The establishment of the hepatic architecture is a prerequisite for the development of a lobular pattern of gene expression

Hepatocytes of the adult mammalian liver show a remarkable heterogeneity in their enzyme profile, which is related to the microvascular architecture of the hepatic tissue. Hepatocytes surrounding the terminal branches of the portal (afferent) vein have an enzymic phenotype that differs from that of hepato-cytes around the smallest branches of the (efferent) hepatic vein. This zonal heterogeneity, which is thought to be necessary for the homeostatic function of the organ, has been demonstrated for enzymes involved in ammonia and carbohy-drate metabolism (reviewed by Jungermann and Katz, 1989; Meijer et al., 1990; Gebhardt, 1992). The expression patterns of ornithine-cycle enzymes and glutamine synthetase (GS [EC 6.3.1.2]) provide the most striking examples of a zonal distri-bution of enzymes in the liver. Carbamoylphosphate synthetase (CPS) I (EC 6.3.4.16), the enzyme with the highest control coefficient of the ornithine cycle (Meijer et al., 1990), is found in a large periportal compartment (Gaasbeek Janzen et al., 1984), whereas GS is found only in a small pericentral com-partment of one to three hepatocytes (Gebhardt and Mecke, 1983; Bennett et al., 1987; Gaasbeek Janzen et al., 1987; Smith and Campbell, 1988). A similar heterogeneous distribution pattern has since been reported for their respective mRNAs (Kuo et al., 1988; Moorman et al., 1988; Smith and Campbell, 1988) which shows that this intercellular compartmentation of gene expression is regulated at the pretranslational level.

The exceptional stability of this distribution pattern in vivo and in vitro has tempted many investigators to speculate whether the emergence of this pattern during development is the consequence of an irreversible differentiative step resulting in two separate (sub-)lineages of hepatocytes (Gebhardt and Mecke, 1983; Bennett et al., 1987; for review see Gebhardt, 1992), or whether it results from a reversible but inflexible dif-ference in the functional state of hepatocytes within a single lineage (reviewed by Häussinger et al., 1992). However, the observations that coculture of periportal hepatocytes with a liver epithelial cell line leads to the accumulation of GS protein in these previously GS-negative hepatocytes (Schrode et al., 1990) and that CPS and GS can be coexpressed in the same hepatocytes of postnatal liver (de Groot et al., 1987; Schöls et al., 1990; Gebhardt et al., 1991) seem to rule out the differen-tiation hypothesis.

Further support for this assumption might be obtained from transplantation studies (Gebhardt et al., 1989; Lamers et al., 1990) in which it was demonstrated that individually settled hepatocytes adapt to their ectopic environment with the selective expression of either the GS (pericentral) or the CPS (periportal) enzymic phenotype, irrespective of their lobular origin. Thus, individual hepatocytes accumulate GS rather than CPS protein when transplanted to the interscapular fat pad, whereas they accumulate CPS protein when transplanted to the spleen. However, after becoming embedded in a lobule-like structure, intrasplenically transplanted hepatocytes were found to develop the same positional polarity for the GS enzymic phenotype as observed in the adult liver in situ (Lamers et al., 1990). This observation shows the dependence of the pos-itional expression of GS on a well-developed hepatic architec-ture. A similar conclusion was reached by Lamers et al. (1987) in a comparative study on the developmental appearance of ammonia-metabolising enzymes in altricial and precocial rodents. Together, these observations prompted us to consider a role for the hepatic architecture in the specification of the enzymic phenotype of hepatocytes.

To test this hypothesis, we have transplanted single-cell sus-pensions of embryonic day (E) 14 rat hepatocytes, in which the zonal expression pattern of both enzymes is not yet established (Gaasbeek Janzen et al., 1987, 1988; Moorman et al., 1990b), into the spleen. Intrasplenic transplantation of isolated hepato-cytes results in the ectopic formation of a lobule-like structure with the morphological and functional characteristics of liver tissue in situ (Mito et al., 1979; Kusano and Mito, 1982; Lamers et al., 1990). Not all transplanted hepatocytes take part in this neo- organogenesis, a fraction remaining present as dispersed cells. Thus, this method allows one to compare simultaneously the phenotypic fate of individually settled cells with those that develop a ‘lobular’ structure. By using a com-bination of immunohistochemistry and in situ hybridisation we demonstrate that embryonic rat hepatocytes when transplanted to the spleen acquire their terminally differentiated phenotype only in a hepatic context, whereas the individually settled hepatocytes retain a protodifferentiated phenotype. A zonal pattern of gene expression only develops if hepatocytes become embedded in a lobular structure.

Adult inbred Wistar rats (WAG/RijCpbHsd) were purchased from Harlan CPB (Zeist, The Netherlands) and kept in a controlled 16 hours light -8 hours dark cycle. The rats were fed a standard diet (AM-II diet; Hope Farms, Woerden, The Netherlands) and water ad libitum. Rat embryos at 14 days of gestation and adult female rats (weighing 200-250 g) were used as hepatocyte donors and recipients, respec-tively. Embryonic age was calculated from dated matings taking the day of conception as day 0.

Animal welfare was in accordance with the institutional guidelines of the University of Amsterdam.

Single-cell suspensions were prepared from embryonic livers as described (Lamers et al., 1984), except that DMEM-F12 culture medium with 15 mM Hepes (Gibco BRL, Paisley, UK; pH 7.4) was used. Briefly, after incubation in DMEM-F12 containing 1 mM EGTA and 0.25% (v/v) trypsin (Gibco BRL) for 15 minutes at 37°C, livers were mechan-ically dispersed by aspiration into a wide-bored Pasteur pipette. The cells were washed twice with DMEM-F12 supplemented with 10% (v/v) fetal calf serum (Gibco BRL) and 0.01% (w/v) DNAse I (Boehringer, Mannheim, Germany) by centrifugation (45 g) for 5 minutes at 5°C. Finally, the cells were resuspended in cold DMEM-F12 and kept on ice until injection. Liver cell transplantation was carried out under diethylether anaesthesia by direct injection of 250μl cell suspension with a 25-gauge needle into the inferior part (apex) of the spleen.

At 2, 5, 9, 14 and 21 days after transplantation four recipient rats were killed by decapitation. Their spleen was rapidly removed, fixed for in situ hybridisation (Moorman et al., 1993) and albumin (ALB) immunohistochemistry (Guillouzo et al., 1982) in 4% (w/v) freshly prepared formaldehyde by immersion for 4 hours at room tempera-ture, or, for all other immunohistochemistry, in a mixture of methanol, acetone and bidistilled water (2:2:1 [v/v]) overnight at 4°C (Gaasbeek Janzen et al., 1984). Processing of tissue for in situ hybridisation and immunohistochemistry was performed as described before (Gaasbeek Janzen et al., 1984; Moorman et al., 1993). Examination of spleens for the presence of intrasplenic hepatocytes was performed on haema-toxylin-azophloxin (HA)-stained, 5-to 7-μm thick sections.

In situ hybridisation was performed on serial sections (5-7 μm) with cDNA probes and/or cRNA probes. For cDNA probes, cDNAs (listed in Table 1) were labelled with [α-³⁵S]dCTP (1000 Ci/mmole; Amersham, Buckinghamshire, UK) to a specific activity of approxi-mately 2.2×10 ⁸ dpm/μg DNA, according to the multiprime labelling method (Feinberg and Vogelstein, 1983, 1984) with the modification that 50 mM Tris-HCl (pH 7.2) was used as buffer (Moorman et al., 1993). Sense and antisense cRNAs were synthesised from each cDNA insert by in vitro transcription (Melton et al., 1984) in the presence of [α-³⁵S]UTP (1000 Ci/mmole; Amersham) using T3, SP6 or T7 RNA polymerases (Gibco BRL) as appropriate. The synthesised cRNA strands were hydrolysed in 40 mM NaHCO ³, 60 mM Na ₂CO ₃ (pH 10.2) at 60°C (Cox et al., 1984), to an average fragment length of 150 bases. After phenol/chloroform purification cRNA probes were ethanol precipitated and dissolved in TE buffer (10 mM Tris-HCl, 1 mM EDTA; pH 8.0) with 10 mM 1,4-Dithiothreitol (Merck, Darmstadt, Germany). Probe size was estimated by electrophoresis of labelled fragments through a 4% polyacrylamide gel.

Prehybridisation, hybridisation and autoradiography were carried out as described by Moorman et al. (1993) with minor modifications. cRNA probes were dissolved at a concentration of approximately 6×10 ⁴ cpm/μl hybridisation mix. Hybridisation and washing pro-cedures were carried out at 51°C, while an incubation with RNAse A (Sigma, St. Louis, MO; 10 μg/ml in 0.5 M NaCl, 10 mM Tris-HCl (pH 8.0), 5 mM EDTA) was included after the hybridisation step to reduce background. Slides were exposed for 10 days at 4°C, using Ilford nuclear research emulsion G5 (Ilford, Mobberly Cheshire, UK). Finally, sections were lightly counterstained with 0.01% toluidine blue or 0.02% nuclear fast red.

Although the signal-to-noise ratio was generally lower when cRNA probes were used, no differences in the hybridisation pattern itself were observed. For that reason, we have combined the results.

Alternating serial tissue sections (5-7 μm) were incubated with rabbit antisera specific for rat liver CPS (Charles et al., 1980), GS (de Groot et al., 1987), glutamate dehydrogenase (GDH; EC 1.4.1.3) (Miralles et al., 1982; kindly provided by Dr S. Grisolía) and ALB (Nordic, Tilburg, The Netherlands), and a goat antiserum specific for rat liver α-fetopro-tein (AFP; Selten et al., 1981; a gift from Dr S. H. Yap). A mouse mon-oclonal antibody raised against purified vimentin from porcine eye lens (Osborn et al., 1984) was obtained from Dako (Dako AS, Copenhagen, Denmark) and included to detect cells of mesenchymal origin. Three different serial dilutions were used per antibody.

Antibody binding on sections was visualised according to the indirect unlabelled antibody peroxidase anti-peroxidase (PAP) method (Sternberger et al., 1970) with 0.05% (w/v) 3,3′-diaminoben-zidine tetrahydrochloride (Sigma) dissolved in 25 mM imidazole, 1 mM EDTA (pH 7.0) as chromogen, and 0.01% (v/v) H ₂O ₂ as per-oxidase substrate. Incubations were performed at room temperature as described previously (Gaasbeek Janzen et al., 1984, 1985) with minor modifications: preincubations with normal sheep serum were substituted by addition of 0.5% (w/v) Natrinon milk powder (Nutricia, Zoetermeer, The Netherlands) to the original TENG-T buffer (10 mM Tris-HCl, 5 mM EDTA, 150 mM NaCl, 0.25% (w/v) gelatin, 0.05% (w/v) Tween-20 (Merck); pH 8.0); To further minimise nonspecific binding of immunoglobulins (Igs), all antisera were diluted in phosphate-buffered saline (PBS) with a high salt concentration (0.5 M NaCl in PBS; pH 7.4). Goat anti-rabbit Ig serum was prepared as described (Gaasbeek Janzen et al., 1984). Donkey anti-goat IgG(H+L) serum, rabbit anti-mouse IgG(H+L) serum, and the appropriate PAP immunocomplexes were purchased from Nordic.

In situ hybridisations on RNAse-treated sections (20 μg/ml RNAse A, for 30 minutes at 37°C) and hybridisations with pBR322 or sense cRNA strands were used as negative controls. Similarly, hepatocytes never stained positively after substitution of one of the antisera by preimmune serum or PBS during immunohistochemistry.

The specificity of the reactions in intrasplenic hepatocytes was ascertained using in situ hybridisation and immunoperoxidase staining of embryonic and adult livers, in which the hybridisation and staining patterns are well known (data not shown). Control tissue was processed and treated for in situ hybridisation and immunohisto-chemistry as described for spleens.

Contours of the tissue structures of interest were manually copied onto acetate sheets using a projection microscope/camera lucida setup and fed into a computer program for three-dimensional reconstruction with the help of a digital drawing tablet (Verbeek et al., 1995). Three-dimen-sional reconstructions were displayed using application programs written in Sun Vision software running on a Sparc 2 workstation (Sun Microsystems Inc., Mountain View, CA) (Verbeek et al., 1993).

In this study, we have used E14 embryos as hepatocyte donors. At this age the hepatic primordium has developed into a large, lobed organ. Embryonic hepatocytes are present in irregular, loosely arranged, anastomosing cell plates of three-to-five hepa-tocytes thickness (Fig. 1A). These immature hepatic plates (Elias, 1955) are surrounded by large sinusoidal spaces with numerous haematopoietic foci. Further proliferation and organisation of the hepatocytes into one-cell-thick plates produces the final liver architecture (Elias, 1955), which in the rat is attained in the second postnatal week (LeBouton, 1974). Thus, since the hepa-tocytes at E14 are not yet embedded in a mature lobular archi-tecture, they can serve as a suitable source of cells to assess the effect of an architectural context on the phenotype of hepatocytes.

In the spleen, embryonic hepatocytes settle in the red pulpa and remain present either as dispersed cells or gradually develop into cellular agglomerates, which establish a lobule-like organ-isation (Fig. 1B-F; cf. Lamers et al., 1990). In the present study we observed that these organotypic agglomerates did not become recognisable until 9 days after transplantation. In 2- and 5-day-old transplants, hepatocytes were present as single cells or in small cell clusters (Fig. 1B). From 9 days after trans-plantation onward, agglomerates were formed. However, agglomeration of hepatocytes could be accelerated by increas-ing the number of transplanted cells (Fig. 2). With proceeding time after transplantation, sinusoid-like capillaries, which drain on venules, developed within these agglomerates and the archi-tectural organisation of the hepatocytes gradually changed to that of an adult liver in situ. Within agglomerates of 9-day-old transplants, hepatocytes were predominantly present in clusters of three to five cells (Fig. 1C) and sometimes formed irregular networks resembling the architectural organisation of a fetal liver (cf. Elias, 1955). By day 14 after transplantation, hepato-cytes had become arranged into irregular cell plates of two-to-three hepatocytes thickness. This kind of architecture is remi-niscent of the organisation in the liver in situ around birth (cf. LeBouton, 1974). At 21 days after transplantation, hepatocytes within agglomerates were present in one-cell-thick plates sur-rounded by sinusoids and radially aligned to the draining venules (Fig. 1D,E). The plates in the periphery of the agglom-erates, however, showed a more tortuous alignment. Thus, the hepatocytes within these agglomerates had become embedded in an architecture that resembles that of a mature liver lobule in situ (see Elias, 1955). The straight and tortuous sinusoids seem to identify the central and peripheral portions of the agglomerates as pericentral and periportal zones, respectively (LeBouton, 1974). Reconstruction of serial tissue sections (Fig. 3A) showed that these ‘pericentral’ hepatocytes, while being enveloped by the hepatocytes of the ‘periportal’ zone, formed a spatial continuum around the branches of the splenic vein. Hepatic agglomerates were encapsulated by a continuous layer of fibrous cells (see e.g. Figs 1C, 8A, thin arrows), which stained positively for vimentin (Fig. 2A) and created a separate hepatic compartment within the spleen. The source of these cells (i.e. whether they originated from the transplanted liver cells or from splenic cells) was not investigated. Simultane-ously with the development of hepatic agglomerates an extensive formation of bile ducts could be observed to develop at the transplantation site (see e.g. Fig. 8A; cf. Lamers et al., 1990; Notenboom et al., unpublished data).

During normal rat liver development, AFP and ALB are both expressed in hepatocytes from E11 and E12 onwards, respec-tively (Moorman et al., 1990a; Shiojiri et al., 1991). Accord-ingly, in situ hybridisation showed that essentially all hepato-cytes of 2-day-old transplants contained AFP and ALB mRNA transcripts (Fig. 4A,B). Immunostaining gave similar results, that is all hepatocytes displayed a uniform staining pattern for both proteins at this stage as well as in the 5-day-old trans-plants (shown for AFP in Fig. 5D,H). In 9-day-old transplants, AFP and ALB mRNA transcripts were still demonstrable in virtually all hepatocytes, irrespective of their topographical position, but in comparison to the previous stage fewer AFP transcripts were detected, whereas cellular ALB transcripts were more abundant (data not shown). Immunostaining of 9-day-old transplants visualised the same developmental trend in AFP and ALB expression as seen by in situ hybridisation. Nev-ertheless, all hepatocytes showed a positive staining reaction for AFP (Fig. 5L). Although AFP expression was still present in the dispersed hepatocytes and in hepatocytes within the smaller agglomerates of 14-day-old transplants, within the larger agglomerates it was only detectable in hepatocytes near the (draining) venules, that is, it had disappeared from the peripheral cells (Fig. 6A,B). In 21-day-old transplants, AFP expression (Figs 4C, 6C) was no longer detected in hepatocytes within agglomerates with a well-developed lobular organisa-tion. However, AFP mRNA transcripts (Fig. 4C, arrowheads; compare the small arrowheads with those in Fig. 4D) and protein (Fig. 6D) remained demonstrable outside the agglom-erates in the dispersed hepatocytes. No interhepatocyte differ-ences were observed for ALB expression, which remained con-stitutively high in all intrasplenic hepatocytes from 9 days after transplantation onward (Fig. 4D).

During normal rat liver development, GDH is expressed in hepatocytes as soon as organogenesis of the liver starts at about E10.5 (Gaasbeek Janzen et al., 1988; Moorman et al., 1990b). CPS expression, however, only begins in hepatocytes at E14 (Gaasbeek Janzen et al., 1988; van den Hoff et al., 1994). Although GS mRNA is found in all rat hepatocytes from E13 up to E16 (Moorman et al., 1990b), GS protein is only con-vincingly demonstrated after GS expression becomes restricted to the pericentral cells starting at E18 (Gaasbeek Janzen et al., 1987). In agreement with their developmental age, most if not all hepatocytes of 2-day-old transplants contained GDH, CPS (Fig. 7B) and GS (Fig. 7A) mRNA transcripts. Immunostain-ing showed that all hepatocytes contained GDH protein (Fig. 5A), but that only a few hepatocytes had accumulated CPS protein to a detectable level (Fig. 5B, arrows). Intense immunostaining for GS was present in large myeloid cells (cf. Gaasbeek Janzen et al., 1987) with lobed nuclei (probably monocytes) that were randomly distributed throughout the splenic parenchyma (Fig. 5C, arrow). GS protein, however, could not be demonstrated in hepatocytes at this stage (Fig. 5C). In 5-day-old transplants, all hepatocytes displayed a positive staining reaction for GDH (Figs 2B, 5E) and a strong positive reaction for CPS (Fig. 5F). GS protein, however, only accu-mulated in agglomerating hepatocytes (Fig. 2D; compare with Fig. 5G). In situ hybridisation and immunostaining of 9-day-old transplants showed that all hepatocytes still expressed GDH (Fig. 5I) and CPS (Fig. 5J). At this stage, however, substantial interhepatocyte differences in the expression of GS became clearly visible. Virtually all hepatocytes still contained GS mRNA transcripts, but the hepatocytes near the draining venules of the agglomerates usually displayed a stronger hybridisation signal for GS than the remaining hepatocytes. Immunostaining demonstrated that all hepatocytes within agglomerates had become GS-positive (Fig. 5K), but hepato-cytes surrounding venules stained more intensely for GS than the peripherally situated cells that were only faintly stained. The dispersedly settled hepatocytes contained GS mRNA, but no protein (Fig. 5K, arrowheads; compare with those in Fig. 5I,J,L). Within agglomerates of 14-day-old transplants, GS expression became confined to hepatocytes surrounding the draining venules, whereas these perivenular hepatocytes often revealed a weaker expression of CPS than the peripheral hepa-tocytes (Fig. 8C,E; cf. Lamers et al., 1990). Finally, in most of the GS-expressing hepatocytes within the architecturally well-developed agglomerates of 21-day-old transplants, CPS mRNA transcripts and protein were no longer detected (Figs 7D, 8D, thick arrows; compare with those in Figs 7C and 8F, respec-tively). Coexpression of CPS and GS within agglomerates was only occasionally observed in cells at the boundary between both domains. Although the dispersed hepatocytes never stained positively for GS protein (Fig. 8E,F, small arrowheads; compare with those in Fig. 8C and D, respectively), GS mRNA transcripts (Fig. 7C, arrowheads; compare with those in Fig. 7D) were still detected in most of these cells. No alterations were observed for GDH expression when compared to the previous stages (Figs 1D,E, 8B).

The objective of this study was to provide experimental evidence that the establishment of a hepatic architecture is a prerequisite for the development of a lobular pattern of gene expression. Since the hepatic architecture in the intact liver in situ cannot be manipulated experimentally to a sufficient degree, we have used the technique of transplantation of a single-cell suspension of hepatocytes to allow the development of a three-dimensional lobular architecture. The spleen has been reported to represent the most suitable location for hepatocel-lular transplantation, probably because its stromal architecture (i.e. that of the red pulpa) closely resembles that of the liver (Mito et al., 1979; Kusano and Mito, 1982). For a controlled analysis of the phenotypic fate of the transplanted hepatocytes, we took advantage of a previously observed feature (Lamers et al., 1990), i.e. the fact that part of the hepatocytes developed into organotypic agglomerates with a lobule-like organisation, whereas the rest remains autonomous upon transplantation. Thus, we have compared the enzymic phenotype of the indi-vidually settled hepatocytes with those that participated in the development of a three-dimensional hepatic architecture. This comparison reveals that only those hepatocytes that participate in hepatic morphogenesis (i.e. lobule formation) undergo a normal developmental programme of gene expression.

We have summarised the expression patterns of the genes studied in relation to the spatiotemporal aspects of lobular mor-phogenesis in Fig. 9. This analysis shows that the GDH and ALB genes are expressed in all transplanted hepatocytes, irre-spective of whether they are present as dispersed cells or part of an hepatic-like structure. In contrast, the expression of the AFP gene, which in the liver of mice and rats ceases in the periportal hepatocytes after the first postnatal week and subse-quently also in the pericentral hepatocytes (Gleiberman et al., 1983; Gleiberman and Abelev, 1985; Poliard et al., 1986; Moorman et al., 1990a), only follows this natural course in transplanted hepatocytes undergoing lobular morphogenesis. The dispersed hepatocytes, in contrast, continue to express the gene. Our observations are in good agreement with those of Gleiberman et al. (1983), who demonstrated that the cessation of AFP expression during postnatal liver development coincides with the establishment of the definite hepatic plate organisation. Hence, the reexpression of the AFP gene in both regenerating liver (Gleiberman et al., 1983) and primary monolayer cultures (see Gleiberman and Abelev, 1985), and the continued expression of AFP in the dispersed hepatocytes within the spleen appear to reflect the loss c.q. absence of the mature lobular architecture.

This relationship between phenotypic fate and lobular archi-tecture is further underscored when the expression of CPS and GS is considered. The expression of CPS increases progress-ively after transplantation, both in the dispersed hepatocytes and in those that belong to agglomerates. However, only in the agglomerates that have acquired a lobular structure, does CPS expression stop in the perivenular hepatocytes. This discontin-uation of CPS expression in rat hepatocytes normally occurs in the preweaning period (Gaasbeek Janzen et al., 1985; Moorman et al., 1990b). Initially, all transplanted hepatocytes contain GS mRNA transcripts, but no protein is found. This is in agreement with previous observations that in rat liver, GS protein is normally not detected until the perinatal period (de Groot et al., 1987; Gaasbeek Janzen et al., 1987). Although we cannot exclude the possibility that GS protein is degraded immediately after synthesis, we believe that translation of GS mRNA is sup-pressed until hepatocytes engage in the process of lobular devel-opment. The observation that accumulation of GS protein (Fig. 2D) in transplanted hepatocytes entirely depends on the cellular environment, shows that the onset of GS protein accumulation is not a cell-autonomously regulated event. In keeping with this observation, many of the dispersed hepatocytes continued to synthesise GS mRNA transcripts upon transplantation, but never acquired the GS-positive enzymic phenotype. Accumu-lation of GS protein in all hepatocytes prior to its confinement to the pericentral hepatocytes can be observed in the intrasplenic agglomerates in the rat and in prenatal mouse liver (Bennett et al., 1987; Notenboom et al., 1995), but never in prenatal rat liver (Gaasbeek Janzen et al., 1987). At present, we do not know why rat hepatocytes in our experimental model and mouse hepatocytes in situ behave differently, in this respect, from rat hepatocytes in situ. Concomitant with the overt formation of a lobular architecture, GS expression becomes confined to the hepatocytes surrounding the efferent venules. The restriction of GS expression to this (‘pericentral’) subpop-ulation of hepatocytes is normally observed around birth (Gaasbeek Janzen et al., 1987; Moorman et al., 1990b). Com-parable with the situation during normal liver development (Lamers et al., 1987), the appearance of this pericentral, GS-positive compartment precedes that of the ‘periportal’, CPS-positive compartment (Fig. 9). Interestingly and in accordance with the unit-concept of a metabolic liver lobule (Lamers et al., 1989), the pericentral compartment is more or less centrally located, whereas the periportal compartment is peripherally located within an organotypic agglomerate (Fig. 3).

In conclusion, our data show that intrasplenic hepatocytes exhibit a similar developmental programme of gene expression as noticed with hepatocytes in situ when positioned within a hepatic architectural context. If such an architectural framework is lacking, the cells retain the phenotype that is seen prior to the development of the hepatic lobule. These findings have considerable consequences for our understanding of the topographical aspects of the regulation of gene expression and the temporal aspects of the establishment of the terminally dif-ferentiated hepatocellular phenotype.

The apparent requirement of an architectural framework for the emergence of zonal hepatocyte heterogeneity has several implications. Various lines of evidence have shown that deter-minants of the hepatic architecture that are responsible for the development of zonal differences in the adult hepatocellular phenotype, cannot be traced to: (i) two separate (sub-)lineages of hepatocytes (de Groot et al., 1987; Schöls et al., 1990; Schrode et al., 1990; Gebhardt et al., 1989, 1991; this study), (ii) the age of the hepatocytes, that is the time since the cells underwent their previous DNA replication (Schöls et al., 1990; Bralet et al., 1994), or (iii) a selective (noradrenergic) inner-vation of hepatocytes which develops simultaneously (Lamers et al., 1988). Furthermore, we have recently shown that the size of the liver lobule (Wagenaar et al., 1993a) and the concen-tration of extrahepatic blood-borne factors (Wagenaar et al., 1993b) are not responsible for zonal differences in gene expression. These latter two conclusions are underlined by the present study. First, the same advanced state of adult peripor-tal and pericentral phenotypes is found throughout the organ-otypic agglomerates, irrespective of their cross-sectional size (Fig. 3B). Second, all hepatocytes, both the dispersed ones and those in an architectural context, are supplied with systemic blood, but only in lobular hepatocytes a typical enzymic zonation develops. Our findings therefore strongly suggest that factors regulating the onset and the maintenance of hepatocyte heterogeneity originate within the liver and are produced by the hepatocytes themselves. These data suggest that the phe-notypic diversity between hepatocytes depends on the direction of the sinusoidal bloodstream, which in turn is imposed on the hepatocytes by the hepatic (angio-)architecture. In this hypoth-esis the phenotype of the upstream hepatocytes is determined by modulating signals in afferent blood and nervous input (Jungermann and Katz, 1989), whereas the phenotype of the downstream hepatocytes is regulated, in addition, by factors that are secreted or released by the upstream hepatocytes.

A further implication of our experimental findings is that the establishment of the terminally differentiated hepatocellular phenotype is realised via a multistep process. The first step is taken when hepatocytes differentiate from the embryonic foregut and acquire the capacity to express hepatocyte-specific genes, such as those for AFP (Shiojiri et al., 1991) and ALB (Cascio and Zaret, 1991). The rate at which these (Sellem et al., 1984; Cascio and Zaret, 1991) and other hepatocyte-specific genes (Lamers et al., 1984; Dingemanse et al., 1994) can be expressed, is initially limited when compared to the adult state. This limited biosynthetic capacity prompted Rutter to mark this developmental stage for pancreatic exocrine cell development as the ‘protodifferentiated state’ (Rutter et al., 1968; see for data on the liver, Lamers et al., 1984). Transition of the proto-to the differentiated state has frequently been associated with the pronounced decrease in proliferative activity that occurs in the late fetal period (see Rutter et al., 1968; Tsanev, 1975). Our present results show that an undis-turbed development towards the terminally differentiated state requires, in addition, the enlistment of hepatocytes in a hepatic context. Outside such a structural context, maturation becomes arrested and hepatocytes linger in the protodifferentiated state. The acquisition of the capacity to express hepatocyte-specific genes at the mature (maximal) rate is referred to as the process of maturation (Lamers et al., 1984). Our present data clearly show that maturation is not a gradual process as we had expected, but can be traced to a well-defined developmental event, that is the establishment of a hepatic organisation. Other functional and structural features also point to the fetal period as a crucial time point for normal liver development. Thus, the disappearance of the intercellular heterogeneity of the concen-tration of hepatocyte-specific proteins in fetal hepatocytes (Dingemanse et al., 1994; Notenboom et al., 1995) can be regarded as a functional parameter that marks the end of the protodifferentiated state, while the formation of bile ducts (van Eyken et al., 1988; Shiojiri et al., 1991) can be interpreted as a salient feature of the emerging lobular architecture.

We are grateful to Dr J.-L. Danan for the AFP cDNA clone pRAF 65, and Drs S. Grisolía and S. H. Yap for providing the antiserum to glutamate dehydrogenase and α-fetoprotein, respectively. We thank Drs Frits Michiels, Johan Offerhaus and David Wilson for advice with the preparation of the manuscript; Fons Verbeek for converting the contour model of the three-dimensional reconstruction into a voxel display, and Wilfried Meun and Cars Gravemeijer for the photo-graphic service rendered.