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).
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).
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.
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 Fnip2T/T, Fnip2T/C, and Fnip2C/C mice were obtained at the expected Mendelian ratios from heterozygous breeders (Additional file 1: Fig. S4B). Macroscopically, Fnip2C/C and Fnip2T/C knock-in mice were indistinguishable from those expressing the ancestral T allele in homozygosity.
There are no commercially available antibodies against mouse Fnip2 to assess a potential difference in Fnip2 protein levels in Fnip2T/T versus Fnip2C/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 Fnip2T/T and Fnip2C/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 Fnip2T/T and Fnip2C/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 Fnip2T/T and Fnip2C/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 Fnip2T/T and Fnip2C/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 Fnip2T/T and Fnip2C/C MEFs. Consistently with impaired binding of the Fnip2C variant to miR-81b-5p, the decay of Fnip2 mRNA levels was significantly slower in Fnip2C/C cells, in comparison to that of Fnip2 mRNA in Fnip2T/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 Fnip2C variant.
As mentioned, Fnip2C/C mice were viable, macroscopically indistinguishable, and fertile. While no differences in body weight were seen between adult Fnip2T/T and Fnip2C/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 Fnip2C/C male mice, and in female Fnip2C/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 Fnip2C variant in mice, we performed a transcriptomic analysis of two metabolically relevant organs, liver, and visceral WAT, from Fnip2T/T and Fnip2C/C mice. Strikingly, while only 9 genes were differentially expressed in livers, 4795 genes were differentially expressed in WAT from Fnip2C/C versus Fnip2T/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 Fnip2C/C versus Fnip2T/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 Fnip2T/T mice (Fig. 5E, F; Additional file 1: Fig. S5E and S5F), consistently with the increased fat content observed in Fnip2T/T mice.
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.
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.