Defective Olfactomedin-2 connects adipocyte dysfunction to obesity

Abstract

Olfactomedin-2 (OLFM2) is a pleiotropic glycoprotein emerging as a regulator of energy homeostasis. We here show the expression of OLFM2 to be adipocyte-specific and inversely associated with obesity. OLFM2 levels increase during adipogenesis and are suppressed in inflamed adipocytes. Functionally, OLFM2 deficiency impairs adipocyte differentiation, while its over-production enhances the adipogenic transformation of fat cell progenitors. Loss and gain of function experiments revealed that OLFM2 modulates key metabolic and structural pathways, including PPAR signaling, citrate cycle, fatty acid degradation, axon guidance and focal adhesion in 3T3 cell lines and primary human adipocytes. On the molecular level, OLFM2 deficiency in differentiated adipocytes predominantly downregulates genes involved in cell cycle. Extending these findings in vivo, both whole-body Olfm2 knockout and adipose-specific Olfm2 depletion in mice resulted in impaired adipose cell cycle gene expression, with the latter also displaying fat mass accretion and metabolic dysfunction. Collectively, our results underscore a critical role for OLFM2 in adipocyte biology, and support a causative link between reduced adipose OLFM2 and the pathophysiology of obesity.

Fig. 1. Adipose OLFM2 is adipocyte-specific and opposite to obesity.

a Gene expression dynamics (microarray) of Olfactomedin 1–4 (OLFM1-4) in the SC adipose tissue of 16 women with obesity after weight loss. b OLFM2 gene expression levels (real time-PCR) in an extended sample of 24 female patients with obesity following bariatric surgery, and 24 age-matched women without obesity (BMI < 30 kg/m²). See in Table S3 for additional details. Boxplots show expression of SC OLFM1-4 in c 50 female patients with obesity at the baseline, and after 2 (n = 49) and 5 (n = 38) years of surgery-induced weight loss (post-WL), and in 28 age-matched women without obesity (NO), and in d 14 healthy volunteers before and 24 h after an intravenous bolus of 3 ng/kg LPS, as retrieved from publicly available microarray datasets. See expression of others biomarkers in Supplementary Fig. S1b, c. OLFM3 gene expression was not detectable in the dataset of ref. . RMA, Robust multiarray analysis. Below, bean plots show OLFM2 gene expression (real time-PCR) in e SC and f omental (OM) adipose tissues of an independent sample of subjects with and without obesity, while g scatter dot plots show the association of SC (upper) and OM (lower panel) OLFM2 with BMI (further details in Tables S4 and S5). Individual values for men (blue) and women (red) are provided in gender glyphs. h Additional scatter dot plots associate OLFM2 with biomarkers of adipocyte function. Two-sided Spearman’s rank correlation coefficients (r) and p values are shown for each association. i OLFM2 transcript levels were also measured (real time-PCR) in stroma-vascular cells (SVC) and mature adipocytes (MA) isolated from paired SC and OM samples of adipose tissue, biopsied from 12 women with obesity while being bypassed, and in j the single-cell RNA sequencing of 14 individuals with obesity undergoing bariatric surgery, as explained in ref. . RPKM, Reads per kilobase per million reads. The box plots show center line at the median, upper and lower lines bound at 75th and 25th percentiles, respectively, and whiskers at minimum and maximum values. Statistical significance was determined by two-tailed Fisher’s exact t-test (cross-sectional comparisons), or two-tailed paired t-test (longitudinal changes). r.u. stands for relative units. *p < 0.05, **p < 0.01 (two-sided Student’s test). Source data are provided as a Source data file.

Fig. 2. OLFM2 is a marker of adipocyte commitment.

a Pipeline diagram of in vitro differentiated human preadipocytes (PA) growing into lipid-containing mature adipocytes (MA). DM-2 (first) and AM-1 (second week and thereafter) stand for differentiation and adipocyte media, respectively. b Expression dynamics of OLFM2 during adipogenesis (n = 3 biological replicates for each day of hormonal stimuli), and c dynamic changes with regard to unstimulated precursor cells (n = 4/group/time point) and the expression of the others olfactomedin family protein members. d OLFM2 (red) immunofluorescent staining captures in human PA (n = 4), differentiating adipocytes (Day 7; n = 4), and MA (n = 4 biological replicates). Nuclei are stained in blue (DAPI). White arrows point at the apparent perinuclear red signal in differentiating adipocytes. e, f MA showed less OLFM2 immunofluorescent staining signal when challenged with macrophage LPS-conditioned media (MCM; n = 4), as confirmed by g measures of gene expression in MA challenged with macrophage media (MM) or MCM (n = 4 biological replicates/group). The scale bars denote 100 μm length. h The scatter plot displays the expression (microarray) of transcripts coding for OLFM1-4, adiponectin (ADIPOQ), and pro-inflammatory interleukins (IL6 and IL1β) in inflamed Simpson Golabi Behmel syndrome (SGBS)-derived adipocytes, as explained in ref. . Red ink in labels depicts increased levels, and green labels show transcripts decreased in inflamed adipocytes. Blue labels for transcript with no significant variation. Bar plot shows expression of OLFM2 in human PA and MA from i lean (BMI < 25 kg/m²) and j sex, age and fat depot-matched donors with obesity, and changes in MA when challenged with TGFβ for 24 h (n = 4 biological replicates/group). k Variations in MA responding to LPS in the media (n = 6 replicates/group). In bar plots, results are presented as mean ± S.E.M. The box plots in c show center line at the median, upper and lower lines bound at 75th and 25th percentiles, respectively, and whiskers at minimum and maximum values. r.u. stands for relative units. *p < 0.05, **p < 0.01 (One-way ANOVA and Tukey’s multiple comparisons test). Source data are provided with the article as a Source data file. (l) Number of transcription factors (TF) affecting human olfactomedin coding genes, and m OLFM2 gene region ChIP-seq peaks for human datasets and TF in WashU EpiGenome Browser v40.6. Only signal values for PPARG, CEBPB, and CREB1 (listed in red in Fig. 2l) are represented. n Mean TPM (transcript per million), detectability (TPM > 0.1%), and heat map of pairwise Spearman correlation coefficients for OLFM2 (ENSG00000105088) and reference genes for different populations of adipose tissue resident cells, in the case of adipocytes (AC), ADIPOQ, LIPE and PLIN1. The lack of values for mesothelial cells (MesoC) and neutrophils (NP) in SC adipose tissue is due to the low representation of these cell types, as explained in ref. .

Fig. 3. OLFM2 gain and loss of function tunes adipogenesis.

a Pipeline diagram of differentiating 3T3-L1 cells, and b changes affecting OLFM2 protein and transcript levels during adipogenic conversion (n = 4 biological replicates/time point/experiment). c IF images show OLFM2 distribution in 3T3-L1 PA when transforming into lipid-containing MA. The scale bars denote 50 μm length. d Different treatments applied to 3T3-L1-derived MA underscored significant variations affecting Olfm2 gene expression levels (n = 4/group). e Engineered 3T3-L1 PA and MA cell cultures were obtained by means of lentivirus-mediated plasmid transfections, and expressed higher (GoF) or lower (LoF) amounts of OLFM2 coding transcripts, as shown in f (n = 6/group). These modifications increased the adipogenic transformation of 3T3-L1 Olfm2_GoF cells, while OLFM2_LoF preadipocytes differentiated less that control, as demonstrated by g the expression levels of a hub of genes related to the adipocyte phenotype (n = 4/group), measures of h Oil Red O (n = 8/group) and i fluorescent lipid staining (n = 4 wells/group; the scale bar denotes 100 μm length), and j, k protein amounts (n = 4/group) of different adipogenic markers and OLFM2 (western blot). The box plots show center line at the median, upper and lower lines bound at 75th and 25th percentiles, respectively, and whiskers at minimum and maximum values. In bar plots, results are presented as mean ± S.E.M. r.u. stands for relative units. *p < 0.05, **p < 0.01 (One-way ANOVA (plus Tukey’s test p values), and Two-way ANOVA after controlling for FDR in multiple comparisons). Source data are provided as a Source data file.

Fig. 4. Proteomics of OLFM2 gain and loss of function models.

Mass spectrometry-based quantitative proteomics were conducted in engineered 3T3-L1-derived MA and PA obtained as illustrated in Fig. 3e. a Multivariable analysis (PCA) shows the clustering of non-targeting controls (Ctrl) and models of gain (GoF) and loss of function (LoF) in MA and PA. Volcano plots of up (red) and down-regulated (blue) proteins in b MA and c PA with synthetically altered OLFM2 levels. Dots in gray show proteins meeting our exclusion criteria in a Bayesian moderated t-test (nominal p ≥ 0.05), when assuming equal variance in comparisons of n = 5 biological replicates/group. Horizontal p value thresholds show the number of DAPs (adjusted p < 0.05) after correcting for multiple comparisons with the false discovery rate (fdr) and Bonferroni (bon). Labels show some key gene symbols. Metabolic pathway analysis bubble plots created by applying MetaboAnalyst 5.0 to DAPs identified in d MA and e PA (additional details provided in Table S6). Statistical significance was acknowledged for Holm-Bonferroni adjusted p < 0.05 (dashed blue line). f The Heat map shows variations in 26 proteins (genes) significantly (Bayesian moderated t-test adj. FDR p < 0.05) modulated in both MA and PA, strikingly opposed to (4), or directly compelled by (22) OLFM2 levels (see also in Fig. S6c, d). Color represents row z-scores calculated for each cell type and replicate by subtracting the mean and then divide by the standard deviation of each column. g Gene Ontology (GO) enrichment analysis applied to this subset of 26 DAPs.

Fig. 5. Impaired OLFM2 alters adipocyte transcriptomes.

a Scheme illustrating the treatments applied to accomplish our 3 (si-OLFM2) and 6 (sh-OLFM2) days-lasting OLFM2 loss of function assay in human MA. Defective OLFM2 was first checked at b gene expression (n = 4/group; two-sided Student’s test p value when comparing each modification vs. their respective control) and c protein (n = 3/group; One-way ANOVA) levels. Results are presented as mean ± S.E.M, and source data are provided as a Source data file. For transcriptome profiling (multiarrays), 3 biological replicates of treated cells and 4 controls were used in each experiment. d, e Volcano plots, f, g PCA, and h, i hierarchical clustering show the impact of impaired OLFM2 in MA transcriptomes. Inked dots in Volcano plots depict genes with adjusted p < 0.05 (Bayes moderated t-statistics after correcting for FDR multiple comparisons) and fold-change (FC) ≥ 1.2 (red) and FC ≤ −1.2 (blue). Green dots show genes related to the G2M checkpoint hallmark. Interpretation of genes differentially expressed in j si-OLFM2 and k sh-OLFM2 adipocytes when compared to their respective controls (si-NTC and sh-NS) was carried out by means of Gene set enrichment analysis (GSEA), which highlighted significant variations happening in both experiments, such as those affecting l G2M checkpoint (decreased) and m Tnf signaling via NFκB (increased). n Heat maps and Venn diagrams resume significant changes in a list of 68 common genes, 41 up and (including OLFM2) 27 down-regulated in each approach.

Fig. 6. Olfm2-null adipose transcriptomes show impaired fat cell cycle.

a We ran the transcriptomic analysis of epigonadal omental (OM) and inguinal subcutaneous (SC) adipose tissues of 10 diet-induced obese (DIO) Olfm2-null (Olfm2_KO) and wild-type (Wt) mice. Principal component analysis (PCA) shows b OM and c SC sample distribution in two-dimensional charts. Correlation plots picture transcript changes in d OM and e SC adipose tissues. Inked dots depict genes with FDR p < 0.05 and FC ≥ 1.2 (red) or FC ≤ −1.2 (blue), and green dots show genes related to the G2M checkpoint hallmark. GSEA summarizes the hallmarks affected in f OM and g SC fat. h Genes ascribed to G2M checkpoint and/or being E2F targets were broadly compromised. Below, heat maps show changes affecting a hub of 50 transcripts related to cell cycle and compromised in i Olfm2-null adipose tissues, as well as in our j human adipocyte cultures with impaired OLFM2 (si-OLFM2 and sh-OLFM2 cells vs controls). Heat maps represent z-scores based on gene-counts for each sample using the package Bioinfokit (2.0.8) for Python. Gene symbols inked in blue depict transcripts with FDR p < 0.05 in at least one fat depots and/or experimental approach in vitro. Venn diagrams summarize coincidences within genes significantly k down (31) and l up-regulated (12 transcripts) in Olfm2-null mice and human adipocytes with defective OLFM2 signal.

Fig. 7. Impaired adipose Olfm2 results in an obese phenotype.

a The largest depots of white adipose tissue were surgically exposed and injected with our lentiviruses carrying sh-Olfm2 (Olfm2_KD) or a non-silencing (NS) plasmid control. For the molecular assessment of changes affecting these adipose tissues, half of the specimens were terminated 2 weeks after treatment (n = 10; 1:1). Here, b impaired Olfm2 transcript levels were confirmed, paralleling the expression of a hub of gene markers for inflammation and fibrosis (increased), as well as cell cycle (decreased) in omental (OM) fat pads. c Also in OM, but primarily in d subcutaneous (SC) adipose tissue, impaired expression of gene markers for triglyceride metabolism (Lpl), axon guidance (Ephb2, Slit2), and focal adhesion (Pdgfra, Pdgfrb) ran together with the down-regulation of early adipocyte progenitor markers (Zfp423, Klf5, Klf14) in Olfm2_KD mice. To evaluate the physiological consequences of impaired OLFM2 in adipose tissues, the remaining animals (n = 10; 1:1) were maintained under a standard laboratory chow (NC) and monitored for 5 weeks, before being terminated. Notably, e the Olfm2_KD group depicted a higher growth, leading to f increased (p = 0.17) final body weight, although g not significant differences were observed in terms of food intake. Accompanying theses variations, h impaired adiponectin and i increased basal glucose levels were suggestive of obesity-related impaired metabolic homeostasis affecting our adipose-specific Olfm2 knocked-down mice, as further indicated by j glucose tolerance tests (GTT) and k area under the curve (AUC). l Representative images of the hematoxylin–eosin staining of SC and OM adipose tissues administered with either sh-Olfm2 or NS lentiviral particles in animals fed a normal chow. The scale bars denote 50 μm length. The morphometric analysis of adipocytes indicated higher adipocyte area in the Olfm2_KD group. Data in error bars are expressed as mean ± S.E.M. The box and violin plots show center line at the median, upper and lower lines bound at 75th and 25th percentiles, respectively, and whiskers at minimum and maximum values. *p < 0.05, **p < 0.01 (Two-way ANOVA for multiple comparisons by controlling the FDR (Benjamini, Krieger & Yekutieli) q value, or two-sided Student’s test p value). Source data are provided as a Source data file.

Fig. 8. SNPs in the OLFM2 region are associated with anthropometric traits.

a Meta-analyzed associations between genetic variants of OLFM2 and anthropometric traits mapped on PhenoScanner (https://phenoscanner.medschl.cam.ac.uk). The illustrated SNPs (x axis) are lead SNPs for the presented anthropometric traits (y axis). Different shades of blue are used to distinguish between the traits. b Heatmap matrix of pairwise linkage disequilibrium statistics of lead SNPs for anthropometric traits identified in c the OLFM2 locus (chr19: 9,853,718–9,936,515). An apparent LD block, consisting of rs2303100, rs8112411, rs889122, rs1862471, and rs12979274, and led by rs2303100 missense variant, shows the strongest associations with BF% (P = 8.3E-05, 2.0E-04, 1.3E-03, 2.3E-04 and 1.9E-04) and Trunk fat ratio (P = 3.2E-05, 4.7E-04, 6.2E-05, 1.6E-04 and 1.8E-04, respectively). Publicly available reference haplotypes from the 1000 Genomes Project were used by LDlink (https://ldlink.nih.gov/) to calculate population-specific measures of linkage disequilibrium (LD). FORGEdb scores, ranging between 0 and 10, are used for predicting regulatory genetic variants, and are calculated using different regulatory DNA datasets, including data for transcription factor (TF) binding and chromatin accessibility (https://forgedb.cancer.gov). eQTL data were derived from Open Targets Genetics (https://genetics.opentargets.org/). Source data are provided as a Source data file.

Fig. 9. Alterations in OLFM2 signaling are connected to obesity.

Framework proposal laid out during this research, which consequentially connects obesity to changes in OLFM2 levels during adipogenesis and upon the inflammatory activation of adipocytes. In turn, impaired OLFM2 may compromise adipocyte phenotype in obesity through the disruption of paramount mechanisms in adipocyte function, as differentiation, cell cycle and focal adhesion, to the point of exacerbate the pathogenesis of adipocyte hypertrophy and impaired metabolism in mice. Coincidentally, scrutiny of OLFM2 gene variants (GWAS) and the metabolic phenotype of our Olfm2-null mouse model further suggests physiological implications connecting OLFM2 to obesity-related traits. Created in BioRender. Gómez Serrano, M. (2025) https://BioRender.com/0l2biim.