β-glucuronidase use as a single internal control gene may confound analysis in FMR1 mRNA toxicity studies

Relationships between Fragile X Mental Retardation 1 (FMR1) mRNA levels in blood and intragenic FMR1 CGG triplet expansions support the pathogenic role of RNA gain of function toxicity in premutation (PM: 55–199 CGGs) related disorders. Real-time PCR (RT-PCR) studies reporting these findings normalised FMR1 mRNA level to a single internal control gene called β-glucuronidase (GUS). This study evaluated FMR1 mRNA-CGG correlations in 33 PM and 33 age- and IQ-matched control females using three normalisation strategies in peripheral blood mononuclear cells (PBMCs): (i) GUS as a single internal control; (ii) the mean of GUS, Eukaryotic Translation Initiation Factor 4A2 (EIF4A2) and succinate dehydrogenase complex flavoprotein subunit A (SDHA); and (iii) the mean of EIF4A2 and SDHA (with no contribution from GUS). GUS mRNA levels normalised to the mean of EIF4A2 and SDHA mRNA levels and EIF4A2/SDHA ratio were also evaluated. FMR1mRNA level normalised to the mean of EIF4A2 and SDHA mRNA levels, with no contribution from GUS, showed the most significant correlation with CGG size and the greatest difference between PM and control groups (p = 10−11). Only 15% of FMR1 mRNA PM results exceeded the maximum control value when normalised to GUS, compared with over 42% when normalised to the mean of EIF4A2 and SDHA mRNA levels. Neither GUS mRNA level normalised to the mean RNA levels of EIF4A2 and SDHA, nor to the EIF4A2/SDHA ratio were correlated with CGG size. However, greater variability in GUS mRNA levels were observed for both PM and control females across the full range of CGG repeat as compared to the EIF4A2/SDHA ratio. In conclusion, normalisation with multiple control genes, excluding GUS, can improve assessment of the biological significance of FMR1 mRNA-CGG size relationships.


Introduction
The prevalence of FMR1 Premutation (PM: CGG 55-199) alleles in the general population has been reported to be as high as 1 in 150 females and 1 in 450 males [1]. Maternally transmitted PM alleles have the propensity to expand in future generations to full mutation (CGG !200) alleles that cause fragile X syndrome (FXS) [2]. FXS is a common cause of intellectual disability and co-morbid autism (reviewed in [3]). PM alleles have also been associated with adult onset Fragile X-associated disorders: Fragile X-associated tremor/ataxia syndrome (FXTAS: 40% males and 8-16% females over 50 years old) and Fragile X-associated primary ovarian insufficiency (FXPOI:~20% females) [4,5]. Pathogenic mechanisms suggested to cause PM related disorders include reduced FMR1 protein expression (FMRP), elevated levels of non-coding RNA (FMR4, FMR5, FMR6) [6,7], mitochondrial dysfunction [8] and CGG repeat-associated non-AUG translation [9]. The mechanism most extensively studied in the context of Fragile X-associated disorders is FMR1 mRNA gain of function toxicity [10].
FMR1 mRNA levels in the blood have been reported to be elevated 2-8 fold in PMs as compared to individuals with normal FMR1 alleles (<45 CGG repeats). This finding has been replicated across many different cell types in humans and in various CGG knock-in animal models [10,11]. It is thought that this PM-specific FMR1 mRNA excess is indirectly associated with increase in CGG size in the PM range and the formation of intranuclear inclusion bodies and late-onset neurodegeneration [12]. Indeed, PM-size ribo-rCGG repeat containing FMR1 mRNA can induce formation of intranuclear inclusions in Purkinje neurons of the cerebellum in transgenic mice, also known as nuclear foci [13]. Moreover, inclusion bodies that stain positive for FMR1 mRNA have been found in both the central nervous system of men with FXTAS [14][15][16] and in mice with 'knock in' PM alleles [17,18].
FMR1 mRNA levels in peripheral tissues have been significantly correlated with brain changes associated with FXTAS and with subtle motor signs in adult PM carriers at risk for FXTAS [19][20][21]. However, other studies failed to identify any relationship between the FMR1 mRNA level and similar clinical outcome measures, even in cases where relationships between the phenotype and CGG size were present [22,23].
The first study to propose the RNA toxicity mechanism used real-time PCR, normalising FMR1 mRNA levels with GUS as a single internal control [10]. This study reported that normal peripheral blood leukocytes maintain comparable levels of FMR1 and GUS mRNA, although no data was presented examining variability of GUS mRNA level as compared to other internal control genes in the control and PM groups.
While GUS is commonly used as an internal control gene for RT-PCR normalisation across different settings, especially in plant biology studies, its transcription stability in mammalian systems has been variable between studies and cell types [24][25][26][27][28]. Analysis of GUS mRNA stability as compared to other genes in lymphoid malignancies or B and T cell enriched and stimulated leukocyte fractions found that GUS had less stable transcription when compared to other internal control genes [29]. More recently, the geNorm approach was used to determine the most stably expressed genes in peripheral blood mononuclear cells (PBMCs) of PM and control males [30], to determine the 'optimal' method for FMR1 mRNA normalisation. The EIF4A2 and SDHA mRNA average was found to be the optimal normalisation method. This normalisation method was then applied to PM and control females, where correlations to the phenotype were also investigated [30].
This study expands on the previous findings by examining: (i) how the choice of the normalisation method impacts the strength of the previously published correlation between FMR1 mRNA level and CGG triplet expansion size in PM females and an age and IQ matched group; (ii) the variability in internal control gene mRNA levels including EIF4A2, SDHA and GUS when compared between PM females and an age and IQ-matched control group; (iii) relationships between CGG triplet expansion size and the mRNA levels of the chosen internal control genes.

Participants
All study participants provided signed informed consent and the study procedures were consistent with the Declaration of Helsinki and approved by the Southern Health Ethics Committee (project 10147B). Participants included 35 PM females and 35 age-and IQ-matched control females recruited as part of previous studies [30]. Groups were matched on height, body mass index (BMI), age and Wechsler Abbreviated Scale of Intelligence (WASI) Full Scale IQ score (see details in [30]). Participants were English speaking with no history of epilepsy or of a serious head injury and had normal (or corrected) vision and hearing, and no sign of colour blindness or intellectual disability (as assessed using the WASI Full Scale IQ score).

CGG triplet expansion sizing
Four millilitres of venous blood were collected in ethylene diaminetetraacetic acid (EDTA_ tubes (BD, worldwide) from all participants. DNA was extracted using the BIO ROBOT M48 DNA Extractor (Qiagen Inc., Hilden, Germany). The CGG sizing was performed using the Asuragen AmplideX™

RNA extraction and mRNA analysis
One million peripheral blood mononuclear cells (PBMCs) were isolated per participant from venous blood using Ficoll gradient separation as previously described [7], with PBMC pellets frozen at -80˚C in RLT buffer for total RNA extraction. Total RNA was purified using the RNeasy extraction kit, as per manufacturer's instructions (Qiagen Inc., Hilden, Germany). NanoDrop ND-1000 Spectrophotometer was used to determine RNA concentrations in triplicate, with purity assessed using the A260/A280 ratio (expected values between 1.8 and 2). Each RNA sample was then diluted to 5 ng/μl, with 2 μl RNA added for cDNA synthesis performed using the Multiscribe Reverse Transcription System (20 μl total), 50 units/μl of the reverse transcription enzyme (Life Technologies, Global).
EIF4A2 and SDHA primer/probe mixes were obtained from PrimerDesign (PerfectProbe ge-PP-12-hu kit) and used at concentration of 2 μM, with sequences not disclosed by the manufacturer. The FMR1-5 0 and FMR1-3 0 target gene and the internal control gene dynamic linear range (DLR) common to all the assays was determined to be 1 to 40 ng/μl total RNA input in a 20 μl cDNA reaction. This was determined from a series of doubling dilutions of RNA (160-0.5 ng/ul) of a selected control PBMC sample. All assays showed optimal performance within the DLR, with PCR efficiency ranging between 92 and 94% and coefficient of correlation of greater than 0.98. For all assays in this study, samples were quantified in arbitrary units (au) in relation to the standard curves performed on each plate and had to be within the DLR to be included in further analyses. The mean FMR1 5'and 3' mRNA levels was normalised to: (i) GUS alone (FMR1/GUS); (ii) mean of EIF4A2 and SDHA mRNA levels (FMR1/2IC); (iii) mean of GUS, EIF4A2 and SDHA mRNA levels (FMR1/3IC). GUS mRNA levels were normalised to 2IC (EIF4A2 and SDHA); while EIF4A2 and SDHA mRNA was expressed as a ratio (EIF4A2/ SDHA).
Two separate cDNA reactions were performed for each RNA sample, with each cDNA analysed in two separate RT-PCR reactions. The summary measure for mRNA level for each participant was represented by the mean of the four outputs. FMR1 mRNA results were not obtained for two PM females and two control females from the 70 participants because there was either insufficient RNA extracted or because the results failed the 5 0 and 3 0 FMR1 mRNA quality control step [33].

Statistical analyses
The Shapiro-Wilk normality test was used to check normal distribution for each of the FMR1 mRNA datasets, separately for each group. The data was then transformed if normality was not achieved. FMR1/GUS data was transformed using a natural logarithm function while for all other data, reciprocal function was used The Generalised estimating (GEE) method was then used for the inter-group comparison, taking into account correlation within family in the PM cohort. For the relationship between each mRNA level and CGG size, piecewise linear regression was used to find a threshold, resulting in two different slopes, above and below the threshold. Analyses were carried out using STATA software. See S1 Table for raw data.

Intergroup comparisons of FMR1 mRNA levels between PM and control groups
Three different methodologies for normalisation of FMR1 mRNA levels in blood were compared between PM and control groups: FMR1/GUS, FMR1/3IC and FMR1/2IC. FMR1 mRNA levels were significantly elevated in the PM group compared to the control group for all three normalisation approaches (FMR1/GUS: p = 2.7×10 −5 ; FMR1/3IC: p = 1.2×10 −9 ; FMR1/2IC: p = 3.4×10 −11 ). However, the most significant difference in mean values between the two groups was observed using the FMR1/2IC approach (that does not use GUS mRNA for normalisation) (Fig 1). The choice of normalisation strategy also influenced the proportion of PM females that exceeded the maximum control value for each plot (Fig 1: broken horizontal lines), with only 5 exceeding this value for FMR1/GUS (15%) and 6 for FMR1/3IC (18%) compared to 14 females for FMR1/2IC (42%) where GUS had been omitted from the normalisation equation. GUS/2IC and EIF4A2/SDHA values were not significantly elevated in the PM group compared to controls (Fig 1D and 1E; Table 1). However, the GUS/2IC value showed much greater variability between individuals with interquartile range twice as large in both control and PM groups as compared to the EIF4A2/SDHA value.

Relationships between mRNA levels and CGG triplet repeat size in PM and control female samples
The influence of FMR1 normalisation approach on FMR1 mRNA-CGG relationships in PM and control groups was assessed. In the combined cohort of PM and control females, piecewise linear regression was used to find a threshold CGG repeat size where two different slopes could be differentiated. Below this threshold, CGG repeat size was not significantly correlated with FMR1/GUS data (p = 0.287), but was significantly correlated with data for FMR1/3IC (p = 0.017) and FMR1/2IC (p = 0.001). The strength of the relationship also varied above the threshold depending on the FMR1 normalisation approach that had been used. In particular, effect size of the FMR1 mRNA-CGG relationship was higher with a smaller p-value for FMR1/ 2IC (regression coefficient (β) = 5.07, p = 1.3 × 10 −6 ) and FMR1/3IC (β = 3.25, p = 3 × 10 −5 ) than it was for FMR1/GUS (β = 2.05, p = 0.001). These analyses demonstrate that removal of GUS from FMR1 mRNA normalisation and/or dilution of the GUS contribution by addition of the two other internal control genes improves the FMR1 mRNA-CGG size relationship and also increases the slope for this relationship. In contrast, GUS/2IC and EIF4A2/SDHA values were not significantly correlated with CGG size (Fig 2D and 2E; Table 1). GUS/2IC value also showed greater variability between individuals across the full range of CGG repeat size (from control to PM) as compared to the EIF4A2/SDHA value. ; (E) SDHA/EIF4A2 mRNA ratio variability between groups is also presented. Note: Piecewise linear regression was used to find a threshold in A, B and C, resulting in two difference slopes above and below the threshold (the CGG repeat threshold is presented in bold italics). The regression coefficients and standard errors for these relationships are described in Table 1. https://doi.org/10.1371/journal.pone.0192151.g002

Discussion
This study demonstrates that correlation between FMR1 mRNA levels and CGG triplet repeat size is weakened when GUS is used as an internal control gene for mRNA analysis studies in PBMCs of PM females. Furthermore, using GUS mRNA as a sole control in the denominator led to an underestimation of FMR1 mRNA levels for PM females when compared to agematched controls. Stronger correlations were observed between FMR1 mRNA and CGG triplet repeat size with the slope of the relationship increased when FMR1 mRNA levels were normalised without GUS or when contributions from GUS were minimised by inclusion of the two internal control genes EIFA2 and SDHA. PM GUS mRNA levels (GUS/2IC) were significantly correlated with gait and verbal intelligence scores in PM but not control females [30]. However, in this study the GUS/2IC values: (i) did not significantly correlate with increased CGG triplet repeat size, and (ii) did not significantly differ between control and PM groups. GUS/2IC values were found to be more variable than the EIF4A2/SDHA ratio values in both PM and control groups between individuals. This suggested that GUS mRNA could be confounding if used to normalise FMR1 mRNA levels in RT-PCR experiments (Figs 1 and 2). This is consistent with the earlier stability study of mRNA levels for a panel of genes using the geNorm approach in another cohort [30]. This study showed that EIF4A2 and SDHA mRNA levels were more stable than that of GUS in control and PM males (Supplementary Figure 1 in [30]).
In this study, FMR1 mRNA level normalised to GUS alone resulted in an underestimation of FMR1 mRNA in samples that had high levels of GUS/2IC mRNA. On the other hand, for samples with very low levels of internal control gene mRNA, FMR1 levels were over-estimated. The inter-individual variability in GUS mRNA level detected in blood in both the control and PM groups could be due to biological or environmental factors that directly or indirectly influence the activity of the β-glucuronidase enzyme [34,35]. Furthermore, GUS mRNA stability level has been reported to be influenced by gender, age and cell type [36][37][38]. This limitation has been partly addressed here through the use of different internal control genes (i.e., EIF4A2 and SDHA) Normalisation of FMR1 mRNA levels in PM and control PBMCs with EIFA2 and SDHA (FMR1/2IC) has been demonstrated in previous studies examining relationships between FMR1 mRNA level and phenotype [39][40][41]. Specifically, in twenty PM females without FXTAS, higher FMR1/2IC values correlated with mean diffusivity within the middle cerebellar peduncle determined by diffusion-weighted imaging. FMR1/2IC value was also significantly correlated with poor performance on the Paced Auditory Serial Addition Test scores indicating executive dysfunction and/or slow processing speed [39].
In summary, these findings demonstrate that GUS normalisation in PBMC studies masks the relationship between FMR1 mRNA level and CGG triplet repeat size. It also artificially decreases values for FMR1 mRNA level in PBMC PM data, as compared to values normalised with EIF4A2 and SDHA levels. These findings may not apply to other tissue types, FMR1 alleles other than PM or other age groups. Instead, the most stable combinations of optimal internal control genes should be determined for each setting separately, using validated approaches such as geNorm [42]. This is consistent with the fact that there are no ideal internal control genes across all settings. This is now accepted widely in other fields of genetics [29,43] and should be implemented for investigations of Fragile X-associated disorders. Utility of absolute quantification of mRNA determined through methods that do not rely on internal control normalisation, including competitive PCR [44] and Droplet Digital PCR [45], may circumvent this problem in future studies.
Supporting information S1