RBM4a-regulated splicing cascade modulates the differentiation and metabolic activities of brown adipocytes

RNA-binding motif protein 4a (RBM4a) reportedly reprograms splicing profiles of the insulin receptor (IR) and myocyte enhancer factor 2C (MEF2C) genes, facilitating the differentiation of brown adipocytes. Using an RNA-sequencing analysis, we first compared the gene expressing profiles between wild-type and RBM4a−/− brown adipocytes. The ablation of RBM4a led to increases in the PTBP1, PTBP2 (nPTB), and Nova1 proteins, whereas elevated RBM4a reduced the expression of PTBP1 and PTBP2 proteins in brown adipocytes through an alternative splicing-coupled nonsense-mediated decay mechanism. Subsequently, RBM4a indirectly shortened the half-life of the Nova1 transcript which was comparatively stable in the presence of PTBP2. RBM4a diminished the influence of PTBP2 in adipogenic development by reprogramming the splicing profiles of the FGFR2 and PKM genes. These results constitute a mechanistic understanding of the RBM4a-modulated splicing cascade during the brown adipogenesis.

Alternative splicing (AS) constitutes a prevalent mechanism in expanding the genetic diversity of eukaryotic cells 1 . Approximately 90% of human genes generate more than one transcript by undergoing this meticulous process 2,3 . Spatiotemporal expression profiles of AS events control cell differentiation and specification 4,5 . The interplay between trans-factors and corresponding cis-elements within transcripts precisely manipulated tissueand stage-specific splicing events 6 . Altering expression levels of splicing factors is an efficient and dominant mean to fine-tune alternative splicing profiles 7,8 . It is imperative to obtain comprehensive insights into AS events in the genome-wide era 9 .
Adipose tissue is an endocrine organ that participates in energy homeostasis. White and brown adipocytes (BAs) were identified based on their macroscopic appearance 10 . BAs dissipate fatty acids in the form of heat to maintain the body temperature, which implies its therapeutic potential for combating obesity 11 . The developmental process of BAs is still debated. Transcriptome analyses and lineage tracing suggest the existence of the same precursor cells for myocytes and BAs 12 . A BA-related protein network was uncovered, but the comprehensive mechanism is largely unknown 13 . AS constitutes a molecular mechanism which modulates adipocyte development. AS events, including of IR and MEF2C transcripts, encode adipocyte-specific isoforms which facilitate brown adipogenesis 14,15 . Interest is building in examining the AS-mediated mechanism involved in the development of BAs in terms of fat metabolism.
RNA-binding motif protein 4a (RBM4a) is a multifunctional protein that regulates various AS events in differentiating or malignant cells [16][17][18] . RBM4a was shown to reprogram the tissue-specific splicing profiles that facilitates the differentiation of myocytes and brown adipocytes 14,16 . An increase in RBM4a induces a relatively high level of the MEF2Cγ -protein which constitutes a feed-forward circuit toward upregulating RBM4a during brown adipogenesis 15 . In this study, results of the deep RNA-sequencing showed the differential gene expressions in RBM4a −/− brown adipose tissues (BATs) compared to the wild type counterparts. The ablation in RBM4a with concomitant increases in the Nova1, PTBP1, and PTBP2 transcripts was originally noted in BAs. Nova1 and Differential expression profiles of splicing factors in WT and RBM4a −/− BAs. To validate the analytic results of RNA sequencing, total RNAs prepared from RBM4a −/− BATs or the WT counterparts were subjected to RT-PCR assays. Figure 1 showed the increases in PTBP2 −10 and PTBP1 −11 transcripts, the nonsense-mediated decay (NMD) substrates 21 , which were noted during the development of BATs (Fig. 1a,  lanes 1 and 2). The relative levels of PTBP2 +10 and PTBP1 +11 transcripts were sustained in RBM4a −/− embryonic (E13.5) and postnatal (P0) BATs (lanes 3 and 4) compared to that of WT littermates. The expression of Nova1 transcripts was unchanged in embryonic and postnatal RBM4a −/− BATs, whereas a gradual decrease in Nova1 transcripts was observed during the development of BATs (Fig. 1a). Immunoblotting assays showed the gradual increase in the RBM4a protein with concomitant decreases in expressions of PTBP2, PTBP1, and Nova1 proteins (Fig. 1b, lanes 1-3) which were relevant to the transcript levels. Ablation of RBM4a resulted in the unchanged expression profiles of PTBP2, PTBP1, and Nova1 proteins in embryonic and postnatal BATs (lanes [4][5][6]. Differential expressions of PTBP2, PTBP1 and Nova1 proteins were consistently noted in BATs dissected from adult RBM4a −/− mice and WT littermates (Fig. 1c). Therefore, unlike RBM4a, PTB proteins and Nova1 may possess distinct effect toward the development of BATs.
RBM4a modulates expression profiles of PTBP2, PTBP1, and Nova1 in preadipocytes. In vitro differentiation of C3H10T1/2 cells was subsequently conducted to confirm previous observations. Using RT-PCR (Fig. 2a, left) and quantitative(q)RT-PCR analyses (Fig. 2a, right), the gradual increase in the endogenous RBM4a with concomitant decreases in the PTBP2 +10 , PTBP1 +11 , and Nova1 transcripts was observed during the differentiation of C3H10T1/2 cells. The immunoblotting assay indicated the relevant expression of PTBP2, PTBP1, RBM4a, and Nova1 to the transcript profiles during the differentiating process (Fig. 2b). The direct binding of RBM4a to the CU-elements within PTBP1 exon 11and flanking introns led to PTBP1 exon 11 skipping in myoblast cells 16 . The presence of overexpressing RBM4a or the derived zinc knuckle-mutant (mZn) containing the authentic RNA recognition motifs (RRMs) reduced the relative level of PTBP2 +10 , PTBP1 +11 and Nova1 transcripts (Fig. 2c, lanes 1-3). In contrast, the abundances of PTBP2 +10 , PTBP1 +11 , and Nova1 transcripts showed no  species 23 . The interaction between nELAV and Nova1 3′ -UTR constituted a post-transcriptional mechanism for stabilizing Nova1 transcripts 24 . In addition, PTBP2 was reported to enhance the stability of phosphoglycerate kinase 2 and inducible nitric oxide synthase transcripts by binding to its 3′ -UTRs 25 . We therefore wondered whether PTBP2 participated in the 3′ -UTR-mediated regulation of Nova1 transcripts. The RT-PCR and qPCR results indicated that around 80% of Nova1 transcripts were left with treatment of actinomycin D for 6 h (Fig. 3a, lanes 1 and 3; diamond), but only 36% of the Nova1 transcripts were observed in PTBP2-knockdown cells with the same treatment (lanes 4 and 6; square). This result implied the potential influence of PTBP2 on the stability of Nova1 transcripts. According to the binding tendency of the PTB proteins toward UCUU or CUCUCU motif, the CU-rich elements were noted within the proximal region of the Nova1 3′ -UTR (Fig. 3b, underlined). To determine putative binding site of the PTBP2, the gel-shift assays were performed by incubating the in vitro transcribed and DIG-labeled RNA probes (F1, F2 and F3) with the His-tagged PTBP2. Only the F1 (Fig. 3b, lower, lane 2), but not F2 or F3 probe (lanes 4 and 6), formed ribonucleoprotein complexes with the recombinant PTBP2 protein. The interaction between the F1 probe and recombinant PTBP2 was abolished with guanine nucleotide substitutions within the UCCU motif (mF1, CC to GG, lane 8). The functional relevance of the PTBP2 binding motif (UCCU) was examined by introducing the renilla luciferase reporter containing F1, F2, F3, or mutant F1 fragment into C3H10T1/2 cells. The reduced activity of the F1 fragment-containing reporter was noted in differentiating C3H10T1/2 cells (Fig. 3c, middle, brick color), whereas the activities of other reporters were unchanged in proliferating and differentiating cells (Fig. 3c, middle). Moreover, the statistical analyses of in vivo translation assay showed that only the F1 element-containing reporter, but not the other reporters, exhibited convincing responses to the RBM4a and PTBP2 protein levels. Overexpressing PTBP2 or short hairpin (sh)RNA-mediated ablation of RBM4a preferentially enhanced the activity of the F1-containing reporter. Overexpressing RBM4a or shRNA-mediated knockdown of PTBP2 inversely led to reduced activity of the F1-containing reporter (Fig. 3c, differentiating medium (for 2 and 7 days) were subjected to RT-PCR, qRT-PCR and immunoblotting assays using specific primer sets and antibodies. (c) Total RNAs and cell extracts extracted from C3H10T1/2 cells that overexpressed RBM4a and the derived mutants were subjected to RT-PCR and immunoblotting assays using specific primer sets and antibodies. (d) Total RNAs and cell extracts extracted from RBM4a overexpressing or targeting C3H10T1/2 cells were subjected to RT-PCR, qRT-PCR and immunoblotting assays as previously described. The gels or blots showed in this figure were run under the same conditions and not artificially manipulated. The bar graph represents relative levels of the indicated proteins or transcripts in three independent experiments using TotalLab Quant Software (*p < 0.05; **p < 0.01; ***p < 0.005).
Scientific RepoRts | 6:20665 | DOI: 10.1038/srep20665 right, brick red). Activities of the Renilla reporters that contained the F2, F3, or mF1 fragment were not altered with change of PTBP2 and RBM4a expression (Fig. 3c, right). These results identified RBM4a-PTBP2interplay as a novel posttranscriptional mechanism of the Nova1 expression in BAs.

PTBP2 and Nova1 function as brown-adipogenic repressors. The gradual decreases in PTBP2and
Nova1 proteins throughout the brown adipogenesis ( Imaging results were quantitatively evaluated in terms of extracted oil-red-O using spectrophotometric analysis (Fig. 4c, middle) and counting of oil-red O-stained cells among 100 cells in total (Fig. 4c, lower). The quantitative (a) Empty vector-or PTBP2 targeting vector-transfected C3H10T1/2 cells were mock-treated or treated with actinomycin D. Total RNA and cell extract prepared at indicated time points were subjected to RT-PCR, qRT-PCR, and immunoblotting assays with specific primer sets and antibodies. (b) The diagram presents the sequence of the proximal Nova1 3′ UTR. The mock eluate or recombinant His-tagged PTBP2 protein (2 μ g) eluted from the Ni2+ agarose resin was incubated with 10 nM DIG-labeled probes. The mixtures were fractionated on an 8% native acrylamide gel and transferred to a nylon membrane. The membrane was probed using the horseradish peroxidase-conjugated anti-DIG Fab fragment. The gels or blots showed in this figure were run under the same conditions and not artificially manipulated. (c) The scheme shows the Renilla luciferase reporter containing the distinct fragment within the Nova1 3′ UTR. The intact pRL-Nova1 3′ UTR reporters or the derived mutant were transfected into C3H10T1/2 cells cultured under distinct conditions or cotransfected with the expressing or targeting vectors and the pGL3-basic reference vector into C3H10T1/2 cells. The luciferase assays were performed as described under "Materials and methods" and the bar graphs show relative renilla luciferase activity in three independent experiments. The statistical analyses showed the convincing difference in the activity of F1 reporter, but not other reporters in distinct experiment groups (*p < 0.05; **p < 0.01; ***p < 0.005).  Fig. 1). We therefore wondered whether RBM4a-PTBP2 interplay manipulated the splicing profile of the FGFR2 and PKM genes in BAs. The expression profiles (lanes 1-3) and stabilities (lanes 4-6) of FGFR2 and PKM transcripts remained sustained with the overexpression of RBM4a and PTBP2 (Fig. 5a). RT-PCR results showed the predominant expression of FGFR2 IIIc and PKM2 transcripts in C3H10T1/2 cells as other progenitor cells (Fig. 5b, lane 1). Overexpressing RBM4a and the derived mZn mutant profoundly enhanced the relative level of FGFR2 IIIb (Fig. 5b, upper, lanes 2 and 3), whereas mRRMs mutant completely lost the activity (lane 4). Similarly, the relative level of PKM1 transcripts was elevated in the presence of overexpressing RBM4a and the derived mZn mutant (Fig. 5b, lower, lanes 2 and 3), whereas the mutations within RRMs diminished this phenomenon (Fig. 5b, lower, lane 4). The relatively high levels of the FGFR2 IIIb and PKM1 transcripts were noted in PTBP2-knockdown C3H10T1/2 cells compared to empty vector-transfected cells (Fig. 5c, lanes 1 and 5). In contrast, overexpressing PTBP2 or RBM4a-knockdown enhanced the relative levels of the FGFR2 IIIc and PKM2 transcripts (lanes 3 and 4). These results indicated the differential effect of RBM4a and PTBP2 on programming the splicing profiles of FGFR2 and PKM genes in BAs. The RBM4a-regulated splicing event modulates the brown-adipogenic signaling pathway. FGF10-FGFR2 IIIb interaction activated the downstream ERK1/2 and p38 MAPK pathway, which enhanced UCP1 transcripts and β -oxidation of fatty acids in BATs 29,30 . Controversially, upregulated FGFR2 expression was reported to interfere with adipocytic differentiation of C3H10T1/2 cells 31 . We thus wondered whether alternative splicing constituted a spatiotemporal mechanism in manipulating the effect of FGFR2 on the downstream signal. Using an RT-PCR assay, the predominant expression of FGFR2 IIIb transcripts was noted in postnatal BATs (Fig. 6a, lane 2), whereas the relative level of FGFR2 IIIc transcripts were close during the development of RBM4a −/− BATs (lanes 3 and 4). Synchronous increases in UCP1and Prdm16 transcripts (lane 2) implied their relevance between FGFR2 IIIb-mediated signaling. qRT-PCR result indicated the close expression of UCP1 and Prdm16 transcripts with an unchanged splicing profile of FGFR2 in embryonic and postnatal RBM4a −/− BATs (Fig. 6a, right bar chart). This result was consistently reproduced in response to differentiation of C3H10T1/2 cells. The differentiating condition drove the isoform change of the FGFR2 gene in C3H10T1/2 cells (Fig. 6b, lanes 1-3), whereas the splicing profile of FGFR2 remained unchanged throughout the differentiation of RBM4a-knockdown cells (lanes 4-6). RT-PCR and qRT-PCR results showed the gradual increases in UCP1and Prdm16 transcripts during the differentiation of C3H10T1/2 cells (Fig. 6c, lanes 1-3), but RBM4a silencing substantially abolished this phenomenon (lanes 4-6; qRT-PCR, bar chart). We next validated the influence of RBM4a-PTBP2 interplay on brown adipogenesis-related ERK1/2 signaling. Using immunoblotting assays, we noted that FGF10 treatment, RBM4a overexpression, or PTBP2 knockdown executed similar effects of inducing the phosphorylation of ERK1/2 compared to empty vector-transfected cells (Fig. 6d, lanes 1-3 and 6). RBM4a-knockdown or PTBP2-overexpressing cells inversely exhibited less phosphorylated ERK1/2 proteins (lanes 4 and 5). Moreover, overexpressing RBM4a synergized the FGF10-induced phosphorylation of ERK1/2 (lane 11), whereas overexpressing PTBP2 diminished the effect of FGF10 on ERK1/2 activation (lane 12). These results revealed the influence of RBM4a-regulated splicing cascade on manipulating the BA-related signaling.
RBM4 and PTBP2 exert opposite effects on the energy expenditure of BAs. The expression profiles of PKM gene is spatiotemporally modulated by an AS mechanism 32 . PKM2 is an embryonic or carcinogenic isoform, whereas PKM1 is mainly expressed by differentiated or energy-expenditure cells, including BAs 33,34 . As shown in Fig. 7a, the gradual increase in PKM1 transcript as that of FGFR2 IIIb was only observed in WT BATs (lanes 1and 2), but not in the RBM4a −/− BATs (lanes 3 and 4). Similar results were reproducible in a culture system of the dominant expression of the PKM2 transcript which gradually shifted to PKM1 transcripts during the differentiation of C3H10T1/2 cells (Fig. 7b, lanes 1-3). RBM4a-knockdown cells exhibited a relatively high level of the PKM2 transcript throughout the differentiating process (lanes 4-6). The results of bioenergy analyses showed that RBM4a-overexpressing cells exhibited higher basal oxygen consumption rate (OCR) compared to empty vector-transfected cells, whereas overexpressing PTBP2 inversely reduced the basal OCR of proliferating cells (Fig. 7c, basal respiration, GM). Moreover, the additive upregulation of basal OCR was noted in RBM4a-overexpressing cells maintained in differentiating medium, but no substantial difference in the basal OCR was observed in PTBP2-overexpressing cells under the same condition (Fig. 7c, basal respiration,  DM). Similarly, the maximal and spare respiratory capacities remained unchanged in PTBP2-overexpressing cells cultured in proliferating or differentiating medium (Fig. 7c, maximum respiration and respiratory capacity). Figure 7d showed that the PTBP2-overexpressing cells exhibited less mitochondrial biogenesis than that of vector-transfected cells cultured in proliferating or differentiating medium. In contrast, the relatively elevated mitochondriogenesis was noted in the RBM4a-overexpressing cells cultured under the same conditions. These (a) Total RNA extracted from the embryonic day (E) 13.5, and postnatal day 0 (P0) RBM4a +/+ (wildtype) or RBM4a −/− interscapular brown adipose tissues was subjected to RT-PCR, enzyme digestion and qRT-PCR analyses. (b,c) Total RNAs and cell extract were isolated from the proliferating and differentiating cells at different time points, followed by RT-PCR, enzyme digestion, qRT-PCR and immunoblotting assay with specific primer sets and antibodies. The bar graph presents relative levels of the PCR-amplified transcripts (FGFR2) or a qRT-PCR analysis (Ucp1 and Prddm16) using TotalLab Quant Software (*p < 0.05; **p < 0.01; ***p < 0.005). (d) C3H10T1/2 cells were transfected with an empty vector, expression vectors, or targeting vectors of RBM4a and PTBP2. Before being harvested, transfected cells were mock-treated or treated with FGF10 for 6 h. Cell extracts were isolated from transfectants, followed by immunoblotting assays with anti-ERK1/2, anti-phospho ERK1/2, and anti-Actin antibodies. The gels or blots showed in this figure were run under the same conditions and not artificially manipulated.
Scientific RepoRts | 6:20665 | DOI: 10.1038/srep20665 results indicated the opposite effects of RBM4a and PTBP2 on the metabolic activities of BAs which were relevant to PKM splicing profile and mitochondriogenesis.

Discussion
Posttranscriptional controls constitute complex mechanisms in programming expression profiles or fine-tuning biological activities of tissue-specific factors 35 . Deep RNA-sequencing provides an advance approach to establish a global view of transcript profiles in a spatiotemporal manner 36 . Herein, we performed a deep RNA-sequencing to evaluate the influence of RBM4a on transcript profiles in BAs. The results showed that ablation of RBM4a profoundly affected BA-related genes expressions which subsequently influenced the development or function of BAs.
RBM4a facilitated brown adipogenesis by programming a subset of alternative splicing events 14,15 . In this study, the RBM4a −/− BATs exhibited upregulated levels of PTBP2 and Nova1 transcripts (Fig. 1) which functioned as repressors toward both the development and metabolic signature of BAs (Figs 4 and 7). Overexpressing PTBP2 and Nova1restricted the development of BA-like progenitor cells, and the differentiating process was initiated with a decline in their expression. Similarly, PTBP2-regulated splicing events maintained the biological signature of neuronal progenitor cells and a gradual reduction in PTBP2 was noted during the development of mature neurons 21,37 . Accordingly, PTBP2-programmed splicing profiles may globally silence the inaccurate differentiation of progenitor cells more than maintaining their biological features. PTBP1 was reported to constitute an autoregulatory feedback circuit by enhancing the relative level of PTBP1 −11 and PTBP2 -10 transcripts 38 . Overexpressing PTBP2 increased its abundance by enhancing the utilization of its exon 10, which partially neutralized the repressive effect of elevated PTBP1 in colorectal cancer cells 39 . Nevertheless, RBM4a played a hierarchical role that overrode the crossregulation between PTB proteins in different cells. The RBM4a-PTBP2 interplay constituted a regulatory mechanism that widely modulated the brown adipogenesis. Identification of more RBM4a-, PTBP2and Nova1-specific splicing event can be applied in emphasizing the inference.
Besides being a well-known splicing factor, PTBP2 functioned as a multifunctional RNA-binding protein involved in mRNA localization, IRES-mediated translation, and mRNA stability 40 . Although in vitro studies suggested the binding tendency of PTB proteins towards UCUU or CUCUCU motifs, the interactions between PTB proteins and candidates do not specifically rely on the consensus sequence 41 . For instance, PTBP2 protein enhanced the stability of pgk2 transcripts by binding to the non-consensus CU-elements within the pgk2 3′ UTR 25 . Multiple CU-rich elements shared different sequences additively strengthened the binding of PTBP2 to the pgk2 3′ UTR. In our study, the in vitro binding and functional assay identified UCCU motif within Nova1 3′ UTR as a binding site of PTBP2 protein (Fig. 3). The presence of four RRMs was proposed as a molecular mechanism that largely expanded the binding tendency of the PTB proteins 42 . Herein, we provide another example regarding the effect of the PTBP2 protein on the 3′ UTR-mediated gene regulation As shown in Table 2, complementary expression profiles of the Nova1 and Nova2 proteins were revealed in RBM4a −/− BATs compared to the WT littermates. Nova1 and Nova2 genes are phylogenetically conserved, but their 3′ UTR sequences greatly differ. The highly conserved Nova1 3′ UTR among distinct species suggests that 3′ UTR-mediated regulation may contribute to differential expressions of Nova 1 protein in distinct tissues or particular stage. PTBP2 was demonstrated to be a novel regulator in stabilizing Nova1 transcripts by binding to the UCCU motif within its 3′ UTR in our work. The repressive effects of PTBP2 and Nova1 on the development and functioning of BAs were first revealed. It would be interesting to further evaluate the influence of Nova2, the relatively high expression of which was noted in ordinary BATs compared to RBM4a −/− BATs. To bring the comprehensive insight into the mechanism underlying the differentiation of BAs, the specific candidate of RBM4a, PTBP2, and Nova family proteins in BAs is worthy of further identification by using high-throughput approaches.
FGFR2 isoforms are spatiotemporally generated through AS mechanisms 43 . The abundant level of FGFR2 IIIc was noted in mature white adipocytes 28 . Adipocyte-restricted FGFR2 −/− mice exhibited the impaired hypertrophy of white adipocytes and reduced plasma fatty acid, implying its potential influence on adipogenesis and lipid metabolism 28 . Interestingly, the overexpressing FGFR2 IIIc decreased the adipocytic differentiation of C3H10T1/2 cells which were considered as the progenitors of BAs 31 . Herein, we first revealed the gradual increase in the FGFR2 IIIb transcript during the differentiation of C3H10T1/2 cells and mice BATs ( Fig. 6A and  B). The effect of FGF10-FGFR2 IIIb interplay on ERKs phosphorylation indicated its potential function on a brown adipogenesis-related signaling pathway 29 . Therefore, RBM4a-induced increase in the relatively high level of FGFR2 IIIb may constitute a novel mechanism in promoting the brown adipogenesis. In addition, PKM gene encodes a constitutively active PKM1 isoform in differentiated cells and an embryonic PKM2 isoform which participated in the transition from aerobic respiration to anaerobic glycolysis 34 . The increased PKM2 mRNA was noted in the NIH3T3-L1 cells with the insulin treatment 44 , whereas the predominant expression of PKM1 transcripts was observed in BATs 45 . Nevertheless, the splicing profiles of PKM transcripts during the development of BATs were first revealed in this study (Fig. 7A and B). The splicing patterns of FGFR2 and PKM genes were reprogrammed with the expression profiles of splicing factors, including RBM4a, PTB proteins (Fig 5B,C). The molecular mechanisms involved in the splicing of FGFR2 and PKM genes were worthy of further investigation, which may bring a new insight into the differentiation of BAs.
In conclusion, we first correlated the regulation of two neuronal-specific splicing factors, PTBP2 and Nova1, within RBM4a-modulated posttranscriptional control in BAs. The relatively high levels of PTBP2 and Nova1 in RBM4a −/− BATs suggested their repressive effects on brown adipogenesis. Identification of PTBP2 as a posttranscriptional regulator of Nova1 constituted a novel molecular mechanism for repressing the development of BATs. We also revealed the RBM4a-modulated splicing cascades that widely involved in the differentiation-related signaling and metabolic signature of BAs. Our results continuously elucidated the role that RBM4a plays in brown adipogenesis. Mice dissection. Male RBM4a −/− mice were generated as previously described 19 . Adult mice were fed a regular diet for 8 weeks. After being euthanized, interscapular fat tissues were collected from adult and embryonic mice, weighed, and immediately frozen until the RNA and proteins were extracted.

Methods
RNA extraction, complementary (c)DNA library construction, and sequencing. Total RNA was extracted using the PureLink RNA mini kit (Invitrogen, Camarillo, CA, USA) according to the manufacturer's protocol. The RNA quality was first evaluated using agarose gel electrophoresis and staining with ethidium bromide (EB). The integrity and quantity of the RNA were further assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies, Redwood, CA, USA). The total RNA with a high integrity number (RIN > 8.0) was subjected to library construction. In total, 8 μ g of RNA per sample was applied to construct three cDNA libraries using the NEBNext Ultra RNA Library Prep Kit from Illumina (NEB, Ipswich, MA, USA) according to the manufacturer's instructions. In brief, poly(A) messenger (m)RNA was enriched using oligo(dT)-attached magnetic beads. The poly(A) mRNA was next cleaved into small fragments using divalent cations in NEBNext first-strand synthesis reaction buffer. The cleaved fragments were subjected to synthesis of first-strand cDNA using random hexamer primers and Superscript III reverse transcriptase (Invitrogen). cDNA strands were next synthesized using DNA polymerase I. Overhanging portions of the cDNA duplexes were trimmed to blunt ends and adenylated at the 3′ ends of the DNA fragments. The DNA fragments were ligated with an adaptor containing a hairpin loop and purified. Three microliters of USER Enzyme (NEB) was applied to select cDNA fragments of 150 ~ 200 bp in length and adaptor-ligated before the polymerase chain reaction (PCR). The amplification program was then conducted using high-fidelity DNA polymerase and a universal primer set. The quality of amplicons was assessed using the Agilent Bioanalyzer 2100 system.
Index-coded amplicons were clustered using the cBot Cluster Generation System coupled with a TruSeq PE Cluster Kit v3-cBot-HS (Illumina) following the manufacturer's instructions. The prepared libraries were sequenced on an Illumina NextSeq 500 platform, and 150 ~ 200-bp paired-end reads were generated.
Read mapping, transcript annotation, and quantification. Preliminary reads were cleaned by trimming the adapter sequences and removing poly-N or low-quality sequences (Q < 20). The filtered reads were aligned to the mouse reference genome (GRCm37) using the Tophat v2.0.9 program. Tolerance parameters were the default setting to allow mismatches of fewer than two bases. Aligned reads were next subjected to generation of transcriptome assemblies using the Cufflink program. Mutant loci within the assembled transcripts were identified using SAMtools. These transcriptome assemblies generated from individual samples were merged together using the Cuffmerge utility to provide a standard for estimating transcript levels in each condition. Expression levels and the statistical significance of the merged assemblies were calculated using the Cuffdiff analysis.
Cell culture and differentiation. Mouse C2H10T1/2 fibroblast cells were cultured in Dulbecco's modified Eagle medium (DMEM; Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS) (Invitrogen). To induce adipogenesis, C3H10T1/2 cells were shifted to induction medium supplemented with 20% FBS, 0.5 mM IBMX, 12.7 μ M dexamethasone, and 10 μ g/ml insulin. Forty-eight hours after induction, the induction medium was replaced with differentiation medium (DM) supplemented with 10% FBS and 10 μ g/ml insulin, and it was replenished every 2 days. Plasmid construction. Expression vectors for the mouse Nova1 gene were constructed by placing the coding sequence in-frame into the p3XFLAG-CMV14 vector (Sigma, St. Louis, MO, USA). The mice Nova1 coding region was PCR-amplified using the reverse transcription (RT) product prepared from the total RNA of BATs as the template and then inserted into Hind III/Not I sites of the vector. pRL-Nova1 reporters were constructed by inserting the partial region of mouse Nova1 3′ untranslated region (UTR) fragments (128 nt) into the pRL-SV40 vector (Promega, Madison, WI, USA). Each fragment of the mouse Nova1 3′ UTR was PCR-amplified using a genomic DNA library prepared from mice BATs as the template and then inserted into Xba I/Not I sites of the pRL-SV40 vector. The mutant pRL-Nova1 3′ UTR-F1vector was constructed using the QuikChange site-directed mutagenesis system (Stratagene, Amsterdam, the Netherlands). Sequences of the PCR primer sets are listed in Supplementary Table 1. All constructs were auto-sequenced.
Transient transfection and RT-PCR analysis. C3H10T1/2 cells were grown to 60% ~ 70% confluence, and the indicated plasmid was transfected using Lipofectamine 3000 following the manufacturer's protocol (Invitrogen). After 24 h, total RNA and proteins were separately extracted using the Trizol reagent (Invitrogen). For the RT-PCR assay, 1 μ g of RNA was reverse-transcribed using SuperScriptase III (Invitrogen) in a 10-μ l reaction. The PCR analysis of individual genes was performed using gene-specific primer sets (Supplementary Table 1). The PCR-amplified amplicons of PKM and FGFR2 were then digested with PstI and EcoRV to discriminate products containing PKM2 exon 10 and FGFR2IIIc exons. Densities of the PCR products were determined using TotalLab Quant Software. A quantitative (q)RT-PCR was performed with SYBR green fluorescent dye and gene-specific primer sets (Supplementary Table 2) using an ABI One Step ™ PCR machine (Applied Biosystems, Foster City, CA, USA). The relative mRNA level was quantitated by the Δ Δ -Ct method, and the level of GAPDH mRNA served as the internal control.
In vivo translation assay. C3H10T1/2 cells were seeded in six-well plates (2 × 10 5 cells/well) 24 h prior to transfection. The transfection reaction mixture contained 0.5 μ g of the pRL-SV40 and engineered Renilla luciferase reporters which contained the partial Nova1 3′ UTR fragment, 1 μ g of the effector expression vector, and 0.5 μ g of the pGL3-basic vector (Promega) as the internal control. After 24 h, transfectants were lysed using passive lysis buffer, and cell debris was removed after centrifugation. Activities of the firefly and Renilla luciferases were measured using a dual-luciferase assay kit (Promega) and the Synergy HT multi-mode microplate reader (BioTek, Winooski, VT, USA).

Mitochondrial respiration assay.
A Seahorse XF24 extracellular flux analyzer (Seahorse Bioscience, North Billerica, MA, USA) was used to measure the oxygen consumption rate (OCR; as an indicator of mitochondrial respiration). In brief, 2 × 10 4 C3H10T1/2 cells were seeded in each well of Seahorse XF24 plates with 250 μ l of DMEM and incubated overnight. Prior to the measurement, cells were washed with unbuffered medium and immersed in 675 μ l of unbuffered medium without CO 2 for 1 h. The OCR was assessed in 8-min cycles as recommended by Seahorse Bioscience. The basal and maximal OCRs, and spare respiratory capacity were recorded following injection of complex-specific substrates, including FCCP (2 μ M), rotenone (2 μ M), and oligomycin (2.5 μ g/ml).

Mitochondria analysis.
Proliferating or differentiating C3H10T1/2 cells were subjected to DMEM medium containing100 nM MitoTracker Red FM (Invitrogen) for 45 min at 37 °C. Cells were washed with prewarmed culture medium and visualized with an Olympus IX81 microscope (Olympus, Tokyo, Japan). The signal strength of captured pictures were analyzed with TotalLab Quant Software.
Oil-red-O staining. Proliferating and differentiating C3H10T1/2 cells were washed twice with phosphate-buffered saline (PBS) and fixed with 4% paraformaldehyde for 60 min at room temperature. Cells were washed with PBS twice and rinsed with 60% isopropanol for 5 min at room temperature. Equilibrated cells were stained with a 0.3% filtered oil-red-O solution (Sigma) for 10 min at room temperature. Stained cells were washed with distilled water three times. For extraction of the oil-red-O dye, the culture dish with absolute isopropanol was shaken at room temperature for 2 h. The extract was centrifuged and analyzed at 550 nm using a NanoDrop 2000 spectrophotometer (Thermo).
Statistical analyses. Student's t-tests were performed to determine the significance of blot densitometry. P < 0.05 was considered statistically significant.