Morpho-clinic characteristics of the subjects

The cohort under study (n = 59) was stratified according to absence (S0) or presence (S1-S2-S3) of liver steatosis. Table 1 shows the clinical characteristics of the enrolled subjects. Significant differences (p < 0.05) between groups were observed for gender, BMI, waist circumference, measured resting energy expenditure (mREE), insulin, HOMA-IR, total and LDL cholesterol, liver transaminases, APRI, and the fatty liver index (FLI). All these parameters were higher in the group with steatosis. Table 2 shows the statistical differences when the cohort was stratified by steatosis and gender. In this case, significant differences inside steatosis groups were obtained for FFM, FM, REE in both genders, and liver enzymes, particularly in males (AST, ALT, GGT, and the APRI). Comparing the S0 and the S1-S2-S3 subgroups, the female cohort presented more significant differences in BMI, waist circumference, total and LDL cholesterol, liver enzymes (ALT and GGT), and FLI.

Table 1 Clinical characteristics of the studied populationTable 2 Clinical-biochemical characteristics of the cohort stratified by steatosis and genderSerum proteins associated with the presence of liver steatosis

The PEA assay for 184 target proteins (metabolic and cardiometabolic panels) was conducted on the 59 serum samples. One sample showed alterations in the internal quality control tests and was removed from subsequent analysis. The study identified five circulating proteins associated with the steatosis grade in the liver: Cathepsin O (CTSO), Cadherin 2 (CDH2), Leukocyte immunoglobulin-like receptor subfamily A member 5 (LILRA5), and Serpin B6 (SERPINB6); and Prolyl endopeptidase (FAP). These proteins showed significant differential abundance when comparing groups with presence vs. absence of steatosis (Fig. 1, Supplementary Figure S1 and Table S2), according to p-values adjusted for multiple testing.

Fig. 1

Serum protein expression in obese adolescents with and without steatosis. The volcano plot shows the NPX difference between non-steatotic (n = 21) and steatotic (n = 37) patients on the x-axis and the –log10 of the nominal p-value on the y-axis. All p-values were adjusted for multiple testing using the Benjamini–Hochberg method and − log10 of the p-values was plotted

In individuals with liver steatosis compared to those without, an increase in mean NPX values from 2.99 ± 0.20 to 3.32 ± 0.35 was detected for CTSO (padj < 0.003), from 3.55 ± 0.32 to 3.94 ± 0.40 for CDH2 (padj < 0.007), from 5.41 ± 0.35 to 5.72 ± 0.28 for LILRA5 (padj < 0.021), from 4.42 ± 0.20 to 4.65 ± 0.31for SERPINB6 (padj < 0.034) and from 5.36 ± 0.16 to 5.68 ± 0.40 (padj < 0.007) for FAP, indicating an increased serum expression of certain proteins (Fig. 2).

Fig. 2

Serum protein expression for proteins under study in obese adolescents with and without steatosis. Box plots depict the differences in NPX values for individuals without steatosis (S0, n = 21; Absence) has been compared with those with steatosis (S1, S2, S3, n = 37; Presence). A CTSO protein, B CDH2 protein, C LILRA5 protein, D SERPINB6 protein and E FAP protein, Values were presented as median with their respective 10–90 percentiles. Differences were considered statistical significants at p values less than 0.05. *p < 0.05, **p < 0.01, ***p < 0.001

Afterward, we query the Human Protein Atlas portal (HPA) (https://www.proteinatlas.org/), finding that CTSO cluster as a plasma protein with the highest mRNA expression in the liver, CDH2 shows high protein expression in the liver, kidney, adrenal gland, and heart muscle, LILRA5 in lymphoid tissue and bone marrow, SERPINB6 is enriched in adipose tissue and FAP in connective tissue with higher expression in gallbladder (see Fig. S2). Moreover, a protein–protein interaction (PPI) network was created for the five identified markers in NetworkAnalyst 3.0. PPI was constituted by two subnetworks. The main subnetwork, with a total of 79 nodes and 78 edges is CDH2 protein the component with the highest number of interactions. Then, a second subnetwork was constituted by five total interactors. Then, functional enrichment was performed for all the protein components of the PPI in the Reactome platform. Most of the enriched proteins were clustered in functional activities related to signaling transduction, integration of energy metabolism, cell–cell communication, as well as the immune system and coagulation pathways. (see Figure S3 and Table S3).

Gender differences in the serum proteome of adolescents with obesity

After the initial proteome characterization of the serum samples according to the liver steatosis grade, we investigated the changes in the circulating proteome according to gender. The serum proteome of 28 female and 30 male subjects were compared. Our results show that 6 proteins of the 184 analyzed presented differences according to gender. Figure 3 shows the differences in normalized expression values for the proteins showing changes. CTSO, Fc receptor-Like 1 (FCRL1), Aldehyde Dehydrogenase 1 Family Member A1 (ALDH1A1), and FAP showed increased expression values in males (p = 0.001, p = 0.002, p = 0.002, and p = 0.002, respectively). On the other hand, CXADR Like Membrane Protein (CLMP), and Amyloid Beta Precursor Like Protein 1 (APLP1) serum expression were higher in females than males (p = 0.0007 and 0.002, respectively). Our results indicated that CTSO and FAP were affected not only by steatosis grade but also by the gender of the subject (Fig. 3 and Table S4).

Fig. 3

Serum protein expression differences according to gender for the proteins under study. Box plots show the variations according to gender, females (n = 28,) and males (n = 30). A CTSO protein, B FCRL1 protein, C ALDH1A1 protein, D FAP protein, E CLMP protein, and F APLP1 protein. Values were presented as median with their respective 10–90 percentiles. Differences were considered statistically significant at p-values less than 0.05. *p < 0.05, **p < 0.01, ***p < 0.001

Correlation of steatosis grade with the clinic-biochemical parameters and serum protein expression

Multiple linear correlation analyses were performed to investigate the associations among the clinic-biochemical parameters, the serum proteome, and the steatosis grade. As shown in Table 3 and in the correlation plots (Figure S4) the most significant variables associated with steatosis were weight (ρ = 0.62, p < 0.0001), BMI (ρ = 0.50, p < 0.0001), waist circumference (ρ = 0.60, p < 0.0001), plasma insulin (ρ = 0.55, p < 0.0001), HOMA-IR (ρ = 0.55, p < 0.0001), GGT (ρ = 0.51, p = 0.0001) and the FLI (0.54, p < 0.0001). Moreover, among the proteins showing changes in their expression with steatosis, we observed CDH2 (ρ = 0.52, p < 0.001), CTSO (ρ = 0.44, p < 0.001), and LLRA5 (ρ = 0.45, p < 0.001) presenting the most significant correlations. Additional correlation analyses for steatosis degree and the rest of the parameters were performed when the cohort was also stratified by gender. Correlation matrix plots demonstrated, as an example, that CDH2 is associated with steatosis degree (rho = 0.46, p = 0.013), weight (rho = 0.43, p = 0.021), BMI (rho = 0.41, p = 0.030), insulin (rho = 0.38, p = 0.044), and HOMA-IR (rho = 0.41, p = 0.03), among other parameters in females (XY n = 28) and with steatosis degree (rho = 0.44, p = 0.017), insulin (rho = 0.41, p = 0.023), HOMA-IR (rho = 0.41, p = 0.025), and AST (rho = 0.65, p < 0.001), ALT (rho = 0.61, p < 0.001), GGT (rho = 0.38, p = 0.037); among others in males (XY n = 31) (Figure S5 A-C, and supplementary information Tables S5–S7).

Table 3 Significant correlations among biomolecular data and steatosis degreeCombination of serum proteins and clinical-biochemical parameters to distinguish liver steatosis

The potential of the individuated proteins as biomarkers of liver steatosis was also assessed. We performed ROC analysis for each candidate marker, which showed a statistical correlation with the steatosis grade. AUC and cut-off values based on the Youden index, which maximizes sensitivity and specificity, are displayed in Table 4. Afterward, to obtain the best diagnostic performance for steatosis, a logistic regression model analysis (with forward selection with switching mode) considering all the variables under study was performed. The model includes 33 variables (32 numeric and gender as categorical). Continuous numerical variables were age, weight, height, BMI, waist-circumference, FFM, systolic PA, diastolic PA, REE, fasting glucose, insulin, HOMA-IR, triglycerides, total cholesterol, HDL, LDL, PCR, AST, ALT, GGT, platelets, NLR, APRI, FLI, CLMP, CTSO, FCRL1, SERPINB6, APLP1, LILRA5, ALDH1A1, CDH2 and FAP). The statistical analysis demonstrated that the best predictive model for steatosis diagnosis should include the four variables with the following equation: -38.81 + 0.05 * LDL + 3.21 * CDH2 + 3.89 * FAP achieving a diagnostic performance with AUC of 0.91 (95% CI 0.75–0.97), with a sensitivity of 100% and specificity of 84% at a cut-off value determined at > 0.37 (Fig. 4A). Moreover, the performance in diagnostic accuracy for steatosis using the FLI and each one of the components of the logit model was compared (Fig. 4B). Statistics associated with the figure reveal that the logit model achieved better diagnostic performance for liver steatosis than the components separately, particularly when comparing with the AUC of LDL (p < 0.0008), CDH2 (p < 0.0137) and FAP (0.0119), but not against FLI (p = 0.0845). In addition, Fig. S5 depicts the correlations between the variable outcome steatosis degree obtained with the logit model and several clinic-biochemical parameters for the overall cohort.

Table 4 Ranked list of the AUC values calculated for each of the biomarker candidates and the FLIFig. 4

Receiver operating characteristic (ROC) curves for steatosis diagnosis in obese adolescents. A Logistic regression (logit) model combining the best 4 variables for steatosis outcome prediction. The logit model has the following equation: steatosis = -38.81 + 0.05*LDL + 3.21*CDH2 + 3.89*FAP, achieving a diagnostic performance with an area under the curve of 0.91 with a sensitivity of 100% and specificity of 84%. B ROCs for logit model and for each component of the model equation separately