.1), GhUBQ14 (26.4), and GhPP2A1 (27.8), while GhTBL6 (30.3) and GhTMN5 (30.6) were the poorest expressing genes. SDs widely ranged (1.54–3.28) and ranked highest to lowest GhUBQ7 > GhUBQ14 > GhPP2A1 > GhACT7 > GhTBL6 > GhTMN5.

Fig. 1
figure 1

Threshold cycle (Ct) and ∆Ct values for candidate reference genes of interest in this study. (a) Boxplots represent the median (black horizontal line), 25th and 75th percentile (boxes), and minimum and maximum (vertical lines) of raw Ct values for each refence gene. Comparison of ∆Ct values between all (b) uninfiltrated and VIGS-infiltrated plants, (c) uninfested and aphid-infested plants and (d) plants 14- and 21 days post-infestation. Horizontal lines with asterisks indicate significant differences in ∆Ct values from linear mixed effects models comparing experimental conditions for each reference gene. Benjamini–Hochberg adjusted p values are denoted by: ** < 0.01; *** < 0.001.

Full size image

Wide ranging SDs indicated considerable expression variation due to the experimental conditions examined. Therefore, Ct values for each reference gene were normalized to ∆Ct values to rank genes by ∆Ct SDs (Supplementary Table S3) and compare how experimental conditions influenced transcript abundances. Linear mixed effects modeling revealed that there were no significant differences in ∆Ct values between uninfiltrated plants and VIGS-infiltrated plants for any of the candidate reference genes (Fig. 1b). Aphid infestation status did result in a significant difference in ∆Ct values for only GhTMN5 (Fig. 1c). Infestation duration (14- or 21 days post-infestation) had the most significant effect on ∆Ct values with significant differences found for GhACT7, GhPP2A1, and GhTBL6 (Fig. 1d). The significant effect of aphid infestations and infestation duration on ∆Ct values required further investigation to elucidate sources of variation and identify stably expressing reference genes across these conditions.

Expression stability algorithms of candidate reference genes across experimental conditions

geNorm, BestKeeper, and NormFinder are independently developed algorithms frequently used to identify stably expressing reference genes across experimental conditions or manipulations4,5,9. Each algorithm computes different measures of stability and incorporates different sources of variation when evaluating expression stability. In this study, we were interested in using these algorithms to evaluate expression variation introduced by VIGS infiltrations and aphid-mediated herbivory stress over time. Since the algorithms vary in their calculations and reported metrics, we separately provide summaries of each analysis below. Complete ranking statistics for all algorithms are provided in Supplementary Table S3.

geNorm analysis: The geNorm algorithm calculates gene stability scores (M) by averaging pairwise variation (defined by the log-transformed expression ratio) of a candidate reference gene individually against all other reference genes. This is performed in elimination series where the gene with the highest M score is removed at each iteration until two reference genes remain.

We used geNorm to compare gene expression stability between our experimental conditions infiltration status, infestation status, and infestation duration (Fig. 2). The geNorm analysis comparing uninfiltrated plants to VIGS-infiltrated plants did not reveal notable differences in stability score values for the ranked candidate reference genes within these two conditions (Fig. 2a). For both conditions, candidate reference genes were ranked least stable to most stable GhUBQ7 > GhUBQ14 > GhTBL6 > GhTMN5, and finally GhACT7 and GhPP2A1 having the lowest pairwise variation (Fig. 2b,c). Infestation status, comparing uninfested plants with cotton aphid-infested plants, did reveal an interesting pattern of expression stability whereby infested plants had overall lower expression stability scores (i.e., greater stability) compared to uninfested plants (Fig. 2d). Nevertheless, they resulted in the same ranking of candidate reference genes as infiltration status with the least stable to most stable GhUBQ7 > GhUBQ14 > GhTBL6 > GhTMN5, and finally GhACT7 and GhPP2A1 (Fig. 2e,f). Infestation duration showed differences in stability scoring with 21 days post-infestation having higher M scores (i.e., lower stability) compared to plants at 14 days post-infestation (Fig. 2g). Both 14- and 21 days post-infestation showed the same gene ranking scores with the least stable to most stable GhUBQ7 > GhUBQ14 > GhTBL6 > GhTMN5 and finally GhACT7 and GhPP2A1 (Fig. 2h,i).

Fig. 2
figure 2

Expression stability analysis of candidate reference genes using geNorm’s stability score M. (a, d, g) Comparing average pairwise M score rankings between (a) uninfiltrated and VIGS-infiltrated plants, (b) uninfested and infested plants, and (g) plants at 14- and 21 days post-infestation, as a gene is eliminated sequentially in the ranking analysis. Candidate reference genes were ranked based on (b, c) infiltration status, (e, f) infestation status, and (h, i) infestation duration. Lower M scores indicate greater stability (i.e., less variability); genes are more stable left to right.

Full size image

BestKeeper analysis: Like geNorm, BestKeeper can be used to rank the expression stability of candidate reference genes between experimental conditions. Gene ranking is based on the coefficient of variance (CV) and standard deviation (SD) that, combined, provide estimates of expression stability; candidate reference genes with lower CV and SD are more stably expressed. Unlike geNorm, BestKeeper was used to narrow the ranking to a single stably expressing reference gene.

We used BestKeeper to compare gene expression stability between the experimental conditions (Fig. 3). This resulted in notably different gene rankings compared to geNorm. Gene rankings for all experimental conditions were nearly uniform with uninfiltrated (Fig. 3a), VIGS-infiltrated (Fig. 3b), infested (Fig. 3d), and plants at both time points (Fig. 3c,d) were identical and ordered from least to most stable GhUBQ7 > GhUBQ14 > GhACT7 > GhPP2A1 > GhTMN5 > GhTBL6. Uninfested plants (Fig. 3c) had a slightly different ranking GhUBQ7 > GhACT7 > GhUBQ14 > GhPP2A1 > GhTMN5 > GhTBL6.

Fig. 3
figure 3

Expression stability analysis of candidate reference genes using BestKeeper’s coefficient of variance (CV value; line plot) and standard deviation (SD value; barplot) of crossing points (i.e., Ct) values. Ranking of candidate reference genes based on (a, b) infiltration status, (c, d) aphid infestation status and (e, f) infestation duration. Lower CV and SD values indicate lower variation in expression.

Full size image

NormFinder analysis: NormFinder is distinct from geNorm and BestKeeper in that it is a model-based approach rather than pairwise comparison approach. The model incorporates intergroup variation in addition to intragroup variation to measure and rank expression stability of candidate reference genes. Intragroup and intergroup stability scores for each candidate reference gene are calculated independently and then combined additively for a final stability score; lower stability scores indicate greater expression stability.

We used NormFinder to rank candidate reference genes based on intragroup and intergroup variation of experimental conditions infiltration status, infestation status, and infestation duration (Fig. 4). For infiltration status, we compared uninfiltrated plants to VIGS-infiltrated plants to identify stably expressed reference genes within and between these groups. NormFinder identified reference genes to be least to most stably expressed GhUBQ7 > GhUBQ14 > GhTBL6 > GhTMN5 > GhACT7 > GhPP2A1 (Fig. 4a). NormFinder analysis of infestation status, comparing uninfested versus infested plants, revealed a nearly identical gene ranking to infiltration status with the exception of the UBQ genes with GhUBQ14 being less stable than GhUBQ7 (Fig. 4b). An analysis of infestation duration that compared 14 days post-infestation to 21 days post-infestation resulted in a ranking dissimilar to infiltration and infestation status with least to most stable genes GhTBL6 > GhUBQ7 > GhACT7 > GhUBQ14 > GhPP2A1 > GhTMN5 (Fig. 4c).

Fig. 4
figure 4

NormFinder expression stability scores of candidate reference genes evaluating intra-group and inter-group variation. Stability scores based on (a) infiltration status (uninfiltrated plants versus VIGS-infiltrated plants), (b) cotton aphid infestation status (uninfested versus infested plants), and (c) infestation duration (14 versus 21 days post-infestation). Lower scores indicate greater combined intragroup and intergroup expression stability.

Full size image

Determining optimal number of reference genes for normalization

As mentioned above, expression normalization using more than one reference gene is considered more accurate as it better corrects for the many sources of variation in a qPCR experiment, including biological variation and technical errors. However, there are practical limits to the number of reference genes that can be evaluated. Therefore, we implemented geNorm’s pairwise variation (Vn/Vn+1) calculation to determine the minimum number of reference genes sufficient to accurately normalize expression in this study. Briefly, geNorm calculates a normalization factor (NF) for each gene and evaluates pairwise variation (Vn/Vn+1) of NFs in a sequence adding a new gene each iteration. Pairwise variation values < 0.15 indicate that the addition of new reference gene in that sequence does not significantly contribute to expression variation.

Pairwise variation (Vn/Vn+1) was assessed for all samples and experimental groups in this study (Fig. 5). For all samples and conditions, pairwise variation values beginning with two genes (V2/V3) had values far below the recommended 0.15 threshold. Pairwise variation for all samples, samples within groups for infiltration status, and samples within groups for infestation duration, showed similar patterns whereby increasing the number of reference genes from two to three slightly reduced the value whereas increasing from three and up to six reference genes increased the value (Fig. 5). Uninfested samples showed a sequential decrease in pairwise variation values up to five reference genes but increased at six reference genes. Infested plants showed a sequential increase in values up to six reference genes. Nevertheless, the results from the pairwise variations analysis indicate that two reference genes are sufficient for accurate normalization under all experimental conditions given that values were all well below 0.15.

Fig. 5
figure 5

geNorm’s pairwise variation (Vn/n+1) to determine the optimal number of reference genes for expression normalization under different experimental conditions in this study. Pairwise variation is measured between normalization factors (NFn) and NFn+1 sequentially with the addition of a new reference gene. V < 0.15 is the recommended cutoff and indicates the inclusion of an additional reference gene does not contribute to more accurate normalization.

Full size image

Rank aggregation analysis of reference gene stability results

The ranking methods implemented in the analyses so far resulted in different candidate reference gene rankings for expression stability across the experimental conditions. Therefore, we performed a ranked aggregate analysis of results from the ∆Ct method, geNorm and BestKeeper, for each of the experimental conditions (Fig. 6). For all ranked aggregate analyses, both UBQ genes consistently ranked last and in order GhUBQ7 and GhUBQ14 (Fig. 6). Uninfiltrated (Fig. 6b) and uninfested samples (Fig. 6d) resulted in the same ranking GhUBQ7 > GhUBQ14 > GhACT7 > GhTBL6 > GhPP2A1 > GhTMN5. All samples in this study (Fig. 6a) and VIGS-infiltrated samples (Fig. 6c) had the same ranking GhUBQ7 > GhUBQ14 > GhTBL6 > GhACT7 > GhTMN5 > GhPP2A1. Infested samples had a unique ranking GhUBQ7 > GhUBQ14 > GhTBL6 > GhACT7 > GhPP2A1 > GhTMN5 (Fig. 6e). Lastly, both timepoints for infestation duration had identical rankings GhUBQ7 > GhUBQ14 > GhTBL6 > GhTMN5 > GhACT7 > GhPP2A1 (Fig. 6f,g).

Fig. 6
figure 6

Weighted rank aggregation of all candidate reference genes implemented by RankAggreg’s brute force (BF) algorithm for different experimental conditions in this study. The analysis combined ranked results from ∆Ct, BestKeeper, and geNorm (data shown in grey). Mean rankings for each gene are shown in black. The brute force algorithm weighted ranking is shown in red.

Full size image

Validating selected reference genes for relative expression in VIGS herbivory studies

Results from each independent stability analysis and the rank aggregate analysis suggested that choosing the proper reference genes for accurate normalization is nontrivial and depends on the experimental conditions being evaluated. Because we were interested in testing what effect cotton aphid feeding had on phytosterol biosynthesis with further applications for VIGS studies, we prioritized choosing reference genes that would allow us to accurately compare GhHYDRA1 transcription levels between uninfested and aphid-infested plants under either uninfiltrated (wild-type) or VIGS-infiltrated conditions. As such, we conducted relative expression analysis of GhHYDRA1 using our best performing reference genes GhACT7 and GhPP2A1 (Fig. 7c,d) and contrasted the results with relative GhHYDRA1 expression normalized using our worst performing reference gene GhUBQ7 (Fig. 7a,b) for these conditions. It is important to note that although GhACT7 and GhPP2A1 were consistently ranked highly stable within each timepoint (Fig. 6f,g), normalization using the E-∆∆Ct method was performed separately for each time point as the ∆Ct (Fig. 1d), NormFinder (Fig. 4c) and ranked aggregation (Fig. 6a–e) analyses indicated reduced stability when these genes were evaluated over both timepoints. This resulted in distinct reference ∆Ct values for ∆∆Ct calculations from normalization using GhACT7/GhPP2A1 (14 DPI: 3.5; 21 DPI: 3.25) and normalization using GhUBQ7 (14 DPI: 3.57; 21 DPI: 4.0). Linear mixed-effect models were then fit to the expression data from either normalization approaches to test for what effect aphid infestations, VIGS infiltrations, and time, had on relative GhHYDRA1 expression and to compare effects between the normalization approaches.

Fig. 7
figure 7

Relative expression (E-∆∆ct) analysis of GhHYDRA1 comparing the worst performing reference gene GhUBQ7 (a, b) and best performing paired reference genes GhACT7 and GhPP2A1 (c, d). Normalized expression was compared between uninfested and aphid-infested uninfiltrated and VIGS-infiltrated plants. Reported p values are from pairwise post-hoc tests using estimated marginal means from fixed effects models. DPI; days post-infestation.

Full size image

Normalization using GhACT7/GhPP2A1. Results from a mixed ANOVA revealed a significant effect of infestation status (F(1, 29.882) = 19.7072, p = 0.00011) on relative GhHYDRA1 expression with infested plants exhibiting significantly higher expression. Importantly, infiltration status did not have a significant effect (F(1,29.910) = 0.9562, p = 0.34), nor did days post-infestation (F(1,29.10) = 0.0270, p = 0.87). Lastly, there was no significant interaction between infestation status and time (F(1,29.986) = 2.1973, p = 0.15). To further explore the significant effect of aphid infestation on relative GhHYDRA1 expression, we performed post-hoc pairwise comparisons using estimated marginal means with Bonferroni correction. Aphid infestation had a significant effect (p = 0.0001) on relative GhHYDRA1 expression at 14 days post-infestation (Fig. 7c), but not at 21 days post-infestation (p = 0.06) (Fig. 7d). At 14 days, infestation had a large and significant effect on relative GhHYDRA1 expression (Cohen’s d = 0.81) with infested plants having a ~ 2.4-fold increase in expression. At 21 days, the effect was weak (Cohen’s d = 0.34) and nonsignificant.

Normalization using GhUBQ7: We applied the same linear mixed effects model structure to GhHYDRA1 expression data normalized using GhUBQ7. Unlike the model using GhACT7/GhPP2A1-normalized GhHYDRA1 expression data, model residuals were non-normal (W = 0.809; p < 0.001). A mixed ANOVA still revealed a significant, albeit weaker, effect of infestation status (F(1,30.37) = 4.7597, p = 0.04) on GhHYDRA1 expression. Infiltration status (F(1,30.41) = 0.7931, p = 0.3801) and days post-infestation (F(1,30.370) = 0.1068, p = 0.75) were non-significant. Lastly, there was no significant interaction between infestation status and days post-infestation (F(1,30.862) = 0.1279, p = 0.7230). Pairwise comparisons using estimated marginal means with Bonferroni correction revealed no significant difference (p = 0.06) in relative GhHYDRA1 expression at 14 days post-infestation between infested and uninfested plants despite an over-estimated ~ 4.0-fold increase in GhHYDRA1 expression in aphid-infested plants compared to uninfested plants (Fig. 7a). Lastly, no significant difference (p = 0.26) was found at 21 days post-infestation (Fig. 7b).

The difference aphid infestations had on relative GhHYDRA1 expression at 14 days post-infestation, depending on normalization methods, highlighted the importance of choosing the correct reference genes for accurate interpretation of hypothesis testing outcomes (i.e., a significant versus non-significant result). To assess the robustness and stability of our selected reference genes GhACT7/GhPP2A1, we applied a bootstrapping approach. We resampled (n = 1000) our relative GhHYDRA1 expression values at 14 days post-infestation to compare standard deviation (SD) distributions between GhACT7/GhPP2A1-normalized expression and GhUBQ7-normalized expression (Fig. 8). Mean SD from relative GhHYDRA1 expression normalized using GhACT7/GhPP2A1 was significantly different from mean SD normalized using GhUBQ7 (t(1013.2) = − 99.25, p < 0.0001). GhACT7/GhPP2A1-normalized expression had an acceptable mean SD of 0.62 compared to a mean SD of 4.95 from GhUBQ7-normalized expression.

Fig. 8
figure 8

Bootstrap approach comparing standard deviation distributions of resampled (n = 1000) relative GhHYDRA1 expression values normalized using our best paired candidate reference genes GhACT7/GhPP2A1 (pink) and worst performing reference gene GhUBQ7 (blue).

Full size image

Discussion

Virus-induced gene silencing (VIGS) is an essential component of the plant molecular biology toolkit, offering a simple and cost-effective means to transiently silence genes. It is widely used for investigating gene function in crops and has been particularly valuable for Upland cotton as its complex and redundant allotetraploid genome long hindered assembly of high-quality genomes indispensable to functional genomics studies39. VIGS remains a key reverse genetic tool in cotton to identify genes important to agronomic traits like fiber development and elucidating genetics of cotton responses to environmental stresses40,41. VIGS studies investigating insect herbivory responses are particularly underrepresented in cotton with most studies primarily focused on pathogens as biotic stressors (see42,43,44,45 for examples) despite insect pests imposing significant stress on cotton production globally. Insect pests feed on crops like cotton for nutrients, including phytosterols that are essential dietary precursors to cholesterol46. HYDRA1 (HYD1) is a conserved sterol synthesis gene that catalyzes ∆8-∆7 isomerization during phytosterol biosynthesis47. Silencing HYD1 and mutagenesis of several other phytosterol biosynthesis genes in Arabidopsis has been shown to modify phytosterol composition, negatively affecting insect herbivore development and reproduction, including that of aphids48,49. VIGS can be used to investigate how genes involved in phytosterol biosynthesis and other metabolites (e.g., gossypol) mediate cotton-herbivore interactions. Here, we focused on expression of the cotton HYD1 ortholog GhHYDRA1 in response to aphids as current transgenic technologies (i.e., Bt and VIP3a) do not target these pests.

To accurately normalize GhHYDRA1 expression, we conducted a fully factorial experiment to evaluate the expression stability of six candidate reference genes and identified those with stable expression when cotton is exposed to simultaneous infection of recombinant tobacco rattle virus (TRV) and persistent stress from cotton aphid infestations over time. Candidate reference genes were selected from the literature with a focus on including those frequently used in cotton VIGS studies investigating biotic stress responses. We performed multiple stability analyses (∆Ct, geNorm, BestKeeper, and normFinder) that use different statistical approaches to measure reference gene stability and variation for defined experimental conditions. A ranked aggregation analysis combined results from the independent stability analyses to comprehensively compare reference gene stability across the experimental conditions under investigation in this study. Priority was given to the geNorm and normFinder analyses because when combined they provided the most valuable measures to identify multiple stable and compatible reference genes for experimental conditions while also accounting for intra- versus intergroup expression variation.

We overwhelmingly found that the commonly reported reference genes GhUBQ7 and GhUBQ14 were unsuitable for normalization due to highly variable expression across the experimental conditions evaluated50,51,52,53. It is important to note that in the current study the primer pairs used for both target genes GhUBQ7 and GhUBQ14 were the best performing pairs with respect to reaction efficiencies (102.7% and 99.9%, respectively). Therefore, the variation in their expression is likely due to true biological variation rather than technical or experimental error. Although studies investigating reference gene stability in other systems have identified UBIQUTINs as stable across other experimental conditions, their use as reference genes in future cotton VIGS studies where RT-qPCR is used to measure gene expression should be reconsidered54,55,56.

GhTMN5 and GhTBL6 expression stability was also evaluated as previous transcriptome studies suggested they have potential as stable reference genes to compare expression between transgenic and non-transgenic cotton31. GhTBL6 stability ranking varied across the algorithms used in this study. GhTBL6 performed well in the BestKeeper analyses when considering infiltration and infestation status but performed poorly for these conditions in the geNorm and normFinder analyses. These inconsistencies suggested that GhTBL6 is likely not a suitable reference gene for herbivory studies using VIGS in cotton. GhTMN5 performed generally well for the algorithm analyses but results from the ∆Ct analysis suggested that aphid herbivory may influence its expression. Therefore, GhTMN5 may not be suitable for normalization when comparing expression between uninfested and cotton aphid-infested plants. While our results indicate that GhTBL6 and GhTMN5 are mostly unsuitable reference genes for the experimental conditions in this study, their expression stability should continue to be investigated as they have been used for expression normalization in other experimental contexts in cotton57.

ACT7 has been identified as a stable reference gene especially under stress in other plant systems, including Gossypium arboreum, that performed well in all the algorithm analyses58,59,60,61. geNorm and BestKeeper ranked GhACT7 as highly stable at both time points, suggesting that its expression was not influenced by infiltration or infestation status. These results were supported by the normFinder analysis, which evaluated between-group variation in addition to within-group variation for the stability metric. GhACT7 ranked second in the normFinder analyses for both infiltration and infestation status. The high performance of ACT7 in these analyses was unsurprising given that it is used as a reference gene for expression normalization in many plant systems62,63,64. PP2A1, used in other plant systems, performed like GhACT7 in the algorithm analyses, making it a compatible second reference gene for normalization65,66,67. This was particularly supported by the geNorm analysis that paired GhACT7 with GhPP2A1 due to their lowest pairwise variation as the most stable reference genes for all experimental conditions in this study. Of note, it is unlikely that these reference genes are coregulated given their known cellular activity. Therefore, both reference genes account for independent sources of variation in gene expression and regulation. Additionally, results from the geNorm pairwise variation analysis indicated that these two reference genes are sufficient for normalization.

To validate the selected reference genes, we evaluated expression analysis outcomes using the best (GhACT7/GhPP2A1) and worst (GhUBQ7) performing reference genes for GhHYDRA1 normalization. Mixed effects modeling of both normalization approaches indicated that TRV infiltrations using a GFP-targeting control vector did not have a significant effect on GhHYDRA1 expression. While post-transcriptional gene silencing is well-known as the sequence-specific knockdown mechanism for VIGS, these results supported that systemic infection by recombinant TRV harboring a GFP trigger fragment did not have indirect effects influencing regulation of GhHYDRA1. Contrary to TRV infiltrations, aphid infestations did have a significant effect on relative expression of GhHYDRA1, but the detection of this effect depended on the selected reference genes used for normalization. Normalization with the commonly utilized GhUBQ7 resulted in high variance in GhHYDRA1 expression, particularly in aphid-infested plants. This reduced statistical power and prevented detection of a significant effect of aphid infestations on GhHYDRA1 expression. In contrast, normalization using GhACT7/GhPP2A1 produced much lower variance and resulted in detection of a significant effect of aphid infestations on GhHYDRA1 expression at 14 days post-infestation. Of note, we did not observe a significant effect on GhHYDRA1 expression at 21 days post-infestation, which could be hypothesized as a response to prevent overproduction of phytosterols on aphid-infested plants. Like humans, plants require tight regulation of sterol biosynthesis where disruptions in sterol homeostasis can be lethal68. Elucidating phytosterol-mediated herbivory responses will require further investigation and was not within the scope of this study.

We took a novel approach to empirically validate the stability of our selected reference genes by resampling GhHYDRA1 relative expression normalized using GhACT7/GhPP2A1 (best performing) and GhUBQ7 (worst performing) to compare standard deviation (SD) distributions of resampled expression values. The significant difference in mean SDs between expression values indicated that normalization using the selected reference genes GhACT7/GhPP2A1 had empirically meaningful outcomes with GhACT7/GhPP2A1-normalized expression values having a significantly lower mean SD. This provided further confidence that normalization using GhACT7/GhPP2A1 resulted in reliable expression analysis of GhHYDRA1 in our validation study showing aphid herbivory having a significant effect on its expression.

From our analyses, we concluded that GhACT7 and GhPP2A1 are reliable reference genes for expression normalization where VIGS is used to investigate expression of genes important to cotton-herbivore interactions. Our results emphasize the need for reference genes and associated qPCR primers sourced from the literature to be tested in pilot studies to confirm their expression stability. This extends to future VIGS studies testing other viral vectors or investigating gene expression responses to other herbivorous pests (e.g., other hemipterans, lepidopterans, parasitic nematodes, etc.). Lastly, a limitation of using GhACT7 and GhPP2A1 is that their expression appeared to become less stable over time. Therefore, GhACT7/GhPP2A1 are likely not reliable references for timeseries studies statistically comparing changes in expression across multiple timepoints. Future RNA sequencing studies potentially incorporating additional multifactorial herbivory stresses can be used to validate these findings in addition to reference gene discovery in cotton.

In summary, we conducted a novel evaluation of candidate reference genes suitable for expression normalization in cotton-herbivore studies using VIGS. We found that GhACT7 and GhPP2A1 were the most stable reference genes for RT-qPCR under these experimental conditions whereas GhUBQ7 and GhUBQ14 were the least stable. Our results highlight the importance of empirically testing assumptions about reference gene stability, which is required for accurate gene expression normalization and comparisons between experimental conditions. Lastly, our findings facilitate future studies using VIGS to investigate genes mediating cotton-herbivore interactions. A better understanding of the genetics underpinning these interactions has the potential to lead to novel pest management biotechnologies.