Developmental Programming: adipose depot-specific regulation of non-coding RNAs and their relation to coding RNA expression in prenatal testosterone and prenatal bisphenol-A -treated female sheep

Abstract

Inappropriate developmental exposure to steroids is linked to metabolic disorders. Prenatal testosterone excess or bisphenol A (BPA, an environmental estrogen mimic) leads to insulin resistance and adipocyte disruptions in female lambs. Adipocytes are key regulators of insulin sensitivity and tissue-specific differences in insulin sensitivity, coupled with adipose depot-specific changes in key mRNAs, were previously observed with prenatal steroid exposure. We hypothesized that depot-specific changes in the non-coding RNA (ncRNA) - regulators of gene expression would account for the direction of changes seen in mRNAs. Non-coding RNA (lncRNA, miRNA, snoRNA, snRNA) from various adipose depots of prenatal testosterone and BPA-treated animals were sequenced. Adipose depot-specific changes in the ncRNA that are consistent with the depot-specific mRNA expression in terms of directionality of changes and functional implications in insulin resistance, adipocyte differentiation and cardiac hypertrophy were found. Importantly, the adipose depot-specific ncRNA changes were model-specific and mutually exclusive, suggestive of different regulatory entry points in this regulation.

1. Introduction:

Developmental insults contribute to increases in non-communicable diseases, including cardiovascular diseases (CVD), diabetes mellitus (DM) and several cancers [1–3]. Inappropriate exposure to endogenous steroids or environmental steroid mimics serves as one such insult [4, 5], that is known to program the metabolic axis [6]. Our studies with sheep demonstrated prenatal exposure to testosterone (T), an endogenous steroid or bisphenol A (BPA), the environmental steroid-mimic produces reproductive and metabolic perturbations that are characteristic of polycystic ovary syndrome-like phenotype in the female offspring (T [7]; BPA [8, 9]), albeit the phenotype is less severe in the BPA model. At the metabolic level, both prenatal T [10, 11] and BPA [8, 12]) models manifested insulin resistance and adipocyte dysfunctions (T [13]; BPA [8]). Specifically, prenatal exposure to excess T, an estrogen precursor, induced dyslipidemia, peripheral insulin resistance, ectopic lipid accumulation, and an increase in the distribution of small adipocytes in female offspring [14]. Prenatal exposure to BPA, a xenoestrogen with estrogenic action, induced insulin resistance, adipocyte hypertrophy and adipose depot-specific disruptions in markers of adipose differentiation in female offspring.

Adipose tissue, composed of adipocytes, is primarily a fuel reservoir that also serves as an endocrine organ, secreting several adipokines that regulate other metabolic tissues like muscle, liver and pancreas. Recent research has established the role of adipose tissue as a dominant regulator of glucose homeostasis and whole-body metabolism [15, 16]. The role of adipose depot as a major contributor to insulin resistance [17] is underlined by the observation that both excess of adipose, as seen in obesity [18] and deficiency of adipose, as seen in lipodystrophy [19] can lead to development of insulin resistance. Similarly, aberrant changes in adipocyte differentiation such as increased adipocyte size is also associated with the development of insulin resistance [20]. Because adipose tissues are compartmentalized into discrete depots and distributed throughout the body, there are functional depot-specific differences in their influence of tissue-specific insulin resistance [21]. Subcutaneous adipose tissue (SAT), the fat beneath the skin and visceral adipose tissue (VAT), the intra-abdominal fat, form the major adipose depots of the body. While SAT favors uptake and storage of lipids and is associated with a metabolically healthy phenotype, VAT promotes lipid turnover and is associated with cardiometabolic disorders [22]. The smaller depots that are associated with specific organ systems such as the epicardiac adipose tissue (ECAT) and perirenal adipose tissue (PRAT) can influence the organs in their proximity [23]. Perturbations in ECAT are associated with obesity-related insulin resistance [24], a risk factor for cardiovascular diseases [25] while disruption in PRAT is associated with cardiovascular diseases and chronic kidney disease risk [26, 27].

Knowledge of adipose depot-specific changes in gene transcription and their regulation are therefore much needed to delineate the roles they play in eliciting organ-specific changes in insulin sensitivity. Our recent studies with prenatal T-treated animals demonstrated adipose depot-specific transcriptional changes manifested as increased proinflammatory genes in VAT, reduced adipocyte differentiation genes in VAT and SAT, increased cardiomyocyte function gene expression in ECAT, and increased vascular related gene expression in PRAT [28]. Our studies with prenatal BPA-treated sheep, which focused only on VAT and SAT, found increased expression of genes involved in adipocyte development and differentiation and thermogenic brown/beige adipocyte development in the SAT, while in VAT proinflammatory genes and genes involved in adipogenesis and maintenance of insulin sensitivity were upregulated [29]. While these results from prenatal steroid (endogenous or environmental)-treated models demonstrate depot-specific regulation, the mechanisms by which these changes are facilitated is unknown.

Non-coding RNAs (ncRNA) like long non-coding RNA (lncRNA) and microRNA (miRNA) exert epigenetic control by regulating gene expression at the transcriptional and post-translational levels and serve as sensors of environmental insults [30]. Dysregulation of lncRNA [31–33] and miRNA [34] are important regulators of the pathological response to environmental exposures. Some lncRNAs are important regulators in adipogenesis and adipocyte metabolism and several miRNAs have been implicated in metabolic diseases [35]. Small nuclear RNA (snRNA) are involved in regulating gene expression by splicing [36] while small nucleolar RNA (snoRNA) guide posttranscriptional modifications on ribosomal RNA and snRNA [37]. snRNA and snoRNA are largely unexplored but are gaining importance for their potential role in adipogenesis and metabolic health [38, 39]. Given the emerging role of ncRNA in metabolic homeostasis, we hypothesized that prenatal exposure to excess T (an endogenous androgen) or BPA (an environmental estrogen-mimic) will induce steroid and adipose depot-specific disruptions in the ncRNA landscape that are consistent with the reported transcriptomic and phenotypic outcomes.

2. Methods

2.1. Animals:

Animal studies were conducted at the University of Michigan Sheep Research Facility (Ann Arbor, MI) using multiparous female Suffolk sheep. Studies were conducted following Institutional Animal Care and Use Committee of the University of Michigan approved protocol that met the requirements of the National Research Council’s Guide for the Care and Use of Laboratory Animals and the Animal Welfare Act. Two cohorts of sheep were included in this study, one examining the effects of prenatal T treatment, and one examining effects of prenatal BPA treatment. Animals from both cohorts (Cohort 1: control and prenatal T-treated; Cohort 2: control and prenatal BPA-treated) were co-inhabited under similar conditions, fed a similar maintenance diet to prevent obesity, and potential phytoestrogen exposure via diet was similar across treatment groups as described earlier [40].

2.2. Prenatal T Treatment:

Between gestational days 30–90, which is the sexually dimorphic window when fetal males see an increase in testosterone naturally, 100mg T propionate (~1.2mg/kg; Millipore Sigma, St. Louis, MO) suspended in corn oil was administered intramuscularly twice a week to the prenatal T-treated Suffolk sheep. Control animals did not receive any vehicle treatment, since our prior studies demonstrated no effects of corn oil in sheep [41]. From this cohort, four control and prenatal T-treated female sheep ensuring mother as the experimental unit were randomly selected for the current study. The effects of prenatal T-treatment on peripheral insulin sensitivity, adiposity, tissue specific changes in mediators of insulin sensitivity and transcriptome analysis of coding RNA from this cohort have been published [5, 13, 42–44].

2.3. Prenatal BPA Treatment:

Pregnant Suffolk sheep were randomly assigned to control and BPA treatment groups. Between days 30 and 90 of gestation, which is the sexually dimorphic window, control animals received vehicle (corn oil) and the BPA-treated group received 0.5mg/kg/day BPA (purity ≥ 99%, catalog number 239658; Aldrich Chemical Co, Milwaukee, Wisconsin) solubilized in corn oil [45]. Both injections of vehicle and BPA were administered daily, subcutaneously. The average BPA level achieved in the umbilical artery with this dose was ~2.6 ng/ml [45] and this concentration is within the range reported in human biomonitoring studies [46–48]. From this cohort, four control and prenatal BPA-treated female sheep, ensuring mother as the experimental unit, were randomly selected, and used in the current study. The effects of prenatal BPA-treatment on insulin sensitivity, adiposity, mediators of insulin sensitivity and transcriptome analysis of coding RNA from this cohort have been previously published [8, 12, 49].

2.4. Tissue Collection:

Animals from the T cohort were ovariectomized at the end of the second breeding season to remove confounding effects from differing steroid background. For Cohort 1, adipose depot samples were collected during the artificial follicular phase, as described previously [50]. Early follicular phase levels of estradiol was achieved by inserting a 1 cm estradiol implant subcutaneously [51]. An artificial luteal phase of the estrus cycle was initiated by placing two controlled internal drug-releasing implants containing progesterone (CIDR-G; InterAg, Hamilton, New Zealand) subcutaneously. After 14 days, progesterone implants were removed, four 30-mm estradiol implants inserted subcutaneously to mimic follicular phase and 18 hours later animals were euthanized by barbiturate overdose (Fatal Plus; Vortech Pharmaceuticals, Dearborn, MI) and tissues collected. Tissues from the prenatal BPA-treated cohort were collected from animals euthanized by barbiturate overdose during natural follicular phase following synchronization with two injections of PGF2a administered 11 days apart (27h after administration of second PGF2a). In the prenatal T cohort, four adipose depots were collected: VAT from the omental fat surrounding the ventral sac of the rumen, SAT from under the skin in the sternal region, PRAT from around the kidney, and ECAT from between the myocardium and the serous/visceral layer. In the prenatal BPA cohort, only SAT and VAT were collected. (Supplementary Table S1). Samples were immediately flash-frozen and stored at −80°C.

2.5. RNA extraction, library construction and sequencing:

Adipose tissue was homogenized with liquid nitrogen and dissolved in Trizol reagent (Life Technologies, Carlsbad, CA) and total RNA isolated as per manufacturer’s recommendations. Residual DNA was removed using the RNeasy binding column, treating with RNAse free DNAse (Qiagen, Germantown, MD) then eluting RNA in nuclease free water. RNA purity and integrity were measured with the Agilent 2100 bioanalyzer (Agilent Technologies, Santa Clara, CA). Libraries for ncRNA were prepared using the NEBNext smallRNA kit (New England BioLabs, Ipswich, MA) at the University of Michigan Advanced Genomics Core as per manufacturer’s recommendations. Sequencing was performed on an Illumina Nextseq platform.

2.6. Data Processing and Quality Control:

First, raw fastq files were 5’ trimmed with cutadapt (v3.2) [52] to remove the adapter sequence ‘AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC’ as per manufacturer’s recommendation. Additionally, reads that were of low quality and <17bps were removed from the sequencing data. FastQC (v0.11.5) [53] was performed on both the raw and trimmed reads, with reports summarized using MultiQC [54]. The trimmed reads were aligned to the sheep reference genome (Oar_rambouillet_v1.0) using the Spliced Transcripts Alignment to a Reference (STAR) program (v2.6.0c) [55]. Quality metrics of reads mapped to the genome were evaluated using Quality of RNA-Seq Tool-Set (QoRTS, v1.3.6) software [56]. FeatureCounts (v1.6.1) software [57] was used to count the number of reads of each ncRNA. For featureCounts, annotation files were subset to each class of ncRNA (miRNA, lncRNA, snRNA, and snoRNA). Subsequent analyses were stratified by class of ncRNA.

2.7. Differential RNA Expression:

Differential expression of ncRNAs were evaluated using the DESeq2 (v1.24.0) package in R statistical software, which performs linear regression modelling counts on a negative binomial distribution [58]. Normalization was done with default package settings. To evaluate similarity between the prenatal T cohort and the BPA cohort (artificial follicular phase vs. natural follicular phase), we compared expression levels between the SAT and VAT samples of the control animals across the two cohorts. To evaluate adipose depot-specific differences, control samples of each tissue type were compared against all the other tissue types within cohorts. In both cohorts, we examined differential impact of treatment stratified by adipose tissue types [4 adipose types in prenatal T (VAT, SAT, ECAT and PRAT) and 2 adipose types in prenatal BPA cohorts (VAT and SAT)]. Analysis was done separately for the prenatal BPA data and the prenatal T data, in addition to stratification by ncRNA class. The differentially expressed ncRNA for each evaluation was visualized using the EnhancedVolcano package [59]. Univariate analysis using volcano plots with cut-off value of adjusted p-value < 0.05 and absolute log-fold change > 0.5 was used to identify the significant ncRNA.

2.8. Multivariate analysis:

Data reduction was done using SIMCA version 17 (Umetrics, Umea, Sweden) software. Multivariate modeling was applied to the imported normalized counts for ncRNA from adipose tissues. Unsupervised principal component analysis (PCA) was performed to get an overview of the data and identify potential outliers and trends in the data. The principal components (PC) were displayed in two-dimensional and three-dimensional score plots to allow visualization of the distribution and grouping of the samples. Hoteling’s T-square statistic, a multivariate extension of the Student’s t-test, was employed to draw a tolerance ellipse around the sample cluster and any data outside the ellipse was considered outliers. No outliers were found in the data sets.

Differences between control and treatment groups were visualized using the supervised approach of orthogonal projections to latent structure discriminant analysis (OPLS-DA) using SIMCA version 17 software (Umetrics, Umea, Sweden). The variations arising from treatment were the first predictive component and variations not related to treatment were explained by the orthogonal components. In this model, the status of treatment (control and prenatal T or BPA) was the outcome (Y) variable, and the ncRNA expression data was the predictor (X) variable and association between these are highlighted by filtering the orthogonal variation. The differences in the transcriptome profiles were summarized in the OPLS-DA score plots that were developed for the first predictive component and the first orthogonal component and each point on the score plot represents one animal. Q² and R² values indicate the robustness of the model with values above 0.5 indicating a good model. The generated OPLS-DA models were validated using a 100-iteration permutation test to avoid over-fitting and false-positives. The model was considered valid if all Q² values from the permuted data set are lower than the Q² values on the actual dataset. The variables of importance in projection (VIP) scores reflect the contribution of the ncRNA to the model and a cut-off of VIP>1 was used in this study.

2.9. Correlation between ncRNA and coding RNA:

Data on coding RNA from the same samples have been previously published [29, 60]. We matched ncRNA from the current study with this previous data. The Differential Gene Correlation Analysis (DGCA) package [61] was used to determine putative ncRNA-mRNA pairs. As an exploratory analysis liberal cutoff criteria were used for consideration in analysis. Any ncRNA with adjusted p-value < 0.1 in the current study, and coding RNA from the previously published studies with adjusted p-value < 0.1 and absolute log fold change > 1.0 were evaluated using correlation matrices. Among these ncRNA and mRNA, ncRNA-mRNA pairs that met the Pearson correlation coefficients with p-values <0.05 across control and treatment groups were retained and visualized. Box plots for the gene expression levels were generated using the web tool BoxPlotR [62].

2.10. Data and code availability:

The new ncRNA sequencing data are publicly available through the National Institutes of Health’s Genome Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/;accession number GSE219111). The re-analyzed coding RNA sequencing data remain publicly available through the Genome Expression Omnibus (accession numbers: GSE158436 for T-treatment and GSE142222 for BPA-treatment). Code to conduct the current analysis is publicly available through GitHub (https://github.com/bakulskilab/Developmental-Programming-adipose-depot-specific-regulation-of-non-coding-RNAs).

3. Results:

3.1. Adipose tissue specific differential expression in controls:

The total number of non-coding RNA identified in the control vs. treatment groups of prenatal T-treated BPA-treated cohorts are listed in Supplementary Table S2. Comparison of ncRNA expression across the four adipose depot types in control samples from the prenatal T-model found VAT to be the least distinct, when compared to the other three adipose types, with no ncRNA reaching significance. Also, snRNA patterns were not significantly different among the other three adipose depot types (Figure 1, left panel). In contrast, expression levels of lncRNA, miRNA, and snoRNA in ECAT, PRAT, and SAT differed from the other three adipose types. Differential expression analysis results from all depots of prenatal T cohort are included in Supplementary Table S3. Comparison of ncRNA profiles in the SAT and VAT depots in controls from the prenatal BPA-model identified 9 lncRNAs, 25 miRNAs, 45 snoRNAs and 2 snRNAs that were differentially expressed between these two depots (Figure 1, right panel). Differential expression analysis results for SAT and VAT comparison in the prenatal BPA cohort are listed in Supplementary Table S4.

Figure 1: Differential expression of ncRNAs in adipocyte tissue types in controls.

Volcano plot representation of differential expression of lncRNA, miRNA, snoRNA, and snRNA in ECAT, PRAT, SAT and VAT of the prenatal T-treated cohort is represented in the left panel and that between SAT and VAT in the prenatal – BPA-treated cohort in the right panel. ncRNA are plotted by log₂ fold change on the x-axis and −log₁₀ adjusted P values on the y-axis. Pink points represent the genes that have an absolute log₂ fold change greater than 0.5 and adjusted P values less than 0.05. Yellow dots represent genes that met the absolute log₂ fold change greater than 0.5 but did not meet the P-adjusted cutoff of less than 0.05, blue dots represent genes that met P-adjusted cutoff of less than 0.05 but did not meet absolute log₂ fold change greater than 0.5. Grey dots represent genes that did not meet either cutoff criterion.

3.2. Prenatal-T treatment effect:

The effect of prenatal T-treatment in each adipose depot is discussed below by ncRNA class. Differential expression analysis in ECAT identified ten significant miRNAs. Specifically, prenatal T-treatment increased the expression of six miRNAs (miR-99A, miR-10B, miR-143, miR-3959, miR-199A, and miR-218A), while decreasing expression of four other miRNA (miR-150, miR-485, miR-433, and miR-543) in ECAT. In PRAT, prenatal T-treatment decreased the expression of the miRNA, miR-133. In VAT, prenatal-T treatment increased the expression of the miRNA, miR-3955 and four snoRNA (LOC114117654, LOC114110216 and LOC114110211 of the SNORA18 family and LOC114112592) (Figure 2).

Figure 2. Prenatal -T treatment related differential expression of ncRNA by adipose tissue type:

Volcano plot representation of differential expression. a. miRNA in ECAT, b. miRNA in PRAT, c. miRNA in VAT, and d. snoRNA in VAT. ncRNA are plotted by log₂ fold change on the x-axis and −log₁₀ P adjusted values on the y-axis. Pink points represent the genes that have an absolute log₂ fold change greater than 0.5 and P adjusted values less than 0.05. Grey dots represent genes that did not meet P-adjusted cutoff of less than 0.05 and absolute log₂ fold change greater than 0.5 and yellow dots represent genes that met the absolute log₂ fold change greater than 0.5 but did not meet the P-adjusted cutoff of less than 0.05.

Multivariate analysis identified overall T-treatment specific differences based on ncRNA class. Although unsupervised PCA on the first two dimensions did not show clear separation between the control and T-treatment, the 3-D PCA score plot showed clear clustering of the control samples in miRNA, snRNA and snoRNA in the ECAT and miRNA in PRAT (Figure 3). This was also confirmed by the supervised OPLS-DA models with R² and Q²Y values above 0.6 indicating a robust model and validated by a permutation test. SAT did not show a separation of samples based on treatment in any of the ncRNA classes (Figure 4-left panel). miRNA and snoRNA in VAT showed separation between the control and prenatal T-treated groups, based on validated OPLS-DA models (Figure 4 – right panel). Other tissues and ncRNA classes did not meet significance (Supplementary figure S1). Table 1 lists all non-coding RNA with differential expression by T treatment with adjusted p-value < 0.1. Complete list of all ncRNAs is included in Supplementary Table S5.

Figure 3. Overall patterns of ncRNA expression in ECAT and PRAT in response to prenatal T-treatment:

PCA analysis - 2D and 3D score plots and Orthogonal projections to latent structure discriminant analysis (OPLS-DA) score plots on miRNA, snoRNA and snRNA.

Figure 4. Overall patterns of ncRNA expression in SAT and VAT in response to prenatal T-treatment:

Principal Component Analysis (PCA) - 2-Dimensional(2D) and 3-Dimensional (3D) score plots and Orthogonal projections to latent structure discriminant analysis (OPLS-DA) score plots on lncRNA, miRNA, snoRNA, and snRNA.

Table 1. Differential expression of non-coding RNAs in ECAT, PRAT and VAT in prenatal T-treated cohort of sheep. (No differences were evident in SAT).

ECAT miRNA
Gene	Control	Prenatal T	log2FC	SE	P-value	P-adj#
miR-99A	49268.53	101039	0.91	0.3	2.97E-04	0.02
miR-150	4801.36	2435.59	−0.83	0.3	7.32E-04	0.02
miR-485	236.96	107.25	−0.96	0.35	5.66E-04	0.02
miR-10B	509.3	1272.02	1.05	0.46	1.54E-03	0.03
miR-143	13823.15	35820.35	1.11	0.46	1.13E-03	0.03
miR-3959	34.56	107.86	1.28	0.58	1.40E-03	0.03
miR-433	160.52	78.3	−0.83	0.36	1.87E-03	0.03
miR-199A	8914.48	16225.99	0.72	0.3	2.53E-03	0.03
miR-218A	106.18	247.73	0.95	0.45	2.53E-03	0.03
miR-543	244.51	124.22	−0.78	0.35	2.77E-03	0.03
miR-148A	2479.8	4846.34	0.74	0.39	7.23E-03	0.06
miR-LET7B	129311.9	67193	−0.69	0.39	8.60E-03	0.07
miR-152	436.75	836.57	0.67	0.4	1.18E-02	0.08
miR-29B-1	34.52	67.9	0.7	0.44	1.22E-02	0.08
miR-29B-2	34.52	67.9	0.7	0.44	1.22E-02	0.08
miR-LET7F	11419.44	23705.63	0.72	0.48	1.40E-02	0.08
miR-381	2.29	8.93	0.68	1.34	1.56E-02	0.09
miR-10A	245.72	459.34	0.61	0.41	1.86E-02	0.09
PRAT miRNA
miR-133	1039.3	209.15	−1.98	0.64	7.95E-05	0.008
VAT miRNA
miR-3955	2.19	13.15	2.17	0.85	4.58E-04	0.047
VAT snoRNA
LOC114117654	0.18	13.2	4.43	1.47	1.02E-04	0.013
LOC114110216	0.18	13.26	4.43	1.5	1.27E-04	0.013
LOC114110211	2.1	24.4	3	1.02	9.88E-05	0.013
LOC114112592	0.18	13.26	4.43	1.5	1.27E-04	0.013
LOC114114240	1.72	18.63	2.71	1.22	6.62E-04	0.05

^#-ncRNAs with adjusted P values < 0.1 are represented in the table

Developmental Programming: adipose depot-specific regulation of non-coding RNAs and their relation to coding RNA expression in prenatal testosterone and prenatal bisphenol-A -treated female sheep

Correlation between ncRNA and coding RNA: Differential gene correlation analyses revealed that the expression of several coding RNA was significantly correlated with that of ncRNA. In ECAT, expression of miR-3959 was correlated with that of HOXB6, while miR-381 correlated with TNRC6C expression and miR-485 with LMOD1 expression (Figure 5 -left panel). In VAT, miR-3955 expression correlated with ITGB2 expression while snoRNA of the SNORA25 family - LOC114110211 expression correlated with TRDN and UGGT1 and LOC114114240 expression correlated with that of LMTK3 (Figure 5- right panel).

Figure 5: Correlation between coding and ncRNA in response to prenatal T-treatment in ECAT and VAT:

Left panel shows box plots representing expression level of ncRNA and mRNA and right panel shows line plots representing correlation of expression between ncRNA and mRNA in ECAT and VAT. Correlation between ncRNA-mRNA, for each pair is represented in orange line for Control animals and as a blue line for prenatal T-treated animals.

3.3. BPA treatment effect:

Analyses of ncRNA classes In VAT found one lncRNA (LOC105602517) to be marginally increased (adjusted p-value < 0.1) and two snRNAs (LOC114117093 and LOC114117060) to be significantly increased in the BPA-treated group (Figure 6 – top panel). Multivariate analysis by PCA and OPLS-DA of miRNA expression showed significant separation between control and BPA-treated groups in VAT (Figure 6 – bottom panel). Other ncRNA classes in VAT and all ncRNAs in SAT did not meet significance criteria (Supplementary Figures S2 and S3). Full results for all ncRNA are included in Supplementary Table S6. The ncRNA that had significant (adjusted p-value < 0.05) or marginal (adjusted p-value<0.1) changes in expression were not correlated with coding RNA expression levels. Log-fold change effect estimates between prenatal T treatment and BPA treatment were not strongly correlated (Supplementary Figure S4).

Figure 6: ncRNA expression in VAT in response to prenatal BPA treatment:

Top panel- Volcano plot representation of differential expression of lncRNA in VAT and snRNA in VAT. ncRNA are plotted by log2 fold change on the x-axis and −log10 P adjusted values on the y-axis. Pink points represent the genes that have an absolute log2 fold change greater than 0.5 and P adjusted values less than 0.05. Grey dots represent genes that did not meet P-adjusted cutoff of less than 0.05 and absolute log₂ fold change greater than 0.5 and yellow dots represent genes that met the absolute log₂ fold change greater than 0.5 but did not meet the P-adjusted cutoff of less than 0.05. Bottom panel - Principal Component Analysis (PCA) - 2D and 3D score plots and Orthogonal projections to latent structure discriminant analysis (OPLS-DA) score plots on miRNA

3.4. Model effect:

Comparison of ncRNA expression profiles in the SAT and VAT depots of controls from the two cohorts (artificial follicular phase [T-treatment cohort] and natural follicular phase [BPA-treatment cohort] identified only one significant miRNA – miR-374b in the VAT (Supplementary Figure S5) and none in the SAT. Complete list of all ncRNA are listed in Supplementary Table S7.

4. Discussion:

In parallel with our earlier findings of adipose depot-specific mRNA changes in sheep exposed prenatally to T (an endogenous steroid) or BPA (an environmental steroid mimic) [28, 29], adipose depot-specific changes in ncRNA profiles were also evident. While specific relationships between all individual ncRNA and coding genes could not be established, owing to the underlying limitations of working with the marginally annotated sheep genome [63], these findings point towards a possible depot-specific regulation of transcription at the epigenetic level, involving ncRNA. This is the first report of the impact of exposure to inappropriate maternal steroid milieu (T or BPA) on multiple ncRNAs namely, miRNA, lncRNA, snRNA and snoRNA at the level of various adipose tissue depots. The implications of the adipose depot-specific changes in the ncRNA profile are discussed below.

4.1. Depot specific differences in ncRNA expression among control animals

4.1.1. Controls in the prenatal T- model:

Expression profiling of ncRNA in the various adipose depots in control animals found the least differences in the VAT and the most in the ECAT depot when compared to the other depots. This parallels our previous results from the transcriptional profiling of the coding RNAs in the 4 different adipose depots from the same animals, that showed VAT had the least and ECAT had the largest number of differentially expressed coding genes relative to the other adipose tissue depots [28]. The ECAT transcriptome is known to have a unique profile compared to other adipose depots [64, 65] and consistent with this, it also has a unique ncRNA profile. This may relate to the fact that ECAT is physiologically [66] and functionally divergent from the other fat depots, serving as energy source of the myocardium with cardioprotective properties [67]. ECAT also had the largest number of snoRNA that were differentially expressed compared to the other adipose depots, although their specific function in ECAT is unknown. Functionally, snoRNA are well-conserved, housekeeping molecules that maintain ribosomal maturation and translation [68] and several snoRNAs play a role in metabolism in adipose tissue [69, 70]. ECAT and SAT had the greatest number of differentially expressed miRNA compared to the other adipose depots. There were 10 miRNA – miR- 103, miR-127, miR- 133, miR- 152, miR- 329A, miR- 329B, miR- 369, miR- 370, miR-411A, miR-487B that were differentially expressed in both ECAT and SAT depots. miR-103, a key player in white adipose tissue differentiation and function [71], and miR-152, positively correlated with adipogenesis [72], were over-expressed in SAT and downregulated in ECAT compared to the other adipose depots. In contrast, miR-127, which attenuates adipogenesis [73] and miR-133, which regulates adipocyte browning [74] were over-expressed in ECAT and downregulated in SAT. The other identified miRNAs had no documented functional relationship with adipocytes. The change in direction of expression of these key ncRNA involved in adipogenesis and adipose browning between the ECAT and SAT corroborate the underlying functional differences between these two depots. While SAT is predominantly composed of white adipocytes, ECAT has a higher proportion of brown adipocytes. snRNA, which plays an important role in intron splicing and other mRNA pre-processing [75], did not differ between the 4 depots, suggestive of a similar functional role in all adipose tissue depots. In terms of lncRNA, PRAT showed the most changes relative to the other adipose depots, although the functions of these in PRAT remains to be identified. In general, lncRNA appear to play a role in adipogenesis [76] and has been implicated in insulin resistance [77] and type 2 diabetes [78].

4.1.2. Controls in the prenatal BPA model:

Comparison of ncRNA profiles in the SAT and VAT revealed differences in expression of all the four types of ncRNA. Amongst the miRNAs that were downregulated in VAT compared to SAT, miR-10B [79] and miR-26B [80] suppress adipogenic differentiation and miR-152 is positively correlated with adipogenesis [72]. Upregulated miRNAs in VAT include miR-432 known to inhibit adipose differentiation [81], miR-127 that attenuates adipogenesis [73], and miRs-200b [82] and −218A [83] that facilitate the proliferation but suppress the differentiation of preadipocytes. In contrast, miRNA upregulated in SAT include miR-148A [84] and miR-21 [85] that promote adipogenic differentiation, miR-17 -a part of a cluster that accelerates adipocyte differentiation [86], miR-221 [87], miR-23A [88], and miR-27A [89] that reduce adipocyte lipid storage and differentiation, and miR-29A that plays a pivotal role in glucose transport and lipid metabolism [90]. While both represent white adipose depots, these differences in expression of various miRNAs in VAT and SAT likely reflect their functional differences. This is consistent with the key differences that prevail between VAT and SAT in their gene expression profiles [29], lipid metabolism, adipokine secretion [91] and insulin sensitivity [92]. The functional role of the lncRNA, snoRNA and snRNA identified to be differentially expressed between the two depots is unknown.

4.2. Comparison of the impact of T (an endogenous steroid) and BPA (an environmental steroid mimic) on ncRNA profile in adipose tissue depots

The functional similarities between the prenatal T and BPA models led us to suspect possible overlap in ncRNA profiling in the VAT and SAT depots of both the models, with expansion to ECAT and PRAT in the prenatal T model.

4.2.1. Prenatal T model:

The depot-specific changes in expression of several ncRNA parallel their functional differences. With regard to changes in ncRNA expression in the ECAT, many of the dysregulated miRNAs are involved in glucose metabolism and insulin resistance, which are contributors to cardiometabolic disorders [93]. Phenotypically, prenatal T-treated females exhibit hypertension [94] and left ventricular hypertrophy along with an increased expression of genes involved in insulin signaling in left ventricular tissue [95]. Several miRNAs altered by prenatal T excess have been associated with cardiac hypertrophy. For instance, overexpression of miR-99A, which is regulated by insulin [96] and plays an important role in glucose metabolism [97, 98], is known to attenuate cardiac hypertrophy [99, 100] . Similarly, miR-143 over-expressed in ECAT, is involved in insulin resistance in brown adipose tissue [101] and adipocyte differentiation [102] and is also over-expressed in obesity-induced cardiac hypertrophy in female mice [103]. Another upregulated miRNA, miR-10B, is upregulated in ventricular tissue of hypertrophic cardiomyopathy patients [104] . Other members of the miR-10b family are differentially expressed in the ECAT of coronary artery disease patients [105]. In addition, upregulation of this mRNA is shown to be associated with obesity associated dyslipidemia and adipokine expression [106] and reduction of thermogenic brown adipocyte differentiation in humans [107], features that promote insulin resistance. Among the downregulated miRNAs in ECAT, miR-150 is associated with cardiac hypertrophy [108]. The downregulated miR-485, also observed in the ECAT of patients with coronary atherosclerosis [109] is known to suppress hypertrophy in cardiomyocytes [110]. In line with the transcriptome profile of ECAT in prenatal-T treated sheep, which showed dysregulation of genes involved in cardiomyocyte function, the miRNA dysregulated in ECAT like miR-433 [111], miR-199 [112] and miR-143 [113] also play pivotal roles in cardiomyocyte function. Thus, the miRNA profile in ECAT is consistent with the observed phenotypic changes involving hypertrophy [95]. None of the other ncRNAs examined including lncRNA, snRNA, and snoRNA were impacted by prenatal T excess in the ECAT.

In PRAT, the decreased expression of miR-133, a well-known brown adipogenesis inhibitor [74, 114], is consistent with the increased expression of the thermogenic gene UCP1 indicative of increased browning [43]. The functional role of the miR-3955 and the 4 snoRNAs that are overexpressed in VAT tissue of prenatal T-treatment is unknown. Interestingly, SAT which showed an upregulation of chromatin modification pathways in the transcriptome analysis [28] did not have any significant differentially expressed ncRNA in response to prenatal T-treatment. Prenatal T-treatment did not result in a difference in the expression of any snRNA or lncRNA in any of the adipose depots, indicating the possible involvement of other epigenetic mechanisms of regulation like DNA methylation or histone modification.

4.2.2. Prenatal BPA model:

Although transcriptome analyses of adipose depots in response to prenatal BPA treatment had identified upregulation of chromatin modeling pathways and mRNA processing pathways in VAT and RNA splicing pathways in SAT [29], similar to findings in the prenatal T model, BPA treatment had no effect on any of the ncRNA in SAT depot. In contrast, the VAT showed increased expression of one lncRNA and two snRNA (no impact on miRNA), the roles of which remain to be functionally characterized. ECAT and PRAT were not studied in this model.

Comparing the impact of prenatal impact of T and BPA in VAT and SAT depots (only depots studied in both models), there was no impact of either treatment on any of the ncRNA studied in the SAT. In contrast, in the VAT, prenatal T resulted in overexpression of miR-3955 and 4 snoRNAs while prenatal BPA exposure resulted in the overexpression of one lncRNA and two snRNAs with no overlap between the two models, indicative of a mutually exclusive outcome in the ncRNA profile. While prenatal T had no effect on snRNA and lncRNA, prenatal BPA had no effect on miRNA and snoRNA expression, suggestive of mediation via different steroid receptor systems and involvement of other epigenetic modulators.

4.3. Correlation between coding RNA and ncRNA

Non-coding RNA play an important role in regulating the coding RNA expression by several mechanisms [115] and correlation of miRNA and mRNA expression has helped in comprehending the biological mechanisms and regulatory processes involved in metabolic disorders [116, 117].

4.3.1. Prenatal T model:

In ECAT, the changes in expression of three miRNAs were correlated with coding RNA expression. The expression of miR-3959, which has an important role in energy metabolism [118] was correlated with that of HOXB6, identified as a pre-adipocyte precursor gene [119]. miR-381 expression which has a role in brown adipocyte differentiation [120] correlated with expression of TNRC6C which plays a role in RNA-mediated gene silencing by miRNA [121] and circular RNA degradation [122]. miR-485 has a role in adipocyte differentiation [123] and its expression correlated with that of LMOD1, which is involved in smooth muscle differentiation and associated with coronary artery disease [124, 125]. In VAT, the expression of one miRNA and two snoRNAs correlated with the expression of coding genes. miR-3955 expression correlated with that of ITGB2, known to be upregulated in obesity [126], type 2 diabetes [127] and insulin resistance [128]. ITGB2 encodes an integrin beta chain that is a part of integrin heterodimers that regulate white adipose tissue insulin sensitivity and brown fat thermogenesis [129]. The expression of the snoRNA LOC114110211 correlated with TRDN and UGGT1 gene expression. While TRDN is upregulated in obesity [130] and diabetes [131], UGGT1 is a glycoprotein glucosyltransferase whose expression is associated with obesity in humans [132] and increased in obese - type 2 diabetes rat model [133], indicating a potential role for LOC114110211 in regulating metabolism. LOC114114240 expression correlated with the expression of a tyrosine kinase LMTK3, which phosphorylates the estrogen receptor [134], shown to regulate insulin sensitivity [135], beiging of adipocytes [136] and has a central role in energy homeostasis [137].

4.3.2. Prenatal BPA model:

There were no significant correlations between the ncRNA and any of the significant differentially expressed coding RNA identified from the earlier total transcriptome analysis. This may be because of the limited number of ncRNA that were identified as differentially expressed in this model.

4.4. Impact of prenatal T vs. BPA on ncRNA profiles- differences

There was very little overlap relative to the impact on prenatal T vs. BPA on VAT and SAT ncRNA expression pattern, although these animals manifested similar metabolic outcomes in terms of their insulin resistance albeit, of different severity. These differences in ncRNA profile may relate to the modes of action of T and BPA. The androgen, T, signals through androgen receptor but has the potential to be aromatized and activate estrogen receptor. Indeed, prenatal T-treatment increases estradiol levels in the female fetuses [138]. On the other hand, BPA is a weak estrogen and shown to be an androgen antagonist [139]. Both models manifest hyperinsulinemia that have the potential to work through insulin receptor [140]. Alternatively, the T model used an artificial follicular phase model where animals are ovariectomized and replaced with steroids mimicking that seen in the ovary intact model. Another difference was that the T model received twice weekly intramuscular injections of T and no vehicle treatment for controls, while BPA model received daily subcutaneous injections of BPA with controls receiving daily vehicle treatment. However, comparison of ncRNA profile to address the underlying differences in the controls of the two cohorts found no difference in the SAT and marginal difference in VAT (1 miRNA) indicating the two cohorts (ovary intact and artificial follicular phase models) are comparable and vehicle treatment differences does not exist. Our earlier study has also indicated the absence of phenotypic changes in response to vehicle treatment in sheep [41]

4.5. Strengths and Limitations

Adipose depots are recently gaining significance for their pivotal role in regulating energy metabolism and insulin resistance [141, 142]. In addition to the established roles of VAT and SAT, ECAT and PRAT have emerged as important risk factors for cardiometabolic disorders [26, 143]. While VAT and SAT have been extensively studied, molecular studies on ECAT and PRAT from humans are scarce due to the limited availability of these tissues for molecular analyses, that has resulted in a limited understanding of the underlying biology of these adipose depots. There is a gap in research on studies comparing the different adipose depots apart from a few in mouse models [144, 145] and a meta-analysis [146]. Our study gives a comprehensive understanding of the ncRNA expression profile in all the four adipose depots in response to prenatal T treatment and in VAT and SAT on prenatal BPA treatment. The observed depot-specific ncRNA changes reiterate that molecular and functional differences exist between the various adipose depots. Although both models manifest an insulin resistance phenotype, the adipocyte impact from ncRNA in VAT and SAT is mutually exclusive. Another strength is correlating the ncRNA expression changes with the coding RNA changes in the same samples to delineate their potential contribution to regulation of coding genes [28, 29]. An important advantage of this study is the use of a large precocial animal model that has a translational relevance to humans [147], owing to similarities in their developmental trajectories.

A major limitation of this study is the non-availability of comprehensive annotation details for sequences from sheep and lack of sufficient functional characterization. The miRbase, which is a comprehensive database of miRNA including precursor and mature mRNA, includes 4571 miRNAs for humans, 3212 miRNAs for mouse [148] but only 259 for sheep. The same is true for the other classes of ncRNA. This has been a major deterrent to identify more differential ncRNA and make functional correlations of the observed ncRNA differences in our study, as some of the differential ncRNA that we identified were uncharacterized. To facilitate future sheep annotation studies, we made our raw sequencing data public. The use of a small number of samples for the analyses may have also limited the identification of more differentially expressed ncRNAs. Paradoxically, a smaller number of snoRNA that were identified in the earlier total transcriptome analysis of the coding genes in the prenatal -T and -BPA cohorts [28, 29] were not identified in the current study. This may be due to the difference in the RNA extraction and sequencing analysis procedures for the two analyses.

4.6. Translational significance:

Metabolic disorders like diabetes [149], hypertension [150], and cardiovascular disorders [151] have been on the rise in low- and middle-income countries. Different adipose depots have specific roles to play in the development of metabolic disorders [152, 153] and adipose depot measures have been identified as early risk factors for cardiovascular disorders [154, 155], hypertension [156], and diabetes [157]. In fact, transplantation of PRAT from healthy donors has been proposed as a therapeutic option to treat metabolic disease [158]. While structural, functional and molecular differences between VAT and SAT [159] have been identified in humans and mouse models [160], similar comparisons in other depots including ECAT and PRAT are minimal in humans and are restricted to indirect measurements of fat depots with imaging and spectroscopy [160]. In this context, our study that document differences in the ncRNA profile across four different depots that parallel the direction of changes in coding RNA profile in a sheep model of PCOS phenotype, is of relevance to organ-specific regulation exerted by specific adipose depots. The literature available on VAT depots in humans is primarily from obese individuals undergoing bariatric surgery [161] or cardiac or abdominal surgery [162]. In contrast, our data from animals raised on maintenance diet to avoid confounding raising from obesity [14] may be of a broader applicability.

5. Conclusions:

This study illustrates that prenatal T and BPA treatments program adipose depot-specific ncRNA changes in sheep. We identified changes in ncRNA that link to the previously reported transcriptional changes in these models. This also suggests prenatal exposure to the endogenous steroid, testosterone and environmental steroid mimic, bisphenol A could have different epigenetic mechanisms of action on the different adipose tissue depots. Overall, our study presents a comprehensive overview of the ncRNA landscape of the different adipose depots.

Supplementary Material

1

S1. ncRNA expression in ECAT and PRAT in response to prenatal T treatment: PCA analysis - 2D and 3–3D score plots on lncRNA in ECAT and lncRNA, snoRNA, snRNA in PRAT.

2

S2. ncRNA expression in SAT in response to prenatal BPA treatment: PCA analysis - 2D and 3D score plots on lncRNA, miRNA, snoRNA and snRNA.

3

S3. ncRNA expression in VAT in response to prenatal BPA treatment: PCA analysis - 2D and 3D score plots on lncRNA, miRNA, snoRNA and snRNA.

4

S4.Effect estimates of prenatal BPA and T treatments on ncRNA: Scatter plots comparing log 2-fold change effect estimates for prenatal BPA and T treatments. Comparisons done for lncRNA, miRNA, snRNA, and snoRNA in SAT and VAT.

5

S5.Non-coding RNA differential expression by model: Volcano plot representation of differential expression of lncRNA, miRNA, snRNA, and snoRNA in SAT and VAT between the control samples of the prenatal T (artificial follicular phase model) and prenatal BPA (natural follicular phase model) cohorts. ncRNA are plotted by log₂ fold change on the x-axis and −log₁₀ P adjusted values on the y-axis. Pink points represent the genes that have an absolute log₂ fold change greater than 0.5 and P adjusted values less than 0.05. Grey dots represent genes that did not meet P-adjusted cutoff of less than 0.05 and absolute log₂ fold change greater than 0.5 and yellow dots represent genes that met the absolute log₂ fold change greater than 0.5 but did not meet the P-adjusted cutoff of less than 0.05.

6

Supplementary Table S1. Study design of prenatal T and prenatal BPA cohorts.

7

Supplementary Table S2. Number of the different ncRNAs expressed in the various adipose depots of prenatal T- and prenatal BPA-treated cohorts.

8

Supplementary Table S3. Differential expression of ncRNAs by adipose depot in controls (prenatal T cohort). Each adipose type was compared to all other types, in control animals.

9

Supplementary Table S4. Differential expression of ncRNAs by adipose depot in controls (prenatal BPA cohort). VAT was compared to SAT, in control animals.

10

Supplementary Table S5. Differential expression of ncRNAs by prenatal T treatment. Comparisons done within adipose depot type and class of ncRNA.

11

Supplementary Table S6. Differential expression of ncRNAs by prenatal BPA treatment. Comparisons done within adipose depot type and class of ncRNA.

12

Supplementary Table S7. Differential expression of ncRNAs between prenatal T and prenatal BPA cohorts. Each adipose depot was compared to the corresponding depot between the prenatal T and prenatal BPA cohorts in control animals.

Highlights:

Prenatal testosterone (T) excess programs adipose depot-specific changes in ncRNA.

Prenatal T induced ECAT miRNA changes is consistent with its cardioprotective role.

Prenatal T induced VAT ncRNA changes link to metabolism regulating coding genes.

Prenatal bisphenol A (BPA) programmed ncRNA changes was specific to VAT not SAT.

ncRNA changes in pre-T and BPA models indicate different mechanisms of action.