Identi ﬁ cation and characterization of the long non-coding RNA NFIA-AS2 as a novel locus for body mass index in American Indians

U.S


INTRODUCTION
Obesity is a primary health concern that affects people from all ethnic backgrounds but there are considerable differences in its prevalence. A notable portion of this variation is due to the response of genes to an obesogenic environment which promotes weight gain during a high-calorie state [1][2][3][4][5]. Such genes can be identified by studying populations with a high prevalence of obesity, minimal genetic admixture, and less environmental variability. American Indians living in a southern Arizona community exhibit these features [6,7], and many of these community members have participated in a longitudinal study of health in which height and weight were measured to calculate body mass index (BMI), a classical parameter to assess obesity. To identify obesity susceptibility genes in this community, our prior genetic studies have included replication of established BMI loci as well as genome-wide association studies to uncover new loci. However, a major challenge with identifying new loci for a trait in a relatively isolated population is that the sample size in a population isolate is limited; thus, genes with moderate effect sizes which influence polygenic diseases/traits will rarely achieve genome-wide significance. While well-established variants for BMI that show moderate associations are considered valid contributors in small samples, it is difficult to interpret the credibility of unreported variants that associate with BMI with comparably moderate statistical evidence. The focus of the current study is to identify loci for BMI in our American Indian sample that also have some effect in the well-powered Genetic Investigation of ANthropometric Traits (GIANT) consortium dataset. This strategy identified several known genes for BMI; however, among the genes never implicated with BMI was NFIA-AS2, a primate-specific long noncoding RNA (lncRNA).

METHODS Imputation and association analyses
Genotypes and BMI for American Indians were obtained from a community-based longitudinal study in Arizona from 1965 to 2007, where study exams included measurements of a 75-g oral glucose tolerance test for type 2 diabetes (T2D) status defined by 1997 ADA criteria [8] and BMI. Since T2D is very common in this population and T2D progression and its treatment can affect BMI, the measure of BMI analyzed in the current study was the highest BMI recorded at any longitudinal examination at age ≥15 years when the volunteer was determined to be free from diabetes {N = 5870, mean age = 29.7 ± 11.4 years, mean BMI = 35.2 ± 8.4 kg/m 2 ; among these N = 2571 were male, mean age = 29.1 ± 11.4 years, mean BMI = 33.9 ± 8.0 kg/m 2 ; and N = 3299 were female, mean age = 30.2 ± 11.5 years, mean BMI = 36.1 ± 8.5 kg/m 2 }. The study was approved by the Institutional review board of the National Institute of Diabetes and Digestive and Kidney Diseases (National Institutes of Health) and all participants provided written informed consent.
Study volunteers were previously genotyped using a custom genotyping array (Affymetrix) designed specifically for this population based on wholegenome sequence data. Variants on this array that passed all quality control metrics (N = 515,723) captured 91% of all common variants with minor allele frequency (MAF) ≥ 0.05 and 56% of low-frequency variants (MAF = 0.01 − 0.05) across the genome [4,9]. These 515,723 single nucleotide polymorphisms (SNPs) were used as a scaffold for imputation using the reference panel of 296 American Indians which were genotyped by whole-genome sequencing. For the reference panel, sequencing reads were aligned to the Human genome build Hg38 using bwa; [10] genotype calling was made using two walkers HaplotypeCaller and GenotypeGVCF of GATK version 3.5 [11]. For imputation, we removed the genomic variants with more than 2 alleles from the reference panel. We adopted the best practices for pre-phasing and imputation as suggested in the IMPUTE2 website (https://mathgen.stats.ox.ac.uk/impute/impute_v2.html). In brief, we phased our genotype data by estimating haplotypes by SHAPEIT [12] using the duoHMM method which incorporates pedigree information in our data for a more accurate estimate of the haplotypes. The genotypes with Mendelian errors were set to missing, which could be imputed back together with the missing calls, using other samples as reference. The phased data was divided into 5 Mb segments and the missing genotypes in the reference panel were typed with IMPUTE2 executed using default parameters [13]. The imputed allele dosage data was transformed to the long text format understood by SAS version 9 [14]. Imputed variants with MAF ≥ 0.05 and imputation INFO score ≥ 0.5 were analyzed for association with rank normalized BMI values using normalizing inverse Gaussian transformation. The association between BMI and genotype dosages (assuming additive model) was adjusted for age, sex, birth year, first five principal components, and was analyzed using linear "mixed" model that accounted for genetic relationships among individuals (estimated by determining identity by descent (IBD) across genotyped markers among all pairs of individuals) as previously described [4]. Variants that showed association with BMI in American Indians at a P-value < 0.001 were considered for meta-analysis with the BMI results in the GIANT consortium database [15]. GIANT BMI association data was downloaded from the consortium portal (https://portals.broadinstitute.org/collaboration/giant/ index.php/GIANT_consortium_data_files). In total, 2600 variants that were present in both the American Indian and GIANT datasets were metaanalyzed by combining P-values using Stouffer's method. P-values for heterogeneity were also determined based on Cochran's Q statistic, calculated from the effect sizes and their standard errors.

Tissue expression
Total RNA of human pituitary, adrenal and adipose, and cDNAs of brain, hypothalamus, skeletal muscle, liver, kidney, and pancreas were purchased from Takara Bio (San Jose, CA, USA). RNA was converted to cDNA using ProtoScript II First Strand cDNA Synthesis Kit (New England Biolabs Inc., Ipswich, MA, USA). Primers specific for NFIA-AS2 transcript variant 1 were used for amplification. PCR products were sequenced for confirmation.

Cellular localization
RNA from nuclear and cytoplasmic fractions of human brown preadipocytes was purified using a commercial extraction kit (Norgen Biotek Corp., ON, Canada) and NFIA-AS2 presence was screened by RT-PCR. Different RNA markers were analyzed to check any cross-contamination between the fractions: 5S rRNA (nuclear), β-actin (cytoplasmic), and 28S rRNA (both nuclear and cytoplasmic). No amplification was seen in the RT (-) samples. Assay was repeated on 4 independent days.

RNA decay
Expression plasmids for NFIA-AS2 rs1777538 T and C allele-specific transcripts and eGFP were purchased from GenScript (Piscataway, NJ, USA). Each NFIA-AS2 plasmid (1000 ng) was co-transfected with eGFP (1000 ng) (to monitor transfection efficiency) in PAZ6 cells using Lipofectamine 3000 (Invitrogen, Waltham, MA, USA). After 24 h, global transcription was inhibited using 15 μg of actinomycin D (Sigma-Aldrich, St. Louis, MO, USA) per ml of complete growth medium. Cells were harvested before actinomycin D treatment (time 0) and at 3-, 6-, and 24-hours posttreatment, and NFIA-AS2 and TBP expression were measured by qRT-PCR on a QuantStudio 12K-Flex Real-Time PCR system (Applied Biosystems, Waltham, MA, USA). The percentage of remaining transcripts was calculated at each time-point by 2 -ΔCt method and transcript half-lives (T 1/2 ) were computed from rate of decay using the expression lnð2Þ Àðslope of the graphÞ . Assays were done in triplicate and performed on 3 separate days.

MicroRNA mimic assays
We used the lncRNASNP2 database [16] to determine whether rs1777538 disrupts a microRNA binding site. To measure endogenous expression levels of miR4270, miR6754-5p, and miR4441 in HEK293 cells (ATCC, Manassas, VA, USA) and brown preadipocytes (ABM, BC, Canada), we isolated total RNA including miRNAs using miRNeasy Kit (Qiagen, Hilden, Germany) which was reverse-transcribed using TaqMan Advanced miRNA cDNA Synthesis Kit (Applied Biosystems) and miRNA expression was measured by qRT-PCR using TaqMan Advanced miRNA assays and TaqMan Fast Advanced Master-Mix (Applied Biosystems). For binding assays, Mission miRNA mimics (50 nM) for miR4270, mir6754-5p (Sigma-Aldrich, St. Louis, MO, USA) were co-transfected with NFIA-AS2 rs1777538 T or C allele-specific expression vector (500 ng) and eGFP (500 ng) in HEK293 cells cultured in EMEM (ATCC) supplemented with 10% FBS (Gibco, Thermo Fisher Scientific). Transfections were performed in an antibiotic-free medium, and 24 h post-transfection HEK293 cells were collected, and NFIA-AS2 rs1777538 T and C allele-specific gene expression were measured by qRT-PCR using a TaqMan expression assay (Thermo Fisher Scientific). To further examine the effects of rs1777538 on miR4270 and mir6754-5p binding, 50 bp fragments containing the putative miRNA target site with either the T or C allele were cloned into the psiCHECK-2 miRNA target expression vector (Promega, Madison, WI, USA) at the 3' end of the Renilla luciferase gene. The constructs were co-transfected with either miR4270 or miR6754-5p miRNA mimics (Sigma-Aldrich) or with Mission miRNA, negative control 1 (Sigma-Aldrich) into HEK293 cells using lipofectamine 3000 (Thermo Fisher Scientific). Forty-eight hours post-transfection, the cells were analyzed for Renilla luciferase activity and normalized with firefly luciferase activity using the Dual-Luciferase Reporter Assay System (Promega). Transfections were done in triplicate and luciferase assays were performed 3 time on different days.

NFIA-AS2 overexpression
Uninduced PAZ6 brown preadipocytes at Day zero (D0) and differentiating preadipocytes at D4, D6, and D14 were co-transfected with 1000 ng of either the NFIA-AS2 T or C allele expression plasmids and 1000 ng of eGFP P values for the American Indian BMI associations were adjusted for age, sex, birth-year, the first five genetic principal components and identity by descent. Meta-analysis was performed using Stouffer's method based on combining P values. Freq P frequency of the protective allele, Effect P effect size for the protective allele, P het the posterior probability of heterogeneity. a P protective allele for obesity, R risk allele for obesity. Fig. 1 Regional BMI association results for the NFIA-AS2 locus in American Indians show stronger associations after meta-analysis with the GIANT data. Rs1777538 (shown as purple diamond) presented stronger association following meta-analysis. SNP coloring is based on linkage disequilibrium (LD) with rs1777538 in 1000 Genomes AMR population. Genetic recombination rates (cM/Mb) are shown as blue spikes. The lower panel displays the genomic segment presented in the regional association plot and the flanking region showing 13 protein-coding genes that exist in NFIA-AS2 neighborhood, UCSC Genome Browser build 38 [44].
plasmid. Differentiation and gene expression analysis at D0, D4, D6, and D14 was conducted as described in the previous section. Expression of neighboring genes was normalized with LMNB1. Overexpression experiments were conducted on 3 separate days in triplicates per experiment.

RESULTS
Genome-wide association analysis Association analysis of approximately 10.5 million imputed variants in 5870 American Indians identified 7733 variants that associated with BMI with a P < 0.001. A threshold of P < 10 −3 was selected for further study because this is the level of significance obtained for many well-established BMI genes (e.g., variants in FTO) in this sample. Among these 7733 variants with moderate evidence for association with BMI in American Indians, 2600 were also present in the GIANT dataset from multiple ancestries (N = 240,608) ( Supplementary Fig. 1). A meta-analysis of the American Indian and GIANT BMI association data for these 2600 variants was conducted (Supplementary Table 1) and the lead variants for the top genes are shown in Table 1. Among top signals, six of the genes (FTO, TMEM18, PRKD1, GRP, NLRC3, and MC4R) are well-established GWAS loci for BMI [9], while another gene (MAP2K3) has been previously reported as a BMI locus in American Indians which replicates in GIANT [18]. Furthermore, the gene RORA has been implicated in severe obesity in humans and mice [19]. Among the remaining signals, the T2D knowledgeportal provides strong SNP replication data for rs17522122 in AKAP6 (P = 10 −23 ) and rs9823593 in LSAMP (P = 10 −10 ), and some support for rs4950106 in MIR137HG (P = 10 −5 ). Thus, among our top genes for BMI, the only truly novel locus is NFIA-AS2, and it was prioritized for follow-up. The genomic region flanking the lead NFIA-AS2 variant rs1777538 showed significant improvement in association with BMI after meta-analysis (Fig. 1). Genotypes for rs1777538 were verified by direct genotyping in 5479 American Indian samples using a TaqMan genotyping assay (Applied Biosystems). The concordance for imputed genotyping versus direct genotyping was 99.7%. The minor allele C of rs1777538 is more common in the American Indian as compared to the GIANT datasets (frequency = 0.16 vs 0.04) and each copy of the C allele resulted in an average BMI decrease of 0.117 SD (about 1 kg/m 2 ) in American Indians, as compared with 0.022 SD in Europeans.
NFIA-AS2 is expressed in tissues relevant to obesity NFIA-AS2, an uncharacterized primate-specific lncRNA, maps antisense to the long protein-coding transcript of NFIA, within intron 1. NFIA-AS2 has 3 transcripts, variant 1 (NR_110617.1, NFIA-AS2 v1) and variant 2 (NR_110618.1) and variant 3 (ENST00000665665.1; Supplementary Fig. 2). Rs1777538 (T/C) detected in our meta-analysis, is located in exon 2 of NFIA-AS2 v1, hence only the v1 transcript isoform of NFIA-AS2 was investigated in our study. NFIA-AS2 is expressed in various human tissues that play a role in the development of obesity ( Fig. 2A). These tissue cDNAs were from commercial source and not related to American Indian ancestry. In particular, NFIA-AS2 has high levels of expression in brown preadipocytes. Unexpectedly, while examining NFIA-AS2 tissue expression we detected a novel alternative transcript isoform that is highly expressed in human pituitary and lacks exons 3 and 4. Furthermore, we also examined the expression of the genes flanking NFIA-AS2 (Fig. 1, N = 13) in various tissues in mouse and observed the highest neighboring gene expression in the brown adipose tissue characterized by an abundance of polyA + RNA sequencing reads (ENCODE consortium, Supplementary  Fig. 3). Together, our data supported brown preadipocytes as a biologically relevant cell line for our assays.

NFIA-AS2 is localized in the nucleus
We investigated the cellular localization of NFIA-AS2 lncRNA in PAZ6 human brown preadipocytes and NFIA-AS2 was found in the nuclear fraction (Fig. 2B) suggesting that it could potentially be involved in regulating gene expression.

Rs1777538 variant affects NFIA-AS2 RNA degradation
The effect of rs1777538 on NFIA-AS2 RNA structure and stability was evaluated. In-silico analysis of NFIA-AS2 RNA secondary structure did not identify a significant difference in minimum free energy (MFE) for RNA harboring either the T or C allele at rs1777538 (MFE = −356.9 kcal/mol and −357.8 kcal/mol respectively, Supplementary Fig. 4) [20]. To assess an impact on stability, we analyzed the rate of decay of NFIA-AS2 following transcription inhibition with actinomycin D and observed a significant difference of 4 h in the half-lives between the NFIA-AS2 T and C allele RNA transcripts, where the C allele transcripts had increased degradation compared to the T allele transcripts (T 1/2(C) = 9 h, T 1/2(T) = 13 h, Fig. 3) and there was no significant difference in TBP mRNA half-life (Fig. 3). These results suggest that rs1777538 alters NFIA-AS2 RNA turnover and potentially influences its downstream function.
Rs1777538 variant is predicted to disrupt a microRNA binding site One mechanism whereby a variant could affect RNA decay is via an effect on miRNA binding (miRNA target loss or gain). We found that rs1777538 is located within the seed region for a putative target site for 3 miRNAs miR4270, miR6754-5p, and miR4441 (Fig. 4A), and the rs1777538 T allele is predicted to alter the miRNA binding site (miRNA target loss). This predicted miRNA binding site was not found anywhere else in the NFIA-AS2 RNA sequence (data not shown). These results may underlie the increase in RNA degradation for the NFIA-AS2 transcript containing the C allele A RT-PCR gel shows NFIA-AS2 is expressed in various human tissues that play a role in the pathogenesis of obesity and obesity-related traits. The arrow is pointing to a novel alternative NFIA-AS2 transcript identified in the pituitary which is missing exons 3 and 4. B Results for the nuclear and cytoplasmic fractionation of PAZ6 brown preadipocytes show that NFIA-AS2 transcripts are localized to the nucleus. Genes of interest were PCR amplified from the isolated nuclear and cytoplasmic RNA fractions and PCR products were run on 1.5% agarose gel. No amplification was seen in the RT (-) samples. Nuclear marker: 5S rRNA; cytoplasmic marker: β-actin; nuclear and cytoplasmic marker: 28S rRNA. The experiment was repeated 4 times on separate days. L, 100 bp DNA ladder.
(unaltered miRNA binding site) compared to the T allele transcripts (altered target site). This prediction was experimentally examined in HEK293 cells which have a very low endogenous expression of NFIA-AS2 (data not shown). Also, expression levels for miR4270 and miR6754-5p were very low (average Ct = undetermined and 35.7, respectively) while miR4441 was highly expressed (average Ct = 19.1). To avoid spurious results due to high expression of endogenous miR4441, only miR4270 and miR6754-5p miRNA were investigated. NFIA-AS2 expression plasmids carrying either the rs1777538 T or C allele were cotransfected with miR4270 and miR6754-5p miRNA mimics in HEK293 cells. We found that both miRNAs target NFIA-AS2 and significantly decreased expression levels compared to the control (no miRNA) (Fig. 4B). There was also a modest difference between the rs1777538 T and C alleles with miR4270 (P = 0.04) presenting a higher impact on the transcripts carrying the C allele. To further investigate the effects of rs177538 on miR4270 and mir6754-5p binding efficiency, 50 bp fragments containing the predicted miRNA binding site with either the T or C allele were cloned into a miRNA target reporter vector. In vitro analysis using these reporter constructs co-transfected with either miR4270 or miR6754-5p miRNA mimics showed that the cloned 50 bp region can function as a transcriptional regulatory element as indicated by an approximately 2-fold increase in luciferase activity compared to the empty reporter vector (Fig. 4C). There also appears to be a modest difference in luciferase activity between the rs1777538 T and C alleles. The two miRNAs do not have a significant effect on the expression of the luciferase gene containing the predicted miRNA binding site (Fig. 4C). This could be due to the confounding effects of the increased expression levels due to the putative cisregulatory element located in the 50 bp fragment.

NFIA-AS2 is not a miRNA sponge for its neighbors
LncRNAs that bind miRNAs can upregulate gene expression by preventing the interaction of bound miRNA with its target mRNAs, thus serving as a miRNA sponge. NFIA-AS2 is reported to sponge miR-665-3p [21]. Furthermore, analysis of the sequence around rs1777538 in the ENCODE Regulation database supports that this variant is positioned in a genomic region highly enriched with the histone modifications H3K27Ac and H3K4Me1, a signature characteristic of active enhancers (Supplementary Fig. 2). Given that, we studied the 'miRNA sponging' ability of NFIA-AS2 in cis for 13 genes that reside in a 3 Mb region flanking NFIA-AS2 (Fig. 1). We inspected the changes in their endogenous gene expression in the presence of miR4270 and miR6754-5p in HEK293 cells. None of the investigated genes were targeted by miR4270 and miR6754-5p, reducing the likelihood that 'miRNA sponging' is an important mechanism for NFIA-AS2 activity ( Supplementary Fig. 5).

NFIA-AS2 affects neighboring gene expression in brown adipocytes
We examined the expression pattern of 13 neighboring genes during brown adipogenesis. Human brown preadipocytes were differentiated over 14 days (Fig. 5A) and brown adipogenesis was verified by analyzing the expression of marker genes (Supplementary Fig. 6). Investigation at 4 different Days (D)/stages: D0 (preadipocytes), D6 (differentiating preadipocytes), D10 (immature adipocytes), and D14 (terminally differentiated adipocytes) indicated differential expression of all genes during adipogenesis except for C1orf87, KANK4, and ANGPTL3, which were not expressed at any stage in PAZ6 cells (Fig. 5B). Immediate neighbors of NFIA-AS2, CYP2J2, NFIA, TM2D1, and INADL were strongly upregulated Fig. 3 RNA degradation kinetics for NFIA-AS2 harboring either the rs1777538 T or C allele. Global transcription in brown preadipocytes was blocked with actinomycin D (ActD) at a 0-hour time point. Cells were collected at 3, 6, and 24 h post-ActD treatment for RNA isolation and subsequent measurement of NFIA-AS2 transcript levels. TBP was used as a control. Transcript levels for the ActD treated cells are shown relative to untreated cells (0 h). Results are from 3 independent experiments each done in triplicate and shown as mean+/− SE. Fig. 4 The rs1777538 T > C variant destabilizes NFIA-AS2 affecting a miRNA target site. A Rs1777538 T > C variant disrupts a miRNA target site. The rs1777538 variant is located within the seed region (underlined) for a putative target site for 3 miRNAs miR4270, miR6754-5p, and miR4441 [figure adapted from lncRNASNP2 database (16)]. The rs1777538 C allele is shown in orange font and underlined. B Both miRNA mimics miR4270 and miR6754 significantly decrease NFIA-AS2 levels in HEK293 cells. Data was normalized with eGFP (transfection efficiency) and TBP. C Relative luciferase activity for the binding of miR4270 and miR6754-5p to NFIA-AS2 in an allele-specific manner. The miRNA target site constructs carrying rs1777538 alleles (C) or (T) were cotransfected with either a human miRNA mimic or a negative control in HEK293 cells. Observed luciferase activity was normalized with respect to the empty vector (E). Results are from 3 independent experiments each done in triplicate and shown as mean + SD. Data points per group are shown as hollow circles. Statistical analysis was carried out by Student's t test. *P < 0.05; **P ≤ 10 −5 .
upon differentiation and showed the highest expression at D10, except NFIA which had slightly lower expression than the D6 and D14. Genes distant to NFIA-AS2, L1TD1 and DOCK7 were downregulated in mature brown adipocytes compared to preadipocytes; while FGGY, USP1 and ATG4C were modestly upregulated upon induction of differentiation but downregulated in mature adipocytes (Fig. 5B). NFIA-AS2 gene expression did not change during differentiation (data not shown).
To investigate the effects of rs1777538, T and C allele-specific NFIA-AS2 transcripts were transiently overexpressed at days D0, D6, D10, and D14. We did not observe any considerable changes in neighboring gene expression at the preadipocyte stage (D0) (Fig.  6A). Once the cells were shifted to the maturation phase (D6), overexpression of both NFIA-AS2 T and C allele transcripts induced expression for all genes except for NFIA which showed some reduction in expression with the T allele (Fig. 6B). Meanwhile, NFIA-AS2 C allele transcripts resulted in a greater magnitude of upregulation in nearby gene expression compared to the T allele transcripts, however, the differences did not reach statistical significance. (Fig. 6B). At D10, there was an observable decline in both the upregulation and differential expression for the neighboring genes and this was especially evident for L1TD1 (Fig. 6C). However, in mature brown adipocytes (D14) the variable effects of the NFIA-AS2 allele-specific transcripts on gene expression were opposite compared to D6. Contrary to D6, at D14, overexpression of the NFIA-AS2 T allele transcript led to an increase in gene expression for many neighboring genes compared to the C allele transcript (strongest allelic difference observed for L1TD1) (Fig. 6D). These results indicate that the effects on gene expression by NFIA-AS2 may be stage-specific (temporal) and that NFIA-AS2 may have different functional roles for the D6 (immature adipocytes) and D14 (mature adipocytes) stages to upregulate gene expression.

DISCUSSION
Our meta-analysis of genetic variants that nominally associate with BMI in both American Indians and individuals from the GIANT consortium identified several loci that have been extensively studied for their role in BMI and several others that have been statistically associated with BMI but warrant a biologic follow-up for a potential role in body weight. The current report is focused on follow-up of one of these loci, NFIA-AS2, since this locus had never been linked to BMI yet it was among the strongest signals in the American Indian data. NFIA-AS2's namesake NFIA has been linked to brown adipocyte differentiation [22,23] and prior studies have suggested involvement of other lncRNAs in obesity [24][25][26][27][28]; however, NFIA-AS2 has never before been reported to affect body weight.
Our functional studies suggest that the BMI-associated variant rs1777538 may affect NFIA-AS2 RNA decay (turnover rate) which in turn may alter cellular function. In addition, we show that the variant is in a cis-regulatory element (CRE) and may alter a transcription factor binding site where the T allele results in lower gene expression. The putative CRE is predicted to interact with the promoters of long isoform of NFIA (ENST00000371191.5) and NFIA-AS2 ( Supplementary Fig. 2); and the rs1777538 variant is reported to be an eQTL for NFIA-AS2 in tibial nerve tissue where the TT genotype is also associated with lower gene expression ( Supplementary Fig. 7). Despite a greater number, C-allele transcripts showed higher RNA degradation compared to the transcripts with the T allele. This increased decay of NFIA-AS2 RNAs with C allele could be in part due to stronger binding of miRNAs to their predicted target which encompasses the C/T variant at rs1777538, e.g., miR4441 which is highly expressed in brown preadipocytes (average Ct= 19.8). However, because of the small difference in binding efficiency for miR4270 between the T and C alleles of NFIA-AS2 transcripts, we do not think that differential miRNA binding is the sole reason for increased RNA decay. Of the obesity-related tissues we examined, NFIA-AS2 is highly expressed in brown preadipocytes. LncRNAs play an important role in brown adipose metabolism (e.g., adipogenesis, non-shivering thermogenesis) including Blnc1, H19, and lnc-dPrdm16 among others [28] and they can function in cis (locally) by regulating the spatiotemporal expression of neighboring proteincoding target genes [24,29]. We demonstrated that overexpression of NFIA-AS2 T and C allele-specific transcripts during brown adipogenesis influenced the expression of nearby genes in a temporal manner at D6 and D14. Moreover, we also observed differential effects on gene expression between the NFIA-AS2 T and C allele transcripts particularly in mature brown adipocytes (D14).
The differential regulation of DOCK7 at D14 is intriguing. A previous study has shown that Misty mice which harbor a premature Fig. 6 Effects of NFIA-AS2 rs1777538 variant on neighboring gene expression during brown adipogenesis. Uninduced brown preadipocytes (D0) and differentiating preadipocytes at D4, D6, and D14 were transfected with either the NFIA-AS2 T or C allele expression plasmids. Expression levels for the nearby genes were measured 24 h post-transfection for D0, D6, D10, and D14 and are shown relative to the empty vector control (E). All gene expression levels were normalized with LMNB1. The differentiations and transfections were performed 3 times on separate days and done in triplicate per experiment. Data is shown as means + SD. Data points per group are shown as hollow circles. Significance was calculated using the Student's t test. *P = 0.01-0.05; # P = 0.003-0.009. stop codon in the Dock7 gene lack brown fat pads [30]; and, in the second pair of later studies, it was also shown that Misty mice have reduced brown adipose innervation and impaired function (thermogenesis) [31,32]. Among others, CYP2J2 expression has also been linked to brown adipogenesis [33], while ATG4C belongs to a family that includes similar proteins involved in brown adipocyte metabolism [34]; we noticed a constant variation in ATG4C expression whenever NFIA-AS2 was overexpressed during differentiation. The connection between the gene, L1TD1 which showed the largest differential regulation in mature adipocytes, and brown adipose metabolism is unclear. L1TD1 functions as an RNA-binding protein in various cellular processes including RNA degradation, splicing, translation, and protein trafficking [35,36]. Rs1777538 C-allele which predicted leanness in our association studies and caused higher NFIA-AS2 degradation, demonstrated significant reduction in L1TD1 expression in brown adipocytes (D14). We speculate that L1TD1 might be involved in the decay of NFIA-AS2 transcripts. The only report, to date, linking L1TD1 and obesity is a study that examined genome-wide DNA methylation patterns in preadipocytes isolated from lean individuals and individuals with obesity and T2D, they observed differential methylation at L1TD1 promoter (↑ methylation in lean group) and a 2-fold increase in L1TD1 expression for the group with obesity and T2D [36]. Considering the functional effects of NFIA-AS2 on multiple genes in the neighborhood, we suspect that the overexpression of the affected genes influences the entire differentiation, and the response may not be necessarily mediated by the direct action of NFIA-AS2 lncRNA.
In addition to human brown preadipocytes, NFIA-AS2 is also highly expressed in the pituitary and moderately expressed in the hypothalamus and adrenal gland (Fig. 2A). These 3 tissues form the hypothalamic-pituitary-adrenal (HPA) axis which assumes a central role in the neuroendocrine control of obesity. HPA axis controls the stress response in the body by regulating the production and secretion of neurohormones that influence energy intake and body weight [37]. In general, a higher BMI is associated with a larger volume of pituitary gland [38] and a hyperactivity of HPA axis [39]. Given the high expression of NFIA-AS2 in HPA tissues and strong association with BMI, future studies can explore the biological role of NFIA-AS2 in HPA activity.
In conclusion, we identified a lncRNA, NFIA-AS2, as a novel BMI locus using a multiethnic meta-analysis that included American Indians from southern Arizona and individuals from the GIANT consortium. Functional characterization of lead NFIA-AS2 signal rs1777538 showed that its minor allele (C) promotes NFIA-AS2 RNA decay and downregulates several neighboring genes during later stages of brown adipogenesis. Several groups have suggested that both brown adipose tissue and lncRNAs could potentially be used as therapeutic targets for the treatment of obesity and obesity-related diseases [24,[40][41][42][43]. The prospect of utilizing rs1777538 allele-specific NFIA-AS2 lncRNAs targeting brown adipose or other obesity-related tissues as obesity therapies is very intriguing and requires additional in-depth studies.

DATA AVAILABILITY
The data generated and/or analyzed in the current study is available from the corresponding author upon reasonable request. The summary results of all the variants from our meta-analysis that achieve a P < 10 −5 are available in Supplementary Table 1.