A comparison of normalization methods for the expression of genes associated with oxidative stress in the liver of sheep

Abstract

Background

Normalization of data obtained from RT-qPCR studies is important for accurate interpretation of the results. The two most common methods of normalization that are used are the reference gene method or an algorithm-only approach, such as NORMA-Gene. Here, we assessed the impact of normalization using reference genes or the NORMA-Gene method on the expression results of five target genes that are related to oxidative stress (CAT, GPX1, GPX3, PRDX1, and SOD1) in the liver of sheep that had been exposed to three dietary treatments.

Results

The reference genes selected as the most stable across samples for normalization were HPRT1, HSP90AA1, and B2M. Interpretation of the effect of the treatment on the expression of GPX3 differed significantly between the methods of normalization. NORMA-Gene was better at reducing the variance in the expression of the target genes than was any of the other normalization methods.

Conclusions

We demonstrated that NORMA-Gene can provide a more reliable normalization method that requires less resources than the use of reference genes for studies on gene expression in livestock.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12863-025-01345-y.

Background

The reverse transcription-quantitative real-time polymerase chain reaction (RT-qPCR) is a fundamental technique that can be used to quantify various types of ribonucleic acid (RNA). The RT-qPCR technique involves the extraction of RNA from a biological sample that is reverse transcribed to generate complementary deoxyribonucleic acid (cDNA). The cDNA is then amplified to newly synthesized DNA during qPCR [1]. The qPCR produces an amplification curve with an established threshold and generates a cycle threshold value for each sample when its amplification curve crosses that threshold. The cycle threshold value that is generated for each of the samples is used for quantification by absolute or relative methods [1]. Systematic variations that occur during RT-qPCR from sample quality, reagents, assays, pipetting, or instruments, can introduce variation in the expression of a gene to the extent that variations caused by an experimental treatment can be interpreted incorrectly [2]. To account for the variability that is due to factors other than the actual expression of a given gene, normalization is used. If normalization works as intended, any effect of an experimental treatment can be estimated accurately [3–6].

The most common method of normalization is to use the geometric mean of one or more reference genes to normalize the expression of a target gene [7]. An ideal reference gene will have the same expression across all samples and be unaffected by the experimental treatment [2, 8, 9]. The most suitable reference genes for normalization will depend on the species, the biological sample type, and the experimental conditions [4]. The “Minimum Information for Publication of Quantitative Real-Time PCR Experiments Guidelines” recommend that at least two reference genes be used and validated across experimental samples [3]. However, many published studies use only a single reference gene, or reference genes that have not been validated on the experimental samples [10]. When unsuitable reference genes are used, the expression of a target gene can be overestimated or underestimated, and the effect of an experimental treatment can be interpreted incorrectly [11].

Several algorithms have been developed to assist with the selection and validation of references genes for normalization that are expressed at a constant rate and unaffected by the experimental treatment. The algorithms include the comparative delta cycle threshold method [12], geNorm [7], NormFinder [13], and BestKeeper [14]. Each algorithm estimates the variance of the cycle threshold for each reference gene across the experimental samples [10]. A reference gene with low variance of expression across samples is considered to be more stable than one with higher variance [10]. The online tool RefFinder [15] and the BruteAgregg R-studio software package [16] can also be used to aggregate the results from the other algorithms and to provide an overall ranking of the reference genes from the most to the least stable expression.

NORMA-Gene is an alternative method of normalization that does not require the use of reference genes [17]. The NORMA-Gene algorithm requires the expression data of at least five genes and uses a least squares regression to calculate a normalization factor that is used to reduce the variation in the expression of the genes across the experimental samples [17]. It has been claimed that the NORMA-Gene method reduces variance in expression data better than does normalization using reference genes [17]. The NORMA-Gene method has the additional benefit that it requires fewer resources and less time than does the use of reference genes because there is no need to run the additional RT-qPCRs to validate the chosen reference genes for the experimental samples. NORMA-Gene has been used for normalization in studies on several insect species [18, 19], fish [20–22], hamsters [23], and humans [24, 25]. To our knowledge, the NORMA-Gene method has not been used in ruminant species such as sheep.

In the present study, we compared the effectiveness of using reference genes or NORMA-Gene to normalize the expression of five target genes in the liver of sheep. The samples were collected from sheep that had been fed three dietary treatments for six weeks, sheep were either fed at maintenance, fed above maintenance, or fed below maintenance. The metabolic changes that are induced by under and over feeding have been shown to affect the level of oxidative stress in sheep [26–28]. Therefore, we tested five target genes (CAT, GPX1, GPX3, PRDX1, and SOD1) whose products have antioxidant mechanisms that are associated with regulating levels of oxidative stress. We also tested nine reference genes that have previously been used in sheep (ACTB, B2M, GAPDH, HMBS, HPRT1, PPIA, HSP90AA1, SDHA, and YWAZ) to select the most suitable refence genes to use for normalization [29–32]. We tested the hypothesis that the NORMA-Gene method would be a better method of normalization and offer practical advantages compared to the use of reference genes when assessing the expression of genes related to oxidative stress in sheep.

Materials and methods

Animal ethics approval

Liver samples were collected from sheep as approved by The University of Western Australia Animal Ethics Committee (2020/ET000194). The experiment was conducted in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (8th Edition, 2013).

Animals and experimental model

The liver samples that were used in this study were obtained from 34 female Merino sheep that had been in an experiment where they received three dietary treatments that were implemented to induce metabolic changes. At the time that the liver samples were collected the sheep in each of the three treatment groups (n = 12) were either fed at maintenance (control), fed at a maximum of 20% above their maintenance requirements (above), or fed at a maximum of 20% below their maintenance requirements (below). The maintenance requirement of each sheep was calculated using the following equation: Metabolic energy for maintenance (megajoules/day) = (0.09 × liveweight) + 1.4 (e.g., 50 kg sheep = 5.9 MJ/day) [33]. The sheep were housed in individual pens in an indoor shed and provided water ad-libitum. The sheep were fed twice a day with a 70:30 oaten chaff and lupin mix (9.0 megajoules metabolizable energy, 114 g crude protein, and 888 g dry matter per kg) and a mineral powder supplement (Topstock All-Season Minerals, Narrogin WA, Aus).

Collection and storage of liver tissue

On the day of tissue collection, each sheep was exposed to an isolation box test as a psychological challenge [34]. The isolation box test involved placing each sheep into a wooden box (H: 1.5 m x L: 1.5 m x W: 0.75 m) for one minute during which they were isolated from their conspecifics. Sixty minutes after the isolation box test, the sheep was killed using an injection of Lethabarb^® (Virbac, Aus). Within 15 min of death being confirmed, a sample of the liver was collected by cutting a section of the liver (approximately thumb size) with a scalpel blade and immediately transferring the sample into a cryotube. The samples were snap-frozen in liquid nitrogen and then stored at −80 °C until further processed.

RNA extraction and DNase treatment

Total RNA was extracted from each liver sample (n = 34) using a TissueRupter II (Cat.9002755, Qiagen, Aus) and the QIAzol^® Lysis Reagent (Cat.79306, Qiagen, Aus). The concentration and purity of each extracted total RNA sample was measured using the NanoDrop^® ND-1000 (Thermo Fisher Scientific, USA) (Supplementary material 1). The genomic DNA was removed from each total RNA sample using RQ1 RNase-Free DNase (Cat.M6101, Promega, Aus) in the T100TM Thermal Cycler (Cat.1861096, Bio-rad, Aus). Each DNase reaction contained 7 µL solution of 2 µg total RNA and RNase-free water; 1 µL of RQ1 DNase 10X Reaction Buffer; and 2 µL of RQ1 RNase-Free DNase that was incubated at 37 °C for 30 min. Then 1 µL of RQ1 DNase Stop Solution was added and the reaction was inactivated at 65 °C for 10 min.

Primer design

The forward and reverse primer for each gene (either a reference gene or a target gene) was designed using the messenger RNA (mRNA) sequence from GenBank and the Primer BLAST software (NCBI, USA). Each primer was designed to be a PCR product between 70 and 200 base pairs; melting temperature between 57 and 60 °C; GC% values between 50 and 70%; and to span across an exon-exon junction. The specificity of each primer was verified through sequencing of the PCR product as conducted by the Australian Genome Research Facility (Perth, WA) and using the Basic Local Alignment Search Tool (BLAST, NCBI, USA) program. For each gene, a single peak on the qPCR melting curve (Supplementary material 2) confirmed that each primer pair amplified a unique product, and no primer-dimers were generated.

For this study, target genes that are associated with antioxidant activity were selected because metabolic changes in sheep have been linked to changes in the level of oxidative stress [26, 27]. The five target genes selected were (Table 1): catalase (CAT); glutathione peroxidase 1 (GPX1); glutathione peroxidase 3 (GPX3); peroxiredoxin 1 (PRDX1); and superoxide dismutase 1 (SOD1). The reference genes that we selected had been tested as reference genes in other studies on sheep tissue and blood samples [30, 31, 35–46]. The nine reference genes were (Table 1): actin beta (ACTB); β 2 microglobulin (B2M); glyceraldehyde-3-phosphate dehydrogenase (GAPDH); hydroxymethylbilane synthase (HMBS); hypoxanthin phosphoribosyl transferase 1 (HPRT1); peptidylprolyl isomerase A (PPIA); heat shock protein 90 alpha family class A member 1 (HSP90AA1); succinate dehydrogenase complex flavoprotein subunit A (SDHA); and tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta (YWHAZ).

Table 1.

Primer sequences and annealing temperatures for qPCR analysis

Target genes
Primers	Gene description	Primer sequence (5’ → 3’)	Start-Stop	Tm/C°	Length
CAT XM_004016396.5	Catalase	F: TTCGCTTCTCCACTGTTGCT	376–395	61.8	172
		R: TGGCTGTGGATAAAGGACGG	547 − 528
GPX1 XM_004018462.5	Glutathione peroxidase 1	F: AACCAGTTTGGGCATCAGGAAA	290–311	62.5	122
		R: CCATTCACCTCGCACTTTTCG	411 − 391
GPX3 XM_015096153.3	Glutathione peroxidase 3	F: TCCTAGCCACCCTCAAGTATGT	408–429	62.5	137
		R: CGAGGTAGGAGGACAGGAGT	544 − 525
PRDX1 XM_004001918.4	Peroxiredoxin 1	F: CCTCTTTCTCTGGAACTGCTGATA	27–50	62.5	188
		R: GGGGCACACAAAGGTGAAGT	214 − 195
SOD1 NM_001145185.2	Superoxide dismutase 1	F: GCCAAAGGATGAAGAGAGGCAT	303–324	65.4	177
		R: TCATTTCCACCTCTGCCCAA	479 − 460
Candidate reference genes
Primers	Gene description	Primer sequence (5’ → 3’)	Start-Stop	C°	Length
ACTB NM_001009784.3	Actin beta	F: ACCGCAACCAGTTCGCCAT	70–88	64.5	187
		R: TCATCCCCCACGTAGGAGTC	256 − 237
B2M XM_027971438.2	β 2 microglobulin	F: CCCAAGATAGTTAAGTGGGATCG	355–377	62.4	195
		R: CAAGTAGGGCCCAAGGTAGA	549 − 530
GAPDH XM_060411595.1	Glyceraldehyde 3-phosphate dehydrogenase	F: GGCTGCCAGAACATCATCC	846–859	60	130
		R: CTCCAGGCGGCAGGTCAGA	969–987
HMBS XM_004016091.5	Hydroxymethylbilane synthase	F: CCTACCTGCCGAACACAGC	128–146	65.6	199
		R: CCCGTGGTGGACATAGCAATTAT	326 − 304
HPRT1 XM_015105023.3	Hypoxanthin Phosphoribosyl transferase 1	F: CTGTGGCCAGCTGCTTAGTA	771–790	64.5	198
		R: TCAGTCAATAGTGGTGTGGTTT	968 − 947
HSP90AA1 XM_027957416.2	Heat shock protein 90 alpha family class A member 1	F: GCCAGTTTGGTGTCGGGTTT	455–474	62.5	148
		R: CCCATCGGTTCTCCTGTGTCA	602 − 582
PPIA NM_001308578.1	Peptidylprolyl isomerase A	F: CTTGGGCCGCGTCTCTTTTG	49–68	63.4	154
		R: GAAGTCACCACCCTGGCACA	202 − 183
SDHA XM_027980212.2	Succinate dehydrogenase complex flavoprotein subunit A	F: CGTTCGACAGGGGAATGGTC	1687–1706	63.6	162
		R: CGTACTCGTCAACCCTCTCC	1848 − 1829
YWHAZ NM_001267887.1	Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta	F: TGAACTCCCCTGAGAAAGCC	545–564	62.3	148
		R: TCCGATGTCCACAATGTCAAG	692 − 672

RT and qPCR

Each sample of total RNA sample was synthesized into complementary DNA (cDNA) through reverse transcription using a T100TM Thermal Cycler (Cat.1861096, Bio-rad, Aus). Each reverse transcription reaction contained 0.5 µL of random primers (Cat.C1181, Promega, Aus), 5.5 µL total RNA (1 µg total RNA), and 8 µL RNase-free water and was run at 70 °C for 5 min and then chilled to 4 °C for 5 min. Then 5 µL of M-MLV RT 5X reaction buffer (Cat.M1705, Promega, Aus), 1.3 µL of dNTP Mix (Cat.U1511, Promega, Australia), 1 µL of M-MLV RT(H-) Point Mutant (Cat.M3682, Promega, Australia), and 3.7 µL of RNase-free water was added to each reaction. Then the reaction was run at 25 °C for 10 min, 55 °C for 50 min, and 70 °C for 15 min. Following reverse transcription, each cDNA sample was purified using the QIAquick^® PCR Purification Kit (Cat.28104 and 28106, Qiagen, Aus) and diluted with 30 µL RNase-free water by following the manufacture’s protocol.

The qPCR analysis was performed using the QuantiNova™ SYBR^® Green PCR Master Mix (Cat. No. 208056, Qiagen, Aus) with Bio-Rad^® CFX (Qiagen, Aus). The same cDNA that was extracted from each liver sample was used to estimate the expression of each gene. Each qPCR reaction was prepared in duplicate on a 384 well plate and contained 1 µL of cDNA template (33 ng of total RNA), 5 µL of SYBR Green PCR Master Mix, 1.4 µL of 5 µM Oligo-dT forward and reverse primer (see Table 1), and 2.4 µL of RNase-free water. The qPCR reaction was run for 45 cycles at 95 °C for 1 min for activation, 95 °C for 1 s for denaturation, then at an annealing temperature for 15 s (see Table 1), then 72 °C for 5 s for extension, and then finally a melt curve from 60 to 95 °C with 0.5 °C increments and a 5 s hold. A negative control of RNase-free water and non-RT control samples was run on each 384 well plate.

The relative expression of each gene was calculated by absolute quantification using the standard curve of each gene on each plate. A total of ten samples were used to generate the standard curve for each gene. The PCR product of each gene from the RNA that was extracted from liver samples, was washed up with QIAquick PCR Purification kit (Cat.28104, Qiagen, Aus). The purified PCR product for each gene was added with 100 µL RNase-free water to make the top standard and then serial diluted using a 1:10 ratio to generate an additional nine standards. Those ten standard samples were then run using qPCR on the respective plates. The standard curve was plotted against the log of the dilution factor of the standard samples and the cycling value. A minimum 98% amplification efficiency for each standard curve was calculated using the slope of the linear regression line and the coefficient of determination (R²) (Supplementary material 2).

Normalization using reference genes

The expression of each reference gene was estimated in the sheep liver samples using RT-qPCR (see methodology above for Primer design, RNA extraction, and RT-qPCR). The stability of expression for each reference gene was then analyzed across the experimental samples using the comparative delta cycle threshold method [12], geNorm [7], NormFinder [13], and BestKeeper [14]. Then the RefFinder software [15] and the BruteAggreg R-studio software package [16] were used to aggregate the results from the other algorithms to rank the reference genes from the most to the least stable expression across the samples. The three candidate reference genes that had the most stable expression across the samples were selected and used to normalize the expression of the target genes.

Normalization using NORMA-Gene

The NORMA-Gene macro-based Excel workbook (Microsoft) was used for normalization [17]. The effectiveness of the NORMA-Gene algorithm has been reported to improve when more genes are used in the algorithm [17], so we compared the accuracy of the algorithm using five (the target genes only) and eight (the five target genes plus the three reference genes that were selected) genes.

Statistical analysis

The data were analyzed using SPSS (Version 16.0, SPSS Inc., Chicago, IL). Before statistical analysis, the normality of the studentized residuals of all of the variables was analyzed using a Shapiro-Wilk test and the homogeneity of variance of the variables using Levene’s Test for Equality of Variances was checked. Effects and differences were defined as significant when P ≤ 0.05. When the residuals of a variable were not normally distributed, the data were either log or square root transformed to achieve a normal distribution for statistical analysis. The expression of the three selected reference genes and the target genes were analyzed using a general linear model with the feed treatment as the fixed factor. The change in variability of the normalization methods was determined by comparing the pre- and post-normalization variance and the standard deviation of the genes. Figures were created using GraphPad Prism software (Version 10.0 for Windows, GraphPad Software, Boston, Massachusetts, USA).

Results

Selection of reference genes for normalization

Comparative delta cycle threshold method

The comparative delta cycle threshold (∆Ct) method compares all possible combinations of the reference genes to calculate the level of deviation in the ∆Ct among the samples. A reference gene with the low mean ∆Ct and standard deviation indicates a low variability in the expression of that gene across the samples [12]. The mean ∆Ct and standard deviation of each reference gene, in order from lowest to highest, was HPRT1 (2.56 ± 3.12), HMBS (2.60 ± 3.84), HSP90AA1 (3.20 ± 4.24), B2M (3.29 ± 4.31), YWHAZ (3.30 ± 4.44), PPIA (3.69 ± 5.47), SDHA (4.60 ± 5.52), ACTB (4.56 ± 6.30), and GAPDH (7.02 ± 10.05) (Fig. 1).

Fig. 1.

Variability in the expression of the candidate reference genes using the comparative delta cycle threshold method. The symbols represent the mean delta cycle threshold and standard deviation of each reference gene among the samples compared with all the other reference genes. A lower delta cycle threshold indicates that a gene is expressed more stably across samples

GeNorm

The geNorm M value represents the stability of gene expression as measured for each reference gene using a value that is derived from the average pairwise variation of each gene with all the other genes. The gene with the lowest M value is considered to have the most stable expression across the samples [7]. The M value for each reference gene, in order from lowest to highest, was HPRT1 (1.56), B2M (1.66), HSP90AA1 (1.70), HMBS (1.83), YWHAZ (1.99), SDHA (2.22), GAPDH (2.39), ACTB (2.46), and PPIA (2.56) (Fig. 2a). The geNorm V value uses pairwise variations to determine the optimal number of reference genes for normalization. A geNorm V value of 0.15 and below is considered optimal for normalization [7]. The nine reference genes combined did not reach the optimal value of 0.15, however the geNorm V value did decrease when additional genes were added (0.54) (Fig. 2b).

Fig. 2.

Variability in the expression of the candidate reference genes using the geNorm algorithm. a shows the geNorm M value, which represents the calculated stability of expression value for each reference gene that is derived from the average pairwise variation of each gene with all the other genes. A lower M value indicates that a gene is expressed more stably across samples. b shows the geNorm V value, which determines the number of reference genes needed to reach the optimal geNorm V score of < 0.15 (green line)

NormFinder

The NormFinder algorithm estimates the variation of the expression of each reference gene among the samples between and within the treatment groups to calculate a stability value. A lower value indicates that the gene has a more stable expression across the samples [13]. The refence gene with the lowest stability value was HPRT1 (0.83) (Fig. 3). The remaining references genes, in order from the lowest to highest stability value, were HSP90AA1 (0.94), B2M (0.96), PPIA (1.11), ACTB (1.15), GAPDH (1.19), SDHA (1.23), YWHAZ (1.33), and HMBS (2.75). The best combination of reference genes, with a stability value of 0.74, was HPRT1 and HSP90AA1.

Fig. 3.

Variability in the expression of the candidate reference genes using the NormFinder algorithms. The bars show the NormFinder stability value based on the variation of the expression of each reference gene between and within the treatment groups among the samples. A lower stability value indicates that a gene is expressed more stably across samples

BestKeeper

The BestKeeper algorithm uses numerous pair-wise correlation analyses to determine the reference gene that has the smallest standard deviation of the cycle threshold, which indicates that the reference gene has the most stable expression across the samples [14]. The reference gene with the smallest standard deviation of the cycle threshold was YWHAZ (1.38). The remaining reference genes, ranked from lowest to highest standard deviation of the cycle threshold, were HPRT1 (1.55), HMBS (1.64), HSP90AA1 (1.70), B2M (1.74), SDHA (1.89), GAPDH (2.04), PPIA (2.43), and ACTB (2.44) (Fig. 4).

Fig. 4.

Variability in the expression of the candidate reference genes using the BestKeeper algorithms. The bars show the cycle threshold standard deviation of each reference gene calculated using the BestKeeper algorithm. A lower cycle threshold standard deviation indicates that a gene is expressed more stably across samples

RefFinder

The RefFinder online tool integrates the results from the comparative delta cycle threshold method, geNorm, NormFinder, and BestKeeper to calculate an overall geomean stability value for the expression of each reference gene. The reference gene with the lowest geomean stability value is considered to have the most stable expression across samples [15]. The ranking order of the reference genes, from most to least stable expression across the samples, was calculated using the different algorithms in RefFinder. When using the comparative delta cycle threshold method, the ranking order was HPRT1, HSP90AA1, B2M, PPIA, GPADH, ACTB, SDHA, YWHAZ, and HMBS. Using geNorm, the ranking order was HPRT1, B2M, PPIA, HSP90AA1, GAPDH, ACTB, SDHA, YWHAZ, and HMBS. With NormFinder, the ranking order was HPRT1, HSP90AA1, B2M, GAPDH, PPIA, ACTB, SDHA, YWHAZ, and HMBS. With BestKeeper, the ranking order was HPRT1, HSP90AA1, B2M, GAPDH, YWHAZ, PPIA, ACTB, SDHA, and HMBS. The overall comprehensive ranking order of the candidate genes by RefFinder, from most to the least stable expression across the samples, was HPRT1, B2M, HSP90AA1, PPIA, GAPDH, ACTB, YWHAZ, SDHA, and HMBS (Fig. 5).

Fig. 5.

The RefFinder comprehensive stability ranking value of each candidate reference gene. The bars represent the calculated geomean ranking value, in which the lowest value indicates the gene that was expressed the most stably across the samples, and the highest value indicates the gene that was expressed the least stably across the samples

BruteAggreg

The BruteAggreg R-studio package aggregates the rankings of the reference genes from the comparative ∆Ct method, geNorm, NormFinder, BestKeeper, and RefFinder and determines a single consensus ranking for the reference genes based on the stability of the expression of the genes across the samples [16]. The BruteAggreg function using Spearman’s rank correlation coefficient ranked the references genes, in order from most to the least stable expression, as HPRT1, B2M, HSP90AA1, HMBS, YWHAZ, GAPDH, SDHA, ACTB, and PPIA. The BruteAggreg function using Kendall’s rank correlation coefficient ranked the references genes, in order from most to the least stable expression, as HPRT1, HSP90AA1, B2M, YWHAZ, HMBS, PPIA, SDHA, GAPDH, and ACTB.

Expression of the target genes without normalization

For the raw expression data without normalization there was no significant difference between the treatment groups in the expression of any of the target genes in the samples from the liver (CAT: P = 0.57, GPX1: P = 0.58, GPX3: P = 0.38, PRDX1: P = 0.74, and SOD1: P = 0.77) (Fig. 6a).

Fig. 6.

The expression of GPX3 (mean ± SEM) in the liver samples from sheep. Graph (a) shows the expression of GPX3 in the liver of the sheep is shown without normalization (raw expression data). Graph (b) shows the expression of GPX3 after normalization using reference genes. Graph (c) shows the expression of GPX3 after normalization using NORMA-Gene with five genes (target genes only). Graph (d) shows the expression of GPX3 after normalization using NORMA-Gene with eight genes (five target genes and the three selected reference genes). Non-transformed data is shown in each graph. The sheep fed at maintenance are represented by the white bar (Control), the sheep fed below their maintenance requirements are represented by the spotted bar (Below), and the sheep fed above their maintenance requirements are represented by the striped bar (Above). Statistically significant differences between the treatment groups in relative expression of GPX3 is shown by the lowercase letters above the bars

Expression of the target genes using the three selected reference genes for normalization

Overall, the three reference genes that were consistently ranked as having the most stable expression across the samples from the liver using the various algorithms were HPRT1, HSP90AA1, and B2M. Statistical analysis validated that there was no significant difference between the treatment groups in the expression of HPRT1 (P = 0.94), HSP90AA1 (P = 0.84), or B2M (P = 0.59).

When the expression of the target genes was normalized using the reference genes there was a significant effect of the treatment on the expression of GPX3 in the liver (P < 0.05; Fig. 6b). The expression of GPX3 was significantly higher in the liver of the sheep that had been fed above their maintenance requirements than it was in the sheep that had been fed at maintenance (P < 0.05). There was no significant difference between the sheep that had been fed below their maintenance requirements and the other treatment groups (P = 0.23). There was no significant difference between the treatment groups in the expression of the other target genes (CAT: P = 0.71, GPX1: P = 0.08, PRDX1: P = 0.23, and SOD1: P = 0.07).

Expression of the target genes using NORMA-Gene for normalization with the five target genes

When the expression of the genes was normalized with NORMA-Gene using only the five target genes there was no significant difference between the treatment groups in the expression of any of the target genes (CAT: P = 0.23, GPX1: P = 0.43, GPX3: P = 0.09, PRDX1: P = 0.4, and SOD1: P = 0.8) (Fig. 6c).

Expression of the target genes using NORMA-Gene for normalization with eight genes (five target genes and three reference genes)

When the expression of the genes was normalized with NORMA-Gene using eight genes (the five target genes and the three reference genes) there was no significant difference between the treatment groups in the expression of the three reference genes (HPRT1: P = 0.62, HSP90AA1: P = 0.77, and B2M: P = 0.09) or the target genes (CAT: P = 0.52, GPX1: P = 0.91, GPX3: P = 0.14, PRDX1: P = 0.22, and SOD1: P = 0.84) (Fig. 6d).

Comparison of the variance in target gene expression using the different normalization methods

Normalization of the expression of the five target genes using the selected reference genes increased the standard deviation and variance of the target genes compared to the raw expression data (Table 2). When NORMA-Gene was used to normalize the expression of the target genes with either five (only the target genes) or eight (the five target genes and the three reference genes) genes, the standard deviation and variance of the five target genes was lower across the samples than it was in the raw expression data. NORMA-Gene with eight genes resulted in a lower standard deviation and variance in the expression of CAT, GPX1, and PRDX1 than when only five genes were used, but for GPX3 and SOD1 it was the converse, and the standard deviation and variance was higher when eight genes were used than it was when five genes were used.

Table 2.

Standard deviation and variance of the amount of expression of the target genes before and after application of each normalization method (N = 34)

Before normalization
	CAT	GPX1	GPX3	PRDX1	SOD1
Std. Deviation	423	0.26	0.51	0.04	1.82
Variance	179, 047	0.07	0.26	0.001	3.32
After normalization with selected reference genes
Std. Deviation	9228	1.38	1.47	0.10	3.51
Variance	85, 170, 021	1.89	2.16	0.01	12.36
After normalization with NORMA-Gene 8 genes
Std. Deviation	365	0.21	0.14	0.03	0.27
Variance	133, 449	0.04	0.02	0.001	0.07
After normalization with NORMA-Gene 5 genes
Std. Deviation	378	0.40	0.08	0.06	0.18
Variance	143, 011	0.16	0.007	0.004	0.03

Discussion

Our results confirmed that NORMA-Gene was a better method of normalization for the expression data of genes in the liver of sheep than the use of reference genes. The NORMA-Gene method reduced the variability in the expression of the target genes across the samples in our study, whereas normalization using the reference genes increased the variability of the target genes. Other studies have reported a similar effect of the different normalization methods on the variability of gene expression in other species, sample types, and under different experimental treatments from our study [23, 24]. Even though the refences genes that we used for normalization were experimentally validated, they showed limited effectiveness at reducing the variation in the expression of the target genes. Our results also support that the NORMA-gene method adequately normalized gene expression using the required minimum of five genes, but there was minimal improvement with the use of additional genes. Notably, when we compared the expression of one of our target genes (GPX3) between our experimental treatments, we obtained different outcomes when we used reference genes than when we used NORMA-Gene to normalize the expression data. These contrasting outcomes raise issues about the validity of the two normalization methods to correctly identify any effect of an experimental treatment on the expression of a target gene.

In our study, the stability of the expression of the nine reference genes across the samples was evaluated using six different algorithms (comparative delta cycle threshold method, geNorm, NormFinder, BestKeeper, RefFinder, and BruteAggreg). In general, those methods gave very similar outcomes. HPRT1 had the most stable expression across the samples by all the algorithms except for the BestKeeper algorithm, which had it with the second most stable expression. Depending on the algorithm used B2M or HSP90AA1 was the gene identified to have the next most stable expression across the samples. Other studies have shown that HPRT1, B2M, and HSP90AA1 all have stable expression in liver samples from sheep and have used them as reference genes for normalization [36, 37]. Our geNorm analysis on the nine reference genes did not achieve the optimal geNorm V value of 0.15 or lower. The developers of the algorithm recommend that when the optimal geNorm V value is not reached, then the three reference genes with the lowest geNorm M values should be used for normalization [7].

The fact that our data did not achieve the optimal geNorm V value could have been due to the screening of an insufficient number of reference genes, the inherent instability of the reference genes that we assessed, the biological variability in our samples, or variability in the extraction and RT-qPCR process. The evaluation of additional references genes or RNA-sequencing to pre-screen for suitable reference genes could have improved the chances of achieving an optimal geNorm V value, however, was not possible in our study due to resource constraints. In our study we assessed a similar number of reference genes to several other studies [29, 36, 47, 48] and used genes that have previously been validated as suitable reference genes in sheep. Our geNorm results illustrate the challenge that is presented in attempting to select suitable reference genes that will provide robust normalization for an experiment, even when previously validated genes are used.

The differences in the ranking of stability of the reference genes by the various algorithms that we used were likely because of the different ways that the rankings are calculated by each algorithm. It is recommended and practiced in other studies on sheep to use several algorithms and a consensus ranking to select suitable reference genes for normalization [31, 36, 37]. However, our results demonstrated that even when reference genes have been experimentally validated, they may still be ineffective at reducing the variation in the expression of target genes. Given that the selected reference genes did not differ between our three treatments, the inefficiency of the reference genes for normalization might be due to systematic variations during sample preparation or processing during RT-qPCR. Any systematic variations could have increased the variability of the expression of each reference gene and consequently the variability of the target genes after normalization [49].

The NORMA-Gene method of normalization was more effective at reducing the variation in the expression of the target genes across the samples than was the use of reference genes. In our study with liver samples and in previous studies using samples from other tissues under different experimental treatments [23, 24]. In human leukocyte samples, NORMA-Gene was better than using two common reference genes (ACTB and B2M) [24]. In that study, normalization using the two reference genes increased some of the variation of the expression of the target genes across the human leukocyte samples compared to the raw expression data. Another study with tissue samples from three species of Hamsters showed that NORMA-Gene reduced the variance, while normalization with a single reference gene (GAPDH) increased the variance of the expression of the target genes [23]. Normalization with reference genes is designed to reduce the within and between variation of gene expression across samples based on the assumption that their expression is unaffected by the treatment and measured without error [7]. In the present study, the three reference genes we selected to use for normalization were not affected by the experimental treatment, therefore variability in preparation of the samples during RT-qPCR seems to be the most probable reason for an increase in variability of target genes after normalization using the reference genes [8]. Conversely, the studies with human leukocytes and hamster tissue samples did not experimentally validate the reference genes prior to use so it is possible that the expression of these reference genes was affected by the experimental treatment [23, 24]. The NORMA-Gene method, being an algorithm-only method, does not require reference genes so is not affected by systematic variations or treatment effects. The better performance of NORMA-Gene at reducing the variation in the expression of the target genes in this and other studies supports its preferential use for accurate and reliable normalization of gene expression.

The present study was the first study, to our knowledge, to apply the use of the NORMA-Gene method to normalize the expression of genes in sheep and compare it against the traditional method of using reference genes. Heckmann et al., (2011) advised that at least five genes are needed for reliable normalization with the NORMA-Gene method and that additional genes will only incrementally improve the accuracy of normalization [17]. Our results confirmed that increasing the number of genes from five to eight resulted in negligible reduction in the variation of target gene expression, and it made no difference to the interpretation of the results.

In our study, the method of normalization changed the outcomes of the analysis for the expression of the target genes in the liver samples between the three treatment groups. When we analyzed the raw expression data for each of the target genes, there was no difference in the expression of any of the target genes between the treatments. Normalization using reference genes indicated that the sheep fed above their maintenance requirements had a significantly lower expression of GPX3 in their liver than the sheep fed at maintenance. Knowing where the truth lies is not helped by the lack of studies that have evaluated the effect of overfeeding on genes associated with oxidative stress. Studies that have restricted the feed of sheep to 50–55% below their maintenance report conflicting results with restriction either increasing or decreasing the expression of genes that are associated with mediating the level of oxidative stress [27, 50]. However, these other studies in sheep used only one reference gene that was not validated across the experimental samples before it was used for normalization. A review of the use of reference genes showed that when reference genes are unvalidated across experimental samples, incorrect interpretation of the expression of the target genes can occur [10]. By contrast, when we used NORMA-Gene to normalize expression, we found no significant effect of the experimental treatment on the expression of any of the target genes. Admittedly, the P values for some were suggestive of a trend, but if we accept P < 0.05 as an acceptable rate for Type I errors, then there was no difference between the treatment groups. This version of our results is consistent with a review of rodents studies, which showed that dietary restriction in majority of cases did not result in changes to the expression of genes with antioxidant mechanisms, such as CAT, GPX, and SOD, in the liver of rodents [51].

The question is really, where does the truth lie. If we normalize the expression of our target genes by using the expression of reference genes, then GPX3, which has antioxidant potential that is protective against oxidative stress, was activated in sheep that were overfed. On the other hand, if we use the NORMA-Gene method, then none of the target genes were activated in the overfed sheep more than the other treatment groups. Our results using NORMA-Gene are likely closer to the truth because it was better at reducing the variation in the expression of genes across the samples than the use of reference genes for normalization. There was a 40% difference in the energy intake between the sheep on the two dietary treatment groups at the time that samples of liver were collected from each sheep for the present study. However, our previous study by Babington et al., (2025) showed no difference in levels of oxidative stress, measured by the percentage of thiol oxidation in blood, among the treatment groups at the time of sample collection [52]. Together our results suggest that the metabolic discrepancy did not affect genes with antioxidant mechanisms or levels of oxidative stress in sheep, which is consistent with those results from the rodent studies [51].

Other limitations that should be considered in the interpretation of our results are related to the experimental design and possible influence of technical variation. We used a relatively small sample size and only used samples of liver from the sheep which could limit the generalizability of our results to other contexts. Technical variation could have been introduced in our study, which we accounted for by using standardized inputs of RNA quantities, the same reverse transcription kit, samples in randomized batches, and used the same pooled sample for the creation of each standard curve. The improved results of the NORMA-Gene method at reducing variance compared to reference genes could be partially explained by the algorithm being better at reducing the technical noise across samples. It would have been beneficial to have included RNA spike-ins and inter-plate calibrators during RT-qPCR to better account for technical variability in this study. However, we believe our results still provide evidence towards the ongoing discussion of the reliability of normalization strategies for studies using RT-qPCR. Additionally, we have validated that NORMA-Gene can provide a reliable method of normalization for studies on the expression of genes in livestock that choose not to use reference genes. Future research could seek to confirm our results in additional contexts, including species, sample types, and experimental treatments.

In addition to more accurate normalization, the NORMA-Gene method required fewer resources and less time than did the use of reference genes for normalization. In very few studies have the reference genes been validated across samples before they are used for normalization [10], even though it is recommended that any reference gene being used is experimentally validated for each experiment. Evaluating several reference genes in addition to the target genes requires more time and resources because of the additional RT-qPCR for the reference genes followed by selection of the best reference genes for normalization using the various algorithms. Conversely, the NORMA-Gene method does not require reference genes and thus does not require any additional RT-qPCR. A limitation of NORMA-Gene is that it has been validated as a method for the normalization of datasets that contain at least five genes [17]. Therefore, experiments with a smaller number of genes still rely on using reference genes for normalization even though it may incorrectly estimate the effect of an experimental treatment. Suitable normalization for studies that are unsuitable for NORMA-Gene need to be developed because it is clear from our results that there are challenges with using reference genes for normalization.

Conclusion

In summary, we validated that NORMA-Gene can provide a more accurate and reliable method of normalization and hence interpretation of the expression of genes in sheep than using reference genes. Additionally, NORMA-Gene required less time and resources than the use of reference genes for normalization. Based on our results, we would recommend that the NORMA-Gene method be used for the normalization of gene expression in eligible studies instead of using reference genes.

Abbreviations

RT-qPCR

Reverse transcription-quantitative real-time polymerase chain reaction

DNA

Deoxyribonucleic acid

cDNA

Complementary deoxyribonucleic acid

RNA

Ribonucleic acid

mRNA

Messenger ribonucleic acid

CAT

Catalase

GPX1

Glutathione peroxidase 1

GPX3

Glutathione peroxidase 3

PRDX1

Peroxiredoxin 1

SOD1

Superoxide dismutase 1

ACTB

Actin beta

B2M

β 2 microglobulin

GAPDH

Glyceraldehyde-3-phosphate dehydrogenase

HMBS

Hydroxymethylbilane synthase

HPRT1

Hypoxanthin phosphoribosyl transferase 1

PPIA

Peptidylprolyl isomerase A

HSP90AA1

Heat shock protein 90 alpha family class A member 1

SDHA

Succinate dehydrogenase complex flavoprotein subunit A

YWHAZ

Tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein zeta

Authors’ contributions

Conceptualization, S.B.; methodology, S.B.; formal analysis, S.B. and L.D.; investigation, S.B. and L.D.; data curation, S.B.; writing—original draft preparation, S.B.; writing—review and editing, A.J.T., S.K.M. and D.B.; visualization, S.B.; supervision, L.D., A.J.T., S.K.M. and D.B.; funding acquisition, A.J.T. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by Meat and Livestock Australia grant P.PSH.1232, The University of Queensland, and The University of Western Australia.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Liver samples were collected from sheep as approved by The University of Western Australia (UWA) Animal Ethics Committee (2020/ET000194). The experiment was conducted in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (8th Edition, 2013)

Consent for publication

Not applicable.

Competing interests

Sarah Babington conducted this research independently from her work with RSPCA Australia. The other authors declare no competing interests. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.