Selection of Ideal Reference Genes for Gene Expression Analysis in COVID-19 and Mucormycosis

ABSTRACT

Selection of reference genes during real-time quantitative PCR (qRT-PCR) is critical to determine accurate and reliable mRNA expression. Nonetheless, not a single study has investigated the expression stability of candidate reference genes to determine their suitability as internal controls in SARS-CoV-2 infection or COVID-19-associated mucormycosis (CAM). Using qRT-PCR, we determined expression stability of the nine most commonly used housekeeping genes, namely, TATA-box binding protein (TBP), cyclophilin (CypA), β-2-microglobulin (B2M), 18S rRNA (18S), peroxisome proliferator-activated receptor gamma (PPARG) coactivator 1 alpha (PGC-1α), glucuronidase beta (GUSB), hypoxanthine phosphoribosyltransferase 1 (HPRT-1), β-ACTIN, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) in patients with COVID-19 of various severities (asymptomatic, mild, moderate, and severe) and those with CAM. We used statistical algorithms (delta-C_T [threshold cycle], NormFinder, BestKeeper, GeNorm, and RefFinder) to select the most appropriate reference gene and observed that clinical severity profoundly influences expression stability of reference genes. CypA demonstrated the most consistent expression irrespective of disease severity and emerged as the most suitable reference gene in COVID-19 and CAM. Incidentally, GAPDH, the most commonly used reference gene, showed the maximum variations in expression and emerged as the least suitable. Next, we determined expression of nuclear factor erythroid 2-related factor 2 (NRF2), interleukin-6 (IL-6), and IL-15 using CypA and GAPDH as internal controls and show that CypA-normalized expression matches well with the RNA sequencing-based expression of these genes. Further, IL-6 expression correlated well with the plasma levels of IL-6 and C-reactive protein, a marker of inflammation. In conclusion, GAPDH emerged as the least suitable and CypA as the most suitable reference gene in COVID-19 and CAM. The results highlight the expression variability of housekeeping genes due to disease severity and provide a strong rationale for identification of appropriate reference genes in other chronic conditions as well.

IMPORTANCE Gene expression studies are critical to develop new diagnostics, therapeutics, and prognostic modalities. However, accurate determination of expression requires data normalization with a reference gene, whose expression does not vary across different disease stages. Misidentification of a reference gene can produce inaccurate results. Unfortunately, despite the global impact of COVID-19 and an urgent unmet need for better treatment, not a single study has investigated the expression stability of housekeeping genes across the disease spectrum to determine their suitability as internal controls. Our study identifies CypA and then TBP as the two most suitable reference genes for COVID-19 and CAM. Further, GAPDH, the most commonly used reference gene in COVID-19 studies, turned out to be the least suitable. This work fills an important gap in the field and promises to facilitate determination of an accurate expression of genes to catalyze development of novel molecular diagnostics and therapeutics for improved patient care.

INTRODUCTION

Understanding changes in gene expression is critical to develop diagnostics, therapeutics, and prognostic approaches against infectious agents (1–6). In this context, real-time quantitative PCR (qRT-PCR) is often preferred to monitor gene expression due to its high sensitivity and specificity and broad quantification range. However, for qRT-PCR an endogenous control or reference gene from the same sample is essential for an accurate assessment of gene expression. Not only must a suitable reference gene have an adequate and consistent expression, but more importantly, the reference gene should show minimal expression variability between samples and under experimental conditions. Nonetheless, the reference genes may often demonstrate significant variability in expression, compromising outcomes and even producing inaccurate results (7–9). Therefore, validation of reference genes in any new experimental system is of utmost importance.

Housekeeping genes, due to their constitutive and stable expression, often qualify as reference genes in qRT-PCR studies (10–12). Indeed, housekeeping genes like glyceraldehyde-3-phosphate dehydrogenase (GAPDH), hypoxanthine guanine phosphoribosyl transferase 1 (HPRT-1), cyclophilin A (CypA), ribosomal protein L13A (RPL13A), and β-ACTIN are among the most commonly used reference genes or internal controls for qRT-PCR analysis (13, 14). For viral infections including COVID-19, β-ACTIN, GAPDH, and 18S rRNA (18S) are popularly used as internal controls (4, 15–18). However, most of these genes do not consistently manifest stable expression under various experimental conditions (19–21). This is critical during infections like those of SARS-CoV-2 that manifest a spectrum of disease ranging from asymptomatic through mild and moderate to severe, each with distinct clinical features (22–24). Moreover, onset of acute coinfections, especially those with fungi (mucormycosis), during SARS-CoV-2 infection or following recovery may further impact the stability and expression of the reference genes. Notably, over 45,000 people in India have developed COVID-19-associated mucormycosis (CAM), which has a fatality rate of >50% and was declared an epidemic in India (25). Unfortunately, despite the high mortality of CAM and the ongoing COVID-19 pandemic, there is not a single study focused on identifying the appropriate reference genes and investigating the impact of COVID-19 or CAM on commonly used reference genes.

In this study, we have addressed these knowledge gaps by evaluating the gene expression stability of nine widely used housekeeping genes in peripheral blood mononuclear cells (PBMCs) isolated from patients having a spectrum of COVID-19 and patients who developed CAM. Our results clearly show significant variations in the expression of housekeeping genes during SARS-CoV-2 infection or CAM, and identify the most appropriate reference gene establishing the significance of the study.

RESULTS

Melt curve analyses indicate specificity of the gene primers.

From clinically confirmed cases of COVID-19 and CAM, we isolated PBMCs, obtained good-quality RNA, and performed cDNA synthesis. All primer pairs used in the study yielded the best amplification efficiency at an annealing temperature of 60°C (see Fig. S1 in the supplemental material). We observed a single amplification product for each primer pair as evident from the presence of a single peak in the melt curve analysis (Fig. 1). This highlighted the absence of nonspecific binding of primer pairs, justifying their suitability for further downstream applications.

FIG 1.

Assessment of dissociation characteristics of candidate reference genes using melt curve analysis. (a and b) Melt peak of TBP, CypA, B2M, 18S, PGC-1α, GUSB, and HPRT-1 (a) and GAPDH and β-ACTIN (b). (c and d) Melt curve of TBP, CypA, B2M, 18S, PGC-1α, GUSB, and HPRT-1 (c) and GAPDH and β-ACTIN (d). A single melt peak and a single melt curve for each individual gene indicate the generation of a single amplicon following primer amplification. RFU, relative fluorescence unit.

Analysis of candidate reference genes reveals heterogeneity in gene expression based on disease severity in COVID-19 and CAM patients.

We next investigated the changes in the expression of candidate housekeeping genes in patients (n = 56) belonging to seven clinically defined groups (healthy, asymptomatic, mild, moderate, severe, pre-CAM, and post-CAM). All nine candidate reference genes were within the dispersal range of threshold cycle (C_T) value (range, >6.43 to 39.69) and also fold increase in PCR product per cycle, i.e., the amplification efficiency was within the accepted range (90 to 100%).

Based on the severity of SARS-CoV-2 or associated fungal infection, we noticed significant differences in the expression of candidate reference genes, between and within groups across the disease spectrum (Fig. 2). Surprisingly, the most commonly used internal control gene in COVID-19 studies, i.e., GAPDH, showed significant variations within healthy subjects and also between healthy and COVID-19 subjects who had a moderate or severe infection or CAM. For β-ACTIN, even though expression levels were comparable in healthy, mild, and asymptomatic individuals, substantial variations in individuals with moderate or severe infections or CAM were observed. Likewise, another commonly used housekeeping gene, peroxisome proliferator-activated receptor gamma (PPARG) coactivator 1 alpha (PGC-1α), also demonstrated significant variations among asymptomatic infections or CAM.

FIG 2.

Analysis of candidate reference genes reveals significant heterogeneity in gene expression among patients with various disease severities. Boxplots of mean C_T values (y axis) for candidate reference genes (x axis) used in this study are shown. The line across the box represents the median C_T value, whereas whiskers indicate the range of C_T values obtained in a particular group of patients. Data represent a total of 56 subjects (n = 56; male, 39; female, 17; 8 subjects/group) including the healthy individuals.

For 18S and glucuronidase beta (GUSB), the ranges of C_T values were comparable across groups. However, these genes consistently were expressed either very early (18S, 6.43 to 15.73; 95% confidence interval [CI], 10.56 to 11.53) or very late (GUSB, 32.40 to 39.69; 95% CI, 35.88 to 36.94). Ideally, for an internal reference gene, a moderate expression level as exemplified by a C_T value of 15.0 to 30.0 is desirable to accurately represent the quantitative expression of the genes of interest (26). As shown in Fig. 2, only TATA-box binding protein (TBP), CypA, β-2-microglobulin (B2M), and HPRT-1 have comparable mean C_T values across the SARS-CoV-2 spectrum and CAM. Nonetheless, the identity of the most suitable reference gene is not immediately obvious.

Application of statistical algorithms identifies CypA and TBP as the most appropriate reference genes for SARS-CoV-2 infection and associated mucormycosis.

Considering the significant variations in C_T values observed among patients (inter- or intragroup), we next employed popular statistical algorithms such as the comparative delta-C_T method, BestKeeper, GeNorm, NormFinder, and RefFinder for objectively determining the stability of reference genes. These algorithms capture various aspects of expression data including gene stability, inter- and/or intragroup variations, standard deviations (SDs), etc., to indicate a suitable reference gene.

Using the comparative delta-C_T method, we observed that GAPDH and PGC-1α showed the highest standard deviations across COVID-19 and CAM cases as evident from the cumulative standard deviation data for individual genes (Fig. 3a). CypA has the lowest average and the lowest cumulative standard deviation in expression, followed by 18S and TBP. Another tool, BestKeeper, on the other hand indicated CypA and TBP as the two most stable reference genes. We also used popular tools, namely, NormFinder and GeNorm, both of which revealed GAPDH to be the least suitable gene based on the highest arithmetic mean of all pairwise variations (indicated by M value in GeNorm) and the highest stability value, which is inversely related to the stability of the reference gene (i.e., the most stable gene has the lowest stability value). Both of these tools identified CypA as the most suitable reference gene followed by 18S. Remarkably, neither of the two most popularly used internal control genes in COVID-19, i.e., GAPDH and β-ACTIN, showed expression consistency across the patients with various severities of disease. All the above tools identified CypA as the ideal reference gene across the disease spectrum by consensus (Fig. 3).

FIG 3.

Comparison of candidate reference genes based on cumulative ranking score using different statistical algorithms. (a) Intragroup average standard deviations for individual patient groups were obtained using the delta-C_T method and are added to obtain a cumulative average standard deviation (y axis). The highest cumulative average standard deviation (i.e., GAPDH) indicates the lowest gene stability (x axis). (b) The graph represents a cumulative stability value (y axis) obtained by adding intragroup stability value based on coefficient of correlation (r) and standard deviation using the BestKeeper algorithm. The candidate gene yielding the lowest cumulative value (i.e., PGC-1α) indicates the highest stability. (c) In the NormFinder algorithm, intragroup stability values (y axis) based on coefficients of variation are added to obtain a cumulative stability value. The candidate gene yielding the lowest cumulative stability value (i.e., CypA) indicates the most stable reference gene. (d) In the GeNorm algorithm, intragroup expression stability (M value) based on the average pairwise variation of an individual gene with all other genes is used as a control to obtain a cumulative expression stability (y axis). The gene yielding the highest M value (i.e., GAPDH) indicates the lowest stability. (e) In the RefFinder algorithm, intragroup geometric means (y axis) based on geometric mean of ranking values calculated by all algorithms above are added to obtain a cumulative geometric mean of ranking values. CypA yielded the lowest cumulative geometric mean, indicating the highest stability, and therefore emerged as the most suitable reference gene.

We next applied a widely used algorithm, RefFinder (27), that considers the output of four other algorithms (comparative delta-C_T method, BestKeeper, GeNorm, and NormFinder) as input data to yield a final comprehensive ranking based on the geometric mean (GM) of ranking values. Based on RefFinder analysis, CypA emerged as the most suitable gene with the lowest cumulative GM (10.35) (GM has an inverse relation with gene stability; the lower the GM, the higher the stability). This value is less than half of those of the second and third most suitable reference genes identified here, i.e., 18S (GM = 22.92) and TBP (GM = 26.43), respectively. GAPDH has a cumulative GM (53.51) over five times higher than that of CypA and therefore emerged as the least suitable reference gene. β-ACTIN, the other widely used reference gene in COVID-19 and other viral infections, could be ranked only 5th out of the nine candidate genes evaluated (Table 1).

TABLE 1.

Comprehensive ranking based on geometric mean (GM) obtained using RefFinder analysis^a

Gene	GM for group:							Cumulative geometric mean	Comprehensive ranking
	Healthy	Mild	Moderate	Severe	Asymptomatic	Pre-CAM	Post-CAM
CypA	2.00	1.73	1.00	2.21	1.00	1.00	1.41	10.35	1
18S	3.22	1.57	2.00	5.23	2.00	4.47	4.43	22.92	2
TBP	6.48	1.86	3.72	1.97	5.42	3.76	3.22	26.43	3
HPRT-1	1.41	7.00	6.73	2.82	6.65	2.45	1.57	28.63	4
β-ACTIN	4.61	4.23	4.79	5.47	3.41	6.70	3.98	33.19	5
B2M	8.00	6.00	3.57	6.05	5.48	2.45	5.44	36.99	6
GUSB	6.19	4.73	6.19	3.25	6.51	7.14	8.00	42.01	7
PGC-1α	2.51	8.00	9.00	4.90	9.00	8.74	6.19	48.34	8
GAPDH	9.00	9.00	6.40	9.00	3.87	7.24	9.00	53.51	9

^aThe lowest cumulative geometric mean corresponds with the highest comprehensive ranking or suitability.

Gene expression analysis of candidate genes NRF2, IL-6, and IL-15 based on normalization with CypA and GAPDH confirms the suitability of CypA as an internal control.

We next evaluated the relevance of having an appropriate internal control by comparing differences between gene expression of select candidate genes based on normalization with CypA and GAPDH, also previously used in COVID-19 (28, 29). We selected key genes whose gene expression profile have been previously known from RNA sequencing (30–33). This guided us in comparing the qRT-PCR-based gene expression data obtained following normalization with internal control genes. Since no molecular studies were available for CAM patients, we focused on studies featuring COVID-19 severity and selected nuclear factor erythroid 2-related factor 2 (NRF2), interleukin-6 (IL-6), and IL-15 as the candidate genes for this analysis (30–33).

NRF2 is a master regulator for an array of antioxidant responses, and excessive oxidative stress is a well-documented hallmark of COVID-19 pathophysiology (34). However, in SARS-CoV-2 infection, NRF2 expression is dysregulated and not sufficiently induced to mount an effective antioxidant response (35). The relative suppression of NRF2 expression is a viral strategy to hijack host protective responses (35), and pharmacological induction of NRF2 expression is proposed to be an effective strategy against COVID-19 (36).

We evaluated how NRF2 expression would vary in patients based on the choice of the internal control. In our patient cohort, NRF2 expression based on CypA normalization is consistent with the previous reports wherein NRF2 expression is low and relatively suppressed (only 1.65- to 4.2-fold change compared to healthy subjects even in the high-oxidative-stress inflammatory environment [30]). Apparently, there was no statistically significant difference in NRF2 expression across the spectrum of COVID-19 and mucormycosis. However, we observed ~12-fold induction only in the pre-CAM group (Fig. 4). Normalization with GAPDH, however, could not accurately capture the heterogeneity of the COVID-19 spectrum and that of CAM patients, and a significantly higher induction of NRF2 (1.3- to 268-fold in asymptomatic, mild, and moderate groups) was observed in COVID-19 patients (Fig. 4). Moreover, in pre- and post-CAM groups, the NRF2 induction based on GAPDH normalization was ~600-fold and ~230-fold, respectively (Fig. 4). In only the severely sick group, there was no apparent difference in NRF2 expression based on normalization with CypA or GAPDH. Clearly, the normalization of NRF2 with GAPDH would be rather misleading and would not reflect the possibility of NRF2 activation-based therapies proposed by several independent studies (17, 18, 30).

FIG 4.

Gene expression profile of NRF2 based on use of CypA or GAPDH as an internal control. NRF2 expression based on CypA normalization reveals that no significant difference exists in its expression across various patient groups. This is consistent with the RNA sequencing-based expression wherein NRF2 expression is relatively low even in the high-oxidative-stress environment reported for COVID-19 and CAM (30, 31). On the other hand, normalization with GAPDH in our study reveals significant expression heterogeneity that varies from 2-fold to over 600-fold across the spectrum of COVID-19 and CAM cases. This does not match with RNA sequencing-based data from published works (31) and erroneously creates the impression of NRF2 overexpression. Healthy subjects represent the baseline NRF2 expression. Data represent mean ± SEM for assays performed with at least eight subjects in duplicate. Significance is ascertained by using a one-way ANOVA with Kruskal-Wallis test with Dunn’s post hoc corrections (n = 8/group; *, P value < 0.05; **, P value < 0.01; ***, P value < 0.001; ns, nonsignificant).

To further determine the suitability of CypA as an ideal reference gene, we next studied IL-6, which is transcriptionally inhibited by NRF2 to block the proinflammatory responses (37, 38). A negative relationship therefore exists between NRF2 and IL-6, and a high NRF2 expression would lead to suppression of IL-6 expression. Nonetheless, the normalization with GAPDH as a reference gene produced a high-NRF2–high-IL-6 expression phenotype as both IL-6 and NRF2 showed massive induction (>400-fold) in patients with mild infection and in pre- and post-CAM patients. Moreover, GAPDH normalization did not reveal any significant differences in the expression of IL-6 between asymptomatic and severely infected COVID-19 patients (Fig. 5) despite the fact that inflammation is minimal in asymptomatic patients and very high in patients having severe COVID-19 (39).

FIG 5.

Gene expression profile of IL-6 based on normalization with CypA or GAPDH as an internal control gene. With GAPDH, we observed similar expression levels among asymptomatic, mild, moderate, and severe infections. On the other hand, RNA sequencing data indicate a severity-dependent increase in IL-6 expression going from mild to severe infections (32–34). Only IL-6 expression derived from normalization with CypA, therefore, is in agreement with previously published studies. The y axis represents differences in the fold change based on data normalization vis-à-vis CypA or GAPDH in various groups (x axis) of COVID-19 and CAM patients. Significance is ascertained with respect to healthy subjects by using a one-way ANOVA with Kruskal-Wallis test with Dunn’s post hoc corrections (*, P value < 0.05; **, P value < 0.01; ***, P value < 0.001; ns, nonsignificant).

On the other hand, application of CypA as a reference gene indicated an overall higher IL-6 expression and low NRF2 expression in all groups. The highest IL-6 expression was observed in severe COVID-19 patients who had the lowest NRF2 expression. Furthermore, expression of IL-6 in mild, moderate, and severe cases differed significantly, as inflammation levels vary significantly among these groups, with asymptomatic patients having the lowest inflammation and the severely sick reportedly having the highest inflammation (39). Our observations based on IL-6 expression derived from normalization with CypA and not GAPDH were in accordance with the plasma IL-6 levels that we determined to support the role of CypA as an ideal reference gene (Fig. 6a). With GAPDH normalization, even asymptomatic patients exhibited significant induction of IL-6, and the IL-6 expression was also comparable with that of patients having moderate infections and those with CAM. To further delineate the relationship of gene expression with inflammation status, we next determined the plasma levels of C-reactive protein (CRP), which is an established clinical marker of systemic inflammation among COVID-19 patients. We observed a direct relationship between CRP and disease severity (40), with asymptomatic subjects having the lowest CRP levels and severe COVID-19 patients having the highest (Fig. 6b). The CRP and IL-6 levels correlated well with the gene expression derived from CypA normalization. In addition to IL-6 and NRF2, we also scored for IL-15 gene expression and obtained similar results wherein the normalization with CypA, and not GAPDH (Fig. 7), corresponds fully with the RNA sequencing data for IL-15 (31). Thus, in summary, only data normalized with CypA yielded a gene expression pattern that broadly concurs with the clinical status, substantiating its utility as an ideal reference gene in COVID-19 and CAM patients.

FIG 6.

IL-6 and C-reactive protein (CRP) levels in plasma across the spectrum of COVID-19 severity and CAM. Asymptomatic patients have the lowest and patients having severe COVID-19 have the highest IL-6 (a) and CRP (b) levels (x axis). These data are consistent with the gene expression profile derived from normalization with CypA. Significance is ascertained by using a one-way ANOVA with Kruskal-Wallis test with Dunn’s post hoc corrections (n = 68; *, P value < 0.05; **, P value < 0.01; ***, P value < 0.001).

FIG 7.

Gene expression profile of IL-15 gene based on the use of GAPDH or CypA as the internal control gene. Normalization with GAPDH does not concur with RNA sequencing-based data which show that IL-15 expression increases significantly from mild to moderate disease and then returns to baseline expression in the severely sick (31). In the case of normalization with CypA, the expression pattern concurs with RNA sequencing data by demonstrating an increase from mild to moderate groups and a reduction in the severely sick. The y axis represents differences in the fold change based on data normalization in various groups (x axis) of COVID-19 and CAM patients. Significance is ascertained by using a one-way ANOVA with Kruskal-Wallis test with Dunn’s post hoc corrections (*, P value < 0.05; **, P value < 0.01; ***, P value < 0.001).

DISCUSSION

The availability of qRT-PCR instrumentation in clinics/molecular biology laboratories in even low-income countries affords unique opportunities to study gene expression that can recognize and differentiate discrete gene subsets to predict disease outcome or response to therapy. Unfortunately, there is no universal reference gene for qRT-PCR studies, necessitating the evaluation of reference genes for their suitability for an individual condition. The premise of our study is to select an appropriate reference gene that could capture broad-spectrum health manifestations of COVID-19 in patients (asymptomatic, symptomatic, mild, moderate, and severe), including coinfections with mucormycosis, with the final goal of benefiting patient-centric translational research. Both COVID-19 and CAM are recently emerged infections and warrant investigations of molecular mechanisms to develop better management and treatment strategies. Nonetheless, a formal determination of reference genes for qRT-PCR analysis has not yet been investigated.

An ideal reference gene requires that its gene expression should be minimally regulated across the disease spectrum with minimal influence of patient heterogeneity in terms of age/sex or severity, treatment, etc. For this purpose, we evaluated nine housekeeping genes commonly used for normalization of gene expression and determined their expression across patient groups with a broader age profile (age range, male, 17 to 75 years; female, 16 to 70 years; Table 2) in COVID-19 and CAM cases. The candidate reference genes were selected because these genes are constitutively expressed in all cells and perform different functions, thus minimizing the overrepresentation of genes belonging to the same biological pathway. Indeed, a large number of these nine candidate reference genes displayed variations in gene expression based on patient heterogeneity and therefore did not satisfy the prerequisites of a suitable reference gene as substantiated by various algorithms used in our study. Given that a suitable reference gene should display stable expression, only CypA demonstrated consistent expression across all patient groups and accordingly emerged as the most suitable reference gene in this study, followed by TBP.

TABLE 2.

Description of subjects in study cohort used for gene expression analysis^a

Characteristic	Value by gender:
	Male (n = 39)	Female (n = 17)
Age (range in yr)	17–75	16–70
Median age (yr)	45	35.5
No. of individuals with comorbidity
Diabetes	16	3
Hypertension	2	1
Anemia	1	0
Hyperthyroidism	1	1
Malignancy	1	0
Heart disease	1	0

^aPatient cohort of 56 individuals (39 males and 17 females) enrolled from different groups (healthy, asymptomatic, mild, moderate, severe, pre-CAM, and post-CAM) along with their comorbidity status. Overall median age was 42 years.

First, all the patients in our study had a clinically confirmed diagnosis of mucormycosis or COVID-19. The patients’ classification into various categories (mild to severe) was based on well-established international guidelines on clinical severity that are independent of viral load (41). To isolate RNA, we used PBMCs as they are one of the most accessible, commonly used, and easy-to-use cell types, especially in clinical settings. Further, the gene expression profile in PBMCs correlates well with the systemic gene expression changes in the disease condition (42). These features make them an attractive candidate to investigate differential gene expression, particularly for determination of quantitative traits involved in phenotypic outcomes (43).

To minimize the variations in gene expression, housekeeping genes are usually considered reference genes. But even housekeeping genes are involved in a plethora of cellular pathways, which may become dysregulated in patients based on a variety of factors, including severity (44–46). Therefore, heterogeneity in gene expression among patients is natural, expected, and commonly present, thereby necessitating the determination of the most suitable reference gene. GAPDH and β-ACTIN are among the most widely used reference genes for qRT-PCR studies, including those of COVID-19. For CAM patients, however, there is not a single gene expression or RNA sequencing-based study performed to date. In our study, GAPDH emerged as the least suitable reference gene as significant variations in its expression are observed in patients with various severities of disease. This could be because GAPDH, in addition to glycolysis, is also involved in diverse cellular functions including regulation of inflammation (45). Indeed, we observed a differential inflammation status among COVID-19 patients as also exemplified by significant differences in the levels of IL-6 and CRP (46, 47) (Fig. 6a and b). β-ACTIN, another regularly used reference gene, including in COVID-19, was also found unsuitable and was ranked 5th out of the nine candidate genes. The other candidate reference genes had very high expression levels, such as 18S, or very low expression levels, such as GUSB, in both healthy and diseased subjects, making them unsuitable.

Based on our results, CypA has the most stable expression across the spectrum of COVID-19 cases. Consequently, CypA emerged as the most appropriate reference gene in qRT-PCR-based analysis of COVID-19 and mucormycosis. 18S emerged as the second most stable gene, but it is expressed very early, while a medium- to high-level expression is desirable for an ideal reference gene (26), and hence, this limits the utility of 18S as a reference gene in qRT-PCR. On the other hand, TBP consistently showed a moderate expression across the COVID-19 spectrum and has only a marginal difference between its cumulative geometric means (26.43) vis-à-vis 18S (22.92), making it a better choice than 18S. As per the standard guidelines for reference gene selection for qRT-PCR, C_T values of ≤40 are acceptable for determination of quantitative PCR (qPCR) efficiency and intra- and inter-assay variability (48). Further, most qRT-PCR-based COVID-19 diagnostics kits approved by WHO, the CDC, and other national and international agencies consider a C_T of ≤40 as the cutoff value for effective diagnosis. In our study, irrespective of differences in clinical severity among subjects, the overall C_T values were within the acceptable range for each group (dispersal range, 6.43 to 39.69). These variations in C_T values among and within groups of patients reflect a natural heterogeneity regularly observed in the clinic.

The mean 260/280 ratio for RNA in our experiments ranged from 1.94 to 1.98 for healthy, asymptomatic, and mildly infected subjects and 1.73 to 1.79 in the case of CAM and severe COVID-19 patients. There were no statistically significant differences in the mean 260/280 ratio across the groups (see Fig. S2 in the supplemental material). While a 260/280 ratio of ~2.0 is desirable, 260/280 ratio values of ~1.6 to 1.7 have been often used for downstream applications based on the complexity of the samples (49–54). The minor and statistically nonsignificant variations in 260/280 ratio among various groups in our study did not affect the conclusions as determined by reanalysis of data (n = 49 patients), excluding samples with a 260/280 ratio of <1.7 (n = 7 patients) (data not shown).

Selection of unsuitable reference genes may weaken the detection sensitivity of the target genes and even result in inaccurate outcomes (55). To determine how gene expression varies based on normalization with CypA and GAPDH, we looked at genes such as NRF2, IL-6, and IL-15, whose expression profiles based on RNA sequencing were previously reported in COVID-19 patients (30–33). This was because several studies have shown that RNA sequencing results correlate well with the qRT-PCR data (56, 57). Genes were selected based on their role in redox and inflammation as redox dysregulation and oxidative stress are the hallmarks of various infections including COVID-19 (1). NRF2 is a major redox regulator that is significantly induced in response to oxidative stress. Nonetheless, as a component of viral strategy to hijack host protective responses, NRF2 expression is dysregulated during COVID-19, is not sufficiently upregulated, and remains relatively suppressed/minimally induced to mount an effective antioxidant response (30, 35). Based on normalization with CypA, we obtained a relatively stable NRF2 induction profile in mild (<2-fold), asymptomatic (<3-fold), moderate (<5-fold), pre-CAM (<13-fold), and post-CAM (<3-fold) groups. Also, there was no significant difference in the NRF2 expression across various groups based on normalization (Fig. 4). The slightly higher NRF2 levels observed in the CAM group may be a response to mucormycosis, and NRF2 expression returns closer to baseline following reduction in fungal burden postsurgery (post-CAM). On the other hand, normalization with GAPDH highlighted significant expression heterogeneity (2-fold to over 600-fold) across the spectrum of COVID-19 and CAM. Significant upregulation of NRF2 was observed in mild (>260-fold), asymptomatic (>17-fold), moderate (>17-fold), pre-CAM (>600-fold), and post-CAM (>230-fold) groups. This is in contrast to the RNA sequencing studies, showing relatively suppressed expression of NRF2 in COVID-19 and with stimulation of NRF2 expression proposed as a therapeutic strategy (30, 58) Evidently, normalization with GAPDH erroneously creates the impression of NRF2 overexpression, especially in mild infections and CAM. In the severe group only, we obtained similar results from normalization with GAPDH and CypA (<2-fold induction in both cases).

The status of CypA as an ideal reference gene was further confirmed with the help of additional genes IL-15 and IL-6. NRF2 and IL-6 have a negative relationship as IL-6 expression is transcriptionally inhibited by NRF2 (37, 38). Since NRF2 expression levels are low in the COVID-19 spectrum, high IL-6 expression is expected in these groups. Only when we used CypA as a reference gene, we obtained an overall higher IL-6 and low NRF2 expression phenotype. That upon normalization with GAPDH both IL-6 and NRF2 showed massive induction (>400-fold) in the case of patients with mild infection and in pre- and post-CAM patients reinforces the idea that GAPDH is not a suitable reference gene for the COVID-19 spectrum. Analysis of IL-6 levels concurs well with the CypA-normalized IL-6 expression, and those levels were also consistent with the CRP levels, which further supports our findings and the suitability of CypA as an ideal reference gene.

Therefore, it is clear that CypA is the most suitable reference gene to use for COVID-19 and CAM and that only normalization with CypA can accurately capture the expression heterogeneity across the disease spectrum. Use of CypA would yield an accurate gene expression profile to guide development of further therapeutic and prognostic modalities in COVID-19 and CAM. To the best of our knowledge, the current investigation is the only systematic analysis of the stability of housekeeping genes during the course of COVID-19 and mucormycosis. The results clearly establish that expression of a housekeeping gene may vary due to infection and reinforce the idea of careful selection of housekeeping genes post-validation.

MATERIALS AND METHODS

Study subjects.

Subjects of different ages (25 to 60 years) and genders having SARS-CoV-2 infection (mild, moderate, severe, and asymptomatic) and COVID-19-associated mucormycosis, i.e., presurgery (pre-CAM) and postsurgery (post-CAM), along with healthy controls, were considered (total subjects = 56, 8 subjects/group; Table 2) The COVID-19 status of the subjects was confirmed by the presence of viral content determined by using Xpert Xpress SARS-CoV-2 (Cepheid, USA) or the TrueNat system (Molbio Diagnostics, India) approved by the Indian Council of Medical Research (ICMR) for COVID-19 diagnostics. Briefly, the patients were evaluated for the presence of N2 and E genes of SARS-CoV-2 (Xpert Xpress SARS-CoV-2). In the case of TrueNat, initial screening was based on the presence of the E gene, followed by that of the RdRp gene in the case of positive screening results. Mucormycosis was confirmed by performing tissue biopsy and fungal staining (KOH mount). The presence of fungal hyphae on the KOH mount indicated the presence of mucormycosis and was confirmed in the hospital by a clinical mycologist. We collected blood samples at the time of admission of patients by taking informed consent. Individuals in the healthy category were COVID-19 negative as confirmed by qRT-PCR, with no history of COVID-19 and comorbidities (diabetes, influenza, cough, tuberculosis [TB], etc.).

Criteria of disease and clinical severity.

All the cases of COVID-19 and CAM were clinically confirmed, and severity of COVID-19 infection was ascertained per the institutional clinical definitions that follow the same guidelines as recommended by the Indian Council of Medical Research (ICMR) and the Government of India. These guidelines are universal in nature and are consistent with the guidelines of other global bodies such as the U.S. NIH and WHO. The broad categorization of patients is indicated below:

• (a)Mildly symptomatic: proven SARS-CoV-2 infection with mild symptoms, i.e., fever and cough without shortness of breath or hypoxia.
• (b)Moderately sick: proven SARS-CoV-2 infection with moderate disease, i.e., fever with cough and/or shortness of breath with respiratory rate of ≥24/minute or oxygen saturation (SpO₂) of 90% to ≤93% on room air.
• (c)Severe: proven SARS-CoV-2 infection with moderate to severe disease, i.e., respiratory failure requiring mechanical ventilation and clinical evaluation by the attending physician. Respiratory rate of >30/min, breathlessness, or SpO₂ of <90% on room air.
• (d)Asymptomatic: proven SARS-CoV-2 infection but without any apparent symptoms.
• (e)Pre-CAM: confirmed mucormycosis in patients having COVID-19 or immediately after recovery from COVID-19 ascertained by RT-PCR report.
• (f)Post-CAM: patients having CAM and reporting to the hospital at an advanced stage of infection or who were referred cases who had to undergo surgery immediately and were still under treatment for CAM at the time of sample collection.

The design of the study did not influence the clinical care or treatment for any of the patients.

Isolation of PBMCs, RNA isolation, and cDNA synthesis.

Peripheral blood mononuclear cells (PBMCs) were isolated using density gradient centrifugation (Ficoll; HiMedia, India) as described earlier (59). Buffy coat containing PBMCs was removed carefully and washed twice with phosphate-buffered saline (PBS) at 1,500 rpm for 10 min. Approximately 10⁷ cells were transferred to 1 mL TRIzol (Invitrogen, USA) for RNA isolation followed by downstream processing including qRT-PCR. RNA concentration (nanograms per milliliter) and purity (A₂₆₀/A₂₈₀) were determined spectroscopically (see Table S1 in the supplemental material) (Cytation; BioTek, USA). The mean 260/280 ratios for all patient groups were in the range of 1.7 to 2.0 (Fig. S2). cDNA synthesis was carried out using the iScript cDNA synthesis kit per the manufacturer’s instructions (Bio-Rad, USA).

Selection of candidate genes, primer design, and optimization of qRT-PCR conditions.

From literature review, we selected the nine most frequently used housekeeping genes, namely, TBP, CypA, B2M, 18S, PGC-1α, GUSB, HPRT-1, β-ACTIN, and GAPDH, to determine their suitability to act as reference genes (Table S2). Nuclear factor erythroid 2-related factor 2 (NRF2), interleukin-6 (IL-6), and interleukin-15 (IL-15) were used as the candidate genes to determine the suitability of the most and the least suitable reference gene identified in our study. Primer 3 and NCBI-BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi) were used for extraction of gene sequences and primer design. For our study, we designed primers (18 to 24 bp, predicted melting temperature [T_m] = 50°C to 60°C, and GC content of 40 to 60%). We determined the amplification efficiency of individual primer pairs by performing gradient PCR. Briefly, a 20-μL volume of reaction mixture contained 10× PCR buffer, 500 μM deoxynucleoside triphosphate (dNTP), primer pair (500 nM each), 20 ng of the template (cDNA), and 1.5 U of Taq DNA polymerase (Bangalore GeNei, India). A gradient PCR was used with the following parameters: initial denaturation at 95°C for 3 min and 35 cycles of denaturation at 95°C for 30 s, annealing for 30 s at 58°C, 59°C, 60°C, and 61°C, and subsequent extension at 72°C for 30 s, followed by a final extension step for 5 min at 72°C. The amplified products were subjected to 1.2% agarose gel electrophoresis, and the temperature yielding the highest band intensity on the gel represented the best annealing temperature for an optimum amplification.

We used the SYBR green-based assay to quantify gene amplification by using a CFX Opus 96 qRT-PCR machine (Bio-Rad, USA). All the reactions were performed at least in duplicates with the final volume/reaction mixture of 20 μL having the following components: 10 μL of iTaq universal SYBR green supermix (2×), 2 μL of the 1:10-diluted cDNA template with a final concentration of 10 ng/reaction, 0.5 μL each of forward and reverse primer (10 μM), and nuclease-free water to make the final volume. The cycle was programmed for an initial denaturation at 95°C for 5 min, followed by 39 cycles of denaturation at 95°C for 10 s, annealing at 60°C for 10 s, and extension at 72°C for 30 s. The melt curve was obtained by heating the amplicon from 65 to 95°C in a continuous acquisition mode and was analyzed to ascertain the specificity of the primers.

Determination of CRP and IL-6 levels in patients.

C-reactive protein (CRP) and IL-6 levels were determined from plasma per the approved clinical protocols on hospital patients (n = 8 to 12 subjects/group; total subjects = 68). Briefly, CRP in plasma samples was determined by particle-enhanced immunoturbidimetric assay where human CRP agglutinates with latex particles coated with monoclonal anti-CRP antibodies, using the Roche Cobas c-702 analyzer (Roche, Switzerland). The aggregates were determined turbidimetrically, and the analyzer automatically calculated CRP concentration in the sample (milligrams per deciliter) without manual intervention. Likewise, for IL-6 quantification, we used an automatic analyzer (Beckman Coulter, USA) that performs a one-step immunoenzymatic (“sandwich”) assay. Briefly, the plasma sample is placed in a reaction vessel and incubated with paramagnetic particles coated with mouse monoclonal anti-human IL-6, blocking reagent, and the alkaline phosphatase conjugate. After incubation, materials are washed, followed by addition of chemiluminescent substrate. The light generated by the reaction is measured with an inbuilt luminometer. The concentration (picograms per milliliter) of analyte in the sample is determined from a multipoint calibration curve.

Data analysis and algorithms for the selection of the most suitable reference gene.

Statistical analysis was performed using modules available in GraphPad ver. 5.01. The significance was determined using one-way analysis of variance (ANOVA) with Kruskal-Wallis analysis with Dunn’s post hoc corrections (*, P value < 0.05; **, P value < 0.01; ***, P value < 0.001; ns, nonsignificant). Data represented mean ± standard error of the mean (SEM) or standard deviation (SD). Different tools, namely, delta-C_T, BestKeeper, NormFinder, GeNorm, and RefFinder, were used to calculate the stability of each reference gene. All these software programs have inbuilt statistical tools to score for variations and decide on the most preferred genes.

Ethics approval.

This study was approved by the institutional human ethics committee of the All India Institute of Medical Sciences (AIIMS), New Delhi (IEC-435/02.07.2021, RP-34/2021, and IEC-419/02.07.2021).

Data availability.

All data are available from the corresponding authors upon reasonable request.