Folliculin-interacting protein FNIP2 impacts on overweight and obesity through a polymorphism in a conserved 3′ untranslated region

Abstract

Background

Overweight and obesity are defined by an anomalous or excessive fat accumulation that may compromise health. To find single-nucleotide polymorphisms (SNPs) influencing metabolic phenotypes associated with the obesity state, we analyze multiple anthropometric and clinical parameters in a cohort of 790 healthy volunteers and study potential associations with 48 manually curated SNPs, in metabolic genes functionally associated with the mechanistic target of rapamycin (mTOR) pathway.

Results

We identify and validate rs2291007 within a conserved region in the 3′UTR of folliculin-interacting protein FNIP2 that correlates with multiple leanness parameters. The T-to-C variant represents the major allele in Europeans and disrupts an ancestral target sequence of the miRNA miR-181b-5p, thus resulting in increased FNIP2 mRNA levels in cancer cell lines and in peripheral blood from carriers of the C allele. Because the miRNA binding site is conserved across vertebrates, we engineered the T-to-C substitution in the endogenous Fnip2 allele in mice. Primary cells derived from Fnip2 C/C mice show increased mRNA stability, and more importantly, Fnip2 C/C mice replicate the decreased adiposity and increased leanness observed in human volunteers. Finally, expression levels of FNIP2 in both human samples and mice negatively associate with leanness parameters, and moreover, are the most important contributor in a multifactorial model of body mass index prediction.

Conclusions

We propose that rs2291007 influences human leanness through an evolutionarily conserved modulation of FNIP2 mRNA levels.

Graphical Abstract

Background

Overweight and obesity are caused by an excessive and anomalous accumulation of adipose tissue and constitute a risk to health [1]. According to the World Health Organization (WHO), obesity is one of the most alarming health problems globally, excluding the sub-Saharan Africa and Asia. A body mass index (BMI, defined as the following ratio: weight (kg)/height (m)²) between 25 and 30 kg/m² is considered overweight in adults, while a BMI exceeding 30kg/m² is defined as obese state. Moreover, an elevated BMI strongly associates with several metabolic abnormalities, including insulin resistance and metabolic syndrome, themselves prominent risk factors for type 2 diabetes (T2D), cardiovascular diseases (CVDs) [2], and cancer [3, 4]. In addition, obesity and impaired metabolic health recently emerged as key risk factors for COVID-19 [5] and immune system dysfunction [6].

Both genetic and environmental factors contribute to the development of overweight and obesity. Lifestyle habits such as excessive caloric intake and insufficient physical activity are critical drivers of overweight and obesity [7]. In addition, genome-wide associations studies (GWAS), among other genetic approaches to define molecular drivers of obesity, have established strong associations between genetic variants and elevated BMI, being the fat mass and obesity-associated gene (FTO), the first genetic determinant of human body mass [8, 9]. A Genetic Investigation of ANthropometric Traits consortium (GIANT) meta-analysis (including more than 339,000 participants) linked a total of 97 loci with BMI, 56 of which were novel. The 97 loci account for 2.7% of BMI variation, and genome-wide studies estimate that common variants account for more than 20% of BMI variation [10,11,12]. Thus, genetic influence on BMI is still largely unknown, and in addition, most known associations have not been functionally validated.

It is not surprising that both genetic and environmental factors that contribute to obesity modulate cell signaling pathways responsive to nutrients and metabolic hormones. Among these pathways, the mechanistic target of rapamycin complex 1 (mTORC1) signaling pathway concurrently senses energy, nutrients, and growth factors and couples their sufficiency to the execution of anabolic cell growth and division [13]. MTORC1 itself is composed of the mTOR kinase and essential adaptor proteins (Raptor and mLST8), and dozens of proteins within several multimeric complexes participate in convergent regulatory cues that directly and indirectly control activation and inhibition of mTORC1. Substantial experimental evidence supports that mTORC1 controls several cellular and organismal processes that influence BMI, including food intake, insulin signaling, energy storage and consumption, metabolic activity, synthesis and secretion of hormones, and inter-organ communication [13,14,15]. Thus, it is reasonable to postulate that genetic variants affecting components of the mTORC1 pathway may contribute to determine overweight, obesity, and the associated abnormal metabolism.

While several studies have found single-nucleotide polymorphisms (SNPs) within components of the mTORC1 pathway affecting cancer susceptibility [16,17,18], mirror approaches establishing an association with increased BMI are awaited: to our knowledge, no genetic or epigenetic studies have linked SNPs in components of the mTOR pathway with overweight and obesity. The identification of such variants could pave the way toward tailored, precision therapeutic approaches targeting the mTORC1 pathway to combat the pathological states associated to elevated BMI.

Here, we measured BMI and multiple anthropometric and clinical features related to obesity in a cohort of 790 healthy volunteers and analyzed potential associations with 48 manually curated SNPs in metabolic genes with known functional associations to the mTORC1 pathway. We found an association between multiple metabolic parameters related to obesity and a human-specific SNP within an evolutionarily conserved region of the 3′UTR of the folliculin-interacting protein 2 (FNIP2) that affects mRNA stability. Such associations were replicated in a novel knock-in mouse strain genetically tailored to express the human-specific single-nucleotide variant. Moreover, in addition to the genetic association, the levels of FNIP2 mRNA inversely correlate with anthropometric and clinical features of obesity. Thus, we propose that both rs2291007 and FNIP2 blood mRNA levels are linked to obesity.

Results

FNIP2 polymorphism rs2291007 is associated with metabolic and obesity-related phenotypes

We selected a total of 38 genes encoding proteins with metabolic functions linked to the mTOR pathway (Additional file 1: Fig. S1). We then manually curated potentially functional SNPs to perform a customized genotyping chip, according to the following criteria: SNPs located in coding regions, regulatory SNPs of 5′UTR or 3′UTR sites or SNPs in splicing sites. In addition, a threshold of 15% was established in European minor allele frequency (MAF).

We selected a total of 56 SNPs in 25 genes and genotyped them in a cohort of 790 healthy individuals. Their genomic positions are summarized in Additional file 2: Table S2. Evidence of departure from Hardy-Weinberg equilibrium (HWE) was observed for eight SNPs. They were excluded from the analysis, although none of them remained statistically significant after a conservative Bonferroni correction for multiple testing.

Representation of −log10 p-values for additive model for each metabolic characteristic among the 790 Spanish volunteers are detailed in Fig. 1A and Additional file 1: Fig. S2. With a relatively permissive p-value threshold of 0.05, 33 SNPs in 17 genes were associated with 19 of 21 metabolic characteristics analyzed in our population, suggesting the relevance of genetic variations within selected genes in metabolic function and phenotype. Detailed information on rs number, genes, beta, 95% CI, and p-values for each metabolic association are summarized in Additional file 2: Table S3. Four SNPs located in three genes (two in FNIP2 and one in each of FLCN and RPTOR) were associated with metabolic phenotypes (FNIP2 with muscle and fat mass, visceral fatness, weight, BMI, and waist circumference; FLCN with hip circumference and RPTOR with systolic arterial pressure). Among these SNPs, the two located in FNIP2 gene maintained statistical significance after applying the Bonferroni correction for multiple comparisons (Fig. 1A). Both were associated with fat mass (rs17286116, beta=1.31, 95% CI (0.55–2.07), Bonferroni-corrected p-value= 0.04 and rs2291007, beta=1.33, 95% CI (0.59–2.07), Bonferroni-corrected p-value= 0.03). rs2291007 was also associated with muscle mass (beta= −0.64, 95% CI (−0.99 to −0.29), Bonferroni-corrected p-value=0.02) (Fig. 1A).

Fig. 1

Genetic associations. A Representation of −log10 p-values for phenotypic associations with metabolic characteristics in 790 volunteers. Each point represents one SNP within a gene. Associations of the SNPs with body mass index (BMI), weight, visceral fatness, fat and muscle mass, and hip and waist circumference were modeled through linear regression on the SNP adjusted by sex and age. The −log10 p-values following an additive model for the SNPs (effect in the homozygote minor allele is twice as much as that of the heterozygote) were represented. B Gene location and linkage disequilibrium (LD) among rs10857319, rs17286116, and rs2291007. LD was calculated using LDLink (https://ldlink.nci.nih.gov/). C rs2291007 associations with several metabolic characteristics. Associations were modeled through linear regression on the rs2291007 adjusted by sex and age following an additive model. Ninety-five percent confidence intervals were used in beta estimates. Bonferroni (Bonf.) method was applied for multiple test correction of the p-values

We further analyzed the three SNPs detected in FNIP2 genomic region (rs10857319, rs17286116, and rs2291007), located in 5′UTR, intronic, and 3′UTR regions, respectively (Fig. 1B and Additional file 2: Table S2). rs17286116 and rs2291007 share linkage disequilibrium, with a R² value of 0.735 (Fig. 1B), explaining the overlapping associations detected. rs2291007, located in FNIP2 3′UTR region, had the highest beta and lowest p-values. Moreover, conditional analysis identified rs2291007 as the strongest independent signal for this locus, because adding the rs2291007 SNP to the analysis resulted in the other two SNPs (rs10857319 and rs17286116) losing statistical significance in predictive models for fat mass adjusted by sex and age. In addition to the associations with fat and muscle mass, the minor allele T of rs2291007 positively associated with several metabolic phenotypes, including weight and visceral fatness (beta= 2.27, 95% CI (0.91–3.62); p-value=0.001, Bonferroni-corrected p-value= 0.06 and beta= 0.5, 95% CI (0.2–0.81); p-value=0.001, Bonferroni-corrected p-value= 0.08 respectively). We also detected various association trends of rs2291007 with BMI, waist circumference, and hip circumference (beta= 0.68, 95% CI (0.24–1.12); p-value=0.002, beta= 1.77, 95% CI (0.53–3); p-value=0.005 and beta= 1.28, 95% CI (0.04–2.51); p-value=0.04 respectively) (Fig. 1C, Additional file 2: Table S3). No statistically significant correlations of rs2291007 with lipid or glucose profile and heart or nutritional parameters were found (Additional file 1: Fig. S2 and Additional file 2: Table S3).

In summary, our genetic study on 48 loci revealed that minor allele T of rs2291007 in FNIP2 gene is linked to metabolic and obesity-related phenotypes, being associated with elevated fat mass, visceral fatness, weight, BMI, and waist and hip circumferences, and with decreased muscle mass in healthy individuals of European origin.

miR-181b-5p selectively binds the 3′UTR of the FNIP2 T allele

SNP rs2291007 is a C/T variation in the 3′UTR of FNIP2 and chromosome 4 open reading frame 45—C4orf45—gene, on chromosome 4. This genomic region is evolutionary conserved across vertebrates, as illustrated in Fig. 2A. The T allele in rs2291007 is ancestral and the common allele in the African population (Fig. 2B). In the European population, and in our study of healthy individuals with European origin, T is the minor allele of rs2291007 (Fig. 2B).

Fig. 2

Binding of miR-181b-5p to the 3′UTR of the FNIP2 T allele. A TargetScanHuman predictions (http://www.targetscan.org/vert_72/) for rs2291007 genomic region. B rs2291007 frequencies across European (EUR) and African (AFR) populations (https://www.ensembl.org/). C Sfold (https://sfold.wadsworth.org/cgi-bin/index.pl) prediction for binding of miR-181b-5p to the rs2291007 region. D Dual-luciferase assays on psiCHECK vectors, for the binding of miR-181b-5p to rs2291007 region in presence of C or T alleles. Sequences for FNIP2 3′UTR with rs2291007 C or T alleles were cloned into psiCheck2 by ATUM. The relative luciferase activity (Renilla luminescence/firefly luminescence) was determined after transfection using the Dual-Luciferase Reporter assay system representing the translational repression of FNIP2 upon binding to their 3′UTR of miR-181b-5p. Unpaired Student’s t test was used

According to TargetScanHuman predictions (http://www.targetscan.org/vert_72/) [19], the sequence encompassing the SNP is a putative target region for miR-181b-5p binding (Fig. 2A). We in silico tested putative binding of miR-181b-5p to the rs2291007 region using Sfold (https://sfold.wadsworth.org/cgi-bin/index.pl) [20] (Fig. 2C). According to structural predictions, miR-181b-5p binds exclusively the T allele with two possible secondary conformations (Fig. 2C), and no sites for such binding are predicted for the C allele abundant in Europeans. Hence, only the T ancestral allele is potentially subjected to binding to, and thus, to regulation by, miR-181b-5p.

To validate the interaction of this miRNA with rs2291007 region, we performed dual-luciferase assays (Fig. 2D). In agreement with structural predictions, miR-181b-5p downregulated the activity of a reporter construction carrying the T, but not the C, allele. This result strongly suggests that miR-181b-5p selectively controls the expression of the T allele of FNIP2 rs2291007 through binding to the 3′UTR.

Levels of FNIP2 mRNA associate with metabolic and obesity-related phenotypes

Following on the observation that miR-181b-5p binds to the sequence containing the T allele of rs2291007, we sought evidence for a potential functional interaction between miR-181b-5p and rs2291007 to modulate the expression of FNIP2. First, we analyzed FNIP2 mRNA levels in a panel of 965 Catalogue of Somatic Mutations in Cancer (COSMIC) cell lines [21] and found that rs2291007 minor homozygous cell lines (TT) (22.2%) displayed statistically significant decreased expression of FNIP2, as compared to cells carrying at least one copy of the C allele (77.8%) (Fig. 3A), consistently with the predicted loss of negative regulation by miR-181b-5p in the C allele.

Fig. 3

Transcriptomic consequences of rs2291007. A Box plot of the association between gene expression level for FNIP2 and genotype for rs2291007. The box plot shows how the FNIP2 expression values are distributed for each genotype from recessive model of inheritance in COSMIC cell lines. Common/major Homozygous plus Heterozygous alleles (C/C + C/T); Variant/minor Homozygous allele (T/T). Differences between gene expression in two groups were tested through Student’s t test. B Box plot of the association between gene expression level for FNIP2 and genotype for rs2291007 in 161 healthy volunteers. The 2^−ΔΔCt method was applied to calculate the relative gene expression. Differences between gene expression in two groups were tested through Student’s t test and through linear models with adjustment variables. C Associations of FNIP2 expression and phenotypic characteristics related to metabolism and obesity in 161 healthy volunteers. Differences between gene expression in two groups were tested through linear models with adjustment variables. All tests were bilateral. Ninety-five percent confidence intervals were used in beta estimates. Bonferroni method was applied for multiple-test correction of all p-values. D Associations of FLCN expression and phenotypic characteristics related to metabolism and obesity in 161 healthy volunteers. Differences between gene expression in two groups were tested through linear models with adjustment variables. All tests were bilateral. Ninety-five percent confidence intervals were used in beta estimates. Bonferroni method was applied for multiple-test correction of all p-values

Next, we aimed to confirm decreased FNIP2 expression in carriers of the rs2291007 T allele by analyzing peripheral blood mononuclear cells (PBMC) of 161 healthy volunteers. Indeed, we found that homozygous TT individuals (25.46%) also showed lower FNIP2 expression (p-value= 0.045) than individuals carrying at least one C allele (74.54%). When adjusted to sex and age, the associated p-value increased (p-value= 0.1) (Fig. 3B). We also analyzed the expression of FNIP2 and miR-181b-5p in a subset of 89 healthy volunteers, and analysis of co-expression of FNIP2 and miR-181b-5p showed an inverse correlation trend exclusively in carriers of T allele (r²=0.577, p-value=0.18 and r²=−0.777, p-value=0.35), while no trend was observed in carriers homozygous for the C allele (r²=0.004, p-value=0.99) (Additional file 1: Fig. S3A). These results support the existence of a functional relationship between miR-181b-5p and FNIP2 expression by the selective binding to the T allele in rs2299007. We next analyzed the association between the expression of FNIP2 and clinical parameters of healthy volunteers. Interestingly, and consistently with our expectations, the expression levels of FNIP2 strongly associated with decreased weight, BMI, fat mass, visceral fatness, waist, and hip circumferences, systolic (SBP) and diastolic (DBP) blood pressure, blood levels of triglycerides and glycated hemoglobin, and low basal metabolism, and with increased muscle mass, all with high statistically significant p-values (Fig. 3C and Additional file 2: Table S4). These results support FNIP2 expression in blood as a powerful marker of metabolic alterations.

Because FNIP2 is part of a heterotrimeric complex, we also analyzed expression of the other members, FNIP1 and FLCN. FNIP2 gene expression significantly correlated with both FNIP1 and FLCN (Additional file 1: Fig. S3B) and we observed associations between the levels of FLCN expression and increased weight, BMI, fat mass, visceral fatness, waist and hip circumferences, DBP, and blood levels of glucose, leptin, and triglycerides; and decreased muscle mass (Fig. 3D and Additional file 2: Table S4). In contrast, we did not detect any relationship between the expression of FNIP1 or miR-181b and phenotypic characteristics related to overweight or obesity (Additional file 2: Table S4).

A knock-in mouse model for rs2291007

The ancestral T allelic variant is evolutionary conserved in mammals (Fig. 2A) and present in the 3′UTR region of mouse Fnip2 together with the surrounding seed region for miRNA-181b-5p (Fig. 4A). Thus, we decided to genetically engineer the C allelic variant in the mouse genome and to assess its functional impact on mouse weight and fat content. We used CRISPR/Cas9 genome engineering [22] in mouse zygotes to knock-in the T-to-C change. For technical reasons, we substituted one additional nucleotide in +6 position to disrupt the PAM sequence (a G-to-C change) so as to prevent recognition by CRISPR/Cas9 and unwanted sequence retargeting (Fig. 4A). Importantly, this additional change does not alter the seed region nor the predicted binding of miR-181b-5p (Additional file 1: Fig. S4A). Blastocytes of pure C57BL/6 background have raised to founder targeted chimeras, and Fnip2^T/T, Fnip2^T/C, and Fnip2^C/C mice were obtained at the expected Mendelian ratios from heterozygous breeders (Additional file 1: Fig. S4B). Macroscopically, Fnip2^C/C and Fnip2^T/C knock-in mice were indistinguishable from those expressing the ancestral T allele in homozygosity.

Fig. 4

A Fnip2^C Knock-in mouse to model FNIP2 rs2291007. A Experimental strategy to generate a knock-in mouse model mimicking FNIP2 rs2291007C using CRISPR Cas9 gene-editing tool in blastocysts. B RT-qPCR from liver and gonadal white adipose tissue (WAT) obtained from Fnip2^T/T (n=8) and Fnip2^C/C (n=9) mice ad libitum fed. Fnip2 mRNA levels are normalized to β-actin levels and relativized to the levels in Fnip2^T/T. Statistical significance was calculated by using unpaired two-tailed t-test. C Fnip2 mRNA levels were measured by RT-qPCR in mouse embryonic fibroblasts (MEFs) derived from Fnip2^T/T (n=5) and Fnip2^C/C (n=5) mice in complete Roswell Park Memorial Institute medium (RPMI; supplemented with dialyzed serum) or RPMI without amino acids for 2 h. Fnip2 mRNA levels are normalized to β-actin levels and relativized to the levels in Fnip2^T/T +AA. D MEFs derived from Fnip2^T/T and Fnip2^C/C mice were deprived of all amino acids (AA) in RPMI supplemented with dialyzed serum for 45 min and re-stimulated with AA for 5, 10, and 20 min. Whole-cell protein lysates were immunoblotted for the indicated proteins. E MEFs derived from Fnip2^T/T (n=5) and Fnip2^C/C (n=5) mice were treated with 5μg/mL Actinomycin D for 2 and 4h. Fnip2 mRNA levels were measured by RT-qPCR, normalized to β-actin levels, and then made relative to Fnip2 levels in untreated samples. Statistical significance was calculated using 2-way ANOVA with Sidak’s multiple comparison correction. F The area under the curve (AUC) was calculated from E and statistical significance was calculated by using unpaired two-tailed t-test

There are no commercially available antibodies against mouse Fnip2 to assess a potential difference in Fnip2 protein levels in Fnip2^T/T versus Fnip2^C/C cells and organs, so we first measured the levels of Fnip2 mRNA in liver and gonadal WAT samples from ad libitum fed and 16-h-fasted mice. In contrast to the positive association between the C allele and steady-state levels of FNIP2 mRNA observed in human cancer cell lines and in PBMC from healthy volunteers (Fig. 3), we found no association in mouse tissues (Fig. 4B and Additional file 1: Fig. S4C). We next obtained Fnip2^T/T and Fnip2^C/C mouse embryonic fibroblasts (MEFs) and quantified the levels of Fnip2 mRNA in complete culture medium and in medium without amino acids to modulate the expression of Fnip2. Although we observed the expected increase in Fnip2 mRNA levels upon amino acid deprivation, there was no difference between the levels of Fnip2 mRNA in Fnip2^T/T and Fnip2^C/C MEFs (Fig. 4C). Consistently, the regulation of the mTORC1 pathway upon amino acid deprivation and stimulation, revealed by the phosphorylation of S6K1 in threonine 389, and by the upshift in the band corresponding to total levels of the transcription factor EB (TFEB) band caused by mTORC1-dependent phosphorylation, were indistinguishable between Fnip2^T/T and Fnip2^C/C MEFs (Fig. 4D). The absence of evidence for an increase in the mRNA levels in mouse cells expressing the C allele in homozygosity in the assayed conditions does not rule out that, under specific perturbations, such difference may exist; but to exclude the possibility that similar mRNA levels of Fnip2 may be a consequence of lack of expression of the miR-181b-5p, we measured its levels in cells cultured with and without amino acids for 2 h. miR181b-5p was detected at similar levels in Fnip2^T/T and Fnip2^C/C MEFs in both culture conditions (Additional file 1: Fig. S4E). Alternatively, if miR181b-5p is present, a prediction is that the C variant should be more stable than the T variant, regardless of other compensatory mechanisms that may obscure such difference under steady-state synthesis of Fnip2 mRNA. Thus, we halted new synthesis of all mRNA with Actinomycin D, inhibitor of RNA polymerase II activity, and measured the relative decay of Fnip2 mRNA levels in Fnip2^T/T and Fnip2^C/C MEFs. Consistently with impaired binding of the Fnip2^C variant to miR-81b-5p, the decay of Fnip2 mRNA levels was significantly slower in Fnip2^C/C cells, in comparison to that of Fnip2 mRNA in Fnip2^T/T cells (Fig. 4E, F), indicating an increased stability of Fnip2 mRNA in presence of rs2291007 C allele upon an abrupt interruption of new mRNA synthesis. In summary, the T-to-C substitution in mouse Fnip2 does not have obvious effects on mTORC1 activity in the conditions assayed, neither does it lead to a detectable increase in Fnip2 mRNA in steady state, but results in an increased stability of the Fnip2 mRNA, a result that supports a disrupted binding of miR-181b-5p to the Fnip2^C variant.

As mentioned, Fnip2^C/C mice were viable, macroscopically indistinguishable, and fertile. While no differences in body weight were seen between adult Fnip2^T/T and Fnip2^C/C mice (Fig. 5A), we observed a negative correlation between mRNA levels of Fnip2 and mouse body weight (Fig. 5B), in sharp consistency with the association found in human samples (Fig. 3C). Such negative association with body weight was exclusive for Fnip2, as we saw no statistically significant association between the mRNA levels of the other two components of the Folliculin complex, Fnip1 and Flcn, and body weight (Additional file 1: Fig. S5A). Nevertheless, a positive association on the mRNA levels of the three components of the complex (Fnip2, Fnip1, and Flcn) was observed (Additional file 1: Fig. S5B). Strikingly, in agreement with the association found in healthy volunteers (Fig. 1A, C), a significant decrease in the fat content was recapitulated in 5–6-week-old Fnip2^C/C male mice, and in female Fnip2^C/C mice analyzed at 5–6 and 11–15 weeks of age (Fig. 5C, D), and a similar trend was detected in 1-year-old females (Additional file 1: Fig. S5C). This difference does not seem to occur by a selective decrease on specific WAT depots, but from a general reduction in both visceral and subcutaneous depots (Additional file 1: Fig. S5D). To further investigate the consequences of the expression of the Fnip2^C variant in mice, we performed a transcriptomic analysis of two metabolically relevant organs, liver, and visceral WAT, from Fnip2^T/T and Fnip2^C/C mice. Strikingly, while only 9 genes were differentially expressed in livers, 4795 genes were differentially expressed in WAT from Fnip2^C/C versus Fnip2^T/T mice, indicating a comparably larger effect of the expression of the SNP in Fnip2 in adipose tissue, as compared to liver. We next conducted Gene Set Enrichment Analyses (GSEA) in samples from Fnip2^C/C versus Fnip2^T/T livers and WAT. Strikingly, among the top 5 signatures identified in both liver and WAT analyses, “adipogenesis,” “fatty acid metabolism,” and “mTORC1 signaling” were enriched samples from of Fnip2^T/T mice (Fig. 5E, F; Additional file 1: Fig. S5E and S5F), consistently with the increased fat content observed in Fnip2^T/T mice.

Fig. 5

Impact of the engineered T-to-C substitution on mouse adiposity. A Body weight of 11–15-week-old Fnip2^T/T (n=12) and Fnip2^C/C (n=13) females and 10–13-week-old Fnip2^T/T (n=19) and Fnip2^C/C (n=21) males. B Correlation between Fnip2 mRNA levels in livers from Fnip2^T/T (n=14) and Fnip2^C/C (n=15) mice and their body weight. C Fat content was obtained by measuring body composition in 5–6-week-old Fnip2^T/T (n=5) and Fnip2^C/C (n=6) females and Fnip2^T/T (n=8) and Fnip2^C/C (n=9) males. D Fat content was obtained by measuring body composition in 11–15-week-old Fnip2^T/T (n=8) and Fnip2^C/C (n=9) females and 10–13-week-old Fnip2^T/T (n=9) and Fnip2^C/C (n=9) males. In all cases, statistical significance was calculated using an unpaired two-tailed t-test

Thus, altogether, the mouse genetic data strongly support that the 3′UTR of FNIP2 is an evolutionary conserved genetic determinant of lean-fat mass ratio.

Multifactorial genetic model for overweight and obesity risk

Overweight and obesity are complex conditions modulated by several causes, so we designed a multifactorial model to predict BMI taking into consideration the relevance of genetic susceptibility of the FLCN-FNIP complex investigated herein. We derived a linear regression model that included the rs2291007 SNP (in additive form), three gene expression variables (FNIP2, FNIP1, and FLCN), plus sex and age. Importantly, the inclusion of a FNIP2*rs2291007 interaction was significant and increased the optimism-corrected R², so it was accepted in the final model. Figure 6A shows the estimated parameters of this model, and Fig. 6B displays the result of the bootstrap validation. We found that increased FNIP2 expression associates with a decreased BMI (beta=−3.08, 95% CI (−3.92, −2.25), p-value= 2.12×10⁻¹⁵). Conversely, the beta parameter for the FNIP2*rs2291007 interaction is positive (beta=0.636, 95% CI (0.35, 1.52), p-value= 0.03), which means that the inclusion of a T variant results in a significantly less negative slope, and thus to a less dramatic decrease of BMI with higher FNIP2 expression values. The inclusion of miR181-5p expression did not result in an improved model nor did the interactions between gender and rs2291007 (gender*rs2291007) or between gender and FNIP2 gene expression (gender*FNIP2), neither the use of a codominant or dominant assumption for the SNP.

Fig. 6

Multifactorial genetic model for overweight and obesity risk. A Multifactorial model to predict body mass index (BMI). Estimated parameters and 95% CI, plus p-values of variables in the multifactorial model. The linear regression model included the rs2291007 SNP (in additive form), three gene expression variables (FNIP2, FNIP1, and FLCN), plus sex, age, and the interaction between FNIP2 gene expression and rs2291007. B Bootstrap validation of the multifactorial model. Multifactorial model was validated through bootstrap validation, using 2000 bootstrap samples, which correction by optimism (MSE, Mean Square Error). C Variable importance plot for the multifactorial model, using χ² – df (degrees of freedom) as metrics for the importance of each variable

The variable importance plot for the multifactorial model of BMI prediction (Fig. 6C) establishes that the most important variable is the FNIP2 expression, followed by age, FLCN, FNIP1, and the FNIP2*rs2291007 interaction. Collectively, these results provide evidence that rs2291007-FNIP2-Folliculin complex could modulate overweight and obesity.

Discussion

Obesity and overweight are prevalent in developed countries and predispose to several comorbidities and diseases, such as most types of cancer. Recently, the impact that obesity has on the immune system has acquired a novel dimension of importance for the quality of life of the global population [5] because of the positive association between weight and adverse effects of SARS-CoV2 infection. Obesity is preventable, but causes and consequences must be deeply understood to design efficient preventive measures and tools [23]. Many parallel efforts have been undertaken to understand the genetic bases of this disease, including several GWAS studies that have opened a window of knowledge with the early identification of the FTO gene as the first genetic determinant of human body weight [8, 9]. While additional studies have further contributed to pinpointing genetic bases for the predisposition to obesity, this knowledge can explain so far approximately 20% of genetic predisposition. We restricted a genetic window of analysis to components of a metabolically relevant pathway, the mTOR pathway (Additional file 1: Fig. S1), due to its involvement in the sensing of energy, nutrients, and stress, as well as growth factors, and its deregulation in the obesity state [24].

In this work, our analysis of common genetic variations in this selected set of genes in almost 800 individuals of European origin revealed an association of rs2291007 in FNIP2 with BMI and phenotypic characteristics of the obese state (Fig. 1A). The polymorphism rs2291007 is located in the 3′UTR of the FNIP2 locus, and it has not been previously associated in the literature with overweight or obesity. Other variations in the FNIP2 gene have been associated in GWAS studies with several conditions (depression, intelligence; schizophrenia, smoking initiation; coffee consumption measurement) but none related to BMI or metabolic alterations (https://www.ebi.ac.uk/gwas/genes/FNIP2). Notably, germ-line mutations in other members of the folliculin complex, FLCN gene, are associated with Birt-Hogg-Dube syndrome, which is characterized by fibrofolliculomas, renal tumors, lung cysts, and pneumothorax [25]. For first time, this study describes that the minor allele T of rs2291007 in the FNIP2 gene is associated with metabolic and obesity-related phenotypes (elevated fat mass, visceral fatness, weight, BMI, and waist and hip circumferences) and with decreased muscle mass in healthy European individuals (Fig. 1A,B).

The aforementioned SNP is located within a conserved microRNA binding sequence. Typically, microRNAs posttranscriptionally regulate mRNA expression by binding to the 3′UTR of target mRNAs [26]. Because miRNA binding sites are constrained by secondary stabilizing interactions with the target mRNAs, SNPs located in 3′UTR sequences can result in differential miRNA binding, as they may either abolish or create a microRNA target, thus affecting mRNA levels [27]. We have shown, in cultured cells, that miR-181b-5p binds to the rs2291007 region and abolishes the expression of FNIP2 through binding selectively to the T allele (Fig. 2). Consistently, T carriers of rs2291007 showed diminished FNIP2 expression in peripheral blood, and importantly, its expression strongly associates with decreased obesity-related metabolic parameters (Fig. 3).

The conservation of the 3′UTR sequence of the FNIP2 SNP across mammals allowed us to interrogate the functional effect of this T-to-C human-specific nucleotide change in mice by knocking-in the rs2291007 C variant in the mouse genome. The lack of commercial antibodies for mouse Fnip2 protein precluded its quantification, but mRNA levels of Fnip2^C and Fnip2^T could be ascertained. While differences between the two variants were not detected in steady-state conditions, increased levels of the Fnip2^C mRNA were seen upon acute inhibition of RNA pol II activity with Actinomycin D, consistently with the prediction. Future work may uncover the physiological setting in which mRNA stability yields a difference in Fnip2 protein levels, and thus, on the control of mTORC1 activity, which is seen affected in GSEA of samples from Fnip2^C/C and Fnip2^T/T mice (Fig. 5E, Additional file 1: S5E) but not acutely in cells derived from them (Fig. 4D). Alternatively, Fnip2 levels may be more strikingly affecting other functions reported to be controlled by FNIPs [28], or other, still unknown metabolic roles.

Regardless of the timing and molecular cues that provide a functional explanation for the differences seen in carriers of the FNIP2 T versus C SNP in healthy volunteers, there is a remarkable correlation between the observations in human and mouse: Fnip2^C/C mice show the same decrease in fat content as that observed in volunteers, a difference that is more prominent in female mice at all ages. RNA sequencing analyses revealed that the transcriptome of the WAT is largely affected by the presence of the Fnip2 variant, with more than 4000 genes significantly different, in contrast to less than a dozen genes differentially expressed in liver samples. Moreover, samples from both livers and WAT from Fnip2^T/T mice show a dramatic and consistent enrichment in adipogenesis and fatty acid metabolism-related gene sets (Fig. 5E, F; Additional file 1: Fig. S5E and S5F), consistently with the association between the expression of the FNIP2 T allele and increased adiposity found in both species. In addition, the same negative correlation between the levels of human FNIP2 mRNA and body weight is seen when analyzing Fnip2 mRNA levels and mouse weight. This whole recapitulation is more remarkable when considering that rs2291007 has linkage disequilibrium with additional variants in the genomic region of FNIP2, which are not modeled in these novel knock-in mice, thus suggesting that the T-to-C change is the main modulator of human leanness within the FNIP2 locus.

The identification of relevant biomarkers linked to obesity and its comorbidities is an unsolved challenge [29]. PBMCs can serve as a source of molecular biomarkers for diverse metabolic alterations: PBMCs are easily sampled and mirror changes in the expression of metabolic genes in internal organs, thus providing accessible information on changes in the early obese state [30]. Thus, based on our results, we designed a multifactorial model to predict BMI in a healthy population. We propose a peripheral blood biomarker for BMI composed of FNIP2, FLCN, and FNIP1 expression, plus age of the patient and the effect of rs2291007 (Fig. 6). Although our findings were highly consistent and bootstrap validation was implemented, further studies in an independent cohort should be performed in order to validate the multifactorial model for BMI prediction. Prediction of overweight and obesity can be of paramount importance for early therapeutic interventions focused on obesity-related comorbidities.

Collectively, our work supports a critical role for FNIP2 in the control of human leanness.

Conclusions

• We found an association of rs2291007 minor allele T in the FNIP2 gene with metabolic and obesity-related phenotypes and with decreased muscle mass in healthy European individuals.

• T carriers of rs2291007 have decreased expression of FNIP2 in peripheral blood, and importantly, FNIP2 mRNA levels strongly associate with decreased obesity-related metabolic parameters.

• Fnip2^C/C knock-in mice replicate the decreased adiposity and increased leanness observed in human volunteers.

• Primary cells derived from Fnip2^C/C mice show increased mRNA stability.

• We propose a peripheral blood biomarker for BMI composed of FNIP2 expression, age of the patient, and the effect of rs2291007 on FNIP2 expression.

• Together, our works revealed a crucial role of FNIP2 in the control of human leanness.

Methods

Subjects and sample structure

Genotyping was carried out on 790 volunteers belonging to the Platform for Clinical Trials in Nutrition and Health (GENYAL Platform) of IMDEA-Food Institute. Among them, RNA from 161 volunteers was used for gene expression analysis and microRNA studies were performed in 89 individuals. Volunteers, aged between 18 and 72 years, were recruited during the period 2012–2018 and did not suffer from any serious diseases. All contributors signed the consent participation form. These studies were approved by the IMDEA Food Research Ethics Committee and methodologies conformed to the standards set by the Declaration of Helsinki.

Relevant phenotypic information is summarized in Additional file 2: Table S1 [31,32,33,34,35,36,37,38,39,40]. Anthropometric magnitudes were measured under standardized methods. Height (cm) was assessed to the nearest 0.1 cm using a stadiometer (Leicester Biológica Tecnología Médica SL, Barcelona, Spain). Weight (kg), fat mass (%), and muscle mass (%) were estimated using bioelectrical impedance analysis (Body Composition Monitor BF511-OMRON HEALTHCARE, LT, Kyoto, Japan). Body mass index (BMI) was calculated using these estimates and as defined by the Quetelet Index (weight (kg)/height (m)²). The World Health Organization’s criteria (WHO) was employed to catalog the volunteers as normal weight when BMI < 25 kg/m² and as overweight when BMI ≥ 25 kg/m². Waist and hip circumference were measured with a flexible Dry 201 metal tape, with measuring range 0–150 cm and 1 mm of precision (Quirumed, Valencia, Spain). Blood sample extractions, biochemical determinations (lipid and glycemic profiles), and nutritional assessments (dietary and physical activity parameters) were collected following an overnight fasting as previously described [41,42,43,44].

SNP genotyping

Genetic information was obtained from NCBI-dbSNP (https://www.ncbi.nlm.nih.gov/snp/), Ensembl (http://www.ensembl.org/index.html), and GWAS catalog (https://www.ebi.ac.uk/gwas/). Fifty-six SNPs in 25 genes of the mTOR pathway (Additional file 1: Fig. S1 and Additional file 2: Table S2) were selected.

Genomic DNA from peripheral blood was extracted using the QIAamp DNA Blood Mini Kit (Qiagen Sciences, Inc, Germantown, MD, USA). Genotyping was performed with QuantStudio 12 K Flex Real-Time PCR System (Life Technologies Inc., Carlsbad, CA, USA) using TaqMan OpenArray plates following the manufacturer’s instructions and results were analyzed using TaqMan Genotyper software.

MicroRNA experiments

Two in silico tools were used for predicting miR181 structure and binding to rs2291007 region: TargetScanHuman (http://www.targetscan.org/vert_72/) for prediction of microRNA targets [19], and Sfold (https://sfold.wadsworth.org/cgi-bin/index.pl) for statistical folding of nucleic acids and studies of regulatory RNAs [20].

Sequences for FNIP2 3′UTR with rs2291007 C or T alleles were cloned into psiCheck2 by ATUM (Newark, California). HEK-293T cells were transfected using Lipofectamine 2000 (Life Technologies, Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s recommendations. Each reaction contained 100 ng of either psiCheck2-3′UTR-FNIP2-rs2291007-C or psiCheck2-3′UTR-FNIP2-rs2291007-T and 30 nM of miR181b mimic miRNA (mirVana® miRNA mimic, Thermo Fisher Scientific, Waltham, MA, USA) [45]. Relative luciferase activity (Renilla luminescence/firefly luminescence) was determined after transfection using the Dual-Luciferase Reporter assay system (Promega, Madison, WI, USA) as surrogate for the translational repression of FNIP2 upon binding to their 3′UTR of miR181b.

mRNA expression analysis

For human peripheral blood samples, we used TRIzol method or RNeasy Mini Kit (Qiagen, Germantown, MD, USA) following the manufacturer’s conditions to obtain total RNA. Reverse transcription was performed with the High-Capacity cDNA Reverse Transcription kit (Thermo Fisher, Madrid, Spain), following the manufacturer’s instructions. RT-PCR reactions were performed as previously described [46] using QuantStudio 12 K Flex Real-Time PCR System (Life Technologies Inc., Carlsbad, CA, USA), with specific Taqman probes: FNIP2 (Hs01574322_m1), FLCN (Hs00376065_m1), FNIP1 (Hs00382846_m1), GAPDH (Hs02786624_g1). The 2^−ΔΔCt method was applied to calculate the relative gene expression. COSMIC Cell Lines Project (https://cancer.sanger.ac.uk/cell_lines) was used to download genotypes and gene expression values for FNIP2 in 965 cell lines.

For mouse tissues and primary cells, RNA was extracted using TRIzol together with RNeasy Mini Kit (Qiagen, 74106) or Direct-zol RNA Miniprep (Zymo Research, R2051). Reverse transcription was performed using SuperScript IV VILO Master Mix (Invitrogen, 11756500). Quantitative real-time PCR was run in triplicates using GoTaq qPCR Master Mix (Promega, A6001) in a QuantStudio 6 Flex Real-Time PCR System thermocycler (Applied Biosystems). The 2^−ΔΔCt method was applied to calculate the relative gene expression, using β-actin as reference gene.

microRNA expression analysis

Total RNA obtained from human peripheral blood samples and from mouse primary cells was retrotranscribed using TaqMan microRNA reverse transcription kit (Applied Biosystems, 4366596). Real-time PCR was run in triplicates using TaqMan Fast Advanced Master Mix (Applied Biosystems, 4444557) with specific TaqMan probes: hsa-miR-181b (4427975, MI0018778), U6 snRNA (4427975, NR_004394), mmu-miR-181b-5p (4440886, MI0000723), and snoRNA202 (4427975, AF357327).

Mouse genome engineering

To engineer Fnip2 rs2299007 C allele in mice, C57BL/6 mouse blastocysts were injected with Cas9, a single-guide (sg) RNA targeting the sequence of interest and a repair single-stranded oligonucleotide containing the intended mutations flanked by 70 bp of homology arms adjacent to the double-strand break site. Genotyping was performed by specific amplification followed by restriction fragment length polymorphism (RFLP) and/or Sanger sequencing.

sgRNA: GATAAGTGATATGAATGTAT

Repair single-stranded oligonucleotide:

CCGAGCAGAAGTGTCTCAGTGTCCTGTAATGACCTCTTCTAGCATGTTGCAGTTTTATATTTGTTAAGTTGATAAGTGATATGAACGTATACGCAATTGTGTATGTTTTCAAAAAGGACAATGAAAATTTAAAATGTAGCTTCCACACTTGTGCATAATTCCA

Mouse experimentation

All animal procedures carried out at the CNIO were performed according to protocols approved by the CNIO-ISCIII Ethics Committee for Research and Animal Welfare (CEIyBA), under protocol number PROEX 015/18. Mice were housed under specific pathogen-free (SPF) conditions, at 22 °C and with 12-h dark/light cycles. Mice were fed with a standard chow diet (Harlan Teklad 2018). For fasting experiments, mice were placed in clean cages without access to food from 4 pm to 8 am.

Animal imaging

For densitometry analysis, body composition (body weight, fat mass and lean muscle mass) was measured using dual-energy X-ray absorptiometry (DEXA) PIXImus, Mouse Densitometer (GE Lunar co, Madison, WI, USA), software version 1.46. During the measurements, mice were anesthetized with isoflurane. Quality control was performed using a calibrated phantom before imaging.

Mouse embryonic fibroblasts

MEFs were isolated from E13.5 embryos using a protocol consisting of a chemical digestion with trypsin followed by mechanical disaggregation. For amino acid starvation experiments, cells were rinsed 3 times and placed in RPMI medium without amino acids (US Biological, R8999-04A) supplemented with 10% dialyzed serum. Amino acid stimulation was performed adding a cocktail containing the 20 amino acids (with concentration as in RPMI) directly to cell culture plates.

Immunoblotting

Cells were rinsed once with ice-cold PBS and lysed in ice-cold protein lysis buffer containing 50 mM HEPES (pH 7.4), 40 mM NaCl, 2 mM EDTA, 1.5 mM sodium orthovanadate, 50 mM NaF, 10 mM pyrophosphate, 10 mM glycerophosphate, and 1% Triton X-100 and one tablet of complete protease inhibitors (Roche) per 25 ml. Cell lysates were cleared by centrifugation at maximum speed for 10 min. Protein content was measured from extracts with BCA Protein Assay. Protein extracts were denatured by adding sample buffer and boiling for 5 min, resolved by SDS-PAGE and analyzed by immunoblotting. Western blot analyses were performed according to standard procedures. Results were visualized using Odyssey Infrared Imaging System LI-COR Biosciences. The following primary antibodies were used: P-T389-S6K (Cell Signaling Technology, 9234), S6K (Cell Signaling Technology, 2708), TFEB (Bethyl Laboratories, A303-673A), and β-actin (Sigma-Aldrich, A1978).

mRNA stability measurements

For mRNA stability measurements MEFs were treated with 5 μg/mL Actinomycin D for 2 and 4 h to stop transcription. RNA was extracted using Direct-zol RNA Miniprep (Zymo Research, R2051).

Statistical analysis in human studies

Descriptive analysis was implemented for different continuous and categorical variables. Associations between gene expression variables were analyzed through Pearson correlation coefficients and tested through the corresponding test. Differences between gene expression in two groups were tested through Student’s t test, or through linear models with adjustment variables. SNPs were categorized by genotype (homozygote minor allele, heterozygote, and homozygote major allele). Deviations from Hardy-Weinberg equilibrium of genotype frequencies at individual loci were assessed using standard χ² tests. Linkage disequilibrium was quantified through the r² and D’ statistics.

Associations of the SNPs with numerical variables were modeled through linear regression on the SNP adjusted by sex and age. Associations with binary variables were modeled through logistic regression adjusted by sex and age. An additive model for the SNPs (effect in the homozygote minor allele is twice as much as that of the heterozygote) was considered by default, as in general it gave better fits, although in some cases a recessive or dominant model was considered.

A final multifactorial model to predict BMI was validated through bootstrap, using 2000 bootstrap samples, which allowed to correct the performance statistics (R², MSE or Mean Square Error) by optimism in order to assess its predictive power and possible overfit.

This final model was obtained by sequentially adding SNP and gene expression variables to a basal model with only sex and age as predictor variables. A new variable was accepted if it was statistically significant and in addition produced a new model with an increased optimism-corrected R², in order to prevent overfit. In the final model, the relative importance for the different predictor variables was estimated from the χ²- df (degrees of freedom) of the corresponding variables.

All tests were bilateral, with a significance level of 0.05. Ninety-five percent confidence intervals (95% CI) were used in parameter estimates (betas and odds ratios). Bonferroni method was applied for multiple-test correction of the p-values. All statistical analyses were performed using the R statistical software, version 3.6.1 (www.r-project.org).

Statistical analysis in mouse studies

Statistical analyses were carried out with Prism 8 (GraphPad). Experiments including a second variable (e.g., time or nutritional status) were analyzed using 2-way ANOVA with Sidak’s multiple comparison post-test. Where appropriate, the area under the curve (AUC) was calculated. Unpaired Student’s t test was used for single comparisons. Associations between gene expression levels and body weight were modeled using a simple linear regression.