Formation of liver fibrosis patterns is a complex process that can only be marginally investigated experimentally. A digital twin (DT), a computational model of the liver, is warranted, as it permits integration and modulation of multiple mechanisms. We develop a DT on chronic injury-mediated formation of CCl4-induced septal fibrosis based on experimental iterations. This DT closely reproduces the spatial-temporal pattern of hepatocytes, hepatic stellate cells (HSCs), macrophages (Mphs), collagen fibers secreted by activated HSCs, blood vessels and cell-cell communication through literature and quantitative imaging. It simulates formation of septal fibrosis and predicts that attraction of activated HSCs and Mphs is controlled by hepatocytes, damaged by CCl4 intoxication. Meanwhile, undamaged hepatocytes proliferate to replace the dead ones, thereby mechanically compressing the fibrotic network formed by collagen into “wall”-like shapes. Importantly, simulations assuming loss of spatial pattern of CYP2E1 expressing hepatocytes match with experimental data from mice with deleted transcription factor GATA4 in endothelial cells displaying a decreased CYP2E1 expression and disturbed distribution pattern in hepatocytes, thereby validating the DT.
Article
A digital liver twin demonstrating the interplay between biomechanics and cell kinetics can explain fibrotic scar formation
https://doi.org/10.21203/rs.3.rs-3911195/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Chronic liver diseases (CLD) have variant etiologies, including viral and other infections, autoimmune activities, alcohol consumption, mal/overnutrition, cholestasis, drug (ab)use or genetic abnormalities. Irrespective of the underlying etiology, chronicity reflects a long history of parenchymal cell injury and a tissue response switching between regeneration, inflammation, and fibrosis1. Fate decisions of hepatocytes upon perturbation depend on their localisation inside "zones" of the liver lobule, the smallest repetitive anatomical unit. For example, single treatment of mice with CCl4 leads to pericentral hepatocyte damage, since only these cells express the Cytochrome P450-enzymes CYP2E1 and CYP1A2, which metabolize CCl4 to cell damaging products. The consequence of repetitive CCl4 damage is a change of the Cytochrome P450-pattern and a formation of pericentral/septal fibrosis. If, however, the liver suffers from cholestasis, damage is initially caused by toxic levels of bile acids, beginning with the cholangiocytes, progressing via the periportal hepatocytes, and leading to periportal/biliary fibrosis. All CLD have in common that the wound that arises from damage is “repaired” by excessive extracellular matrix (ECM) deposition, which in case of multiple rounds of damage finally leads to tissue scarring towards cirrhosis2, the major risk factor for hepatocellular3. During the past decades, research on liver fibrogenesis mainly focused on molecular and cellular mechanisms3,4, suggesting a widely accepted concept, where injured hepatocytes send out signals, so called damage associated molecular patterns (DAMPs) towards the surrounding liver cells. Liver Mphs produce fibrogenic cytokines such as TGFβ and PDGFβ to activate and recruit hepatic stellate cells (HSC), which produce and deposit ECM proteins5,6. Mphs comprise the liver resident Kupffer cells and recruit monocyte derived Mphs, and have diverse functions in onset, progression and regression of liver fibrosis, by switching between inflammatory and restorative phenotypes7-10. HSCs also display significant plasticity and are able to switch between quiescent (qHSC), differentially activated (aHSC) and reverted subtypes (rHSC)9,11,12. During fibrosis regression in phases without damage, aHSCs are depleted, approximately 50% by cell death and 50% by reversion to quiescence phenotype11. Changes in the cellular phenotype of HSCs are induced by molecular signals and by alterations of the ECM with mechanical consequences on tissue stiffness and viscoelasticity13-15. Despite these and other investigations2,16-18, pattern formation in fibrosis is not well understood. A better understanding of the formation of specific fibrosis patterns, however, would be helpful to design the treatment strategy to prevent liver fibrosis progression to cirrhotic nodule. Eventually, the presentation of fibrosis progression or reversion is tightly orchestrated by spatial-temporal distribution of ECM producing (HSC) and degrading (Mph) cells, and the respective cell numbers and their location will finally also be decisive for the variant fibrosis pattern formation discussed above.
In patients, knowledge on the dynamics of liver disease development, progression and regression can only be collected indirectly via longitudinal blood sampling or by non-invasive imaging methods, like MRI or ultrasound. Blood parameters do not exclusively relate to disease activities in the liver, and imaging can up to date only provide morphological and biomechanical tissue properties at super-lobular scales. Information about the liver micro-architecture is collected from resected or biopsied livers that are available from only very few selected time points. Time course data on the dynamics of cellular fate changes are therefore principally generated from animal disease models, but these still do not permit simultaneous measurement of all possibly involved process variables, therefore often fail in establishing unique cause-effect relationships, due to the complexity of the system. To address this, we develop a computational digital twin (DT) to simulate hypothesized mechanisms and to assist in data interpretation and data acquisition design, where the complexity of the system interplay does not permit unambiguous conclusions. With the DT, we are able to overcome the bottleneck that the system of interest is not sufficiently amenable to manipulations and measurements, as it is the case in human chronic liver disease.
The presented DT is able to resolve the spatial-temporal organization of tissue microarchitecture as determined from histological tissue samples or by intravital imaging, both requiring invasive interventions. It does this by representing each cell individually within a cell (agent)-based simulation approach, either at high resolution resolving cell shape in space regions, where interactions with ECM have to be considered, or at low resolution sufficiently far away from the liver tissue damage zone. High resolution cell models permit a higher accuracy of biomechanics at the expense of longer simulation times19-23, while low resolution models are less accurate, but computationally more efficient24. The DT is based upon consistent use of computing forces of cell and ECM displacement and deformations, whereby ECM is mimicked as linked strings of polymer-like objects similar as in refs25-27. Computational models have already been successfully used to predict cellular and tissue responses on several aspects of acute and chronic liver injury in mice and patients and are able to integrate complex interactions and system components with increasing realism28-36. Some of them resolve tissue microarchitecture, representing cells individually28,29,31,32,34-36, others study cell densities and molecular concentrations averaged over spatial scales at which the individual organization of cells can be neglected within continuum models33, or even using statistical models discarding any space structure30. At the tissue microarchitectural level, some models consider a rigid scaffold structure leading to cellular automaton models, which does not permit gradual cell displacements or deformations, but are computationally highly efficient31,32,34, while others resolve biomechanical cell and tissue properties at the expense of long simulation times but generate simulated tissue structures highly reminiscent of 3D histological image reconstructions28,29,35,36.
The current DT belongs to the last class of models and simulates liver fibrogenesis, a highly dynamic, tightly controlled process that comprises continuous communication between hepatic cell types and structural liver components. It builds upon an earlier highly simplified model that omit HSCs, Mphs, cell-cell communication, ECM production, and focus on regeneration of liver after one dose of CCl4 only28. The present work extends this line towards chronic liver disease and hence represents a fundamental step towards a full liver DT (Supplementary Fig. 1). Many molecular aspects have been experimentally delineated, whereas some key aspects, such as the biomechanical control of the interplay between hepatocytes, HSCs, Mphs and the collagen fibres have not been explicitly considered previously and are subject of the presented work.
(I) Different etiologies of liver damage induce variant patterns of fibrosis
In a first step we compare fibrosis patterns of different etiologies to generate possible hypotheses on the generation of pericentral/septal fibrosis. Liver fibrosis patterns present as septal, biliary or pericellular types. In case of septal fibrosis, liver damage occurs in the pericentral compartment and then bridges between central veins (CV) of contiguous lobules. Repetitive (chronic) CCl4 intoxication is a well-studied animal model for septal fibrosis.
Picro-sirius red (PSR) staining of the liver after 12 CCl4 injections during 6 weeks shows bridging fibrosis between CVs, forming so-called “pseudolobules” (Fig. 1A). In biliary fibrosis, exemplified by the Abcb4KO mouse, collagen deposition starts periportal and bridges towards the adjacent portal veins in progressed disease (PV; Fig. 1B). In CCl4-induced fibrosis, crosslinked collagen fibers form a thin and sharp wall-like structure, whereas in the livers of Abcb4KO mice, collagen fibers remain scattered and build a periportal fibrosis area that bridges as a broader street net between portal fields (Fig. 1B). Early liver fibrosis in human appears in both pericentral and periportal regions (Fig. 1C) according to etiological factors, forming cirrhotic nodules in late disease stages17 (Fig. 1D). The view of the accumulated collagens of septal fibrosis in 3D shows that they form into a wall-shape structure (Fig. 1E). In conclusion, fibrosis patterns vary depending on disease etiology and type of injury. Currently, knowledge on the underlying mechanisms that explain the respective kind of ECM deposition pattern is limited. The sharp localisation of ECM in form of a narrow streak that is connecting two lobules, as present in CCl4 fibrosis, cannot be found in any disease stage of the Abcb4KO-mouse, but is representative for advanced fibrosis in human patients e.g. from NASH (Nonalcoholic steatohepatitis)/MASH (Metabolic dysfunction-associated steatohepatitis) (Fig. 1D). We analyze the spatial distribution of CCl4 metabolising (CYP2E1+) hepatocytes in correlation to collagen deposition during the dynamics of liver fibrosis by (immuno)staining of serial sections with PSR, CYP2E1 and CK19, the latter indicating bile duct epithelial cells representative for periportal locations. Thus, we observe that the fibrotic wall in CCl4-induced fibrosis is generated in the spatial region of CYP2E1+ hepatocytes, which localize around and form stripes connecting CVs, upon repetitive CCl4 injections (Supplementary Fig. 2). This indicates that the observed septal fibrosis pattern formation is associated with the dynamics of metabolic activity (CYP2E1 expression), damage, and proliferation of hepatocytes in response to the toxin-induced injury. This guides us to the main hypothesis, that the sharp fibrotic wall in septal fibrosis may be formed by compression of an initially scattered ECM, driven by mechanical forces that are exerted by the proliferating hepatocytes.
In order to handle the complexity of this scenario, and to be able to continuously integrate and exchange possible drivers of septal fibrosis patterns formation, we select a systems biology approach and develop a minimal liver Digital Twin (DT) that is able to simulate observed and hypothesized process steps leading to fibrotic scar formation, including intercellular signal processing, cell-cell communication and tissue biomechanics. The key components of the DT include different liver cell types, sinusoids, and collagen fibres (Fig. 1F), integrated in a two-hepatic lobule model for the process of fibrogenesis. We select a two-liver lobules DT, since the minimal fibrotic wall unit in septal fibrosis bridges two central veins (Fig. 1F). The micro-architecture and cell composition in the DT space unit is developed as an experimental equivalent of the starting state prior to the first administration of CCl4.
Next, we collect data to construct the spatial-temporal microarchitecture in the DT of liver fibrosis. The pursued strategy is to calibrate the DT with data on acute liver injury after a single dose of CCl4 and predict with the so calibrated DT the remodeling of liver tissue microarchitecture after repeated doses of CCl4. The predictions are then confronted with the experimental findings after repeated CCl4 injections (twice a week at days 3 and 7 of each week for 6 weeks).
(II) Experimental data-driven generation of a computational microarchitectural model for the formation of fibrotic scars resulting from acute drug-induced liver injury (DILI) following a single CCl4 injection.
Below the components of the DT are stepwise introduced, starting with the tissue microarchitecture, then extending to the hepatocytes presenting in various states, e.g. physiological, damaged (or dead), and proliferating. Following this, non-parenchymal cells, such as hepatic stellate cells and macrophages, are incorporated, along with the ECM. Each model element is outlined with its spatial depiction and potential dynamics. The chosen model constituents are crafted based on quantitative imaging of their designated structures.
A) Sinusoidal scaffold:
Previous research indicates that the sinusoidal network within the liver, often referred to as liver capillaries, sustains damage following CCl4 treatment, yet retains its principal structural framework28. Consequently, we utilize this framework as a foundational structure for the two-lobule DT. The sinusoids are conceptualized as semi-flexible chains of small spheres interconnected by springs (depicted in transparent red, Fig. 1F). This model representation enables the formulation of force balance equations for each sphere within the sinusoidal network, similar in structure to those governing the cells. Such a framework facilitates the simulation of sinusoid displacements. When neighboring cells exert compressive forces on a sinusoid, this reduces the apparent thickness of the sinusoidal diameter, thus emulating a compressive effect.
B) Hepatocytes:
To experimentally determine the fate and localization of different liver cell types, we utilize mouse livers from healthy controls and those harvested 3 days after a single CCl4 injection. This serves as a reference state to establish the initial configuration of the dynamic system. Day 3 is chosen, as it marks the initiation of fibrosis development, starting with administration of the second dose of CCl4. Following a solitary CCl4 injection, the liver undergoes complete regeneration of the lesion by day 7 (see ref28).
We quantify the proportions of healthy, necrotic, CYP2E1-expressing, and proliferating (BrdU-labeled) hepatocytes in the reference state. Initially, pericentral hepatocytes experience necrosis due to CCl4 metabolism (8.86±2.45% of the total area shows CYP2E1 positivity). Conversely, midzonal and periportal hepatocytes maintain a healthy appearance throughout the process. This is demonstrated and quantified on day 3 through HE and IgG staining (Fig. 2A, B; Supplementary Fig. 3A). Notably, the highest fraction of necrotic cells is observed 1 day after the initial dose, consistent with prior findings28.
Vanishing pericentral hepatocytes are confirmed by the complete loss of CYP2E1 positivity (Fig. 2A-C), a decrease in Cyp2e1-mRNA expression by day 2, and elevated plasma alanine transaminase (ALT) and aspartate transaminase (AST) levels on days 1, 2, and 3. Enzyme levels normalize as regeneration progresses (Fig. 2D). Necrotic regions nearly close by around day 5. CYP2E1 mRNA and protein expression do not fully return to their original levels and patterns, at least until day 30 in the regenerated liver. Moreover, a layer of midzonal peri-necrotic hepatocytes starts expressing CYP2E1 on day 3, forming a dumb-bell-shaped region connecting adjacent central veins (brownish region around the yellow center line of CYP2E1, Fig. 2A). The number of central veins connected by CYP2E1-positive cells remains elevated until day 30 in the regenerated livers (Fig. 2C). The remaining hepatocyte fraction on day 3 exhibits an increased count of BrdU/Ki-67 positive cells/nuclei, which are not uniformly distributed within the hepatic lobule, but are more prevalent in peri-necrotic areas (Fig. 2A, B, E, F; Supplementary Fig. 3A).
In the DT, each hepatocyte is individually modeled as a single unit, whereby the spatial resolution of the cell shape is made dependent on its proximity to the emerging necrotic lesion (Fig. 2G). Hepatocytes close to the lesion, interacting with the ECM fiber network during fibrosis formation, are simulated using a "deformable cell model (DCM)"23 that explicitly tracks cell shape changes (Fig. 2G, Supplementary Movie 1). Hepatocytes further away, not in contact with newly synthesized collagen fibers, are modeled using a "center-based model (CBM)", where forces on the cell are represented as forces on the cell's center. For simplicity, the shape of a cell in CBM is depicted as a sphere, as this is the shape the cell would assume in isolation.
CYP2E1-positive hepatocytes (yellow) are connecting two CV (Fig. 2G). CYP2E1-negative hepatocytes outside the lesion are classified as damage-resistant cells (green, Fig. 2G). CYP2E1-positive hepatocytes underwent cell death (necrotic cells, brown) following CCl4 injection. A portion of the remaining hepatocytes that proliferate (blue) to close the necrotic lesion is stochastically selected based on the time point after CCl4 treatment and their distance from the lesion. The spatio-temporal proliferation pattern is inferred from quantifying Ki-67-positive cells, as previously performed28. BrdU staining predominantly serves to illustrate a higher number of proliferating hepatocytes surrounding the necrotic lesions.
C) Non-parenchymal cells and the extracellular matrix (ECM):
Next, a spatial-temporal visualisation with a detailed analysis of the kinetics is conducted on HSC, recognized for their involvement in both regeneration and fibrogenesis, following a single CCl4 injection. The assessment encompasses a spatiotemporal approach, utilizing α-SMA, desmin, and PSR as markers to evaluate aHSC and total (aHSC plus qHSC), alongside ECM deposition and macrophage (Fig. 3A-E; Supplementary Fig. 4A). The spatial distribution of α-SMA-positive cells in regions between CVs at day 3 corresponds to a width of about four hepatocyte layers (80 μm, the size of the lesion; Supplementary Fig. 5). Notably, the stained area fraction for both desmin (5.54±0.48%) and α-SMA (3.73±1.57%) peak at day 3. Additionally, RNA expression levels of desmin and α-Sma are found to be upregulated on day 3.
The presence of PSR-positive fibers is mainly observed in the tunica adventitia of larger blood vessels, namely the CV, PV, and hepatic artery. Conversely, small, isolated, and scattered ECM fibers are evident on day 3 within necrotic areas. This observation aligns with the increased mRNA levels of Col1α1 and Col1α2 around this specific time point (Fig. 3D). Notably, an excessive deposition of ECM has not been previously reported in the context of acute toxic liver injury, and our study findings are consistent with this trend. This suggests that the isolated ECM fibers visualized within the necrotic region on day 3 are rapidly degraded following their production, ultimately diminishing over time during the process of liver regeneration.
In the DT, each HSC is conceptualized as a nucleus sphere linked to several chains of elastic springs, which represent the elongated protrusions of the cells. qHSCs are denoted in cyan, while aHSCs are depicted in light brown (dashed red boxes Fig. 3F). In the model, solely the aHSCs are endowed with the capacity to generate collagen fibers. These fibers congregate within the extracellular space, coalescing to form an intricate network (elaborated further in the Supplementary Information).
The ECM network is portrayed as a network of semi-flexible chains that necessitate energy to bend and compress along the bundle, effectively simulating their mechanical attributes (indicated in purple, as shown in Fig. 3F; Supplementary Fig. 6A). The density of the ECM network is contingent on the density of the collagen fibers. When the separation between two fibers, characterized by their shortest distance, falls below a specific threshold, they undergo crosslinking, illustrated by the introduction of an additional node (Supplementary Fig. 6B). The ECM network model is fine-tuned through bending and compression experiments conducted in vitro, in line with the work of Yang et al.41 (2008) and Ferruzzi et al.42 (2019). This calibration process ensures congruence between the modeled ECM network and the documented mechanical responses to bending and compressive stresses (Supplementary Fig. 6C-G).
In the next stage, we postulate that the cumulative ECM deposition is influenced by degradation processes, such as matrix metalloproteinases (MMPs) secreted by resident macrophages (Kupffer cells), which act as ECM modulators. Consequently, we conduct an analysis of the spatiotemporal distribution of macrophage infiltration using F4/80 immunostaining. In healthy livers, F4/80-positive cells exhibited a predominantly uniform distribution within periportal (arrow 1 in Fig. 3A, day 0 of F4/80) and midzonal (arrow 2 in Fig. 3A, day 0 of F4/80) compartments, while in pericentral areas (arrow 3 in Fig. 3A, day 0 of F4/80), there are much fewer F4/80-positive cells (Fig. 3A; Supplementary Fig. 4A, B). However, at days 2 and 3 following CCl4 injection, there are many more F4/80-positive cells in the entire lobule (see exemplarily marked regions by black arrows), Supplementary Fig. 4A). This injury mediated spatial distribution of F4/80-positive cells required more than 16 days to return to a normalized state. Gene expression analysis revealed upregulated Cd68, a macrophage marker, as well as ECM modulators (Tgfβ1, Mmp9, and Timp1), within the first 4 days after CCl4 injection. Mphs engulf the bodies of necrotic hepatocytes43 and play a role in degrading collagen fibers44.
In the DT, macrophages are depicted as elastic spheres (dark blue, Mph, Fig. 3F), which upon contact, phagocytose necrotic hepatocytes and are able to degrade collagen fibers.
In summary, the experimental findings suggest a complex interplay involving hepatocytes (healthy, damaged source of DAMPs, and proliferating), qHSC, aHSC, ECM, and macrophages, forming a minimal set of building blocks to construct a dynamic model for studying the cycles of damage, regeneration/repair, and fibrogenesis after repetitive CCl4 injections (Fig. 3G). Notably, a subset of aHSC is known to revert to a quiescent state by cell communication with macrophages, and present with a “reverted” phenotype that displays a so-called memory response upon challenge with a new insult9,11,12, in this case from the 2nd injection of CCl4.
(III) Implementation of Processes Pertaining to Septal Fibrosis Formation Dynamics & Prediction of Disease Scenario
Subsequent to setting up the experimental design and observations leading to fibrotic wall formation, the DT is subjected to a simulation of six CCl4 injections over a three-week timeline (twice a week), mirroring the experimental schedule. CCl4 injections are performed at days 0 and 3 of each week, enumerating weekdays from 0 to 6. In this scheme, injections (enumerated in parenthesis) occur at days 0 (1), 3 (2), 7 (3), 10 (4), 14 (5), 17 (6), with a last histological analysis at day 20. Each CCl4 injection induces damage to the fraction of CYP2E1+ hepatocytes (Supplementary Fig. 7A), which is always reflected by upregulated AST and ALT levels in the blood of the mice (Supplementary Fig. 7B) compared to the levels in control mice (represented by time point zero in Fig. 2D).
In the experiments, only the cell layer adjacent to the necrotic lesion expresses CYP2E1 at day 3 after one CCl4 injection. The necrotic areas subsequently are restored by hepatocyte proliferation, whereby importantly, the newly generated hepatocytes are all expressing CYP2E1, indicating peri-central metabolic zonation, and therewith, sensitivity for the next CCl4 injection. CYP2E1 positive hepatocytes are now spatially organized to connect two neighboring CVs (note that the stripe of CYP2E1-expressing hepatocytes after 2, 4, or 6 injections of CCl4 is of very similar size; Supplementary Fig. 7A, C, D).
In the DT, hepatocytes filling the necrotic lesion are assumed to express CYP2E1. The reestablished CYP2E1-positive pattern guides the progression of damage over time and space, therewith shaping the spatial DAMP concentrations and ensuing responses, including recruitment of HSC and Mph, deposition of ECM, as well as compensatory hepatocyte proliferation, in order to repeatedly restore the lesion.
The proliferation profile is parameterized based on Ki-67 staining quantification on day 3 after the first CCl4 injection, consistent with previous studies28. Experimental quantification and spatial localization of hepatocyte proliferation after repeated CCl4 injections revealed that the healthy hepatocytes enter the cell cycle for proliferation. The proliferation rate is highest adjacent to the lesion and decreases gradually with increasing distance from the lesion (Fig. 2F).
A) DT predicts intermediate chicken-wire pattern:
In the DT, compromised CYP2E1-expressing hepatocytes are assumed to serve as the source of DAMPs, which among others attract HSC (Supplementary Fig. 8A, B), causing their activation and migration into the lesion. These HSCs eventually produce collagen fibres, initially distributed in a so-called “chicken wire” pattern, and captured by our model as shown in Supplementary Fig. 8C, D. In the experimental context, scattered fibers are observed a week after the second CCl4 injection, which progressively organize into a well-defined fibrotic wall after 4-6 doses. The collagen deposition pattern goes along with the spatial pattern of the desmin and α-SMA expressing cell population (Supplementary Fig. 9A). The temporal pattern of the mRNA expression of the ECM genes collagen 1α1 and 1α2 shows a similar increase as those of Desmin, α-SMA, Cd68, Mmp9 and Timp1 at RNA and protein levels (Supplementary Fig. 9B, C, D). No further accumulation of HSCs is found in response to doses from 4-6 CCl4 injections as shown in protein and mRNA levels of α-SMA.
The mechanism mediating spatial organization of secreted collagen fibres and its dynamic alterations to finally develop fibrotic walls is unknown. We hypothesize that the enzyme CYP2E1 is most probably shaping the overall spatial patterning of the ECM, as well as the occurrence of the ECM producers (aHSC), defined by desmin and α-SMA staining, even though it remains relatively disperse (Supplementary Fig. 9A). To test this hypothesis, we simulate the entire process of repeated CCl4-injections and the downstream events this triggers in the DT. We study six alternative scenarios (S) of collagen fibres deposition: (S1), as single fibres; (S2), as single fibres anchored to the nearest sinusoids; (S3), as crosslinked single fibres; (S4), as crosslinked single fibres anchored to the nearest sinusoids; (S5), as collagen fibres, which are firstly deposited as single fibres and then gradually crosslinked; (S6), as collagen fibres that are firstly deposited as single fibres anchored to the nearest sinusoids and are then gradually crosslinked (Supplementary Fig. 10A). In order to investigate how sensitive the collagen network in each of the scenarios is in response to biomechanical forces exerted by proliferating hepatocytes, we first simulate a simplified lesion without HSCs and Mphs in a stripe inside the two-lobules system with the DT (Supplementary Fig. 10B-D) for a time frame of 21 days.
B) Repeated rounds of CCl4-injection result in fibrotic wall formation in DT simulations through mechanical compression by hepatocyte proliferation:
The total mass of collagen fibres is initialized homogeneously distributed in the lesion region. There is no fibre deposition or degradation during the simulation. All collagen fibres are generated immediately after the first CCl4 injection. To pre-estimate the maximum density that the produced collagen adopts in the striped space, which we assume corresponds to the collagen density required for the formation of the experimentally observed fibrotic wall, different densities as described in the literature27 (0.4, 1.0, and 2.0 mg/mL) are tested in the simulations with the DT. Surprisingly, every time a wall-like fibrotic structure is resulted at day 20 after 6 repeated CCl4 injections. Nevertheless, some of the simulated scenarios exhibit variations in thickness, appearing either thinner or thicker than what is observed in the experimental data (see Supplementary Fig. 10B-G for details). Additional information regarding the extensive testing of various collagen densities can be found in the Supplementary Information section titled "Organizational Structure of the Collagen Network." This supplementary section provides detailed insights into the evaluation process of different collagen densities and their impact on the simulation results. Ultimately, scenario 5, which exhibits a collagen density of 1.0 mg/mL, closely resembles the experimental data and is chosen for further simulations and analysis. This scenario's plausibility stems from the concept that aHSCs generate and accumulate ECM fibers, which subsequently crosslink to form a comprehensive ECM network.
Following the selection of a plausible collagen density scenario (scenario 5) based on its similarity to experimental observations, the next step involved incorporating macrophages into the DT Model. This is done considering the dynamic changes in the numbers and distribution of F4/80-positive-stained Mph (dark blue; Fig. 3F and G). The assumption here is that these Mphs digest collagen fibers upon contact. Consequently, in the DT, collagen fibers start being generated by aHSC after the second CCl4 injection, followed by digestion of these fibers by Mphs within the lesion. After six doses of CCl4 injection in DT (Fig. 4A), the total amount of deposited collagen is experimentally quantified by measuring the PSR positive area over time (Fig. 4B).
In the DT, the interplay of ECM production (α-SMA positive cells), digestion (F4/80 positive cells), and biomechanical forces from hepatocyte division collectively modulate the thickness of the ECM network. This modulation is reflected in the changes of the gyration diameter, which captures the width of the collagen distribution, with the maximum distribution approximately aligned with the fibrotic streak connecting neighboring CV. This gyration diameter decreases from 46µm after the second CCl4 dose to about 20µm after 6 doses of CCl4, ultimately shaping the collagen network into a well-established fibrotic wall (Fig. 4C, D). This intricates interplay leads to the formation of the characteristic fibrotic patterns observed in experimental data (Fig. 4E; Supplementary Movie 2).
C) Repeated rounds of CCl4-injection in the experiments support DT-predicted scenario
In a parallel setup mirroring the chronic liver injury scenario over 3 weeks with two CCl4 injections per week, each CCl4 injection results in cell death of CYP2E1-expressing hepatocytes localized in the lesion (Supplementary Fig. 7A). PSR staining reveals that single collagen fibers are deposited in the liver starting from day 3 after the first CCl4 injection, gradually accumulating in the necrotic region after two doses, and coalescing into thick bundles after 4 doses (Fig. 4E). Following 6 doses of CCl4, a clear wall-shaped fibrosis pattern is established, spanning between, and surrounding the CVs. This sequential order of collagen fiber accumulation, along with the spatiotemporal distribution of fibers and aHSCs, is successfully simulated by the DT model (Fig. 4D, E). aHSCs migrate into the lesion following the first dose and accumulate over time in the region, where the wall-shaped collagen network is formed after 6 doses of CCl4 (Supplementary Fig. 9A). This accumulation of aHSCs is also reflected in the quantification of the α-SMA positive cell fractions over time (Fig. 4F).
As indicated by HE staining (Supplementary Fig. 7A), tissue infiltration of cells with small nuclei is observed in and around fibrotic areas. Some of these cells are Mphs, whose spatial distribution follows a pattern similar to aHSC as shown by F4/80 positivity. The accumulation of positively stained Mphs in the fibrotic region is confirmed by F4/80 staining after two doses of CCl4. After 4 doses, Mphs start to accumulate in wall-like patterns, linking CVs after 6 doses that in the 2D sections look like streaks (Supplementary Fig. 9A). Quantification of F4/80 positive cells indicates that the Mph numbers remain relatively stable after the second CCl4 dose (Fig. 4F). This stability is supported by RNA data, which show no further upregulation of Cd68, Mmp9, or Timp1 expression (Supplementary Fig. 9D).
In the DT, it has been established that Mphs migrate toward the gradient of DAMP concentrations following the first CCl4 dose and locate around the lesion in a scattered manner, even after 6 doses of CCl4 (Fig. 4D). Furthermore, the number of Mphs in the lesion is defined by the F4/80 positive fraction (Fig. 4F), which increases after the first CCl4 injection.
In summary, the spatiotemporal distribution patterns of the collagen network, aHSC, and Mph simulated by the DT align well with the experimental observations. The DT predicts that a gradient of DAMPs derived from CYP2E1-expressing cells determines the localization of ECM deposition around and between the closest CVs.
(IV) Spatial distribution of cells expressing CYP2E1 drives ECM deposition through the release of DAMPs, while the division of hepatocytes shapes the resulting pattern of ECM deposition.
In order to elucidate the origin of the observed patterns in ECM deposition, cells expressing CYP2E1 are pinpointed as the source of DAMPs, both temporally and spatially. In cases of acute liver injury induced by CCl4, the proportion of CV connected by CYP2E1-expressing hepatocytes increases from day 3 to average 2 connections per CV and remains constant until at least day 16 (Fig. 2C, Supplementary Fig. 3B). Surprisingly, in chronic liver injury, the majority of CVs display 3-4 connections with CYP2E1+ hepatocytes after 6 doses), which is 3-4 connections more than 3 days after the first injection (Supplementary Fig. 7D, and 2 connections more than 4 days after the first injection (Fig. 2C). The identified pattern of CYP2E1+ cells is integrated into the DT as a source of DAMPs for subsequent CCl4 injections (Supplementary Fig. 3A and 8A). Within this framework, the model predicts that spatial DAMPs lead to ECM deposition by attracting aHSC and Mph (Supplementary Fig. 3A, 4A, 8B, Supplementary Movie 3).
To test whether the spatial distribution of dying/damaged/stressed hepatocytes influences the shape of the observed collagen network, we conduct perturbation simulations (Fig. 5A). For instance, we consider a scenario with randomly distributed CYP2E1+ hepatocytes after the second CCl4 injection (Perturbation 1; P1, Fig. 5A), as compared to the reference case, where CYP2E1-expressing hepatocytes are confined to the lesion. In this case, the resulting collagen network has a larger diameter and a more homogeneous distribution of collagen fibers (without forming a wall pattern), unlike the reference case (P1, Fig. 5B, C, Supplementary Movie 4).
To experimentally validate the DT simulation for P1, mice with altered spatial and absolute expression of CYP2E1 are tested. This is achieved by deletion of GATA4 from the endothelial cell.In these mice, removal of the transcription factor GATA4 (GATA4LSEC-KO) from endothelial cells45 results in reduced Wnt2 expression. Wnt2 is a well-known upstream regulator of CYP2E1 expression46. Therefore, these mice show a reduction of CYP2E1 positive hepatocyte fraction due to reduction of Wnt2 expression from LSEC (Fig. 5D, E). These mice underwent chronic CCl4 intoxication with2 injections per week for 3 consecutive weeks. The resulting Cyp2E1+ hepatocyte pattern as known from the control mice is disrupted in chronic CCl4-treated GATA4LSEC-KO mice, presenting with a more randomly distributed pattern (Fig. 5F). As predicted, the ECM deposition pattern is also compromised in 75% (6 out of 8) of the tested GATA4LSEC-KO mice, showing a more uniformly distributed perisinusoidal fibrosis, as compared to control mice displaying the septal pattern (Fig. 5F). This perturbation experiment suggests that the spatial distribution of CYP2E1+ cells is responsible for the septal fibrosis patterns that develop in the mouse liver due to chronic CCl4 injections.
We also conduct in silico perturbations targeting hepatocyte proliferation, HSC activation, and Mph phagocytic activity, comparing the outcomes with the reference DT model (Fig. 5A). To assess the impact of biomechanical pressure from dividing hepatocytes on collagen organization, we run the DT simulation by excluding hepatocyte proliferation for 21 days (P2). The model predicts that fewer hepatocytes would be present in the lesion compared to the reference model. Interestingly, the same amount of deposited ECM in the injured regions fails to form a fibrotic wall, as observed in the reference model after 6 doses of CCl4. The collagen network diameter is much larger compared to the reference case (P2; Fig. 5B, C, Supplementary Movie 5). The relationship suggesting that fewer hepatocytes lead to a wider ECM network supports the conclusion that hepatocyte division is a significant driver of the experimentally observed septal fibrosis pattern, likely through biomechanically displacing and compressing the ECM fibres, which is absent in absence of hepatocyte proliferation (Supplementary Fig. 10B, C, D).
In another scenario, we investigate the influence of HSC phenotypes and activities on fibrosis pattern development and shaping. We conduct three independent perturbations. In the reference model, 50% of aHSCs undergo apoptosis while 50% revert to a quiescent state during regeneration (as determined in previous experimental studies11). In one perturbation, we assum that 70% of aHSCs become apoptotic (P3). In a second perturbation, aHSC migration is set to 0% (P4), and in a third perturbation, 100% of aHSCs convert to a quiescent phenotype (P5). These perturbations are simulated with the DT for 3 weeks, involving 6 CCl4 injections. In P3, the total number of deposited collagen fibers is significantly reduced, because 70% of aHSCs are lost after each round of CCl4 injection, resulting in 40% fewer aHSCs in the perturbation model compared to the reference model, thus leading to less collagen deposition (P3, Supplementary Fig. 11; Supplementary Movie 6). In P4, there are fewer collagen fibers distributed along the lesion boundary, because HSC lacking migratory activity remains in their original position (P4, Supplementary Fig. 11; Supplementary Movie 7).
As a consequence, fewer HSCs accumulate in the center of the lesion compared to the reference. In P5, collagen fiber deposition begins earlier compared to the reference model, around day 1.5 versus day 3 (Fig. 5B). This occurs because reverted HSCs are more responsive to DAMP-mediated activation. Perturbations P3 and P4 result in a compromised ECM deposition pattern, whereas P5 leads to the same dense fibrotic wall in the center of the lesion after 6 CCl4 injections. Although collagen fibers are generated earlier as compared to the reference, they are scattered within the lesion. The wall structure does not form earlier. This is evident in the gyration radii, where those of P3 and P4 are much larger, whereas that of P5 is very similar to the reference model. The HSC-directed perturbations do not noticeably affect the number of hepatocytes in the lesion (Fig. 5B, Supplementary Movie 8). In summary, the HSC-directed model perturbations suggest that the HSC phenotype primarily impacts the absolute amount of ECM deposition, while HSC migration is crucial for the formation of the septal fibrosis pattern.
Finally, we assess the phagocytic and migratory effects of Mph and conducted two separate Mph related perturbations with the DT. Mphs are transitioned to Ly6Clow after 5 days following each CCl4 injection, instead of the 12 hours as evident in the reference model (P6), or they are assumed to be deficient in migration ability (P7; Fig. 5A). Simulation P6 results in increased collagen fiber presence in the injured regions without altering the wall structure formation pattern (Fig. 5C; Supplementary Movie 9). Simulating Mph migration deficiency (P7) leads to the formation of a looser and more broadly distributed collagen fiber network (Fig. 5A, Supplementary Movie 10). Furthermore, the amount of deposited collagen increases, because the number of Mphs available in the lesion to degrade collagen fibers is decreased (Supplementary Fig. 11). The gyration radius of the collagen network in simulations P6 and P7 is slightly larger than that in the reference model (Fig. 5). Perturbations P6 and P7 have no impact on the number of hepatocytes in the lesion (Fig. 5B). The perturbation results indicate that Mph activity has a minor effect on the development of the septal fibrosis pattern.
In conclusion, we propose that the septal fibrosis pattern observed in CCl4-induced chronic liver disease in mice is primarily driven by the spatial distribution of CYP2E1+ hepatocytes, HSC activation and migration, and is shaped by mechanical pressure resulting from dividing hepatocytes.
Liver fibrosis is characterized by the accumulation of ECM, in the here considered model linking the CVs of neighboring lobules by fibrotic walls due to injuries by repeated injections of CCl4. A fundamental concept in understanding liver fibrosis is comprehending why and where ECM is deposited in specific patterns that eventually progress to cirrhotic nodules. However, a comprehensive grasp of this pattern has been impeded by its intricate complexity involving numerous components and subprocesses that can currently not be simultaneously monitored in all their aspects through experimentation.
In an effort to provide an alternative to a solely experimental approach, we have devised a digital twin (DT) of a two-lobule liver system. This model represents the fundamental tissue unit for the fibrosis pattern we have investigated. It relies on quantifiable experimental parameters to explore the potential sequence and interplay of subprocesses in the formation of fibrotic walls. The iterative process involves refining the model based on experimental guidance, making predictions, and validating the model through experimentation.
The DT encompasses a minimal selection of critical liver cell types (hepatocytes, hepatic stellate cells, and macrophages) as well as tissue elements (sinusoidal networks and collagen fibers). It serves as a platform to study the development, progression, and pattern formation of septal fibrosis following exposure to CCl4. The DT further elucidates how mechanical interactions and molecular crosstalk between these tissue constituents contribute to the observed septal fibrosis pattern in experiments. To achieve this, the DT meticulously captures the temporal and spatial dynamics of tissue microarchitecture. It directly models cells and the ECM fiber network at a subcellular level of resolution, emulating their rearrangements and remodeling in response to repetitive CCl4 exposure. At the current stage of the DT model, the intricate intracellular mechanisms governing cellular decisions and cell-cell communication have not been explicitly incorporated yet and are planned as follow up steps in the context of integrating models at molecular and cellular scales.
A simulation can be considered to study a process in a well-defined setting under a microscope, similar as in live imaging, the latter, however, due to the duration of the process currently being out of reach. Repetitive exposure to CCl4 leads to iterations of tissue damage caused by the death of CYP2E1-positive hepatocytes, followed by incomplete regeneration through the division of surviving hepatocytes. In the DT model, the region of CYP2E1 positivity is determined based on experimental evidence linking adjacent CV from the second injection of CCl4 onward.
The dynamic process of growing and dividing hepatocytes is stochastically sampled from the spatiotemporal distribution of Ki-67-positive hepatocytes surrounding the lesion over time. Migration of HSC and Mph is modeled by applying a chemotactic force on the cells in the direction of the gradient of DAMPs, as described in studies by Seo et al.47 and Roohani and Tacke48. We mimic active cell migration as in a previous communication28, discarding explicit representation of cell protrusions as those can be shown to generate the same movement dynamics24.
DAMPs are emitted by damaged hepatocytes. After the initial exposure to CCl4, the DAMPs concentrate around CVs. However, during subsequent regeneration phases, observations align with previous research49,50 showing that CYP2E1-expressing cells recover and form a wedge shape connecting CVs. Analyzing connected CVs through CYP2E1-expressing hepatocytes reveal that shortly after fibrosis establishment, one CV is linked to 3-4 neighboring CVs. This may occur when a ("mother") CV associated with one lobule splits into two ("daughter") CVs that are then associated with two (other) lobules: before branching, there is one circular CYP2E1-positive area surrounding the mother-CV, while each daughter CV is also associated with a circular CYP2E1-positive area. Through the tree-like hierarchical branching structure, the CYP2E1-positive zone may still connect both daughter CVs until they are sufficiently far apart from each other (the branching concept is explained in Fig. 3 and 4 in ref51). Consequently, hepatocytes perish along these CV-CV connections after subsequent CCl4 doses, altering the spatial pattern of DAMP sources from around the CV after the first dose to along the CV-CV connection in later doses. Thus, the time point of the second dose, 3 days after the initial CCl4 exposure, is chosen as the model's initial reference point, encompassing critical fibrogenic factors such as the earliest signs of metabolic activity of midzonal hepatocytes (CYP2E1 expression), pronounced cell division, maximal HSC activation, and Mph infiltration. Most model parameters are calibrated based on data at this specific time point. Using the calibrated DT, simulations encompassing 6 consecutive CCl4 injections are performed, with results being compared to experimental outcomes.
The DT simulations accurately replicate the experimentally observed scenario including the exact fibrosis wall pattern formation, suggesting that the spatial distribution of CYP2E1-expressing cells drives pattern formation through DAMP gradients originating from deceased hepatocytes. To test this concept, a mathematical analysis using the DT is conducted, considering a homogeneous distribution of CYP2E1-expressing cells. In such a scenario, the DT predicted dispersed DAMP concentrations resulting in homogeneous ECM deposition, as opposed to forming a linear wall as seen in the reference case. This hypothesis is examined using mice with partially disturbed CYP2E1-expression (GATA4LSEC-KO), where the GATA4 transcription factor plays a role in LSEC specification and differentiation. These mice exhibit a more scattered CYP2E1 distribution, as compared to the control and administering repeated CCl4 doses, similar to the model's setup, leads to a more scattered fibrosis pattern devoid of the expected formation, thereby validating the DT's prediction. This indicates that the spatial distribution of deceased hepatocytes, the source of DAMPs in chronic liver injury, significantly influences the overall fibrotic pattern.
While the pattern of CYP2E1+ hepatocytes shape the fibrosis's spatial arrangement, it does not elucidate the reason behind the sharp localization of fibrotic walls in the middle of the CYP2E1-positive region. The DT proposed that this phenomenon results from successive waves of hepatocyte proliferation after repeated CCl4 administrations. When the dead cell lesion closes, proliferating hepatocytes near the lesion mechanically displace ECM towards the center of the dead cell zone due to mechanical pressure. This prediction is supported by the observation that initially, ECM is scattered within the CYP2E1+ zone where aHSCs deposit it. Over time, the ECM becomes increasingly localized.
The intricate network structure and ECM density could not be directly measured. The DT predicts an appropriate density of around 1.0 mg/mL to explain fibrotic wall formation, a value consistent with the study by Stein et al.27. Different scenarios for the architecture of collagen networks are compared, with the scenario where collagen initially forms individual fibers that gradually crosslink fitting the observed collagen deposition pattern well. This configuration is selected for all subsequent DT simulations. Another DT prediction indicates that if a sufficient number of aHSCs (20% more than in the reference case) are depleted, after each CCl4 injection, the wall-shaped collagen network would not form due to the reduced number of collagen fibers. This aligns with prior experimental research demonstrating that the loss of aHSC leads to diminished fibrogenesis and matrix accumulation51.
In addition to the overall global fibrosis pattern, surprisingly, the model manages to reproduce the local "chicken wire" pattern of collagen fibers observedat early stage of CCl4-induced fibrosis, and as characteristic of steatotic and alcohol related liver disease. This is because aHSCs are generating collagen fibers while the remains of dead hepatocytes are still present, resulting in the fibers enveloping the dead bodies and forming scaffolds with large spaces. This further suggests the model's flexibility for other disease etiologies and confirms its clinical relevance.
In summary, the presented DT model effectively explains the formation of fibrotic walls in septal fibrosis. While other studies have addressed aspects of fibrosis formation, they often overlooked tissue microarchitecture or the influence of mechanical aspects.
Friedman and Hao33 developed a mathematical model aimed at investigating liver fibrosis by representing interactions among various cells and cytokines. However, their model utilizes a "continuum" approach that involves averaging over multiple cell diameters, thus lacking the capability to capture tissue microarchitecture. Additionally, their focus is primarily on cell and molecular kinetics, omitting mechanical aspects. On the other hand, Dutta-Moscato et al.31 and Yoshizawa et al.32 adopted agent-based models that positioned cells on a two-dimensional fixed lattice. Unfortunately, this choice limits their ability to model cell and tissue biomechanics, which, as evidenced by the present DT emerged as a critical factor in fibrotic wall formation. In addition, the work by Dutta-Moscato et al.31 describes the formation of fibrotic streaks along the portal field, which does not align with the potential of CYP2E1-positive hepatocytes driving such a process. In contrast, our research demonstrates that fibrotic walls are formed between CV, as indicated by experiments, inline with the interpretation of histological images by Yoshizawa et al.32.
We firmly believe that the present mathematical model for septal fibrosis signifies a significant stride in developing a comprehensive liver DT for CLD. It serves as the starting point for extending the DT to encompass various etiologies, including biliary fibrosis. Additionally, the gradual integration of more cell types and processes into the model could enable us to simulate the progression towards non-compensated cirrhosis and hepatocellular carcinoma, which represent critical endpoints of CLD that go beyond the fibrosis stage.
The liver DT technology offers a multitude of opportunities. As demonstrated and noted by Drasdo and Zhao53, it can guide the selection and design of experiments. It facilitates testing the interplay of numerous mechanisms at different levels within a strictly defined computer-based environment, spanning from intracellular pathways to tissue and even whole-body levels, as shown by Dichamp et al.35. Similarly, as highlighted in the study addressing paracetamol (acetaminophen) toxicity by Hoehme et al.29, the present DT has the capacity to extrapolate outcomes from in vitro experiments to expected in vivo systems behavior, even across different species.
Although the application of the DT has been demonstrated with non-human data, its prospective implementation in human contexts has the potential to bridge the gap between human tissue cultures, bioengineered experiments, animal disease models and patient-related scenarios. While there might be initial setbacks, these failures can help identify and address critical knowledge gaps concerning mechanisms and parameters, ultimately contributing to a holistic comprehension of human liver disease.
Personalizing the DT for specific patients remains a major challenge. In addition to in vivo extrapolations from in vitro systems, which can involve data that are generated in cells extracted from a patient, or translating mouse data to human disease phenotypes, the DT can be of direct use in clinical settings. Integrating molecules determined at various disease stages that qualify as biomarkers, and using them as indicators, such as in histopathology, could complement evolving high-resolution technologies like high-magnetic field MRI or temporal diffusion spectroscopy.
While artificial intelligence (AI) approaches have their merits, mechanistic DTs offer predictive capabilities for unknown interventions and provide causal chains that elucidate underlying mechanisms for which no data is available. AI and mechanistic approaches can be complementary, with AI potentially informing mechanistic DTs and accelerating their development.
In conclusion, our presented DT offers a valuable framework for understanding fibrotic wall formation in septal fibrosis. This model constitutes a significant advancement in liver disease research and sets the stage for broader applications, bridging experimental and clinical domains and promoting a comprehensive understanding of chronic liver disease.
Limitation
Macrophages produce matrix metalloproteinases (MMPs) to digest the collagen fibres. They further have been reported to contribute to the survival of HSCs54,55, and they phagocytize dead cells7. We here focus on the phagocytizing function of macrophages only and do not consider a survival signal to aHSC provided by macrophages or other immune cells. In our DT, aHSC autonomously does not revert or die before either the lesion is closed, or a certain collagen density is locally reached. As a consequence, depletion of macrophages in the DT does not shorten lifetime of aHSC but only reduces the degradation of the collagen network. Such further refinements will be addressed in future versions of the DT along with further applications in disease progression, as e.g. until HCC.
Animal experiments
Adult C57Bl/6N male mice (Janvier labs; France) are housed three per cage in a temperature-controlled (24°C) room with a 12-h light/dark cycle. Mice are given ad libitum access to water and laboratory diet (Ssniff, Germany). Mice are exposed to carbon tetrachloride (CCl4; Sigma-Aldrich, Cat. no. 319961) in a dose of 1.6 g/kg body weight37,38, either at a single (acute) or repeated doses. To track regeneration and wound healing parameters upon acute intoxication, mice are exposed to a single dose of CCl4. Blood and liver samples are harvested at 0, 1, 2, 3, 4, 6, 16, and 30 days post CCl4 injection. Liver fibrosis is induced by two doses of CCl4 /week for 3 weeks40. Three days after the last CCl4 injection, blood and livers are harvested. All experimental protocols with animals are carried out in full compliance with the guidelines for animal care and are approved by the Animal Care Committee from the German government (Animal permission number: 35.9185.81/G87/10 and 35.9185.81/G-216/16).
Histological staining and quantification. Stainings are performed on paraffin slides (3µm) for hematoxylin and eosin (H&E) and Sirius red staining and with fast green staining40. Further, immunohistochemistry (IHC) using anti-CYP2E1 (Sigma-Aldrich, HPA009128, raised in rabbit, 1:50), anti-mouse F4/80 (BioRad, MCA497, raised in rat, dilution 1:2000), anti-Ki-67 (Cell Signaling Technology, CST#12202, raised in rabbit, dilution 1:200), anti-CK19 (ProteinTech, 14965-1-AP, raised in rabbit,1:100), anti-BrdU (BioRad, MCA2060, raised in rat, dilution 1:250) and anti-α-smooth muscle actin (α-SMA; Abcam, Ab5694; raised in rabbit, dilution 1:100) is performed. Image acquisition is performed using an Aperio 8 slide scanner (Leica). A minimum of 10 high-power fields is captured for image analysis. The positive areas are quantified as a brown area percentage or nuclei number by Image J or manually, respectively (5-7 fields at least per slide, 10X).
Mathematical model: the digital twin
The digital twin considers each individual hepatocyte, HSC, and Mph as individual model units, and considers sinusoids, and extracellular matrix bundles both in space and time. Each hepatocyte and macrophage is represented by one component, while HSCs, sinusoids, and ECM bundles are mimicked as composed of many model components. The movement and interaction of each model component is modeled by a force balance equation. The force balance equations are updated simultaneously for all model components (e.g. hepatocytes, HSCs, Mphs and the collagen fibres) by solving a system of coupled equations, one for each component, to represent all forces exerted on each component 
are the friction force with the environment, deformation force, and adhesive force, respectively (see details in Supplementary Information). In total, half a million equations (total number of model components) had to be solved at each time step and are simulated over three weeks of real time. Due to the long simulation times with one time step of the order of 10-3 s, we have to limit it to one single part of the damage region along the CV-CV axis of two neighboring lobules sections. In addition to direct interactions by physical forces, cells can communicate via diffusive signals. For these, reaction-diffusion equations are solved.
Detailed Methods are described in the Supplementary Information.
Data availability
The data that support the findings of this study are available from the corresponding authors upon reasonable request.
Code availability
The simulation source code is accessed upon request to [email protected] upon signing a non-disclosure agreement since the final license of the current code has not been granted yet.
Acknowledgements
The authors acknowledge the support of the Core Facility for LIMA Live Cell Imaging Mannheim at Microscopy Core Facility Platform Mannheim (CFPM), Medical Faculty Mannheim, Heidelberg University and the data storage service SDS@hd supported by the Ministry of Science, Research and the Arts Baden-Württemberg and the German Research Foundation.
Funding
This study is supported by the BMBF (German Federal Ministry of Education and Research) Project LiSyM (Grant PTJ-FKZ: 031L0043 and 031L0045), LiSyM Cancer (Grant PTJ-FKZ: 031L0257A and 031L0257D), ANR iLite, ANR-16-RHUS-0005, DFG grant INST 35/1503-1 FUGG, DFG project number: 259332240 - RTG/GRK 2099, 394,046,768 - CRC/SFB 1366; 314,905,040 - CRC/SFB-TR 209; 413,262,200 - ICON/EB 187/8-1.
Competing interests
The authors declare no competing interests.
Author contributions
S.H conceived animal studies, blood analysis, and RNA isolation. P.E, S.H and Y.L performed RT-PCR, and IHC staining, as well as manual image analysis. M.W, S.W.K, P.S.K performed animal experiments in GATA4LSEC-KO mice. M.D.L and N.G performed an automated image analysis. J.Z and D.D designed and developed the DT model. J.Z implemented the code and performed simulations and visualized the results. P.V.L and A.B provided the original code of DCM and collagen model. S.H and J.Z wrote the manuscript. M.P.E, J.G.H, S.D, S.D.W, J.B and D.D performed critical revision of the manuscript. S.D, S.H and D.D acquired funding for this study. All authors read the final version of the manuscript.
- Pinzani, M., Rombouts, K. Liver fibrosis: from the bench to clinical targets. Dig. Liver Dis. 36(4), 231-242 (2004).
- Kisseleva, T., Brenner, D. Molecular and cellular mechanisms of liver fibrosis and its regression. Nat. Rev. Gastroenterol. Hepatol. 18(3), 151-166 (2021).
- Moon, H., Cho, K., Shin, S., Kim, D.Y., Han, K., Ro, S.W. High risk of hepatocellular carcinoma development in fibrotic liver: role of the Hippo-YAP/TAZ signaling pathway. International Journal of Molecular Sciences 20, 581 (2019).
- Seki, E., Brenner, D.A. Recent advancement of molecular mechanisms of liver fibrosis. J. Hepatobiliary Pancreat Sci. 22(7), 512-518 (2015).
- Seki, E., Schwabe, R.F. Hepatic inflammation and fibrosis: functional links and key pathways. Hepatology 61(3), 1066-1079 (2015).
- Higashi, T., Friedman, S.L., Hoshida, Y. Hepatic stellate cells as key target in liver fibrosis. Adv Drug Deliv. Rev. 121, 27-42 (2017).
- Ramachandran, P., Pellicoro, A., Vernon, M.A., Boulter, L., Aucott, R.L., Ali, A., Hartland, S.N., Snowdon, V.K., Cappon, A., Gordon-Walker, T.T., Williams, M.J., Dunbar, D.R., Manning, J.R., Van Rooijen, N., Fallowfield, J.A., Forbes, S.J., Iredale, J.P. Differential Ly-6C expression identifies the recruited macrophage phenotype, which orchestrates the regression of murine liver fibrosis. Proc Natl Acad Sci USA. 109(46), E3186-95 (2012).
- Krenkel, O., Tacke, F. Liver macrophages in tissue homeostasis and disease. Nature Rev. Immunol. 17, 306-321 (2017).
- Krenkel, O., Hundertmark, J., Ritz, T.P., Weiskirchen, R., Tacke, F. Single cell RNA sequencing identifies subsets of hepatic stellate cells and myofibroblasts in liver fibrosis. Cells 8(5), 503 (2019).
- Papachristoforou, E., Ramachandran, P. Macrophages as key regulators of liver health and disease. Int. Rev. Cell Mol. Biol. 368, 143-212 (2022).
- Kisseleva, T., Cong, M., Paik, Y., Scholten, D., Jiang, C., Benner, C., Iwaisako, K., Moore-Morris, T., Scott, B., Tsukamoto, H., Evans, S.M., Dillmann, W., Glass, C.K., Brenner, D.A. Myofibroblasts revert to an inactive phenotype during regression of liver fibrosis. Proc. Natl Acad Sci USA 109(24), 9448-53 (2012).
- Filliol, A., Saito, Y., Nair, A., Dapito, D.H., Yu, L., Ravichandra, A., Bhattacharjee, S., Affo, S., Fujiwara, N., Su, H., Sun, Q., Savage, T.M., Wilson-Kanamori, J.R., Caviglia, J.M., Chin, L., Chen, D., Wang, X., Caruso, S., Kang, J.K., Amin, A.D., Wallace, S., Dobie, R., Yin, D., Rodriguez-Fiallos, O.M., Yin, C., Mehal, A., Izar, B., Friedman, R.A., Wells, R.G., Pajvani, U.B., Hoshida, Y., Remotti, H.E., Arpaia, N., Zucman-Rossi, J., Karin, M., Henderson, N.C., Tabas, I., Schwabe, R.F. Opposing roles of hepatic stellate cell subpopulations in hepatocarcinogenesis. Nature 610(7931): 356-365 (2022).
- Huang, H., Chang, M., Chen, Y., Hsu, H., Chiang, C., Cheng, T., Wu, Y., Wu, M.Z., Hsu, Y., Shen, C., Lee, C., Chuang, Y., Hong, C., Jeng, Y., Chen, P., Chen, H., Lee, M. Persistent elevation of hepatocyte growth factor activator inhibitors in cholangiopathies affects liver fibrosis and differentiation. Hepatology 55(1), 161-172 (2012).
- Török, N., Fan, W., Adebowale, K., Li, Y., Rabbi, M.F., Vancza, L., Chen, D., Kunimoto, K., Mozes, G., Li, Y., Tao, J., Monga, S., Charville, G., Wells, R., Dhanasekaran, R., Kim, T., Chaudhuri, O. Extracellular matrix viscoelasticity drives liver cancer progression in pre-cirrhotic NASH. preprint (2022), available at Research Square [https://doi.org/10.21203/rs.3.rs-2087090/v1]
- Caliari, S.R., Perepelyuk, M., Cosgrove, B.D., Tsai, S.J., Lee, G.Y., Mauck, R.L., Wells, R.G., Burdick, J.A. Stiffening hydrogels for investigating the dynamics of hepatic stellate cell mechanotransduction during myofibroblast activation. Sci. Rep. 6, 21387 (2016).
- Hernandez-Gea, V., Friedman, S.L. Pathogenesis of liver fibrosis. Annu. Rev. Pathol. 6, 425-456 (2011).
- Parola, M., Pinzani, M. Liver fibrosis: pathophysiology, pathogenetic targets and clinical issues. Molecular Aspects of Medicine 65, 37-55 (2019).
- Pinzani, M., Luong, T.V. Pathogenesis of biliary fibrosis. Biochimica et Biophysica Acta - Molecular Basis of Disease 1864 (4), 1279-1283 (2018).
- Odenthal, T., Smeets, B., Van Liedekerke, P., Tijskens, E., Van Oosterwyck, H., Ramon, H. Analysis of initial cell spreading using mechanistic contact formulations for a deformable cell model. PLoS Computational Biology 9(10), e1003267 (2013).
- Kim, M., Siberberg, Y.R., Abeyaratne, R., Asada, H.H. Computational modeling of three-dimensional ECM-rigidity sensing to guide directed cell migration. Proc. Natl Acad Sci USA 115(3), E390-399 (2018).
- Winkler, B., Aranson, I.S., Ziebert, F. Confinement and substrate topography control cell migration in a 3D computational model. Communications Physics 2, 82 (2019).
- Van Liedekerke, P., Neitsch, J., Johann, T., Alessandri, K., Nassoy, P., Drasdo, D. Quantitative cell-based model predicts mechanical stress response of growing tumor spheroids over various growth conditions and cell lines. PLoS Computational Biology 15(3), e1006273 (2019).
- Van Liedekerke, P., Neitsch, J., Johann, T., Warmt, E., Gonzalez-Valverde, I., Hoehme, S., Grosser, S., Kaes, J., Drasdo, D. A quantitative high-resolution computational mechanics cell model for growing and regenerating tissues. Biomechanics and Modeling in Mechanobiology 19, 189-220 (2020).
- Van Liedekerke, P., Palm, M.M., Jagiella, N., Drasdo, D. Simulating tissue mechanics with agent-based models: concepts, perspectives and some novel results. Computational Particle Mechanics 2, 401-444 (2015).
- Ban, E., Franklin, M., Nam, S., Smith, L.R., Wang, H., Wells, R.G., Chaudhuri, O., Liphardt, J.T., Shenoy, V.B. Mechanisms of plastic deformation in collagen networks induced by cellular forces. Biophys. J. 114(2), 450-461 (2018).
- Ronceray, P., Broedersz, C.P., Lenz, M. Fiber networks amplify active stress. Proc. Natl Acad Sci USA 113(11), 2827-2832 (2016).
- Stein, A.M., Vader, D.A., Jawerth, L.M., Weitz, D.A., Sander, L.M. An algorithm for extracting the network geometry of three-dimensional collagen gels. Journal of Microscopy 232, 463-475 (2008).
- Hoehme, S., Brulport, M., Bauer, A., Bedawy, E., Schormann, W., Hermes, M., Puppe, V., Gebhardt, R., Zellmer, S., Schwarz, M., Bockamp, E., Timmel, T., Hengstler, J.G., Drasdo, D. Prediction and validation of cell alignment along microvessels as order principle to restore tissue architecture in liver regeneration. Proc. Natl Acad Sci USA 107(23), 10371-10376 (2010).
- Hoehme, S., Hammad, S., Boettger, J., Begher-Tibbe B., Bucur, P., Vibert, E., Gebhardt, R., Hengstler, J.G., Drasdo, D. Digital twin demonstrates significance of biomechanical growth control in liver regeneration after partial hepatectomy. iScience 26(1), 105714 (2023).
- Lara, J., Lopez-Labrador, F.X., Gonzalez-Candelas, F., Berenguer, M., Khudyakov, Y.E. Computational models of liver fibrosis progression for hepatitis C virus chronic infection. BMC Bioinformatics 15(8), S5 (2014).
- Dutta-Moscato, J., Solovyev, A., Mi, Q., Nishikawa, T., Soto-Gutierrez, A., Fox, I.J., Vodovotz, Y. A multiscale agent-based in silico model of liver fibrosis progression. Frontiers in Bioengineering and Biotechnology 2, 18 (2014).
- Yoshizawa, M., Sugimoto, M., Tanaka, M., Sakai, Y., Nishikawa, M. Computational simulation of liver fibrosis dynamics. Scientific Reports 12, 14112 (2022).
- Friedman, A., Hao, W. Mathematical model of liver fibrosis. Mathematical Biosciences & Engineering 14(1), 143-164 (2017).
- Adhyapok, P., Fu, X., Sluka, J.P., Clendenon, S.G., Sluka, V.D., Wang, Z., Dunn, K., Klaunig, J.E., Glazier, J.A. A computational model of liver tissue damage and repair. PLoS ONE 15(12), e0243451 (2020).
- Dichamp, J., Celliere, G., Ghallab, A., Hassan, R., Boissier, N., Hofmann, U., Reinders, J., Sezgin, S., Zuehlke, S., Hengstler, J.G., Drasdo, D. In vitro to in vivo acetaminophen hepatotoxicity extrapolation using classical schemes, pharmacodynamic models and a multiscale spatial-temporal liver twin. Front. Bioeng. Biotechnol. 11, 1049564 (2023).
- Zhao, J., Ghallab, A., Hassan, R., Dooley, S., Hengstler, J.G., Drasdo, D. A digital twin of liver predicts regeneration after drug-induced damage at the level of cell type orchestration. Accepted by iScience (2023).
- Zellmer, S., Schmidt-Heck, W., Godoy, P., Weng, H., Meyer, C., Lehmann, T., Sparna, T., Schormann, W., Hammad, S., Kreutz, C., Timer, J., Weizsacker, F., Thurmann, P.A., Merfort, I., Guthke, R., Dooley, S., Hengstler, J.G., Gebhardt, R. Transcription factors ETF, E2F, and SP-1 are involved in cytokine-independent proliferation of murine hepatocytes. Hepatology 52(6), 2127-2136 (2010).
- Hammad, S., Hoehme, S., Friebel, A., Von Recklinghausen, I., Othman, A., Begher-Tibbe, B., Reif, R., Godoy, P., Johann, T., Vartak, A., Golka, K., Bucur, O.P., Vibert, E., Marchan, R., Christ, B., Dooley, S., Meyer, C., Ilkavets, I., Dahmen, U., Dirsch, O., Bottger, J., Gebhardt, R., Drasdo, D., Hengstler, J.G. Protocols for staining of bile canalicular and sinusoidal networks of human, mouse and pig livers, three-dimensional reconstruction and quantification of tissue microarchitecture by image processing and analysis. Archives of Toxicology 88, 1161-1183 (2014).
- Hammad, S., Braeuning, A., Meyer, C., Mohamed, F., Hengstler, J.G., Dooley, S. A frequent misinterpretation in current research on liver fibrosis: the vessel in the center of CCl4-induced pseudolobules is a portal vein. Arch. Toxicol. 91(11):3689-3692 (2017).
- Lopez-De Leon, A., Rojkind, M. A simple micromethod for collagen and total protein determination in formalin-fixed paraffin-embedded sections. J. Histochem Cytochem 33(8), 737-743 (1985).
- Yang, L., Fitie, C.F.C., Van der Werf, K.O., Bennink, M.L., Dijkstra, P.J., Feijen, J. Mechanical properties of single electrospun collagen type I fibres. Biomaterials 29, 955-962 (2008).
- Ferruzzi, J., Sun, M., Gkousioudi, A., Pillar, A., Roblyer, D., Zhang, Y., Zaman, M.H. Compressive remodeling alters fluid transport properties of collagen networks - implications for tumor growth. Scientific Reports 9, 17151 (2019).
- Wen, Y. The role of immune cells in liver regeneration. Livers 3(3), 383-396 (2023).
- Madsen, D.H., Leonard, D., Masedunskas, A., Moyer, A., Jürgensen, H.J., Peters, D.E., Amornphimoltham, P., Selvaraj, A., Yamada, S.S., Brenner, D.A., Burgdorf, S., Engelholm, L.H., Behrendt, N., Holmbeck, K., Weigert, R., Bugge, T.H. M2-like macrophages are responsible for collagen degradation through a mannose receptor-mediated pathway. Journal of Cell Biology 202(6), 951-966 (2013).
- Winkler, M., Staniczek, T., Kuerschner, S.W., Schmid, C.D., Schoenhaber, H., Cordero, J., Kessler, L., Mathes, A., Sticht, C., Nessling, M., Uvarovskii, A., Anders, A., Zhang, X., Figura, G., Hartmann, D., Mogler, C., Dobreva, F., Schledzweski, K., Geraud, C., Koch, P., Goerdt, S. Endothelial GATA4 controls liver fibrosis and regeneration by preventing a pathogenic switch in angiocrine signaling. Journal of Hepaotycte 74, 380-393 (2021).
- Rocha, A.S., Vidal, V., Mertz, M., Kendall, T.J., Charlet, A., Okamoto, H., Schedl, A. The angiocrine factor Rspondin3 is a key determinant of liver zonation. Cell Reports 13(9), P1757-1764 (2015).
- Seo, Y.S., Kwon, J.H., Yaqoob, U., Yang, L., De Assuncao, T.M., Simonetto, D.A., Verma, V.K., Shah, V.H. HMGB1 recruits hepatic stellate cells and liver endothelial cells to sites of ethanol-induced parenchymal cell injury. Am. J. Physiol. Gastrointest. Liver Physiol. 305, G838-G848 (2013).
- Roohani, S., Tacke, F. Liver injury and the macrophage issue: molecular and mechanistic facts and their clinical relevance. Int. J. Mol. Sci. 22(14), 7249 (2021).
- Ghafoory, S., Breitkopf-Heinlein, K., Li, Q., Scholl, C., Dooley, S., Woelfl, S. Zonation of nitrogen and glucose metabolism gene expression upon acute liver damage in mouse. PLoS One 8(10), e78262 (2013).
- Ghallab, A., Myllys, M., Holland, C.H., Zaza, A., Murad, W., Hassan, R., Ahmed, Y.A., Abbas, T., Abdelrahim, E.A., Schneider, K.M., Matz-Soja, M., Reinders, J., Gebhardt, R., Berres, M., Hatting, M., Drasdo, D., Saez-Rodriguez, J., Trautwein, C., Hengstler, J.G. Influence of liver fibrosis on lobular zonation. Cells 8(12), 1556 (2019).
- Teutsch, H.F. The modular microarchitecture of human liver. Hepatology 42(2), 317-325 (2005).
- Wu, L., Zhang, Q., Mo, W., Feng, J., Li, S., Li, J., Liu, T., Xu, S., Wang, W., Lu, X., Yu, Q., Chen, K., Xia, Y., Lu, J., Xu, L., Zhou, Y., Fan, X., Guo, C. Quercetin prevents hepatic fibrosis by inhibiting hepatic stellate cell activation and reducing autophagy via the TGF-β1/Smads and PI3K/Akt pathways. Sci. Rep. 7(1), 9289 (2017).
- Drasdo, D., Zhao, J. An integrative experimental and computational twin modeling approach to understand the clonal dynamics in normal liver. Journal of Hepatology 79(2), 273-276 (2023).
- Duffield, J.S., Forbes, S.J., Constandinou, C.M., Clay, S., Partolina, M., Vuthoori, S., Wu, S., lang, R., Iredale, J.P. Selective depletion of macrophages reveals distinct, opposing roles during liver injury and repair. J. Clin. Invest. 115(1), 56-65 (2005).
- Pradere, J., Kluwe, J., De Minicis, S., J, J.J., Gwak, G.Y., Dapito, D.H., Jang, M.K., Guenther, N.D., Mederacke, I., Friedman, R., Dragomir, A., Alomen, C., Schwabe, R.F. Hepatic macrophages but not dendritic cells contribute to liver fibrosis by promoting the survival of activated hepatic stellate cells in mice. Hepatology 58(4), 1461-1473 (2013).
There is NO Competing Interest.
- SupplementaryMovie1.avi
Supplementary Movie 1
- SupplementaryMovie2.avi
Supplementary Movie 2
- SupplementaryMovie3.avi
Supplementary Movie 3
- SupplementaryMovie4.avi
Supplementary Movie 4
- SupplementaryMovie5.avi
Supplementary Movie 5
- SupplementaryMovie6.avi
Supplementary Movie 6
- SupplementaryMovie7.avi
Supplementary Movie 7
- SupplementaryMovie8.avi
Supplementary Movie 8
- SupplementaryMovie9.avi
Supplementary Movie 9
- SupplementaryMovie10.avi
Supplementary Movie 10
- SupplementaryInformation.docx
Supplementary Information