Adapted approach to profile genes while reconciling Vegf-a mRNA expression in the developing and injured lung

Abstract

During lung development and injury, messenger RNA (mRNA) transcript levels of genes fluctuate over both space and time. Quantitative PCR (qPCR) is a highly sensitive, widely used technique to measure the mRNA levels. The sensitivity of this technique can be disadvantageous and errors amplified when each qPCR assay is not validated. In contrast to other organs, lungs have high RNase activity, resulting in less than optimal RNA integrity. We implemented a strategy to address these limitations in developing and injured lungs. Parameters were established and a filter designed that optimized amplicon length and included or excluded samples based on RNA integrity. This approach was illustrated and validated by measuring mRNA levels including Vegf-a in newborn mouse lungs that were injured by 85% oxygen (hyperoxia) for 12 days and compared with control (normoxia). We demonstrate that, in contrast to contradictory Vegf-a expression when normalized to the least suitable housekeeping genes, application of this filter and normalization to most suitable three housekeeping genes, Hprt, Eef2, and Rpl13a, gave reproducible Vegf-a expression, thus corroborating the sample filter. Accordingly, both short amplicon length and proper normalization to ranked, evaluated genes minimized erroneous fluctuation and qPCR amplification issues associated with nonideal RNA integrity in injured and developing lungs. Furthermore, our work uncovers how RNA integrity, purity, amplicon length, and discovery of stable candidate reference genes enhance precision of qPCR results and utilizes the advantages of qPCR in developmental studies.

gene expression alters mammalian lungs during development, injury, and then recovery from either genetic or environmental defect. The genes encode products such as messenger RNA (mRNA), which can be translated into proteins that subsequently modulate the above processes. Understanding changes in gene expression is important to design new therapies or drugs that can alleviate developmental defects. Since it is not feasible to take lung biopsies from normal human infants, cells or animals such as mice are used to examine development and therapeutic strategies.

There are five stages of normal mammalian lung development: embryonic, pseudoglandular, canalicular, saccular, and alveolar. Gene expression differs not only across the stages but also between the cell types of the evolving lung that govern distal and alveolar regional development and functionality (11). As lung cell type-specific mRNA fluctuates during maturation, mRNA transcript number also decreases over time (40). Injury during development alters gene expression. Furthermore, disruption in any of the five stages generally alters distal lung development results in abnormal formation of air-exchanging alveolus growth, which usually occurs during the alveolar stage (3, 5, 8, 22, 42). The gene expressions we quantify in this paper are mRNA levels. To measure these mRNA level changes, we utilize the widely used and highly sensitive quantitative PCR (qPCR) technique (43).

Proper assessment and validation of a technique is essential to obtaining robust results that minimize variation from developmental changes that occur during lung formation and sample processing. This requires validation of each of the following steps: 1) tissue handling including dissection and storage, 2) total RNA extraction, 3) total RNA purity and integrity check, 4) data collection including primer design, and 5) data analysis.

Challenges toward optimal practice are present in lung RNA preparation. These include the time from when a sample is harvested to its storage impacting total RNA degradation. Importantly, RNases A and H and other RNase activity in the lung make it the third highest tissue with endogenous RNase activity, only behind the pancreas and spleen (24). During the extraction process, lung tissue remains extremely fragile owing to both the high endogenous RNase activity and instable, single-stranded RNA structure, which tends to undergo hydrolysis reactions (10, 39). Because of these challenges, RNA integrity should be determined. Traditionally, RNA integrity was assessed from the examination of 28S and 18S bands by agarose gel electrophoresis. This method has been shown to be inconsistent and subjective while providing indirect information about messenger RNA integrity (20). An RNA integrity number (RIN) calculated from the Agilent Bioanalyzer 2100 represents intactness of total RNA including gradient of mRNA degradation and most ribosomal sizes including 28S and 18S (20, 34). Minimum RINs of either 4.2 or 5 have been suggested in nonlung cell types (12, 14).

In addition, variance in qPCR can occur since total RNA in these cells and RNA type degradation rates differ. Therefore qPCR data has to be normalized to minimize the variance. A single, stable gene has not been shown to exist (33). Therefore, identifying and normalizing by more than a single gene would help yield accurate results (33, 41). Frequently these listed validation steps in lung development and injury are not widely performed.

To our knowledge, this is the first lung development and injury paper that utilizes these steps and demonstrates consistent expression following evaluation. Here we apply and assess the steps listed above as standard practice of qPCR on a lung developmental and injury model. Genetic deletion models of either one or both alleles encoding angiogenic cytokine Vegf-a exhibited vascular defects and are embryonic lethal (7). Literature using similar variants of an injury model during lung development reports conflicting magnitude and direction of Vegf-a expression (2, 18, 25, 28, 29). We determined that implementation of strict RNA parameters for RNA integrity, purity amplicon length, and identification of stable candidate reference genes enhanced qPCR consistency and reproducibility. For normalization of data, reference genes that were independent of each other regarding biological function and chance of bias were identified; they had been used in other lung development and injury studies (21, 26). Six candidate genes (Rpl13a, Hprt, Ppia, Eef2, Gapdh, and Rn18s) were evaluated for their use as reference gene in normalization. In contrast to the contradictory Vegf-a expression when normalized to the least suitable housekeeping genes, application of this filter and normalization to the most suitable three housekeeping genes, Hprt, Eef2, and Rpl13a, gave reproducible Vegf-a expression, thus corroborating the sample filter. Finally, we show that by applying these steps listed above in the developing lung we can circumvent the limitations of qPCR and decreased error amplification when comparing the uninjured to injured developing lung.

MATERIALS AND METHODS

Developmental oxygen injury mice model.

C57BL/6 mice were housed in 12 h light-dark cycles according to the Institutional Animal Care and Use Committee-approved protocol. Beginning on postnatal day 3, pups were exposed to either 21% oxygen (normoxia) or 85% oxygen (hyperoxia) until day 15, n = 6 and n = 6, respectively. Mice dams were kept with their brood until postnatal day 4, when they were subsequently exchanged between normoxia and hyperoxia every 24 h to prevent oxygen toxicity.

Lung organ harvest.

Pups were humanely euthanized; their lungs excised within 3–5 min directly into Eppendorf tubes, snap frozen immediately in liquid nitrogen, and stored at −80°C for up to one and a half years before analysis.

Primer design and in silico validation.

Primers were designed on IDT or Primer3. Primers were designed to span introns toward the 3′ end whenever possible, to amplify 65- to 160-bp products of gene targets, and to have few secondary structures. To check in silico product specificity and secondary structure, primer sequences were entered in NCBI Primer-BLAST for product with no mismatches followed by mFold, respectively.

Experimental primer validation.

The template to determine primer efficiencies was pooled single-stranded cDNA diluted by either tenfold for reference gene candidates or fourfold for Vegf-a. All of the primers had single melt peak curve and single band when run on 2% agarose gel in TAE (Tris-acetate-EDTA) buffer (data not shown).

RNA isolation.

A coronal section through both the right and left lung was made to include lung tissue from the same plane. The lung cuts weighed ∼15 mg and were suspended in TRIzol (Invitrogen cat. no. 15596-026). The tissue in TRIzol was then homogenized in BBX24 Bullet Blender (Next Advance). The aqueous portion of tissue homogenate was transferred to a Qiagen RNeasy Mini Column (Qiagen cat. no. 74104). Total RNA was isolated from the tissue homogenate according to manufacturer's protocol. RNA was eluted from the column in 40 ml of nuclease-free water (Ambion cat. no. AM9937).

RNA purity then quality assessment.

A Nanodrop 2000 Spectrophotometer was used to measure UV absorbance ratios of both 260 nm/280 nm and 260 nm/230 nm, which are expected to be nearly 2 for pure RNA. One microliter of eluted RNA was diluted in 4 μl of nuclease-free water. Two microliters of the diluted RNA were used for spectrophotometer reading.

Quality was assessed by running 2 μl of the diluted RNA samples on RNA Nano Chip of Agilent Bioanalyzer 2100. The output of the Bioanalyzer is a RIN from a scale of 1–10, 1 predicting complete degradation of RNA and 10 predicting no degradation.

cDNA synthesis.

In total 500 ng of each total RNA sample (150 ng/μl) was reverse transcribed by 200 U of SuperScript III (cat. no. 18080-044) reverse transcriptase in 20 μl reaction volume containing 5× First Strand Buffer, 50 ng of random hexamers (cat. no. N808027), 40 U RNaseOUT (cat. no. 10777-019), 0.1 M DTT, and water.

qPCR.

Reverse-transcribed cDNA template was diluted in nuclease-free water to a concentration within the 95–106% primer efficiency range determined. Then 2 μl of this diluted cDNA template was mixed with either 200 nM of each reference gene forward and reverse primer or 100 nM of each Vegf-a forward and reverse primer, 2× SYBR Master Mix (ABI cat. no. 4472908), and nuclease-free water to make a 20-μl reaction volume. A no-template control was run along with these reactions. Reactions were run in technical triplicates on ABI StepOnePlus machine at 50°C for 2 min, 95°C for 2 min, and 40 cycles of 95°C for 15 s and 60°C for 1 min. Following the run, melting curves were run at ramp rate of 0.2°C/s, obtaining single peaks for all products. Products were run on 2% agarose gel dissolved in TAE buffer to verify the expected amplicon size. These steps took into account the MIQE Guidelines (6).

geNormPlus algorithm.

The algorithm premise is that for two genes to be an ideal reference, the expression ratio between them would be the same in all samples. Beginning from a gene pair, the V value is calculated by standard deviation of the ratio of two genes in all samples, followed by inclusion of a third gene, and so on. After including a certain number of genes, the V value would increase since the genes with more variance would be incorporated. The number needed for normalization would be the number of genes included before the V value significantly increased. For a final variability value of expression among samples, the M value is an arithmetic mean of all V. Since the lowest V value indicates lowest standard deviation of ratio, the genes with lowest M values are less deviating, thereby more stable than the others (16).

Data processing.

Quantification cycles (Cq) are discrete logarithmic values obtained during the exponential phase of the time continuum when signal is above background threshold. There is a linear relationship between two logarithm scale Cq values. Residuals with Lowess smoothing were plotted to verify linear relationship. After assessing the relationship, we determined Spearman rank correlation coefficient to assess correlation between efficiency-corrected Cq values of all candidate reference genes. For Vegf-a expression analysis, calibrated relative quantity values (CNRQ) were calculated, which were normalized to one, two, or three reference genes' relative quantity.

A previous study indicated that mRNA transcripts generally approach normal distribution after log transformation (4). CNRQ values were transformed by factor of log base 2 to approximate normal distribution. Base 2 was chosen to represent the doubling of PCR product.

Frequentist statistical analysis.

All tests were chosen a priori having noted that the sample population was too small (i.e., less than 20) to accurately determine any sample distribution. In approximate log-normal distribution, geometric means of normoxia and hyperoxia Vegf-a CNRQ were calculated followed by parametric, unpaired t-test. Log-transformation generally minimizes differences in variance; Lowess smoothing lines of the residuals plotted were fairly flat (shown in results, Fig. 4A). Together these can satisfy the assumption that there are relatively equal variance within each normoxia and hyperoxia distribution. The satisfied assumptions for the unpaired two-sample t-test of Vegf-a levels were that 1) log-transformed CNRQ values of both normoxia and hyperoxia sample distributions were approximately normal, 2) these sample populations were independent, and 3) log-transformed variance was relatively equal. Type I error, α level, was set at 5%, before the study. The null hypothesis for the unpaired t-test that there is no difference of means between normoxia and hyperoxia was rejected when P values were less than the preset α level; these mean comparisons were then considered statistically significant. Bootstrapped 95% confidence intervals were calculated from geometric mean by sampling 10,000 times with replacement from sample distribution.

Fig. 4.

Rpl13a and Eef2 demonstrate the highest correlation coefficient and the lowest M value and are the most stable reference genes for normalization in qPCR of newborn lungs exposed to hyperoxia. A permutation test that samples Cq values from each gene was performed to test statistical significance against the null hypothesis that there is zero correlation. Shown above are correlation plots of efficiency-corrected Cq between every candidate gene pair. Genes are listed in ascending order of the M value rank found in Fig. 2, C and D. In all samples, the highest Spearman correlation coefficient was between Rpl13a and Eef2 (ρ = 0.99, ***P < 0.001) whereas the lowest was between Rn18s and Gapdh [ρ = 0.17, not significant (n.s.)]. The highest correlation coefficients suggest the stability of the 2 genes for normalization; thereby the 2 genes are most appropriate for normalization. A: in Cqs of all samples, correlations between the 2 highest ranked gene pairs are Rpl13a and Eef2. B: the highest correlation coefficient was still obtained between Rpl13a and Eef2 after exclusion of samples with RIN greater than 5. Because of sample exclusion, the sample size decreased, so the number of possible ranks decreased. Hence, Spearman correlation fluctuates. However, permutation test show highest statistical significance between Rpl13a, Hprt, and Ppia. Both lowest coefficients and statistical significance was between any pairing using either Gapdh or Rn18s (**P < 0.01, *P < 0.05).

qBase^Plus Software was used for analysis. Analysis verification followed by plotting was performed by use of Python 2.7.

RESULTS

Characteristics of selected genes studied.

Six genes with known biological functions in lung development and injury were selected as candidates for the normalization step of qPCR data analysis (Table 1). Qualifications as a normalization gene included each gene to have independent biological function to reduce the possibility of coregulation of each other, thereby minimizing variation in the normalization step. A literature search did not result in any papers, suggesting coregulation among these six genes.

Table 1.

Candidate genes for the normalization step of qPCR data analysis

	Gene Symbol	Accession Number	Function
Candidate reference gene name
Ribosomal 18S	Rn18s	NR_003278.3	Structural component of 40S ribosomal subunit
Glyceraldehyde-3-phosphate dehydrogenase	Gapdh	NM_002046	Catalyzes reduction in sixth step of glycolysis
Hypoxanthine guanine phosphoribosyl transferase	Hprt	NM_013556	Catalyzes purine salvage pathway
Eukaryotic elongation factor 2	Eef2	NM_007907.2	Promotes translocation of growing protein peptide during translation
Cyclophilin A	Ppia	NM_021130	Promotes protein folding
60S ribosomal protein L13a	Rpl13a	NM_009438.5	Structural component of 60S ribosomal subunit
Target gene
Vascular endothelial growth factor-A	Vegf-a	NM_001025250.3	Mediates angiogenesis
		NM_009505.4
		NM_001025257.3
		NM_001110266.1
		NM_001110267.1
		NM_001110268.1

Validating similar primer efficiencies and specificity.

After gene selection, primers were designed to target these genes as efficiently and specifically as possible during qPCR (Table 2). To check primer efficiency, known quantities of cDNA were serially diluted and plotted against the Cq that was measured by qPCR (not shown). Cqs, the qPCR output, are the number of cycles needed for the signal measured from theoretical, ideal doubling of transcript amplification to cross a set threshold over set time, making Cqs a discrete value. They indirectly represent transcripts per input, and as a result, lower Cqs can be interpreted as higher abundance of transcripts. Thus the Cq magnitude is limited by the signal measured from transcripts. Primer efficiencies are measured to check not only how specifically and well primers bind to a template, but also the limits of signal. Efficiencies calculated from the slope of these plots were close to 100%, with the range being 95–106%, well within the generally suggested 90–110% range. After qPCR, melting curves of the products were run. There was one peak, suggesting specificity (data not shown). In addition to melting curves, the products were run on agarose gel and visualized to check for one expected-size product (data not shown). Amplicons were predicted by using Primer-Blast and UCSC in silico PCR. Together these results show specific products.

Table 2.

Primers designed to target selected genes

Gene Symbol	Sense Primer Sequence	Antisense Primer Sequence	Amplicon Length, bp	Efficiency	Slope	LOD, Cq	Ta, °C	Tm, °C
Eef2	ACATTCTCACCGACATCACC	GAACATCAAACCGCACACC	135	1.992	−3.34	39.4	60	83.8
Gapdh	CAGAGGCCCTATCCCAACTC	GGTCTGGGATGGAAATTGTG	65	2.000	−3.27	36.5	60	80.6
Hprt	CCCCAAAATGGTTAAGGTTGC	AACAAAGTCTGGCCTGTATCC	76	1.952	−3.44	37.9	60	78.6
Ppia	GCAGACAAAGTTCCAAAGACAG	CATTATGGCGTGTAAAGTCACC	139	2.060	−3.28	36.9	60	79.9
Rpl13a	TCCCTCCACCCTATGACAAG	GTCACTGCCTGGTACTTCC	136	1.999	−3.44	37.9	60	84.7
Rn18s	AAACGGCTACCACATCCAAG	CCTCCAATGGATCCTCGTTA	155	1.995	−3.33	21.2	60	83.2
Vegf-a	TCTCCCAGATCGGTGACAGT	GGCAGAGCTGAGTGTTAGCA	71	1.961	−3.42	36.4	60	76.2

LOD, limit of detection; Cq, quantification cycle; T_a, primer annealing temperature; T_m, amplicon melting temperature.

Highly pure RNA did not suggest high integrity of RNA.

Having evaluated primers to be used for qPCR, RNA was extracted from the lung tissues. Both 260/280 and 260/230 absorbance ratios of the extracted RNA were assessed by Nanodrop 2000 Spectrophotometer (Table 3). Only 260/280 ratios are generally reported in literature for qPCR since obtaining 260/230 ratios near 2.0 can be dependent on tissue type. But for this study, both ratios are greater than or equal to 2.0, indicative of pure RNA. Although the RNA was pure, the ratios do not reflect integrity (20). Table 3 shows RIN of all samples with the integrity of normoxia RNA samples being 5.0 ± 0.3 and hyperoxia RNA samples 8.3 ± 0.1 (arithmetic mean ± standard error of mean).

Table 3.

RINs and absorbance ratios

	RIN	260/280	260/230
Normoxia	3.2	2.0	2.2
	5.3	2.0	2.4
	7.4	2.0	2.3
	3.5	2.0	2.2
	4.6	2.0	2.3
	5.9	2.0	2.1
Normoxia mean	5.0	2.0	2.2
Standard error	0.3
Hyperoxia	7.5	2.0	2.0
	8.8	2.0	2.4
	9	2.0	2.0
	8.1	2.0	2.1
	8.4	2.0	2.3
	8.2	2.0	2.4
Hyperoxia mean	8.3	2.0	2.2
Standard error	0.1

RIN, RNA integrity number.

Expression distributions of candidate reference genes.

Following total RNA extraction, total RNA purity, and integrity check, we performed qPCR to compare mRNA levels in uninjured and injured lungs of newborn mice. Including multiple lung samples per group minimized biological variability. Three technical replicates per lung RNA sample minimize variability of the Cq reading by qPCR. The arithmetic average of the three technical triplicate Cqs of each sample represents a single lung sample. A single Cq representing a single sample was then plotted to assess the overall distribution (Fig. 1A). The width of the violin plot indicates the number of Cq values at a certain magnitude while the length reflects the Cq range. The width increases around 24 Cqs for Rpl13a, representing multiple lung samples having similar transcript abundance. The Cq magnitude of one gene, Rn18s, is ∼10–15 lower than the other five genes. PCR theoretically doubles input, so this indicates that Rn18s, a ribosomal RNA, is ∼2¹⁰- to 2¹⁵-fold more abundant than the other transcripts; this holds true because ribosomal RNA, the most stable of all the RNAs, comprises most of the total RNA that was reverse transcribed.

Fig. 1.

Quantification cycle (Cq) distribution showing differences in mRNA integrity, tissue heterogeneity, and reverse transcription efficiency, which are reflected in Eef2, Hprt, Ppia, and Rpl13a but not Rn18s or Gapdh. The violin plots above show Cq data points of the 6 candidate genes in alphabetical order while representing both the density and spread of Cqs. For each gene, the width of the plot changes based on Cq density while the height indicates the spread of the Cqs. A: Cqs of all samples show are generally clustered together, except for the 2 or 3 distant, apical points representing individual samples, which can be reflected by its integrity (n = 12). Owing to ribosomal transcript overabundance relative to mRNA population, Cqs of Rn18s are much lower while showing minimal difference after removing samples with RNA integrity number (RIN) greater than 5. Gapdh has a longer spread and less clustering of Cqs compared with Eef2, Hprt, Ppia, and Rpl13a and can be attributed to reasons including the differences in integrity, heterogeneity of lung tissue sampling, or reverse transcription efficiencies. B: 3 samples with RIN less than 5 were excluded. Eef2, Hprt, Ppia, and Rpl13a all exhibit similar distributions whereas Gapdh maintains the largest spread and Rn18s does not appear to change at all (n = 9), suggesting that Rn18s does not reflect the change in mRNA. C: Cqs are grouped by oxygen levels: normoxia and hyperoxia. As mentioned before in A, the apical points are samples in normoxia and are much more clustered once removed, as seen in D.

The RIN of three samples were less than the suggested RIN cutoff of 5, which indicated degraded RNA (Table 3). When RNA samples with an integrity RIN of less than 5 were then excluded (Fig. 1B shows the average Cq values following the exclusion), the Cq distribution of Eef2, Hprt, Ppia, and Rpl13a all look similar whereas Gapdh does not. There was little change in distribution of Cq values of Rn18s. Not only was Rn18s overabundant relative to mRNA, but it also was not reflective of RIN, meaning any genes normalized to Rn18s would be overestimated. Plots of Cqs grouped by oxygen levels show that the lower-RIN, degraded RNA samples were in the normoxia group (Fig. 1, C and D).

Two genes sufficient, but three best for normalization across oxygen injury in newborn mice.

Acquiring any lung tissue and then RNA extraction from tissue takes precious time, money, and effort, requiring cost-effective accuracy in analysis. So the least number of genes sufficient for accurate normalization is important, which was determined by geNormPlus algorithm's V value. This V value, further defined in materials and methods, represents the standard deviation of the ratio between a minimum of two genes and three genes. The suggested V value cutoff is 0.15 (41), but this value should be empirically determined. If the V value increases as one gene is included in the comparison, then the included gene adds deviation. Thus it is better not to include the gene. V values are plotted in Fig. 2, A and B. Both before and after RIN-based exclusion, the V values of 2/3 and 3/4 reference genes were relatively similar (shaded bars). Before and after RIN-based exclusion, the V value gradually increased as additional genes were included. From this analysis, a minimum of two reference genes is needed since inclusion of the third candidate gene did not greatly alter V value. The two reference genes were determined by the geNormPlus M values, with a low M value correlating with the least deviation while the highest value has both the greatest variation and the least reliability.

Fig. 2.

Two genes are needed for quantitative PCR (qPCR) data normalization (Rpl13a and Eef2), but 3 are best (Hprt) even after exclusion of samples with RIN less than 5. The required gene numbers are based on V value, which is the standard deviation of ratio between expression of 2 and 3 reference genes, 3 and 4, and so on, in all samples. This increases as less stable genes are incorporated. After determination of the gene number required for normalization, genes were ranked by geNormPlus M value. A smaller value indicates less variation among the compared genes. A: 2 or 3 genes are needed since V increases once a fourth gene is incorporated, which are shaded dark gray above. There is clearly a fifth gene that greatly increases this value with respect to the others. B: by excluding samples with RIN greater than 5, the discrepancy between including a fifth or not diminishes. C and D: the 2 or 3 genes that can be used for normalization are the 3 highest ranks shaded dark gray, namely Rpl13a, Eef2, and Hprt. The 2 genes that remain as lowest ranks are Gapdh and Rn18s.

Eef2, Hprt, and Rpl13a expression exhibit the most stable M values after oxygen exposure.

In contrast to the marked variability of Rn18s and Gapdh M values, Hprt, Eef2, and Rpl13a had the lowest three M values among the tested reference genes across normoxia and hyperoxia (Fig. 2, C and D). Even after exclusion of samples with RIN less than 5, Hprt, Eef2, and Rpl13a remained the most stable gene set. Note that before and after exclusion M values of the three genes are close. This suggests that a gene of interest normalized to any of the three would be very close, and collectively all three would provide the most accurate normalization. If the two genes are the best reference genes for normalization, strong correlation between Cq values can be observed. Cq values are on logarithmic scale; when two logarithmic scale values are plotted against each other, there usually is a linear relationship. To verify that there was a linear relationship between Cq values, residuals of one reference gene mean were plotted against another reference gene's Cq values along with Lowess smoothing (Fig. 3, A and B). Most of the Lowess smoothing of the residuals was generally flat, suggesting that there is linear relationship; however, as seen in residual plot of Rn18s against Rpl13a, Lowess smoothing was not flat. Since not all relationships were linear, correlation between all pairs of reference genes was determined by rank-ordered Spearman correlation coefficient (Fig. 4, A and B). Of all Cq correlation pairs between reference genes in all samples, the largest correlation coefficient was between the two most stable reference genes, Rpl13a and Eef2 (ρ = 0.99, P < 0.001). The smallest correlation coefficient was between the two lowest ranked reference genes, Rn18s and Gapdh (ρ = 0.17). Based on these results, Rpl13a and Eef2 were the two genes used for normalization for further qPCR studies.

Fig. 3.

Eef2 and Rp113a demonstrate a strong linear relationship and relatively equal variance, supporting their efficacy as reference genes. Lowess smoothing of residuals between 2 stable genes will be mostly a flat, straight line, which means that there is a strong linear relationship; the flatness also suggests that there is relatively equal variance in further analysis. Cqs are logarithmic and the relationship between 2 logarithmic values can be linear. To test this, examples of the residuals are shown. A: residuals between the 2 highest ranked genes, Eef2 and Rpl13a, vice versa form a nearly flat line. B: in contrast, there are sharp peaks, exhibiting a nonlinear pattern, between lowest-ranked Rn18s and Rpl13a. Since there was still a general trend of linear relationship, this suggested preference for a rank-based test, i.e., Spearman correlation test, over a non-rank-based Pearson correlation test.

Vegf-a expression decreases in hyperoxia-exposed newborn lungs when normalized to Eef2, Hprt, and Rpl13a.

Previous studies do not show consensus of Vegf-a expression in the newborn model of hyperoxia with these studies relying on a single reference gene for normalization (2, 18, 25, 28, 29). For our study, Vegf-a expression was normalized to three, two, or one reference genes. Vegf-a expression was normalized to Hprt, Rpl13a, and Eef2 to calculate relative quantities. The relative quantities of individual samples in both normoxia and hyperoxia groups were similar (data not shown). Comparing the relative quantities in normoxia against hyperoxia showed threefold downregulation (P < 0.001). In agreement with normalizing to three reference genes, when normalized to two genes, Eef2 and Rpl13a, there was still a 2.98-fold downregulation in hyperoxia compared with normoxia (P = 0.001). Even after filtering of samples with RIN less than 5, Vegf-a expression decreased.

Vegf-a expression in opposite direction when normalized to Rn18s.

When Vegf-a was normalized to single, lowest ranked Rn18s, there was 2.2-fold upregulation in hyperoxia relative to normoxia (Fig. 5, P = 0.099). Although this upregulation was not statistically significant at our a priori set alpha level, this shows a trend that is completely opposite to the direction when normalized to highest ranked Hprt, Rpl13a, and Eef2. Importantly, when samples with a RIN of less than 5 were excluded and normalized to Rn18s, there was reversal in Vegf-a expression where a 0.07-fold downregulation (P = 0.5563) was determined. In contrast to normalizing expression by Rn18s, after samples with RIN less than 5 were excluded, Vegf-a expression normalized to Hprt, Rpl13a, and Eef2 was still statistically significantly downregulated by 2.2-fold (P = 0.006). When normalized to Rpl13a and Eef2, there was 2.1-fold downregulation (P value = 0.01).

Fig. 5.

Vegf-a tissue expression is accurately and consistently represented across oxygen levels normalized to Hprt, Rpl13a, and Eef2. When normalized to either the 2 or 3 appropriate reference genes (Ref.) or Rn18s, there are conflicting Vegf-a mRNA log₂-fold changes between lungs exposed to hyperoxia and normoxia. Vegf-a expression in all samples appears to increase when normalized to Rn18s (P = 0.09, n.s.), but decreases in samples with RIN greater than 5 (P = 0.56, n.s.). This is in contrast to the expression that statistically significant decreased when it was normalized to either the 2 (***P < 0.001) or 3 genes, Hprt, Rpl13a, and Eef2 (***P ≤ 0.001) both in all samples and after excluding samples (**P < 0.01).

DISCUSSION

qPCR remains widely used as an economical technique to determine and verify selected mRNA expression for its high sensitivity (31, 44). Because of the high sensitivity of this technique, we adapted and assessed the steps involved in qPCR to compare transcript levels in uninjured and injured developing lungs. Based on high RNase activity, RIN of the lung RNA was measured.

We then sought a way to minimize the impact of inferior RIN. Previous studies showed that, on one hand, amplicons greater than 400 bp were affected by loss of RNA integrity and suggested using samples with RIN greater than 5. On the other hand, amplicons less than 400 bp were minimally affected by RIN (13, 32). Not only would primer efficiency be minimally affected, but also amplicon length would be helpful to obtain consistent primer efficiency followed by accurate evaluation of gene expression (32, 34). Based on the RIN of our samples, primers were designed to generate amplicon lengths of 65–160 bp long, which is much less than the 400 bp suggested. This primer validation step is one of many that are known to influence cDNA transcription. Other examples include, but are not limited to, the reverse transcription of RNA to cDNA and priming strategy (37, 38). No reverse transcriptase or priming strategy is perfect. This study used random hexamers; oligo(dT)s are an alternative that could be used for isolated poly-A⁺ mRNA. On one hand, poly-A⁺ mRNA enrichment followed by reverse transcription that includes oligo(dT)s can increase signal-to-noise ratio in qPCR. On the other hand, during the enrichment process, there is possible loss of poly-A⁻ tail RNA such as deadenylated mRNAs and nuclear mRNAs or 3′ bias of mature mRNAs that may be of interest. Either case adds additional steps of technical variability.

This study used these validated primers to determine Vegf-a expression in hyperoxia relative to normoxia. Intragroup relative quantities of Vegf-a were similar, suggesting precision despite three samples with RIN below 5. In addition to comparing relative quantities, intersample variability was assessed. Bootstrapping the 95% confidence interval of Vegf-a expressions implies a lower and upper bound into which the mean will fall in 95% of the time. By applying this bootstrap, when normalized to the three reference genes, all normoxia samples overlapped with the confidence interval of RIN-based excluded samples (P = 0.999). No statistically significant difference between the mean expression of all hyperoxia samples and the RIN-based excluded samples (P = 0.09) was identified. Similar to expression normalized to the three genes, when normoxia samples were normalized to a single gene, Rn18s, there was no difference in Vegf-a geometric mean expression (P = 0.999). However, in hyperoxia samples, there was statistically significant decrease in the Vegf-a geometric mean expression (P < 0.00001) when normalized only to Rn18s. This is in direct contrast to the findings when normalized to the three idyllic reference genes. This suggests that normalizing to experimentally evaluated, idyllic reference genes reduces the intersample variation of Vegf-a in the hyperoxia samples even with RIN variation. Furthermore, without adequate and reliable genes for normalization, obtaining a robust and accurate interpretation of the impact of hyperoxia on Vegf-a expression would be difficult. Not only is there a difference in the mean expression in hyperoxia between RIN filtered samples, but also Vegf-a expression was contradictory when normalized to a single gene (Rn18s) compared with the three stable reference genes. These findings suggest that variable RNA integrity, amplicon size, and lack of adequate genes for normalization have given rise to reports conflicting in magnitude and direction of Vegf-a expression using similar variants of an injury model during lung development. On the basis of these results, although a strict RIN cutoff of 5 was set, both generating 65–160 bp amplicons and finding the best reference genes to normalize to minimized the effect of RNA integrity on means of Vegf-a expression.

Proper normalization was possible by evaluating genes with independent biological functions. A dependent function between two candidate reference genes could mean coregulation, producing a multiplicative error if normalized by these genes. Applying the geNorm algorithm requires an assumption of genes having independent biological function for valid results.

The six candidate genes were chosen since their biological functions in cells are known and used in other fields. Tools such as RefGenes (19) were used to scour public access microarray data to propose genes that are similar in expression signal to a gene of interest. However, there was limited knowledge of function for many of the genes, making it difficult to assess independence of biological function. Verifying independence of biological function of any candidate reference gene is difficult when RNA amounts obtained are small. For example, in one of the six candidate reference genes, Eef2, translational function is predominantly based on phosphorylation. Hyperoxia has been speculated to stimulate phosphorylation of Eef2 protein (33). However, in Ross et al.'s (33) cell culture model, neither mRNA transcript levels were examined nor total Eef2 protein levels change. No other literature suggests hyperoxia affects Eef2. Based on the known biological function of Eef2 and our data suggesting that Eef2 transcript levels were not affected by hyperoxia, it was used as a reference gene.

Although Eef2 was used as reference, Gapdh and Rn18s were not because they were the two least stable genes. This finding corroborates with lung-based literature that suggests Gapdh is affected by oxygen level, and this has not shown in any lung development models. For example, hyperoxia has been shown to disrupt glycolysis, including Gapdh levels, whereas hypoxia can stimulate HIF-1, which in turn activates Gapdh (15, 17, 23). In contrast to Gapdh, Rn18s levels have not been shown to be regulated by oxygen levels. However, Rn18s was still unstable across all samples. Two possible reasons that Rn18s was most unstable despite smallest Cq range are as follows: 1) Overabundant Rn18s ribosomal RNA transcript levels would not correlate with mRNA levels in a given lung sample since there are different number and types of cells. Within each cell, ∼80% in a total RNA population is ribosomal RNA, of which 18S ribosomal RNA comprises ∼20%; there is ∼1–3% of mRNA in total RNA population (27, 30). The exact same area of lung cannot be sampled, especially in clinical tissue, meaning that there will be variable amounts of cells being sampled. Less confidence can be made about normalizing to such an abundant transcript that correlates with neither mRNA levels nor RNA integrity. 2) By using the same input cDNA amount for qPCR to quantify gene expression, Cq values of Rn18s were invariably much lower than those of the other mRNA genes. Describing changes in lower transcript mRNA levels would be less sensitive if Rn18s is used for normalization, which is especially true for samples with nonideal RNA integrity. Rn18s represented ribosomal RNA component and would be the last of the RNA types to undergo degradation. Furthermore, RNA polymerase I reverse transcribes Rn18s whereas RNA polymerase II reverse transcribes messenger RNA. The listed reasons above are in contrast to the ideal characteristics of reference genes, which include, but are not limited to, most closely reflecting the gene of interests' mRNA transcript stability, levels, and reverse transcription enzyme; this is to further minimize any variance during the qPCR steps.

Rigorous measures must be employed for validation of RNA integrity. Quantification must include multiple reference genes; using a single gene is highly likely to yield inaccurate results. Failure to include multiple genes yields conflicting results, as we have shown in the case of Vegf-a. In our study, normalizing with both Rn18s and Gapdh resulted in means that were outside bootstrapped 95% confidence intervals of means generated from normalizing with any of the other stable genes, i.e., Hprt, Rpl13a, and Eef2. These findings further suggest that normalizing with more than one but up to three genes, i.e., Hprt, Rpl13a, and Eef2, results in accurate fold change.

Our study suggests that meaningful PCR results can be obtained despite nonideal RNA in lung development. The criteria for meaningful and reproducible qPCR results in lung tissue should include optimization of RNA integrity, both amplicon size and primer location, and analysis including idyllic, appropriate rigorously assessed reference genes. By following these criteria, this rigor minimized the effect of nonideal RNA integrity typically found in lung-derived RNA.

GRANTS

This publication was made possible in part by 5R01HL114977 (M. A. Schwarz) from the National Heart, Lung, and Blood Institute and by the Lilly Endowment, Inc. Physician Scientist Initiative (M. A. Schwarz).

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

D.D.L. and M.A.S. conception and design of research; D.D.L. performed experiments; D.D.L. analyzed data; D.D.L. and M.A.S. interpreted results of experiments; D.D.L. prepared figures; D.D.L. drafted manuscript; M.A.S. edited and revised manuscript.