How predictive QUANTITATIVE MODELING OF

0 downloads 0 Views 795KB Size Report
Jun 9, 2014 - OF TISSUE ORGANIZATION can inform liver disease pathogenesis, ... This is a PDF file of an unedited manuscript that has been accepted for ..... the closure of the necrotic lesion and the lobule architecture (red curves in Fig.
Accepted Manuscript Clinical Application of Basic Science How predictive QUANTITATIVE MODELING OF TISSUE ORGANIZATION can inform liver disease pathogenesis Dirk Drasdo, Stefan Hoehme, Jan G. Hengstler PII: DOI: Reference:

S0168-8278(14)00409-7 http://dx.doi.org/10.1016/j.jhep.2014.06.013 JHEPAT 5214

To appear in:

Journal of Hepatology

Received Date: Revised Date: Accepted Date:

8 April 2014 9 June 2014 10 June 2014

Please cite this article as: Drasdo, D., Hoehme, S., Hengstler, J.G., How predictive QUANTITATIVE MODELING OF TISSUE ORGANIZATION can inform liver disease pathogenesis, Journal of Hepatology (2014), doi: http:// dx.doi.org/10.1016/j.jhep.2014.06.013

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

How predictive QUANTITATIVE MODELING OF TISSUE ORGANIZATION can inform liver disease pathogenesis Dirk Drasdo1,2,# , Stefan Hoehme2, Jan G. Hengstler3 (1) INRIA Paris-Rocquencourt, Le Chesnay, France (2) Interdisciplinary Center for Bioinformatics, University of Leipzig, Leipzig, Germany (3) Leibniz Research Centre for Working Environment and Human Factors (IfADo), Dortmund, Germany (#) Corresponding

Summary: From the more than 100 liver diseases described, many of those with high incidence rates manifest themselves by histopathological changes, such as hepatitis, alcoholic liver disease, fatty liver disease, fibrosis, and, in its later stages, cirrhosis, hepatocellular carcinoma, primary biliary cirrhosis and other disorders. Studies of disease pathogeneses are largely based on integrating -omics data pooled from cells at different locations with spatial information from stained liver structures in animal models. Even though this has led to significant insights, the complexity of interactions as well as the involvement of processes at many different time and length scales constrains the possibility to condense disease processes in illustrations, schemes and tables. The combination of modern imaging modalities with image processing and analysis, and mathematical models opens up a promising new approach towards a quantitative understanding of pathologies and of disease processes. This strategy is discussed for two examples, ammonia metabolism after druginduced acute liver damage, and the recovery of liver mass as well as architecture during the subsequent regeneration process. This interdisciplinary approach permits integration of biological mechanisms and models of processes contributing to disease progression at various scales into mathematical models. These can be used to perform in-silico simulations to promote unraveling the relation between architecture and function as below illustrated for liver regeneration, and bridging from the in-vitro situation and animal models to human. In the near future novel mechanisms will usually not be directly elucidated by modeling. However, models will falsify hypotheses and guide towards the most informative experimental design.

Introduction (Fig. 1) While recent developments have significantly increased our capability to collect information at multiple spatial and temporal scales, research in disease pathogenesis is largely hampered by the difficulty to orchestrate the individual components and mechanisms inferred by traditional ways of analysis into a consistent picture and to infer a complex interplay of 1

components by pure reasoning. Here, mathematical models can play an important role as they formalize relations between components, quantify components and mechanisms, and test their interplay in a virtual setting defined by the modeler thus avoiding possible perturbations by unknown influences that can rarely be excluded in a real biological system. Mathematical models addressing liver pathology or drug effects increasingly integrate different levels of organization [1]. Extra-hepatic contributions are usually addressed by compartment models [2,3], material transport (blood, lymph, bile) in liver or individual lobules by perfusion models considering local averages of concentrations, volume fractions and flow speed [4,5,6,7], Poiseuille-like flow [8] or spatial compartment models [9,10]. Hepatocytes, stellate cells, sinusoidal endothelial cells or other cell types may be included as cell compartments or as individual entities in space [8,11,12,13]. Metabolism, signal transduction or gene expression is usually modeled by systems of ordinary differential equations [9,14,15]. Model parameters, components or mechanisms can readily be modified, suppressed or added and the impact of such changes can be studied on different system observables. However, mathematical models remain abstractions of their biological counterpart. The aim is not an ‘in-silico duplicate’ of the ‘real biological system’ as this would have the same complexity as the original system. The choice of model components, horizontally on the same scale as vertically spanning several scales, should be guided by a scientific question. Ideally all model parameters would be measured simultaneously but we are currently far from such an ideal situation. Therefore, the model parameters shall represent measurable quantities (MQ) with a (known) direct physical or biological meaning and interpretation as this largely facilitates estimation of their range. Heuristic parameters should be avoided, as their values are difficult to estimate. Each parameter that is not experimentally quantified introduces a degree of freedom that has to be explored by simulation, thereby increasing the search space of the model. For this reason, it is useful to construct a minimal model parameterized by MQ compatible with published knowledge. As many liver diseases leave a signature in the composition and architecture of liver tissue in which case pathophysiology and histopathology are inherently linked the focus here is on quantitative (as opposed to qualitative) models involving liver mass and architecture. Fig. 1 sketches how a process chain consisting of experiment, data analysis and mathematical modeling can be iteratively applied to promote our understanding of liver disease pathogenesis. In the next two sections we illustrate the workflow along two examples, the regeneration of liver mass and architecture after CCl4 – induced damage addressing the cell and tissue scale [12], and ammonia detoxification after CCl4-induced damage addressing ammonia metabolism [10]. As our examples show, mechanisms can often only be ruled out if the model is quantitative.

2

Figure 1: Workflow from scientific question to possible answer. Once the scientific question has been defined (A), hypothetical alternative mechanisms which could underlie the disease pathogenesis should be formulated. (B) In a next step, data has to be collected to inform the model quantitatively about initial conditions (the starting state), boundary conditions (values at the border of spatial domains, in/outflows etc.), and the condition(s) at which the model simulation should stop. Histopathology is characterized by the tissue composition (e.g. the cell types and extracellular matrix) and their architecture. It can be quantitatively described by a set {P1} of composition and architectural parameters (D1) [12]. Such parameters can be inferred from a data analysis pipeline of imaging, image processing and analysis (C, [18]), serving as a starting state for the model (E). Both, data of a concrete individual tissue specimen or representative data obtained from the statistical distribution functions over the (architectural) parameters of many individuals can be used. Quantitative information about the disease process can be obtained from series of images in an equivalent way and leads to a second set of parameters {P2} (D2). This parameter set must be explained by the model. (E) The mathematical model should represent the hypothetical alternative mechanisms, and permit testing them in-silico one by one. It introduces a third parameter set {P3}. If even after calibration of each model parameter within its range the model simulations do not quantitatively capture the biologically observed behavior, either important structures, mechanisms or processes that are required to correctly capture the specific in-vivo situation are likely to be missing. In such situations the model needs to be adjusted (F) until finally agreement between model simulation results and experimental observations is achieved. It may occur that different mathematical models, each basing on another hypothetical mechanism, explain the same data quantitatively. In this case it is possible to use the mathematical model to search for an experimental situation in which the different hypothesis would predict different outcomes (G). Such an ‘informative’ experiment permits to select the correct out of several principally possible explanations.

A model on cell and tissue scale: Identification of key mechanisms of regeneration after drug-induced acute damage (Fig. 2) 3

Toxic doses of acetaminophen (APAP) can induce necrosis in the center of the liver lobules [16] (Fig. 2A). A similar pericentral lesion is caused by the hepatotoxic model compound CCl 4. The extent of damage can amount to 30-40% of the total liver volume. In only approximately ten days livers of rodents not only regenerate their original cell numbers but also restore their functional tissue microarchitecture. This includes the process of reorganizing the complex sinusoidal structure with endothelial cells and sheets of hepatocytes. The necessity of coordinating principles is obvious, since the starting situation of the regeneration process is a seemingly ‘chaotic’ dead cell mass. Much research has been done to understand the mechanisms controlling hepatocyte and non-parenchymal cell proliferation during regeneration [16,17]. However, only little is known how cells coordinately act to restore functional tissue microarchitecture. This question is of practical relevance, since the responsible ‘coordinating mechanism’ seems to be efficient under conditions of acute liver damage where a full regeneration is possible. However, it seems to fail under conditions of chronic exposure or inflammation which often lead to irreversibly disturbed tissue architecture and function. The question of the ‘coordinating principle’ that controls restoration of tissue architecture was addressed by a spatio-temporal modeling approach [12]. Possible hypothetical mechanisms (H) that could either alone or in combination explain the efficient restoration of liver architecture were: (H1) hepatocytes could move individually into the necrotic lesion, and close it by cell division, (H2) the border between healthy hepatocytes and central necrosis could be pushed by proliferation pressure towards the necrotic lesion following a mechanical gradient, (H3) hepatocytes could migrate actively, directed towards the necrotic lesion, (H4) cell division could be randomly undirected, or (H5) directed along a portal-central molecular gradient, or (H6) directed along the closest sinusoid. To determine the starting state a spatial-temporal model (STM) of the 3D liver lobule architecture was reconstructed from confocal laser scanning micrographs (Fig. 2B) by a specifically designed image processing and analysis chain ([12], [18], Fig. 2C, supplement). By this technique the static architectural liver lobule parameters such as the size and shape of hepatocytes, diameter, density and branching parameters of sinusoids, the size and variability of liver lobules, etc., were defined. Next, the process parameters describing the damage and regeneration process were determined. This includes the time-resolved quantification of the numbers of hepatocytes, size of lesion, BrdU incorporation and architectural parameters (Fig. 2, caption).

4

Figure 2: Application of the work flow described in Fig. 1 to the example of liver regeneration after CCl4-induced damage (from: [12]). Based on the question (gray, A) how the regeneration of liver mass and architecture could be explained, experiments (orange, B) were performed to parameterize the tissue architecture obtaining parameters {P1} (green, C, volume visualization) and the regeneration process obtaining parameters {P2} (green, C, plots). Confocal laser scanning micrographs of relatively thick (100 µM) tissue blocks were used to reconstruct the lobule architecture as physical cutting necessary for bright field micrographs rendered registration of sinusoids impossible. To establish the model each individual hepatocyte was represented. To inform the model about the regeneration process, the following process parameters {P2} were experimentally determined and included into the model: (i) the number of hepatocytes per mm2 within a lobule to represent liver mass, (ii) the size of the necrotic lesion, (iii) the fraction of interface between hepatocytes and sinusoids as a measure of the ‘functional architecture’ as it informs about the exchange area for metabolites between blood and hepatocytes, and (iv) the fraction and spatial-temporal distribution of BrdU-stained nuclei as a measure of S-phase entry. Parameter (iv) served as a model input parameter, (i-iii) to evaluate the model. Experimental quantification of the process parameters showed that the number of hepatocytes per lobule had a minimum at day two, the necrotic lesion area a maximum at day one, and the fraction of the hepatocyte surface not in contact with another hepatocyte a minimum at day four after CCl4 damage. The recovery of architecture followed that of mass with a delay of approximately one week. The final model was developed in three iterations. The first model M1 assumed: (a) a hepatocyte in isolation can be approximated as a homogeneous, isotropic elastic sphere, cohere to another hepatocyte and (in principle) adhere to sinusoids. (b) Micro-motility is undirected and random. (c) A moving hepatocyte experiences friction with the extra-cellular matrix in the space of Disse and with sinusoids. (d) In the cell cycle, a 5

hepatocyte doubles its volume and then splits into two hepatocyte daughter cells of equal volume. (e) Sinusoids were mimicked by a chain of spheres linked by elastic springs, and anchored in the central vein and the portal field. Hepatocyte movement was calculated solving an equation of motion for each hepatocyte summing contributions from the cells’ micromotiliy and all physical forces exerted on the cell by its environment. For each sphere belonging to a sinusoid a similar equation of motion was solved omitting micromotiliy. As the model was completely parameterized by measurable parameters {P3}, the range of each model parameter could be estimated. Both, tissue architecture {P1} and process {P3} parameters were used to construct an agent-based model (violet, D) for the regenerating liver lobule. The first iteration showed that proliferation pressure (illustrated as a compressed spring in (F1)) is not sufficient to close the lesion and restore architecture (red curves in (E)). Based on this disagreement, the model was adjusted in two steps (light blue, F1-F3). Firstly, in M2 the random undirected micro-motility was replaced by directed migration into the necrotic zone leading to a relaxation of the proliferation pressure (illustrated in (F2) by a stretched spring). As a consequence the lesion could be closed in time but the architecture could still not be restored (blue lines in (E)). Only when the random orientation of cell division (indicated by the two red cells not aligned along the chain of many small red circles representing the sinusoid in the scheme of F2) was replaced by (M3) cell division oriented along the closest sinusoid (represented by two red cells lining along the sinusoid in (F3)) all process parameters could be correctly explained (green curves). The model further correctly predicted the experimentally determined angle distribution of the daughter cells emerging from the same precursor hepatocyte (red, G).

As initial model configuration either a concrete 3D reconstructed lobule was chosen, or a representative liver lobule was established by statistical analysis of the static liver lobule parameters (supplement). The assumptions of this model, named model 1 (M1), are summarized in Fig. 2 (caption) and detailed in the supplement. Hundreds of simulations varying each model parameter within its pre-determined physiological range were performed. The best simulations with M1 including the experimentally determined proliferation rate could explain the number of hepatocytes in the lobule but failed to explain the closure of the necrotic lesion and the lobule architecture (red curves in Fig. 2E). Therefore, the proliferation pressure alone was insufficient to close the lesion eliminating mechanism H2 and requiring model adjustment. Motivated by the observation that hepatocytes at the border to the necrotic lesion stretched lamellipodia into the lesion, in M2 the random micro-motility was replaced by a directed active migration of hepatocytes at the border of the lesion into the necrotic zone (Fig. 2F2, details in supplement). The 2nd model could explain the closure of the lesion but failed to explain the regeneration of liver architecture, hence H3 was also insufficient. For further adjustment it was hypothesized that daughter hepatocytes after cell division undergo a process of ‘hepatocyte-sinusoid alignment’ (HSA) meaning that an axis through the center of the daughter cells is oriented in parallel to the direction of the closest sinusoid (HSA, H5, M3, Fig. 2F3, details in supplement). One of the observations motivating this hypothesis was that almost all hepatocytes were killed in the peri-central region after CCl4 but a remarkable fraction of the sinusoidal endothelial cells survived. 6

Only after inclusion of HSA into the model all process parameters could be correctly explained. This model prediction was experimentally validated by confocal laser scanning micrographs co-stained with BrdU to label daughter cells and ICAM/DPPIV to identify the sinusoids. The same experiments did not support an orientation of cell division along a portal-central morphogen gradient because the sinusoids did not precisely orient in radial direction, ruling out H6. Therefore, the modelling studies have clearly shown that liver sinusoidal endothelial cells (LSEC) spatio-temporally orchestrate liver regeneration [11,12]. The precise molecular mechanisms of this process have later been elucidated by other groups [19,20,21]. LSEC deploy paracrine trophogens to allow liver regeneration [19]. Of particular relevance seems to be the release of the angiocrine factors Wnt2 and HGF from LSEC that have been reported to induce hepatocyte proliferation [21]. Moreover, the expression of Angiopoietin-2 (Ang2) in LSECs is dynamically regulated during liver regeneration. Ang2 down-regulation leads to reduced TGF-β1 release from LSEC, enabling hepatocyte proliferation [20]. In conclusion, LSECs represent key regulators of liver regeneration that not only control proliferation but also serve as ‘guide rails’ for hepatocytes controlling their orientation and architecture in the liver lobule.

An integrated model of ammonia detoxification after CCl4-induced damage (Fig. 3) The metabolic pathways of ammonia detoxification in the liver have been intensively studied [22,23]. In the periportal compartment of the liver lobule ammonia is detoxified by urea cycle enzymes. Moreover, a ring of pericentral hepatocytes metabolizes ammonia by glutamine synthetase. It is well established that ammonia blood concentrations can increase upon induction of liver damage. However, only little is known about ammonia metabolism in damaged livers. Does the extent of decreased metabolism correspond to the fraction of destroyed tissue (H1)? Or are additional metabolic pathways active that do not occur in the healthy liver (H2)? To come closer to an answer a metabolic model (MM) was established by expressing the reactions involved in ammonia detoxification by a set of differential equations to calculate the concentrations of ammonia and its metabolites in the liver vein (‘outflow’) for a given concentration in the portal vein and artery (‘inflow’) (Fig. 3D summarizes the model assumptions, details see supplement; [10]). From mass balance the volumes of the periportal and pericentral lobule compartments enter the MM as reaction volumes. Consequently, destruction of liver tissue leads to a reduction of the volume of the compartments and consequently to reduced capability of periportal and/or pericentral ammonia detoxification. Since the spatio-temporal model described in the previous paragraph [12] simulates the time course of lobule destruction after acute CCl4 toxicity (Fig. 3B1), it was used as a basis and linked to the metabolic model (Fig. 3D,E). The parameters of this model were calibrated with results from experiments, where mouse liver was perfused ante-and retrograde with buffers containing different ammonia concentrations (Fig. 3B2). 7

The agreement was quantitative. In a next step, the so calibrated MM was used to predict the temporal ammonia and glutamine concentrations in the hepatic vein of mice after CCl4induced damage where the experimentally determined concentrations in the portal vein and right heart chamber (representing the portal artery) were used as inflow.

Figure 3: Application of the work flow described in Fig. 1 to an integrated model of ammonia detoxification after CCl4-induced pericentral necrosis (gray, A). In experiment 1 (orange, B) the size of the glutamine synthetase (GS) labeled zone around the central vein and the size of the necrotic lesion after CCl4-induced damage were determined time-dependently. This information served to construct a spatial-temporal tissue model (STM) (violet, D-Right, see also Fig. 2) capable of simulating the pericentral, GS-positive and the periportal GS-negative volumes. Experiment 2 represents ex-vivo antegrade and retrograde perfusion experiments with different concentrations of ammonia to determine the concentrations of ammonia, glutamine and urea in the outflow (orange B). This information has been used to calibrate a kinetic metabolic model (MM) representing the key reactions of ammonia detoxification [24] (violet, D-Left). These are the glutaminase reaction converting glutamine into ammonia and glutamate, and carbamoyl phosphate synthetase fundamentally converting ammonia into urea, both in the periportal compartment, and the glutamine-synthetase catalyzed reaction of ammonia and glutamate into glutamine, in the pericentral compartment. Briefly, the key reactions of ammonia detoxification (Fig. 3, [22,23]) were translated into a metabolic model (MM) for healthy liver comprising of an ordinary differential equation for each glutamine, urea and ammonia in the periportal and the pericentral compartment. In a final step the 8

volumes in the MM after CCl4 administration were calculated from the STM arriving at an integrated metabolic model (IMM) (violet, D). Simulations with this model were compared with a new set of experiments (blue, E) on the concentrations of ammonia, glutamine and urea in the right heart chamber, portal vein and hepatic vein of mice. Quantitative agreement could only be achieved after adding a hypothetical ammonia consuming reaction (light blue, F) which next requires validation (red G). (from [10], details see supplement)

While in healthy liver the pericentral and -portal volumes remain stationary, after CCl4induced damage they change over time. These volumes were calculated from the spatial temporal model (STM) introduced in Fig. 2, and inserted into the MM arriving at an integrated metabolic model (IMM). The IMM consists of two parts, a kinetic equation model with time-dependent volumes yielding the outflow glutamine, ammonia and urea concentrations in the hepatic vein (Fig. 3D-Left), and a spatial temporal visualization of the concentration in the liver lobule (Fig. 3D-Right). It could be shown that despite qualitative agreement no quantitative agreement between IMM and experiments after CCl4-induced damage could be achieved, leading to the conclusion that the reaction scheme is incomplete. The IMM further predicted quantitative agreement if an additional ammonia sink was introduced. Very recent findings triggered by the model prediction identified a candidate for such a reaction that is currently validated [25] and that may prospectively be used to reduce ammonia concentrations in hyperammonemia. The example illustrates that modeling can indicate a missing element in ‘state-of-the art’ concepts and assist establishment of the correct mechanisms.

Conclusion In the near future, we should not expect that models will directly identify novel mechanisms. However, models can be utilized to falsify hypotheses and to guide towards the most informative experiment. It has proven useful and efficient in physics that validation experiments are not necessarily performed by the same consortium that generated the model prediction. Validation experiments may require a different expertise than needed to generate the model. A limitation may be the time required for the iterative cycles of modeling and experimental validation. To speed up these iterations, it is important that workflow, imaging, image processing and analysis [18] as well as modeling tools are being developed in a standardized way well geared to each other. In future, the mathematical models will become increasingly mechanistic, integrating biological, chemical and physical processes at multiple spatial and temporal scales. They will also become more user-friendly so that at least some basic versions can be directly applied by experimentalists.

Funding by Virtual Liver Network, EU-Notox, ANR-IFLOW, EU Passport and Cancersys is gratefully acknowledged. 9

References [1] Holzhuetter H G, Drasdo D, Preusser T, Lipper J, Henney A M. The virtual liver: a multidisciplinary, multilevel challenge for systems biology. Wiley Interdiscip Rev Syst Biol Med. 2012; MayJun;4(3):221-35 [2] Remien C H, Sussmann N L, Adler F R. Mathematical modelling of chronic acetaminophen metabolism and liver injury. Math. Med. Biol. 2013; Jun 20 [3] Ochoa J G D, Bucher J, Péry A R R, Zaldivar Comenges J M, Niklas J, Mauch K. A multi-scale modeling framework for individualized, spatiotemporal prediction of drug effects and toxicological risk. Frontiers in Pharmacol. 2013; 3:204 [4] Ricken T, Dahmen U, Dirsch O. A biphasic model for sinusoidal liver perfusion remodeling after outflow obstruction. Biomech Model Mechanobiol. 2010; 9:435-450 [5] Siggers J H, Leungchavaphongse K, Ho C H, Repetto R. Mathematical model of blood and interstitial flow and lymph production in the liver. Biomech Model Mechanobiol 2014; 13:2, 363-378 [6] Debbaut C, Vierendeels J, Casteleyn C, Cornillie P, Van Loo D, Simoens P, et al. Perfusion characteristics of the human hepatic microcirculation based on 3D reconstructions and computational fluid dynamic analysis. Journal of Biomechanical Engineering 2014; Accepted. [7] Schwen L O, Krauss M, Niederalt C, Gremse F, Kiessling F, Schenk A, et al. Spatio-Temporal Simulation of First Pass Drug Perfusion in the Liver. PLOS Comput. Biol. 2014; 10(3) [8] Wambaugh J, Shah I. Simulating Microdosimetry in a Virtual Hepatic Lobule. PLOS Comput. Biol. 2010; 6(4) [9] Ohno H, Naito Y, Nakajima H, Tomita M. Construction of a Biological Tissue Model Based on a Single-Cell Model: A Computer Simulation of Metabolic Heterogeneity in the Liver Lobule. Artificial Life 2008; 14,3-28 [10] Schliess F, Hoehme S, Henkel S, Ghallab A, Driesch D, Böttger J, et al. Integrated metabolic spatial-temporal model for the prediction of ammonia detoxification during liver damage and regeneration. Hepatology 2014; (Accepted) [11] Hoehme S, Hengstler J G, Brulport M, Schäfer M, Bauer A, Gebhardt R et al. Mathematical modelling of liver regeneration after intoxication with CCl4. Chemico-Biological Interaction 2007; 168, 74-93 [12] Hoehme S, Brulport M, Bauer A, Bedawy E, Schormann W, Gebhardt R, et al. Prediction and validation of cell alignment along microvessels as order principle to restore tissue architecture in liver regeneration. PNAS 2010; 107(23), 10371-10376. [13] Bhattacharya S, Shoda L K M, Zhang Q, Woods C G, Howell B A, Siler S Q, et al. Modeling drugand chemical-induced hepatotoxicity with systems biology approaches. Frontiers in Physiology 2012; 3(462) 10

[14] Chalhoub E, Hanson R W, Belovich J M. A computer model of gluconeogenesis and lipid metabolism in the perfused liver. Am J Physiol Endocrinol Metab 2007; 293(1676-1686) [15] Lerapetritou M G, Georgopoulos P G, Roth C M, Androulakis L P. Tissue-Level Modeling of Xenobiotic Metabolism in Liver: An Emerging Tool for Enabling Clinical Translational Research. Clin Transl Sci. 2009; Jun;2(3):228-37. [16] Godoy P, Hewitt N J, Albrecht U, Anderson M E, Ansari N, Bhattacharya S, et al., Recent advances in 2D and 3D in vitro systems using primary hepatocytes, alternative hepatocyte sources and nonparenchymal liver cells and their use in investigating mechanisms of hepatotoxicity, cell signaling and ADME. Arch Toxicol. 2013; Aug 87(8):1315-530 [17] Michalopoulos G K. . Principles of liver regeneration and growth homeostasis. Compr Physiol. 2013; Jan;3(1):485-513. [18] Hammad, S., Hoehme, S., Friebel, A., von Recklinghausen, I., Othman, A., Begher-Tibbe, B., et al.. 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. Arch. Toxicol., 2014 Vol. 88 (5), pp 1161-1183. [19] Ding B S, Cao Z, Lis R, Nolan D J, Guo P, Simons M, et al. Divergent angiocrine signals from vascular niche balance liver regeneration and fibrosis. Nature 2014; Jan 2;505(7481):97-102. [20] Hu J, Srivastava K, Wieland M, Runge A, Mogler C, Besemfelder E, et al. Endothelial cell-derived angiopoietin-2 controls liver regeneration as a spatiotemporal rheostat. Science 2014; 24;343(6169):416-9. [21] Ding B S, Nolan D J, Butler J M, James D, Babazadeh A O, Rosenwaks Z, et al. Inductive angiocrine signals from sinusoidal endothelium are required for liver regeneration. Nature 2010; Nov 11;468(7321):310-5. [22] Haeussinger D. Hepatocyte heterogeneity in glutamine and ammonia metabolism and the role of an intercellular glutamine cycle during ureogenesis in perfused rat-liver. Eur J Biochem 1983;133:269-275. [23] Gebhardt R, Mecke D. Heterogeneous distribution of glutamine-synthetase among rat-liver parenchymal-cells insitu and in primary culture. EMBO J 1983;2:567-570 [24] Haeussinger D, Stoll B, Stehle T, Gerok W. Hepatocyte heterogeneity in glutamate metabolism and bidirectional transport in perfused rat liver. Eur J Biochem. 1989; Oct 20;185(1):189-95. [25] Ghallab A, et al. Integrated spatial-temporal and metabolic modelling identifies the necessity of alpha-ketoglutarate substitution upon acute liver damage. 2014, In preparation.

11