Introduction

Mitochondria are highly plastic organelles capable of dynamically adapting to cellular conditions. Liver tissue is among the richest in mitochondria, with each hepatocyte containing between 1,000 and 3,000 mitochondria, depending on nutritional state1. In the liver, mitochondria perform diverse metabolic and signaling roles beyond energy production2. Therefore, mitochondrial dysfunction is believed to play a critical role in the development of various liver diseases. Impairments in mitochondrial quality control and oxidative stress contribute to the pathogenesis of conditions such as metabolic dysfunction-associated steatotic liver disease (MASLD), alcohol-associated liver disease, drug-induced liver injury (DILI), viral hepatitis, and hepatocellular carcinoma3. The dynamic nature of mitochondrial subpopulations reflects hepatocyte responses to liver stressors, highlighting the value of analyzing mitochondrial morphology and organelle interactomes.

Studies on adipose and liver tissues have shown that mitochondria adjacent to lipid droplets (LD) exhibit distinct functional roles, proteomes, bioenergetics, and dynamics compared to mitochondria freely dispersed in the cytoplasm4,5,6,7. Benador et al. isolated mitochondria associated with LD, termed peridroplet mitochondria (PDM), from mouse interscapular brown adipose tissue (BAT) using differential centrifugation4. In BAT, cytoplasmic mitochondria (CM) were primarily involved in catabolic processes, particularly fatty acid (FA) oxidation. In contrast, PDM performed anabolic functions, supporting LD expansion. PDM promoted triglycerides (TG) synthesis within LD through ATP generation via the tricarboxylic acid cycle using pyruvate as a substrate4. Thus, both mitochondrial subpopulations may help mitigate lipotoxicity through different mechanisms: FA storage by PDM and FA oxidation by CM8. In white adipose tissue, contact between mitochondria, LD, and the endoplasmic reticulum appears to promote de novo lipogenesis from non-lipid precursors9. In the liver, particularly in the fed state, both PDM and mitochondria-associated membranes, which contact PDM, contribute to the esterification and storage of excess FA in LD, thereby promoting LD growth5. During fasting, FA are preferentially directed to CM and CM-mitochondria-associated membranes for oxidation5. Studies using MASLD models have also supported the anabolic role of PDM6,7,10,11.

Morphologically, PDM from mouse BAT exhibited higher cross-sectional area and aspect ratio values, indicating a larger size and more elongated shape compared to CM4. Similar findings have been reported in murine muscle and liver6,12,13. Age-related increase in PDM size in BAT has also been observed in mice14. Whereas aerobic exercise has been shown to significantly elongate PDM in the MASLD liver11.

Nonetheless, the characteristics of PDM and CM subpopulations have not yet been analyzed in in vivo human liver studies. In this study, we employed transmission electron microscopy (TEM) together with a deep-learning model for automated mitochondrial segmentation to investigate the morphology of two hepatic mitochondrial subpopulations, classified based on LD contact. We further examined their association with clinical markers of liver function.

Methods

Liver biopsy specimen collection

A total of 145 patients who underwent liver biopsy for diagnostic evaluation of liver disease at Nagoya University Hospital (Aichi, Japan) between 2020 and 2024 were included in this study. Biopsies were performed under ultrasound guidance using Doppler mode to avoid major blood vessels. The anterior segment of the right liver lobe was targeted, and the procedure was conducted percutaneously using a 16G Bard Monopty biopsy needle. All patients fasted from the night prior to the procedure, received local anesthesia, and remained on bed rest for 4 hours post-biopsy. Patients were discharged the following day after observation. Blood samples were collected for research purposes alongside routine clinical care, and clinical data were obtained from electronic medical records. Biopsy specimens were dissected in situ for electron microscopy, permeabilized in 10% formalin, and processed according to standard pathology protocols. A portion of each specimen (1–2 mm³) was fixed in 0.1 M cacodylate buffer containing 2.5% glutaraldehyde, followed by post-fixation in a mixture of 1% osmium tetroxide and 0.1% potassium ferrocyanide in 0.1 M sodium cacodylate buffer. Samples were embedded in epoxy resin, sectioned into ultrathin slices, and imaged at 100 kV using a JEOL JEM-1400PLUS transmission electron microscope equipped with an EM-14661FLASH digital camera. In tumor-related biopsies, surrounding non-tumorous liver tissue was collected. Nine patients were excluded from further analysis because of the absence of background liver tissue in their electron microscopy samples.

Ethical approval

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Nagoya University Hospital Ethics Review Committee (Approval No. 2020 − 0229). Written informed consent for sample collection was obtained from all participants prior to the biopsy procedure.

TEM image selection

A dataset of 3,079 TEM images of human liver biopsy specimens from 145 patients was used to examine the interactions between mitochondria and LD. All images were analyzed in a blinded manner, without access to clinical data. To ensure clear visualization of mitochondrial contours, only images captured at magnifications ranging from 2,000× to 6,000×, corresponding to a scale bar of 1–2 μm, were included. Mitochondria in direct membranous contact with LD and continuous outer membranes were classified as PDM, whereas mitochondria without LD contact were categorized as CM. Based on these selection criteria, 82 TEM images from 40 patients were included in the final analysis.

Method validation

To determine the optimal method for segmenting mitochondria from TEM images, manual and automated approaches were compared using the coefficient of variation and segmentation time as evaluation metrics (Supplementary Fig. S1). Four TEM images at 4,000× magnification were selected and rotated 90°. From each image, 10 mitochondria were identified and manually traced using Fiji software (version 2.14.0)15. The same mitochondria were also segmented using a deep-learning model implemented in the Empanada-Napari plugin (version 1.1.1), a Python-based tool16. The morphological parameters of the extracted mitochondrial contours were then quantified. Due to its significantly lower coefficient of variation and shorter working time, the automated Empanada-assisted segmentation approach was selected over manual tracing.

Morphology analysis

Selected TEM images were uploaded to Empanada and segmented separately for PDM and CM. Given the morphological diversity of mitochondria, minor segmentation inaccuracies were observed and corrected manually. The segmented mitochondrial contours were exported to Fiji software for measurement (Fig. 1a). Six morphological parameters were quantified for both PDM and CM: area, perimeter, circularity, roundness, aspect ratio, and solidity. The cross-sectional area, indicating mitochondrial size, was measured in square micrometers. Perimeter denoted the length of the mitochondrial boundary in micrometers. Mitochondrial sphericity was assessed using both circularity and roundness, which range from 0 to 1, with values near 1 representing a nearly spherical shape. Aspect ratio, calculated as length-to-width ratio, was used to quantify mitochondrial elongation17,18. Solidity was defined as the ratio of mitochondrial area to its convex hull area, describing surface irregularity. Lower solidity values indicate more branched or complex forms, while values approaching 1 indicate more uniform and convex shapes19. Corresponding formulas for each parameter are provided in Fig. 1b.

Fig. 1
figure 1

Identification and morphological analysis of two hepatic mitochondrial subpopulations, peridroplet mitochondria (PDM) and cytoplasmic mitochondria (CM), based on lipid droplet interaction. (a) Workflow for analyzing transmission electron microscopy images from human liver biopsy specimens. Images were processed using the Empanada plugin within the Napari environment for deep-learning-based mitochondrial segmentation. Segmented mitochondrial contours were exported to Fiji software for quantitative morphometric analysis. The scale bar in the images represents 2 μm. (b) Schematic illustration of the morphological parameters calculated in Fiji, along with the corresponding formulas. CM, cytoplasmic mitochondria; PDM, peridroplet mitochondria.

Full size image

Patients’ characteristics

The mean age of enrolled patients was 56.3 years (range: 22–89 years), with an equal number of men and women (n = 20 each). Liver disease diagnoses were based on clinical data and histopathological evaluation. Our analysis focused on the most common liver diseases, defining four main groups: MASLD (n = 13), DILI (n = 6), autoimmune hepatitis (AIH, n = 6), and normal liver (NL, n = 2), which included patients with no pathological findings. Comparison using the Kruskal–Wallis test showed no significant differences in median age among the evaluated groups (p = 0.75). The remaining cohort comprised 9 patients with hepatobiliary or metastatic tumors and one patient each with alcohol-associated liver disease, hepatitis B virus infection, focal nodular hyperplasia, and graft-versus-host disease. Due to the limited sample sizes, these cases were not included in subgroup analyses. Detailed results of fibrosis staging (based on the METAVIR scoring system)20, steatosis grading, and clinical laboratory values are presented in Table 1.

Table 1 Patients’ characteristics and clinical data.
Full size table

Statistical analyses

All statistical analyses were performed using GraphPad Prism version 10.4.1 for Windows (GraphPad Software, Boston, Massachusetts, USA) and R version 4.4.1 (R Core Team, 2024). To compare two independent groups, the Mann–Whitney U test was applied. For comparisons across multiple groups, one-way analysis of variance with Bonferroni correction was used. Cohen’s d was calculated to assess the effect size of differences between PDM and CM. Pearson correlation was employed to examine relationships between PDM count, LD count, and age. Associations between clinical parameters and mitochondrial morphological features were evaluated using the maximal information coefficient (MIC), a component of the maximal information-based nonparametric exploration statistics framework, which enables detection of both linear and non-linear relationships21. Black lines in the plots represent median values, which are also summarized beneath each graph. P-values are indicated above the plots. Statistical significance was defined as p < 0.05.

Results

Morphological disparity between PDM and CM

A total of 263 PDM and 3,016 CM were identified in this study. Comparative analysis of data from all enrolled patients revealed that PDM exhibited significantly higher median values for area, perimeter, and aspect ratio compared to CM (Fig. 2a-b). In contrast, circularity, roundness, and solidity were significantly lower in PDM than in CM. We next examined mitochondrial morphology across distinct liver conditions, including NL, MASLD, DILI, and AIH (Supplementary Fig. S2). The total counts of identified PDM and CM for each patient group are presented in Supplementary Table S1. In the NL, MASLD, and DILI groups, the morphological distinction between PDM and CM was consistent with trends observed in the overall cohort. PDM consistently demonstrated a larger area, greater perimeter, and higher aspect ratio, while circularity, roundness, and solidity remained lower than in CM. However, not all differences reached statistical significance within each subgroup.

Fig. 2
figure 2

Morphological distinction between peridroplet mitochondria (PDM) and cytoplasmic mitochondria (CM). (a) Representative transmission electron microscopy (TEM) images (scale bar = 2 μm) and scatter plots comparing six morphological parameters - area, perimeter, circularity, roundness, aspect ratio, and solidity - between 263 PDM and 3,016 CM extracted from human liver biopsy samples. Statistical significance was determined by the Mann–Whitney U test (** p < 0.01; **** p < 0.0001). (b) Magnified electron micrograph illustrating Empanada-assisted segmentation of a PDM in direct contact with a lipid droplet (LD). (c) Scatter plot showing a strong positive correlation (r = 0.8518, p < 0.0001) between PDM and LD counts across 40 patient cases. Each dot represents data from TEM images of one liver biopsy specimen. CM, cytoplasmic mitochondria; LD, lipid droplet; PDM, peridroplet mitochondria; TEM, transmission electron microscopy.

Full size image

To assess the magnitude of morphological differences between PDM and CM, Cohen’s d was calculated as a measure of effect size (Supplementary Table S2). As a reference, d values of 0.2, 0.5, and 0.8 represent small, moderate, and large effects, respectively. Across the entire cohort and individual liver disease groups, area differences showed a small effect. Perimeter differences ranged from small to nearly moderate effect sizes. Circularity, aspect ratio, roundness, and solidity yielded moderate effect sizes in the full cohort and MASLD subgroup. In the NL group, aspect ratio and roundness approached moderate effects, while circularity and solidity showed large effects. These parameters demonstrated moderate to large effects in the DILI group. In the AIH group, all parameters showed small effect sizes, with the exception of solidity, which approached a moderate effect.

Taken together, these results suggest that PDM are significantly larger and exhibit a more elongated morphology, while CM are generally smaller, more spherical, and solid in shape. Notably, no statistically significant morphological differences were observed between PDM and CM in the AIH group.

Correlation between PDM and LD

To investigate the relationship between the number of PDM and LD, the total number of LD visible in TEM images was quantified for each patient. A total of 393 LD were identified, of which 47% were in membranous contact with mitochondria. Pearson correlation analysis revealed a positive association between PDM and LD counts, highlighting their close spatial and functional relationship (Fig. 2c).

Alterations in mitochondrial morphology across liver diseases

We next examined the influence of liver pathology on mitochondrial morphology by comparing PDM and CM parameters across four groups: NL, MASLD, DILI, and AIH. No statistically significant differences were observed in PDM area, perimeter, aspect ratio, or solidity across the groups (Fig. 3a). However, PDM in AIH showed significantly higher circularity and roundness than in DILI, and roundness was also significantly higher than in MASLD. In contrast, CM displayed greater morphological variability across liver conditions (Fig. 3b). In NL, CM exhibited significantly larger area values than those in MASLD and AIH. CM in NL also showed significantly higher circularity than in DILI. CM in DILI and MASLD exhibited significantly higher aspect ratios and lower roundness than CM in NL and AIH. Additionally, CM solidity was significantly reduced in DILI compared to NL, MASLD, and AIH. No significant differences in CM perimeter were found across the four liver conditions.

Fig. 3
figure 3

Comparative analysis of mitochondrial morphology across liver disease groups. (a) Representative transmission electron microscopy (TEM) image (scale bar = 2 μm) and scatter plots showing the distribution of six morphological descriptors (area, perimeter, circularity, roundness, aspect ratio, and solidity) of peridroplet mitochondria among patients with normal liver (NL), metabolic dysfunction-associated steatotic liver disease (MASLD), drug-induced liver injury (DILI), and autoimmune hepatitis (AIH). (b) Corresponding scatter plots and TEM image (scale bar = 2 μm) illustrating the same parameters for cytoplasmic mitochondria across the same diagnostic categories. Data highlight disease-specific trends in mitochondrial shape, including alterations in elongation and sphericity. Statistical significance was determined by one-way analysis of variance with Bonferroni post hoc test (* p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001). AIH, autoimmune hepatitis; DILI, drug-induced liver injury; MASLD, metabolic dysfunction-associated steatotic liver disease; NL, normal liver; TEM, transmission electron microscopy.

Full size image

These findings indicate that PDM maintain relatively stable morphological features regardless of liver pathology, with the exception of a more spherical phenotype observed in AIH. In contrast, CM morphology varied more markedly with disease state. CM in NL were larger and more spherical than those in MASLD and DILI. Similarly, CM in AIH retained greater roundness, though they were smaller in size. In MASLD and DILI, CM exhibited reduced sphericity, and in DILI, mitochondria were also less solid. These results suggest that PDM and CM exhibit distinct structural adaptations to different liver pathologies, with CM showing greater morphological sensitivity to hepatic injury.

Morphological changes in MASLD progression

We next investigated the morphological alterations that mitochondria undergo during the progression of MASLD. The 13 MASLD patients were divided into two subgroups: 3 with simple steatosis and 10 with metabolic dysfunction-associated steatohepatitis (MASH). Mitochondrial morphology was compared between these groups for both PDM and CM. Consistent with previous results, no significant differences were observed in PDM morphology between patients with simple steatosis and those with MASH (Fig. 4a). In contrast, CM in MASH exhibited significantly lower area and aspect ratio values, as well as higher circularity, roundness, and solidity, compared to CM in simple steatosis (Fig. 4b). These findings suggest that CM become smaller, more spherical, and structurally compact during the transition from simple steatosis to MASH, while PDM maintain morphological stability. This highlights the differential sensitivity of PDM and CM subpopulations to progressive metabolic liver stress.

Fig. 4
figure 4

Comparative analysis of peridroplet and cytoplasmic mitochondrial morphology between patients with simple steatosis and metabolic dysfunction-associated steatohepatitis (MASH). (a) Representative transmission electron microscopy (TEM) images (scale bar = 2 μm) depicting six morphological parameters (area, perimeter, circularity, roundness, aspect ratio, and solidity) of peridroplet mitochondria in patients with simple steatosis and MASH. (b) Corresponding scatter plots and TEM images (scale bar = 2 μm) for cytoplasmic mitochondria in the same patient groups. Statistical significance was determined by the Mann–Whitney U test (ns, not significant; * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001). MASH, metabolic dysfunction-associated steatohepatitis; TEM, transmission electron microscopy.

Full size image

Association with clinical data

To examine potential associations between mitochondrial morphology and clinical characteristics, MIC values were calculated. MIC, a component of the maximal information-based nonparametric exploration statistics, identifies both linear and non-linear relationships between variables, with values ranging from 0 (no association) to 1 (perfect association)21. Clinical parameters included steatosis grade, lipid profiles (TG, low-density lipoprotein cholesterol, high-density lipoprotein cholesterol, total cholesterol), and liver enzyme levels (aspartate aminotransferase, alanine aminotransferase, alkaline phosphatase, gamma-glutamyl transpeptidase). MIC results are presented in Fig. 5a.

Fig. 5
figure 5

Relationship between clinical parameters and mitochondrial morphological descriptors. (a) Heatmap displaying maximal information coefficient (MIC) values that quantify the strength of association between clinical parameters (e.g., triglyceride levels, liver enzymes, cholesterol levels, steatosis grade) and morphological metrics of peridroplet (PDM) and cytoplasmic mitochondria (CM). MIC values range from 0 (no association) to 1 (perfect relationship), with lower values in blue and higher values in red. (b) Scatter plot illustrating a significant positive correlation between patient age and CM roundness (r = 0.3442, p = 0.0296), suggesting increased mitochondrial sphericity with aging. (c) Scatter plot showing a significant inverse correlation between patient age and CM aspect ratio (r = -0.3459, p = 0.0288), indicating age-associated mitochondrial fragmentation. CM, cytoplasmic mitochondria; MIC, maximal information coefficient; PDM, peridroplet mitochondria.

Full size image

In this cohort, PDM morphology demonstrated the strongest association with serum TG levels. The highest MIC value (0.48) was observed between PDM perimeter and TG concentration, suggesting a potential role of hepatic PDM in lipid handling. Conversely, associations between PDM features and liver enzyme levels were weak, consistent with their relatively stable morphology across liver conditions. CM morphology showed stronger associations with liver injury markers, particularly alanine aminotransferase, alkaline phosphatase, and gamma-glutamyl transpeptidase. The perimeter of CM showed the strongest relationship with alanine aminotransferase (MIC = 0.46), supporting their morphological responsiveness to hepatocellular stress. Additionally, CM exhibited an association with high-density lipoprotein cholesterol, indicating a possible involvement in liver cholesterol transport.

Correlation with age

To evaluate age-related changes in mitochondrial morphology, we correlated the median values of PDM and CM parameters with patient age. Participants ranged from 22 to 89 years. No significant associations were observed between PDM morphology and age. In contrast, CM morphology demonstrated a positive correlation between roundness and age, and a negative correlation between aspect ratio and age (Fig. 5b-c). These findings suggest that CM undergo age-related morphological adaptations, becoming rounder and less elongated, while PDM remain structurally stable across the adult lifespan.

Discussion

This study presents the first in vivo evidence of morphological heterogeneity and distinct clinical associations of PDM and CM subpopulations in human liver tissue. To our knowledge, no previous research has characterized mitochondrial cytoarchitecture across different human liver pathologies or investigated its relationship with biochemical indicators of liver function.

Mitochondria, which do not engage in vesicular transport, participate in intracellular communication by establishing heterogeneous contact sites, influencing both their function and morphology22,23. Regions of close apposition between organelles, known as membrane contact sites, are tethered by specific proteins and contribute to subcellular compartmentalization24. Mitochondria interact with the endoplasmic reticulum, LD, Golgi apparatus, plasma membrane, endosomes, lysosomes, peroxisomes, and melanosomes25,26. Notably, recent studies have highlighted a close crosstalk between mitochondria and LD, leading to the formation of heterogeneous mitochondrial subpopulations.

Quantitative analysis of liver biopsy samples from 40 patients with varying liver conditions demonstrated a higher abundance of CM compared to PDM. A similar disproportion in mitochondrial subpopulations has been reported in MASLD liver models6. As illustrated in our representative TEM images, hepatic mitochondria displayed a wide variety of forms, consistent with the heterogeneity reflected in morphometric quantification. The results presented in Supplementary Table S3 revealed substantial variability and skewness in morphological parameters, suggesting the influence of multiple regulatory factors affecting mitochondrial shape. Despite this variability, hepatic PDM were consistently larger and more elongated than CM, which tended to be smaller, rounder, and more solid in appearance. When categorized by disease state, both PDM and CM maintained consistent morphological trends across NL, MASLD, and DILI. The observed morphology of PDM is consistent with prior reports and supports the hypothesis that increased elongation may enhance contact surface with LD13,27.

LD are dynamic organelles that, like mitochondria, undergo both fusion and fission processes28. They serve as intracellular reservoirs for TG and facilitate the trafficking of free FA, directing them toward various metabolic fates, including lipid biosynthesis and energy production29. Previous research has demonstrated that hepatic PDM primarily promote LD expansion by supporting TG synthesis5,6,7,10,11. Elevated mitofusin-2 levels, associated with mitochondrial fusion, have been reported in PDM from both BAT and liver4,6,10. Given that mitochondrial fusion enhances ATP production efficiency, the elongated morphology of hepatic PDM observed in this study supports the proposed anabolic role of these mitochondria30. Furthermore, correlation analysis between patient laboratory data and quantified mitochondrial morphology revealed that PDM had the strongest association with serum TG levels, possibly reflecting the abundance or size of hepatic LD. These results, together with the positive correlation observed between the number of LD and PDM, support the hypothesis that increased PDM-LD interaction may indicate a metabolic switch from FA oxidation toward lipogenesis. In comparison, CM showed the strongest associations with liver enzyme levels, particularly alanine aminotransferase and gamma-glutamyl transpeptidase, indicating their greater sensitivity to hepatocellular stress. CM also correlated with high-density lipoprotein cholesterol, suggesting a potential role in liver cholesterol transport. Taken together, PDM and CM showed distinct patterns of association with clinical data, reflecting their involvement in separate pathophysiological pathways.

Excessive LD accumulation in hepatocytes can result from dietary imbalances, toxin exposure, viral infections, and metabolic dysfunction defects31. Hepatic steatosis is the pathological hallmark of MASLD, which is the most prevalent chronic liver condition, affecting approximately 30% of the global population32. This underscores the importance of examining dietary composition and its implications for mitochondrial cytoarchitecture in MASLD. Recent in vitro studies demonstrate divergent mitochondrial-LD dynamics in hepatocytes treated with oleic acid, representative of Mediterranean-style diets, and palmitic acid, a major lipid in Western diets33. In our study, PDM morphology did not differ significantly between MASLD and NL samples. In contrast, CM in MASLD exhibited a smaller median sectional area and were more elongated than those in NL. Further comparison of mitochondrial parameters between patients with simple steatosis and MASH showed that CM became progressively smaller and more spherical. Given the central role of oxidative stress-induced mitochondrial dysfunction in MASH, these morphological changes may reflect enhanced mitochondrial fission34. Supporting this interpretation, increased levels of dynamin-related protein 1 - a key mediator of mitochondrial division - have been observed in mitochondria from MASH livers6,34. Additionally, an inverse shift in the abundance of PDM and CM has been reported during MASLD progression6. PDM numbers increase while CM numbers decline. As fibrosis advances, this trend reverses, with PDM levels decreasing and CM abundance rising6,7. Collectively, our findings, together with previous reports, suggest that mitochondrial subpopulations undergo distinct structural and numerical adaptations that vary with the severity of MASLD.

DILI represents a heterogeneous group of dysfunctions caused by diverse factors that injure liver tissue or the biliary system. In this study, six patients were diagnosed with DILI, including two with hepatic immune-related adverse events. Despite clinical heterogeneity, CM in DILI tended to be more elongated, evidenced by reduced sphericity and increased aspect ratio compared to NL. CM in DILI also showed the lowest median solidity among all groups, suggesting an irregular and convoluted mitochondrial architecture. In contrast, PDM in DILI exhibited no significant morphological alterations.

While PDM and CM showed morphological divergence in MASLD and DILI, such differences were not observed in AIH. Both PDM and CM in AIH trended toward a more granular morphology, and CM were also significantly smaller than in NL. Given the unique immunometabolic context of AIH, further studies are needed to elucidate the mechanisms underlying these mitochondrial patterns.

Although patient heterogeneity extended to both liver condition and age, age-related changes were confined to CM. In contrast to observations in mice BAT14, where PDM hypertrophy occurs with age, hepatic PDM in this study showed no morphological aging trends. Conversely, CM morphology correlated positively with age, suggesting progressive rounding. These findings suggest the disease-specific remodelling of hepatic mitochondria, which may also evolve with age.

TEM enabled high-resolution visualization of mitochondrial ultrastructure and LD contact. To minimize variability, all liver biopsies were performed following a 16-hour fasting period to ensure a consistent metabolic baseline, as fasting is known to induce mitochondrial elongation1. All TEM images were analyzed blindly and segmented using MitoNet, a deep-learning model trained on 1.5 million electron microscopy images and implemented via the Empanada-Napari plugin16. While statistically significant differences in mitochondrial morphology were observed, a limitation of this study is the reliance on 2D electron microscopy sections. Future studies using 3D electron tomography, combined with complementary biochemical or molecular analyses, will be essential to map the spatial organization of hepatic mitochondrial cytoarchitecture and extrapolate these findings to broader physiological outcomes. Nonetheless, our analysis of human liver samples verifies results obtained in preclinical models, which may not accurately reflect the conditions in human tissues in vivo.

In summary, this investigation demonstrated that CM morphology varies significantly across liver diseases and correlates with serum liver enzyme levels. PDM morphology, by contrast, was consistent across NL, MASLD, and DILI, except for AIH. This uniformity, along with weaker associations with liver enzyme levels, supports the hypothesis that LD may exert a stabilizing influence on PDM architecture, the mechanism of which remains to be elucidated. Together, these insights encourage further development of advanced image-based diagnostic tools in liver pathology.