Verification of the diagnosis and inclusion of study participants was carried out at the Regional Clinical Hospital in Kaliningrad. Prior to elective abdominal surgery performed under general anesthesia, participants underwent standard dietary adjustment. All obese patients were required to discontinue medication affecting carbohydrate and lipid metabolism 36 hours before surgery. A fasting blood sample was taken on the morning of surgery. During surgery, liver samples of up to 0.5 cm³ were collected for RNA extraction (immediately placed in 600 µL RNAlater solution (Ambion, Waltham, MA, USA)) and up to 1 cm³ for histologic analysis (immediately fixed in 5 mL neutral buffered formalin (SigmaMed, Moscow, Russia)). The study was conducted in accordance with the World Medical Association Declaration of Helsinki (2000) and the Protocol to the Convention on Human Rights and Biomedicine (1999). The study was approved by the local ethics committee of the Immanuel Kant Baltic Federal University: IKBFU Ethics Committee Conclusion №40 of 26.06.2023. All of the participants provided signed informed consent. All interventions were performed by highly qualified medical professionals and posed no risk to the patients’ health.

Inclusion criteria: Participants were eligible if they were over 21 years of age, had received a diagnosis of their health condition from an official healthcare provider at the Kaliningrad Regional Clinical Hospital and were referred for planned abdominal surgery. Eligible surgeries included hernia repair, gastric resection, and interventions for various gallbladder and bile duct pathologies (e.g., cholelithiasis, cholecystitis, polyposis, cysts, etc.). Aditionally, participants had to declare their willingness to participate in the study and donate a small tissue sample and their blood (approx. 12 mL) by giving informed written consent.

Exclusion criteria: Patients were excluded if they had infectious liver diseases (such as hepatitis), concomitant somatic and infectious diseases in the acute inflammatory stage, known human immunodeficiency virus infection, or any malignant or benign neoplasms. Individuals under 21 years old or those who refused to undergo medical and laboratory examinations during the study or to sign the informed consent form were also excluded.

Biochemical parameters were measured in blood serum (fasting blood glucose, alkaline phosphatase, alanine transaminase, aspartate aminotransferase, total bilirubin, direct bilirubin, indirect bilirubin, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, total cholesterol, triglycerides, total protein, gamma-glutamyltransferase, pancreatic alpha-amylase). The analysis was performed using an automated biochemical analyzer, Furuno CA-180 (Furuno Electric Company, Nishinomiya, Hyogo, Japan) with DiaSys test systems (DiaSys Diagnostic Systems, Holzheim, Hessen, Germany).

Histological analysis of liver biopsies was performed to confirm NAFLD. Liver wedge biopsies (approx. 1 cm³) were obtained intraoperatively by incisional biopsy from the left lobe of the liver and immediately fixed in neutral-buffered formalin. Paraffin sections of the liver, stained with hematoxylin and eosin, were examined by traditional histologic examination using a Leica DM3000 microscope (Leica Microsystems, Wetzlar, Hessen, Germany). The biopsies analyzed contained at least four portal tracts and were deemed informative. The categorisation of liver morphology was bases on the LITMUS NAFLD Atlas [19].

The histologic parameters (steatosis, ballooning, lobular inflammation, portal inflammation, fibrosis) were scored according to an adjusted “Steatosis – Activity – Fibrosis” (SAF) scoring system (the exact analytical parameters used by the histologist in this study are given below) [19]; examples of all conditions mentioned are shown in Fig. 1:

• steatosis—0 points: 5% of the histological image area consists of lipid droplets; 1 point: 5–33%; 2 points: $>$33–66%; 3 points: $>$66%;

• hepatocellular ballooning—0 points: normal cell size; 1 point: pale cytoplasm, normal sized cells; 2 points: pale cytoplasm, enlarged cells;

• lobular inflammation—0 points: none; 1 point: 2 foci (more than 2 cells, clusters); 2 points: 2–4 foci; 3 points: $>$4 foci;

• portal inflammation—0 points: none or minimal; 1 point: mild; 2 points: more than mild;

• fibrosis—0 points: none; 1 point: perisinusoidal; 2 points: in 2 areas; 3 points: without complete partitions; 4 points: with partitions.

Normal liver morphology was defined if the biopsy showed neither steatosis nor fibrosis and exhibited no more than 1 point of lobular, portal inflammation, or hepatocellular ballooning. Steatosis was characterized by a score of $\ge$1 for steatosis, while fibrosis was analogously characterized by a score of $\ge$1 for fibrosis. NASH was diagnosed if the patient had a score of $\ge$1 steatosis, hepatocellular ballooning, and either lobular or portal inflammation [20].

Liver samples were preserved in 600 µL RNAlater solution (Ambion, Austin, TX, USA) and stored at –80 °C until further use. Total RNA was extracted from approximately 100 µL of liver sample using the RNeasy Plus Mini Kit (Qiagen, Hilden, Nordrhein-Westfalen, Germany). The isolated total RNA was eluted in 45 µL of RNase-free water. RNA concentration was measured immediately after isolation using an Implen NanoPhotometer N (Implen, München, Bayern, Germany). Samples were stored at –80 °C until subsequent use in reverse transcription-polymerase chain reaction (RT-PCR).

Quantitative reverse transcription polymerase chain reaction for genes was performed as follows: universal reverse transcription was carried out using the MMLV RT kit (Evrogen, Moscow, Russia) with the addition of the RNase inhibitor RiboCare (Evrogen, Moscow, Russia) according to the manufacturer’s protocol. For quantitative polymerase chain reaction (qPCR), HS SYBR PCR mix (Evrogen, Moscow, Russia) was used. The primer sequences are listed in Table 1. Amplification and reading of qPCR results were performed on a CFX96 thermal cycler (Bio-Rad, Hercules, CA, USA). The annealing of all primers was conducted at 62 °C. After amplification, the melting curves were analyzed to verify the specificity of the reactions.

To normalize gene expression data, the reference gene RPLP0 was used as an internal control. Threshold cycles of the transcripts were converted to relative expression values using the 2^-Δ⁢Ct and 2^{-Δ⁢Δ⁢Ct} methods. The obtained expression profiles were transformed into Log₁₀ to bring the data into a normal distribution. Outliers were identified and excluded using the ROUT method (Q = 1%). Normality of the data distribution was assessed using the Shapiro–Wilk test. If the data conformed to normal distribution, the hypothesis of equality of sample means was tested using Student’s t-tests or one-way ANOVA for comparison across three or more groups. If the data did not follow a normal distribution, further assessment of differences between samples was calculated using the non-parametric Mann–Whitney or Kruskal-Wallis test for comparison of three or more groups. Correlations were determined using Pearson’s method for continuous data and Spearman’s method for discrete data (histologic scores). Differences were considered significant at p 0.05. A p-value less than 0.05 was marked with an asterisk (*), ** p 0.01, *** p 0.001 and **** p 0.0001. Statistical processing was performed using GraphPad Prism 9.3.1 (GraphPad Software, Boston, MA, USA).

The study enrolled 80 subjects (24 men and 56 women), aged 24 to 84 years. As indicated in Table 2, study participants were first classified according to body mass index (BMI) (groups from 1 to 4). Thus, there were 14 individuals in the lean group with a BMI of less than 25 kg/m², 12 subjects with pre-obesity (BMI of 25–29.9 kg/m²), 27 obese (BMI over 30 kg/m²) without type 2 diabetes mellitus (T2DM), and 27 obese subjects with T2DM. Of the 80 participants, 36 underwent histologic analysis of their liver biopsies and were futher divided into four groups (5 through 8) depending on liver morphology: group 5 with normal liver morphology (without steatosis and fibrosis and also without or with minimal inflammation) consisted of 8 subjects, 9 subjects were diagnosed with steatosis (group 6), and 7 study participants had NASH (group 7). Group 8 consisted of 20 subjects with any stage of fibrosis: this group included subjects with or without a diagnosis of steatosis and NASH (but not those with normal liver morphology), and can therefore only be compared with group 5.

The anthropometric and biochemical characteristics for each group are presented in Table 2. Statistical analyses were performed separately for BMI-based groups (groups 1–4) and liver morphology groups (groups 5–7), while the fibrosis group (group 8) was only compared with the normal liver morphology group (group 5). Despite the wide age range, the average age did not differ significantly across groups 1 to 4 (one-way analysis of variance (ANOVA) F(3, 72) = 1.992, p = 0.1228) or between groups 5 to 7 (ANOVA F(2, 34) = 1.939, p = 0.1595). Despite a higher proportion of women in the study, the gender distribution did not differ significantly between the groups (as tested by Chi-square $\chi$² (3, N = 80) = 3.123, p = 0.3730 for groups 1–4 and $\chi$² (3, N = 44) = 0.368, p = 0.9468 for groups 5–8). Statistical tests identified a significant difference between the groups for several biometrical markers, including alanine transaminase, fasting blood glucose, gammaglutamyltransferase, total bilirubin, total protein, total cholesterol, low-density lipoprotein cholesterol and pancreatic alpha-amylase (as detailed in Table 2).

Although ERBB4 and NRG4 are thought to have protective effects in NAFLD, their expression patterns in human liver under normal and pathological conditions remain unclear. Measurement of ERBB4 mRNA levels showed that they were reduced by 2.75-fold in the liver of obese compared to non-obese individuals (t(46.78) = 5.023, p 0.0001, mean difference (MD) –1.460 $\pm$ 0.291 relative expression units (REU)) (Fig. 2A). NRG4 also showed a less drastic but still significant 2-fold reduction in its levels (t(25.43) = 3.531, p = 0.0016, MD –0.9508 $\pm$ 0.269 REU). This trend in ERBB4 expression persisted across more specific groups (pre-obese, obese with and without T2DM) when compared to lean individuals, as indicated by one-way (ANOVA F(3, 74) = 11.78, p 0.0001). The mean differences (MD) of ERBB4 mRNA levels in the liver between the lean group and the pre-obesity group corresponded to 1.299 $\pm$ 0.460 REU (p = 0.0305). The MD between the lean group and the obese subjects without and with T2DM was 2.088 $\pm$ 0.388 REU and 2.079 $\pm$ 0.391 REU, respectively, (p 0.0001), as shown in Fig. 2B.

Liver expression of the NRG4 gene was also significantly down-regulated (one-way ANOVA F(3, 68) = 9.264, p 0.0001) in obese study participants without T2DM (MD = 1.354 $\pm$ 0.296 REU) and with T2DM in (MD = 1.450 $\pm$ 0.300 REU), p = 0.0001 and p 0.0001, respectively (Fig. 2B), so the dynamics are similar to ERBB4 mRNA.

Correlation analysis (Pearson, two-tailed) showed that there was a statistically significant (p 0.0001), strong positive correlation (r = 0.6502) between ERBB4 and NRG4 expression levels in the liver (Fig. 2C). Simple linear regression was used to test if ERBB4 levels significantly predicted NRG4 levels. The fitted regression equation was: Y = 0.4509X – 3.239. Despite relevantly small approximation coefficient, the overall regression was statistically significant (R² = 0.4228, F(1, 70) = 51.28, p 0.0001). Further search for correlations also confirmed a conjecture about a relationship between ERBB4 and NRG4 expression and BMI: there was a moderate negative correlation between ERBB4 (r = –0.5073) and NRG4 (r = –0.4494) at p 0.0001. When divided into groups according to the stage of NAFLD—normal liver morphology (group 5), steatosis (group 6), NASH (group 7) (Fig. 2D), there was no significant change in ERBB4 and NRG4 expression genes across these groups (ANOVA: F(2, 21) = 0.2897, p = 0.7514 and F(2, 18) = 0.5953, p = 0.5619, respectively). The comparison between normal and fibrotic liver also showed no significant difference in ERBB4 and NRG4 expression (t(25) = 0.5148, p = 0.6112 and t(21) = 0.3014, p = 0.7661, respectively).

The expression of major lipogenesis genes in the liver biospecimens was also studied, with the results summarized in Table 3. Only the mRNA of ACACAwas significantly deregulated in the livers of the subjects with increased BMI (Kruskal-Wallis statistic = 20.45, p 0.0001). Other transcripts (FASN (ANOVA: F(3, 63) = 1.342, p = 0.2688), FTO (ANOVA: F(3, 68) = 0.2582, p = 0.8553), SREBF1 (ANOVA: F(3, 65) = 0.3150, p = 0.8145), and TFEB (ANOVA: F(3, 65) = 0.3150, p = 0.8145)) showed no significant change in expression in the studied groups. Moreover, no statistically significant differences in the expression of lipogenesis genes (ACACA, FASN, FTO, SREBF1, and TFEB) were found in groups 5 to 8.

The search for correlations with anthropometric and biochemical parameters revealed several statistically significant correlations (Fig. 3A). Hepatic ERBB4 mRNA levels correlated negatively with BMI (r = –0.5073) and fasting blood glucose (r = –0.3632); hepatic NRG4 mRNA correlated positively with total bilirubin (r = 0.2578) but negatively with BMI (r = –0.4494) and fasting blood glucose (r = –0.3625). The negative correlation of fasting blood glucose with hepatic ERBB4 and NRG4 is particularly noteworthy. There is sufficient evidence that insulin receptor substrates (IRS) are regulated by epidermal growth factor receptors (especially ERBB1 and ERBB2). In several cancer lines, ERBB1-3 has been shown to phosphorylate various IRS, resulting in modulated cellular glucose uptake [21, 22, 23]. However, we found no evidence in the literature to suggest an interaction between ERBB4 and IRS. This could be an effect of heterotypic dimerization with ERBB1-3 or an unexplored property of ERBB4, which warrants further investigation. Correlations with histological parameters are shown in Fig. 3B. Among the studied genes, ERBB4 and FTO demonstrated weak positive correlations with steatosis (r = 0.3331 at p = 0.0472 and r = 0.3692 at p = 0.0410, respectively), and ACACA demonstrated a moderate positive correlation with portal inflammation (r = 0.4206 at p = 0.0324). The correlations between the transcript levels are discussed in detail in the Duscussion section.