The fate of the vitelline and umbilical veins during the development of the human liver

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

2 Department of Morphology, Faculty of Veterinary Medicine, Ghent University, Merelbeke, Belgium

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

3 NUTRIM Research School of Nutrition and Translational Research in Metabolism, Maastricht University, Maastricht, The Netherlands

1 Department of Anatomy & Embryology, Maastricht University, Maastricht, The Netherlands

4 Tytgat Institute for Liver and Intestinal Research, Academic Medical Center, Amsterdam, The Netherlands

Differentiation of endodermal cells into hepatoblasts is well studied, but the remodeling of the vitelline and umbilical veins during liver development is less well understood. We compared human embryos between 3 and 10 weeks of development with pig and mouse embryos at comparable stages, and used amira 3D reconstruction and cinema 4D remodeling software for visualization. The vitelline and umbilical veins enter the systemic venous sinus on each side via a common entrance, the hepatocardiac channel. During expansion into the transverse septum at Carnegie Stage (CS)12 the liver bud develops as two dorsolateral lobes or ‘wings’ and a single ventromedial lobe, with the liver hilum at the intersection of these lobes. The dorsolateral lobes each engulf a vitelline vein during CS13 and the ventromedial lobe both umbilical veins during CS14, but both venous systems remain temporarily identifiable inside the liver. The dominance of the left‐sided umbilical vein and the rightward repositioning of the sinuatrial junction cause de novo development of left‐to‐right shunts between the left umbilical vein in the liver hilum and the right hepatocardiac channel (venous duct) and the right vitelline vein (portal sinus), respectively. Once these shunts have formed, portal branches develop from the intrahepatic portions of the portal vein on the right side and the umbilical vein on the left side. The gall bladder is a reliable marker for this hepatic vascular midline. We found no evidence for large‐scale fragmentation of embryonic veins as claimed by the ‘vestigial’ theory. Instead and in agreement with the ‘lineage’ theory, the vitelline and umbilical veins remained temporally identifiable inside the liver after being engulfed by hepatoblasts. In agreement with the ‘hemodynamic’ theory, the left–right shunts develop de novo.

The liver is first identifiable at ~ 3.5 weeks of human development [Carnegie Stage (CS)10] as a local endodermal thickening in the ventral wall of the distal foregut between the heart cranially and the yolk sac caudally (Lewis, 1912; Severn, 1971). Tissue‐culture and explantation experiments have shown that the transition from foregut into hepatic endoderm requires fibroblast growth‐factor (FGF) signaling between the cardiac mesoderm and endoderm, and bone morphogenetic‐protein (BMP) signaling between the mesenchymal cells of the transverse septum and endoderm (Cascio & Zaret, 1991; Rossi et al. 2001). Upon specification, the shape of the hepatic endoderm cells changes from cuboidal to columnar. The activity of the homeobox gene Hex in this columnar epithelium drives ‘interkinetic nuclear migration’ between the basal and apical regions of the cells during cell division, so that this highly proliferative columnar epithelium becomes pseudostratified (Bort et al. 2006) in a similar fashion as observed in the neural tube (for a review, see Willardsen & Link, 2011). Subsequently, the basal membrane disintegrates in the cranial portion of the liver bud (Bort et al. 2006), which allows the hepatoblasts to invade the transverse septum and form a sponge‐like three‐dimensional continuum with interspaced capillaries from 4.5 weeks of development (CS13) onwards (Thompson, 1908; Lewis, 1912; Elias, 1955; Severn, 1972; Cascio & Zaret, 1991). The cells of the caudal portion of the liver bud that give rise to the extrahepatic bile ducts, gallbladder and ventral pancreas, do not extend into the transverse septum (Severn, 1972).

The role of the vitelline and umbilical veins in the development of the hepatic venous system, their drainage into the systemic venous sinus, and their contribution to the hepatic sinusoidal network are less well understood than the differentiation of endodermal cells into hepatoblasts. It is generally assumed that the intrahepatic portion of the vitelline veins forms both the ‘advehent’ (portal) and ‘revehent’ (hepatic) veins (Ingalls, 1908; Hamilton et al. 1945; Patten, 1953). The so‐called ‘vestigial theory’ hypothesizes that the invading hepatoblasts disperse the intrahepatic portion of the vitelline veins into a sinusoidal network (Minot, 1900; Wake & Sato, 2015) that subsequently remodels into portal and hepatic veins (Lewis, 1904, 1912; Geraudel, 1907; Du Bois & Rouiller, 1963). However, two recent findings question the vestigial theory. First, hepatoblasts require the presence of VEGFR2‐expressing endothelial cells in the transverse septum to invade it (Matsumoto et al. 2001; Lammert et al. 2003). These early endothelial cells originate at least partially from endocardial cells in the venous sinus (Zhang et al. 2016). Secondly, the perivascular mesenchymal and hepatic stellate cells of the liver originate from the Wilms tumor 1 homolog (Wt1)‐positive mesenchymal cells of the transverse septum and not from the walls of the vitelline veins (Asahina et al. 2011). Interestingly, the coronary arteriovenous vascular connections in the cardiac ventricles derive from these same Wt1‐positive cells in the pro‐epicardial region of the transverse septum, demonstrating that these cells can form vessels (Rodgers et al. 2008; Cano et al. 2016). These findings also question the so‐called ‘hemodynamic theory’ that poses that the umbilical blood flow determines the course of the hepatic veins and that this course is, therefore, more predictable near the left‐sided entry of the umbilical vein than near the right‐sided outlet of the hepatic veins (Lassau & Bastian, 1983). The disappearance of the intrahepatic portions of the vitelline and (right) umbilical veins and the concurrent appearance of portal veins and liver segments are the main arguments in favor of the hemodynamic theory (Lassau & Bastian, 1983). However, established embryology textbooks (Hamilton et al. 1945; Patten, 1953; Hinrichsen, 1990; Schoenwolf et al. 2015) all adhere to the vestigial theory.

To resolve these issues, we reinvestigated the developmental changes in the topography of the intrahepatic veins in a closely spaced series of human embryos between 3 and 6 weeks of development using 3D reconstruction software. Our data show that the vitelline and umbilical veins share a common entrance (hepatocardiac channels) into the venous sinus and evolve into large liver veins that remain temporally identifiable, whereas the shunts (portal sinus and venous duct) have a hemodynamic origin by coalescence of sinusoids derived from the venous sinus.

This study was undertaken in accordance with the Dutch regulations for the proper use of human tissue for medical research. We reconstructed 15 human embryos between 3 and 10 weeks of development from the historical collections of the Academic Medical Centre (AMC) of the University of Amsterdam and the Leiden University Medical Centre (LUMC), which together contain over 150 human embryos. Also the human collection of the Carnegie Collection in Washington (DC), which is freely accessible on the internet, was included. The criteria of O'Rahilly as modified in 2010 (O'Rahilly & Muller, 2010) were used to determine the Carnegie Stage (CS) of development. The fate of the umbilical and vitelline veins during liver development is described for the human situation. Because we were aware that most mammalian species differ from human in having a lobated liver (Nettelblad, 1954) and because the corresponding fissures may function as topographic landmarks with respect to the position and fate of the vitelline and umbilical veins, we also studied pig embryos from the collection of the Department of Morphology, Ghent University, Belgium, and mouse embryos from the AMC collection. The atlases of Evans & Sack (1973), Butler & Juurlink (1987), Theiler (1989) and O'Rahilly & Muller (2010) were used to select similar stages and to express developmental age in Carnegie Stage for all species (cf. table 1 in Hikspoors et al. 2016).

Serial sections of human embryos from the AMC and LUMC, which had been stained with either hematoxylin and eosin (H&E) or azophloxine (H&A), were digitized with an Olympus BX51 microscope and the dotslide program (Olympus, Zoeterwoude, The Netherlands). Digital images of serial sections from the Carnegie collection were obtained via the Virtual Human Embryo Project (http://virtualhumanembryo.lsuhsc.edu). These embryos were stained with either H&E or Alum cochineal (i.e. carmine). H&E‐stained serial sections of pig embryos were already available as digital images (Cornillie et al. 2008). Serial sections of mouse embryos from the AMC collection were stained with H&A and digitized with the dotslide program. Processing of the digital images and voxel size calculation for the human, mouse and pig embryos was performed as described previously (Cornillie et al. 2008; Hikspoors et al. 2015).

amira 3D (version 6.2; base package; FEI Visualization Sciences Group Europe, Merignac Cedex, France) was used to generate 3D reconstructions after image loading, alignment and segmentation. All sections were automatically aligned with the least squares method and then adjusted manually to account for curvatures in the sagittal and transverse planes of the embryonic body axis with the help of photographic and magnetic resonance images (MRI) of age‐matched human embryos (http://embryo.soad.umich.edu; http://virtualhumanembryo.lsuhsc.edu; http://www.ehd.org/virtual-human-embryo; O'Rahilly & Muller, 1987; Yamada et al. 2010; Pooh et al. 2011). Thereafter, delineation of the liver contour, blood vessels/networks, and structures of interest was performed manually (Supporting Information Fig. S1A).

Polygon meshes were created from all reconstructed materials and exported via ‘vrml export’ into cinema 4D (MAXON Computer GmbH, Friedrichsdorf, Germany). Cinema 4D allows remodeling of the original amira output with a ‘spline’ function for vessel reconstruction and a ‘polygon’ function for more complex shapes. The accuracy of the remodeling process was secured by simultaneous visualization in Cinema 4D of the original output from amira 3D and the remodeled cinema model (Fig. S1). Subsequently, the cinema 3D‐model was exported via ‘wrl export’ into Adobe portable device format (PDF) reader version 9 (http://www.adobe.com) for the generation of 3D‐ interactive PDF files (Supporting Information Figs S2 and S3).

Diameters and dorsoventral or craniocaudal lengths of structures were measured with the software program cinema 4D, using the ‘spline’ function to draw a longitudinal line between the starting and end point of the structure. Surface area of the triangular liver primordium (CS10, CS11) was calculated as ‘0.5 × base triangle × height triangle’ and the elliptic surface of the liver bud (CS12, CS13) and yolk sac stalk as ‘π × (diameter of short side/2) × (diameter of long side/2)’. Volume measurements of the liver (invading hepatoblasts) and transverse septum were performed via the ‘material and statistics’ function in the software program amira 3D (Fig. S1E,F). The gastrointestinal (GI) tract was reconstructed from the level of the 1st pharyngeal arch artery to the connection of the GI tract with the allantois in all human embryos studied (Fig. S1C). The cranial border of the GI tract is the cutting plane of the histological section.

At CS9 or ~ 26 days of development, the foregut epithelium did not display regional differences in thickness (Fig. 1A). The boundary of the foregut and yolk sac was identifiable because foregut epithelial cells were cuboidal in shape and the yolk‐sac epithelial squamous cells. At CS10 (~ 28 days of development), the epithelium of the foregut underlying both dorsal aortas remained cuboidal (Fig. 1B; open black arrowhead), whereas the epithelium on the ventrolateral sides thickened and acquired a pseudostratified nuclear arrangement. Near the junction of the foregut with the yolk sac stalk, just caudal to the venous sinus, the ventral foregut epithelium had become much thicker, indicating the formation of the liver primordium [Figs 1B (green arrowhead) and ​and2A,D].2A,D]. The transverse septum was identifiable as a flat mass of dense mesenchymal tissue ventrally of the venous sinus and liver primordium, and cranially of the yolk sac stalk that extended along the entire width of the embryo (Figs 1B, ​B,2D2D and S2).

Changes in the epithelium of the caudal foregut of CS9‐CS12 human embryos. CS9 embryos (panel A) had similar thickness of the foregut (FG) epithelium dorsally, laterally and ventrally (black open, black filled and green arrowheads, respectively). At CS10 (panel B), the dorsal epithelium under both aortas (Ao) had flattened, the ventrolateral wall remained cuboidal, whereas the ventral side, just caudal to the venous sinus (VS) and transverse septum (TS), had become columnar. At CS11 (panels C,D), the ventral epithelium had become pseudostratified columnar. At CS12 (panel E), the liver bud (LB) had formed from which hepatoblasts in the cranial part of the bud started to migrate into the transverse septum, whereas no cells migrated from the caudal liver bud (LB' in panel F). Y: yolk sac; open green arrowhead: boundary of the cuboidal foregut and the squamous yolk sac epithelium. Section levels are shown in Fig. 2D–F.

Development of the liver bud. Panels A–C show cranioventral and panels D–F right‐sided views of the gut, liver primordium and transverse septum of CS10‐CS12 embryos. Note U‐shape of ‘triangular’ liver primordium (interrupted red line in panel B) and ‘ballooning’ of the transverse septum (transparent in panels D–F). The increase in surface of the liver primordium during its triangular phase (panels A,B) and the subsequent decrease of the surface of the liver bud (due to hepatoblast egression) and yolk‐sac stalk are shown in (panel G). The logarithmic Y‐axis reveals rapid and steady changes in size. Also note very rapid increase in size of the transverse septum between 28 and 30 days (> 5‐fold/day; panel H). Between day 30 (CS12) and 36 (CS15), the liver (hepatoblasts plus enclosed transverse septum cells) continued to grow exponentially (~ 2.5‐fold/day; panel H; dark green line), while the part of the transverse septum not occupied by hepatoblasts retained a constant volume (panel H).

Between CS10 and CS12 (28–30 days of development), the triangular liver primordium expanded exponentially in lateral and ventral directions (~ 3.6‐fold/day; Figs 1C–E and ​and2B,C,E–G),2B,C,E–G), which resulted in an U‐shaped triangle overlaying the yolk sac stalk at CS11 (Figs 1C,D, ​C,D,2B,E2B,E and S2). Simultaneously, the yolk sac stalk decreased exponentially in diameter (~ 3.1‐fold/day; Fig. 2G) and ceased to exist at ~ 35 days (Soffers et al. 2015). As a result, the transverse septum appeared to ‘rotate’ ~ 60° to assume its typical transverse orientation (Fig. 2F). Furthermore, the transverse septum increased rapidly in size (Fig. 2H), changing from a flat plate at CS10 to a spherical form at CS12 (Figs 2D–F and S2). Concurrently, the initially flat liver primordium transformed into a bud that remained connected with the gut via a narrow stalk (~ 0.2 mm diameter; Fig. 2C,G). The surface of the liver bud decreased ~ 3.1‐fold between CS12 and CS14 (Fig. 2G), concomitant with the migration of hepatoblasts from the cranial region of the bud into the transverse septum (Figs 1E and ​and2H).2H). No epithelial cells migrated from the caudal portion of the liver bud (compare Fig. 1E with F), which formed the gallbladder and ventral pancreas. The volume of the liver (taken as the volume occupied by hepatoblasts and the adjacent transverse‐septum cells) continued to increase exponentially with ~ 2.5‐fold/day between CS12 and CS15 (Fig. 2H), while the part of the transverse septum not occupied by hepatoblasts maintained a constant volume (Figs 2H and S1E,F). In aggregate, these data show exponential growth of the hepatoblasts and their surrounding transverse‐septum cells during the fourth week of development.

The bilateral vitelline and umbilical veins were first identifiable at CS10 as vessels on the yolk sac stalk and at the junction of the lateral body wall with the amnion, respectively. Both vessels entered the venous pole of the heart via a common trunk, the so‐called hepatocardiac channel (Fig. 3), while the contributing vitelline and umbilical veins were separated more caudally by the pericardioperitoneal channels of the body cavity (Fig. 4E). The common cardinal veins only appeared and joined the venous pole at CS12 (Figs 3D,E and ​and4A–H).4A–H). The ventromedial position of the vitelline veins relative to the umbilical veins persisted during CS11, when the expanding liver primordium and transverse septum became interposed between the yolk sac and the venous sinus (Fig. 3A,D). The dorsal expansion of liver and transverse septum continued during CS12 and CS13 and was reflected in the more dorsal position of the vitelline veins (Fig. 3E,F) and the ~ 90° change in position of the common cardinal veins from near transverse at CS12 to near frontal at CS13 (Fig. 3E,F). The left hepatocardiac channel had become largely incorporated into the wall of the venous sinus [Figs 3F (inset), ​(inset),4I–K4I–K and S2] and was, therefore, difficult to discern after this stage. In pig embryos, the left hepatocardiac channel remained identifiable during CS13 (Fig. 5C,D) but had disappeared at CS14, that is, coincident with the formation of the venous duct. In the mouse the left hepatocardiac channel already disappeared at CS13 [Embryonic day (E)11 – Theiler Stage 18; Fig. 6], so that only the right hepatocardiac channel and left‐sided umbilical vein were present when the venous duct formed.

Position of the vitelline and umbilical veins before enclosure in the liver. Panels A–C show ventral and panels D–F right‐sided views. The vitelline veins were located ventromedially to the umbilical veins. Both vessels joined the venous sinus via hepatocardiac channels (panels A–E and F with left‐sided inset). The dorsolaterally expanding ‘wings’ of the liver engulfed the vitelline veins and caused a pronounced dorsal bend in these veins (panel F). Only the dense mesenchyme of the transverse septum enclosed the umbilical veins at CS13 (panel F). Note that the common cardinal veins (panels D–F; yellow arrow) appeared only at CS12 and changed, concomitant with the growth of the dorsolateral wings of the liver, from dorsoventrally to craniocaudally oriented vessels. Interrupted black lines (panels D–F) indicate section levels of Fig. 4.

The hepatocardiac channels. On each side of the body, the umbilical (in parietal mesoderm; pink) and vitelline (in visceral mesoderm; light blue) veins merged (compare panels A,E,I with B,F,J and K) to form a hepatocardiac channel just caudal to the venous sinus. The venous sinus was marked by the presence of myocardium (panels C,D,G,H,J,K; dotted lines). From CS12 onwards, hepatoblasts started to migrate into the transverse septum (compare panels D with F–H and I). Note the dorsal ‘wings’ of the liver with the vitelline veins in panel I. #, body cavity; R, right side; L, left side; CCV, common cardinal vein.

Enclosure of vitelline and umbilical veins into the liver of CS13 pig embryos. Transverse sections. The embryonic pig liver consisted of two dorsolateral lobes containing the vitelline veins and a single ventromedial lobe containing the umbilical veins (panel A). Note that, at CS13, the umbilical veins still had intra‐ and extrahepatic portions (panels A–D). Panel D (ventral view) is a 3D model in which section levels of panels A–C are indicated.

Asynchronous enclosure of umbilical veins into the liver of mouse embryos. Transverse sections. Panel A shows the entrance of the right vitelline (portal) vein into the right dorsolateral lobe of an ED11 (CS13; Theiler Stage 18) embryo. The left umbilical vein is enclosed by the ventromedial lobe of the liver and entered the right hepatocardiac channel via the venous duct (panels C,D), whereas the right umbilical vein still coursed in the lateral body wall to the right hepatocardiac channel (panels A–E). In ED11.5 (CS14; Theiler Stage 19) mouse embryos, both umbilical veins flanked the gallbladder and merged with the portal sinus at the junction of the ventromedial and dorsolateral lobes (panel F). R, right side; L, left side.

Concurrent with the invasion of hepatoblasts into the transverse septum at CS12, small capillaries in direct contact with the venous sinus were observed entering it (Fig. 7A–C). Hepatoblast cords with surrounding capillary network extended cranially, laterally and dorsally during CS12 and CS13 (Figs 4I and ​and7B,C).7B,C). They enveloped the vitelline veins in the dorsal portion of the liver at CS13 (Figs 3F, ​F,4I4I and ​and8A),8A), but the vitelline veins remained identifiable inside the liver as wide, dorsally bent conduits, with the right vein being substantially wider than the left (Figs 7D,E, ​D,E,8A–C8A–C and S2). At this stage, the livers of pig and mouse embryos differed from their human counterparts in that they had formed two dorsolateral lobes and a single ventromedial lobe covering the gall bladder (Figs 5A and ​and6C).6C). In both species, the vitelline conduits were exclusively present in the dorsolateral lobes (Figs 8A–C and ​and9A–D),9A–D), while the hepatocardiac channels were located at the cranial end of the furrows between the dorsolateral and ventromedial lobes (Supporting Information Fig. S3). In all species, the hepatocardiac channels remained outside the boundaries of the liver and transverse septum throughout CS13 (Figs 8A–C and ​and99A–D).

Vessels in the transverse septum. Panels A,B transverse sections and panel C right lateral view of reconstruction of a CS12 embryo; panels D,E sagittal sections of a CS14 embryo. Migration of hepatoblasts into the transverse septum at CS12 concurred with the appearance of small capillaries (dark green (C) and black arrowheads (A,B)) in the mesenchyme of the transverse septum in direct continuity with the venous sinus. Stippled lines in panel C indicate the section plane of panels A and B. Panels D,E The vitelline veins were incorporated into the liver at CS13, but remained identifiable as large conduits. Note that the right‐sided vessel, including the right‐sided hepatocardiac channel (D), was substantially wider than the corresponding left‐sided structures (E). H, heart; Li, liver; VM, ventral mesentery.

Fate of umbilical and vitelline veins in human embryos after enclosure by hepatoblasts. Panels A,D,G are dorsal and remaining panels ventral views. Note the C‐shaped intrahepatic portions of the vitelline veins in the dorsal ‘wings’ of the liver. The intrahepatic part of the left vitelline vein broke up, while that of the right vitelline vein temporarily persists (panels D–F: blue arrow). The umbilical veins became enclosed between CS13 and early CS14 (panels C,F), with the right umbilical vein losing its connection with the liver at CS15 (panel J; red open arrow). The left umbilical vein continued via the venous duct towards the right hepatocardiac channel (panels F,I,K) and via the portal sinus (panels D–I) towards the right vitelline vein. Just caudal to the dorsal pancreas a dorsal intervitelline anastomosis was forming (panels A and D; red asterisks), while more caudally a ventral anastomosis was temporarily present (panel E, red arrow). Black line (panel K): line of Cantlie; black arrows: outgrowing left and right portal branches; yellow arrow (panel F): temporary inter‐umbilical vein anastomosis; PV, portal vein.

Fate of vitelline and umbilical veins in pig embryos after enclosure by the liver. Panels A,E are dorsal, panels B,F right‐sided and remaining panels ventral views. The intrahepatic vitelline veins were found in the dorsolateral lobes. The enclosure of the umbilical vein proceeded from caudal (CS13, panels B–D) to cranial (CS14, panels G–H). The intrahepatic portions of the umbilical veins in the ventromedial lobe were identifiable up to the junction with the dorsolateral lobes. Red asterisk: dorsal intervitelline anastomosis; yellow arrow (G,H): temporary inter‐umbilical vein anastomosis; black arrows: outgrowing left and right portal branches.

Both in human and pig embryos, the liver volume increased 10‐fold between CS13 and CS14 (cf. fig. 7 in Hirose et al. 2012). This growth occurred mostly in the ventromedial lobe and allowed this lobe to enclose the umbilical veins symmetrically. In our human embryos, the umbilical veins had become an entirely intrahepatic structure in early CS14 (Fig. 8F), but in pig embryos, the expanding ventromedial liver lobe initially surrounded the adjacent portion of the umbilical veins only (CS13; Fig. 9A–D). The more downstream portions persisted in the lateral body wall to become enveloped by the ventromedial lobe during CS14 (Figs 9E–H and S3). The umbilical conduits in the ventromedial lobe remained identifiable up to the junction of this lobe with the dorsolateral lobes (the future hilum), but the left conduit was much wider than the right (Figs 9C,G and S3). In the mouse, the uptake of the left‐ and right‐sided umbilical veins in the ventromedial lobe was temporally asymmetric: the left‐sided umbilical vein was enclosed by liver tissue at CS13, whereas the right‐sided umbilical vein temporarily retained its original course (Fig. 6B–E) and became enclosed in the ventromedial lobe at CS14 (E11.5 – Theiler Stage 19; Fig. 6F). As a result, both left‐ and right‐sided umbilical veins could be observed to briefly flank the gall bladder before ending in the liver hilum at CS14 (Fig. 6F). The hepatocardiac channels maintained an upstream plexus in the dorsolateral lobes during the uptake and remodeling of the umbilical veins in the ventromedial lobe.

The rightward repositioning of the sinuatrial junction during CS13 was accompanied by a marked increase in the diameter of the right‐sided venous sinus and a corresponding decrease in the diameter of the left‐sided venous sinus at CS14. Concomitantly, the right hepatocardiac channel increased in diameter, whereas the left‐sided channel narrowed (Fig. 7D,E; Marshall's ligament of the left caval vein will form here, but only at CS23). The course and prominence of the intrahepatic part of the vitelline veins changed accordingly. The left‐sided intrahepatic vitelline vein disappeared as a vascular conduit in early CS14 embryos, and the small left hepatocardiac vein and its tributaries, which persisted temporarily (Figs 7D and ​and8D,E),8D,E), had disappeared in late CS14 embryos (Fig. 8G). The right‐sided channel, in contrast, persisted as a large ‘Ω’‐shaped vascular conduit along the dorsal boundary of the liver until the ‘portal sinus’ had formed in late CS14 embryos (Figs 7D, ​D,8D–F8D–F and S2). The portal sinus developed between the (left) umbilical vein and the right vitelline (portal) vein and remained present in later stages as the transverse part of the left portal vein.

The intrahepatic portion of the left umbilical vein formed anastomoses with the intrahepatic portions of both left and right vitelline veins at early CS14 (Fig. 8D,E). The largest connection is the umbilico‐vitelline shunt at the ventral side of the duodenum (Fig. 8D–I). Concomitant with the disappearance of the drainage of the extrahepatic umbilical veins into the hepatocardiac channels at CS14, the venous duct formed between the cranial end of the left umbilical vein in the liver hilum and the right hepatocardiac channel (Fig. 8F). In the liver hilum, the umbilical vein was, therefore, continuous with the relatively narrow venous duct (Figs 8F,I, ​F,I,10A–C10A–C and S2) and the relatively wide umbilico‐vitelline shunt (Fig. 8G,I). Until CS16, the wall of both the left umbilical vein and the venous duct were covered by an endothelial lining only (Fig. 10D,E), but only the umbilical vein had direct access to liver sinusoids.

Fate of the umbilical vein and venous duct. Panels A–C show the transition of the wide intrahepatic portion of the umbilical vein (UV) into the narrow venous duct (VD), while panels D–F show the entrance of the umbilical vein into the liver. The insets in panels D–G show magnifications of the wall of the umbilical vein (black open arrowhead). The falciform ligament represents the connection of the liver with the ventral body wall (panels D–F: stippled line). The width of the falciform ligament, lesser omentum and the umbilical vein (panel L; dark blue, light blue and pink lines, respectively) hardly changed between 5 and 8 weeks (CS14 and CS23) and became relatively thin structures, because the volume of the liver increased ~ 40‐fold (panel L; light green line) in the same period. The falciform ligament (dark blue arrows) containing the umbilical vein continued into the lesser omentum (light blue arrows) containing the venous duct. Panel I (dorsal view with magnified lateral view) shows a 3D reconstruction at 10 weeks of development with the section planes of panels G–K indicated. Asterisks, outgrowing portal branches; PV, portal vein; PS, portal sinus; TS, transverse septum.

Between CS14 and CS16 the left umbilical vein passed through the transverse septum (ventral mesentery) before entering the liver (Figs 7E and ​and10D,E).10D,E). This connection gradually narrowed to form the falciform ligament between CS18 (6.5 weeks) and 10 weeks of development (Fig. 10F,K). Between CS18 and CS23 (8 weeks), the increase in size of the umbilical vein and the falciform ligament were minor relative to that of the liver (Fig. 10L), demonstrating that growth of the liver rather than narrowing of the ventral mesentery is responsible for the development of the relatively thin falciform ligament (Fig. 10D–F,K,L). The liver not only expanded laterally, but also caudally around the part of the umbilical vein that was located inside the transverse septum until CS16, as evidenced by the increasing thickness of its wall in upstream (caudal) direction (Fig. 10F,J). Eventually, a direct connection of the umbilical vein with the liver parenchyma remained only present near and in the liver hilum (Fig. 10G,I,J). Similarly, cranial expansion of the liver explains why the venous duct became gradually surrounded by mesenchyme of the lesser omentum (Fig. 10A–C,G–J,L).

Portal‐vein branches became identifiable at CS14 (Fig. 8F,G,I). Comparing human with pig (Fig. 9E–H) and mouse embryos (Fig. 6) showed that the intrahepatic part of the right vitelline (portal) vein produced portal branches for the right dorsolateral and ventromedial liver lobe. The portal branches in the left‐sided portion of the ventromedial lobe and the left dorsolateral lobe arose from the left umbilical vein in a spiraling fashion from craniolateral, near the transition of the umbilical vein into the venous duct, to caudoventral near the entrance of the umbilical vein from the falciform ligament (Figs 8F,I,K, ​F,I,K,9H9H and ​and10B,C).10B,C). The gall bladder marked the boundary (line of Cantlie) between the left‐sided portal branches originating from the umbilical vein and right‐sided branches originating from the portal vein (Fig. 8K).

Just caudal to the dorsal pancreatic bud, an extrahepatic dorsal anastomosis developed between the left and right vitelline veins at CS13 (Figs 8A and ​and9A).9A). A well‐formed channel was present at this location at early CS14 (Figs 8D and S2). Just upstream of this intervitelline anastomosis the future portal vein entered the left vitelline vein as a small branch originating on the left side of the developing midgut (Fig. 8B,E,H), whereas the main trunks of the vitelline veins followed the stalk of the yolk sac as two channels that merged at several places (Fig. 8B,E,H). In agreement with the size of the yolk sac, the vitelline vein was still a wide vessel in CS15 embryos (Fig. 8J,K), even though the stalk of the yolk sac had already disappeared at this stage. A much smaller ventral anastomosis temporarily (during CS14 only) connected the right and left vitelline veins more caudal in respect to the established dorsal anastomosis, just underneath the liver (Fig. 8E; red arrow). As soon as the liver had enclosed the right umbilical vein, an interumbilical anastomosis existed briefly (CS14 only) just caudal to the liver (Figs 8F and ​and9G,H;9G,H; yellow arrows). The right umbilical vein itself started to regress, did not connect to the liver anymore at CS15 (Fig. 8J; open red arrow), and had disappeared at CS20.

The caudal portion of the liver bud retained its bud‐like appearance until early CS14 (~ 33 days of development; Fig. 11A,B). Its division into gallbladder and ventral pancreas became apparent at late CS14 (~ 35 days of development; Figs 11C,F–J and S2). Liver, gallbladder and ventral pancreas retained a common connection with the foregut (future greater duodenal papilla) (Fig. 11C; magnification). The primordium of the dorsal pancreas became identifiable opposite the liver bud on the dorsal side of the foregut at CS13 (Fig. 11A,E,F). In late CS14 embryos, the position of the primordia relative to the foregut (duodenum) had not changed, but the plane through stomach, dorsal pancreas, ventral pancreas and gall bladder had acquired a ~ 45° rightward tilted position, with the duodenum forming the pivot (Fig. 11A–J). The portion of the duodenum between its junction with the stomach and the entrance of the common bile duct increased ~ 20‐fold in length between CS14 and CS16, without much increase in diameter (Fig. 12E). The growth of the proximal duodenum and associated dorsal mesentery results in the formation of the cranial secondary intestinal loop (Soffers et al. 2015). The ventral pancreas and associated common bile duct, in contrast, remained attached to the liver (and ventral mesentery) and did not change position. Although the longitudinal growth of the duodenum did not change the plane through the duodenum, stomach and dorsal pancreas (Fig. 11K,L), it did extend the duodenum rightward and ventrally beyond the ventral pancreas and gallbladder, which, due to their persisting attachment to the liver via the ventral mesentery, retained their position. As a result, the angle between the dorsal and ventral pancreas decreased from ~ 180° to ~ 45° and the portal vein became wedged between both pancreatic primordia (Figs 11J,L and ​and12A,C).12A,C). Due to the continuing longitudinal growth of the duodenum after CS15, the ventral pancreas and the proximal part of the dorsal pancreas came in closer proximity and merged at CS18 (Fig. 12B,D,F). The portal vein typically occupied the notch between both portions of the definitive pancreas.

Changes in the topography of the gallbladder and pancreas. The ventral pancreas and gallbladder were indistinguishable portions of the caudal part of the liver bud until late CS14 (cf. panels A and B). The dorsal pancreas was identifiable from CS13 onwards (panels A,B). The plane through the dorsal pancreas, gut, and ventral pancreas plus gallbladder had acquired a ~ 45° rightward tilted position, with the duodenum as pivot (panels C and D; insets: magnified dorsal views). Histological sections in (panels E–L) correspond to the stippled lines indicated in (panels A–D). The ‘compass’ in (panels F–L) indicates the plane through the dorsal pancreas, duodenum, and ventral pancreas relative to the midline, PV, portal vein. In the histological sections, the dorsal pancreas is indicated by a continuous black contour (panels E–G and J–L) and the caudal part of the liver bud by a black interrupted contour (panels F–H).

Fusion of ventral and dorsal pancreas. Panels A,B are ventral views and panels C,D dorsal views. The length of the duodenum between the gastroduodenal junction and the greater duodenal papilla (panel E; purple line) increased rapidly, while the diameter of the duodenum (panel E; grey line) changed much less rapidly. As a result, the angle between the dorsal and ventral pancreas decreased from ~ 180° to ~ 45° between CS15 and CS18, so that the portal vein became wedged between both pancreatic primordia, which fused at CS18. Panel F shows the histology of the fusing ventral and dorsal pancreas (interrupted and continuous black contours, respectively). Interrupted lines in panels A–D: entrance of right vitelline (portal) vein into the liver; arrowheads: common bile duct and dorsal pancreatic duct (note that the gut lumen is made transparent in panels A,B to visualize the pancreatic duct).

The vascular architecture of the liver becomes completely remodeled between 4 and 6 weeks of development (Fig. 13; schematic overview). Until CS11 (4 weeks), the vitelline and umbilical veins bypass the liver and drain directly into the venous sinus via the hepatocardiac channels. In the 5th week (CS12‐14), rapidly dividing hepatoblasts entered the transverse septum and enclosed first the vitelline and then the umbilical veins, while sinusoids entered from the venous sinus to form a sinusoidal network. Comparison with pig and mouse embryos allowed a better topographic description of the fate of the vitelline and umbilical veins in the liver, because the vitelline veins were incorporated exclusively into two dorsolateral lobes (wings) of the liver, whereas the umbilical veins became located in the ventromedial lobe. As the umbilical vein evolved into a left‐sided structure, and the vitelline and hepatocardiac veins into right‐sided structures due to the rightward positioning of the sinuatrial junction, two large vascular left‐right shunts developed between the distal end of the (left) umbilical vein and (i) the right vitelline vein (portal sinus), and (ii) the right hepatocardiac channel (venous duct). Once these anastomoses had formed, portal branches started to develop from the intrahepatic portions of the right vitelline and umbilical veins. The gallbladder proved a reliable marker for this vascular division inside the liver.

Schematic overview of the developing vascular architecture in the liver. The vitelline and umbilical veins enter the venous sinus on each side via a common hepatocardiac channel (panel A1 and B1). The liver develops as 2 dorsolateral lobes or ‘wings’ and one ventromedial lobe, with the gall bladder serving as a reliable landmark of the midline of this ventromedial lobe (panels A2, A3 and C1, C2). Vascular development proceeds asymmetrically due to the rightward repositioning of the sinuatrial junction and the left‐sided dominance of the umbilical vein. Expansion of the hepatoblasts in the dorsolateral lobes engulfs the vitelline veins and confers a dorsal bend on them (panels B2, C1), while the umbilical veins are engulfed slightly later by the hepatoblasts of the ventromedial lobe (panels A3, A4 and C1, C2). The venous duct and portal sinus develop as left‐to‐right connections between the left umbilical vein in the liver hilum and the right hepatocardiac channel and the right vitelline vein, respectively (panels A3, A4 and C1, C2).

Our findings on early liver development concur with earlier studies where thickening of the ventral wall of the caudal foregut was shown to start at CS10 (~ 4 weeks of development; Severn, 1971; Cascio & Zaret, 1991). Our measurements of the diameter of the yolk sac stalk suggest that the regression of this structure brings about the positional change of the transverse septum from near‐frontal to its typical transverse position. Severn's liver primordium differs markedly in shape from the folded triangle we describe, probably because he did not actually reconstruct it. Furthermore and unlike Thompson (1908), he underappreciated the effects of the marked growth of the liver on its transformation from a flat primordium into a globular bud. In agreement with the experimental finding that hepatoblasts require the presence of VEGFR2‐expressing endothelial cells in the transverse septum to invade it (Matsumoto et al. 2001; Lammert et al. 2003), we and others (Severn, 1972; Zhang et al. 2016) observed a network of capillaries amid the hepatoblast cords in the transverse septum that originated from the venous sinus.

The inclusion of pig and mouse embryos in our study allowed us to describe the intrahepatic course of the vitelline and umbilical veins in the human embryo in topographically unambiguous terms. The basic building plan (Bauplan) of the embryonic mammalian liver consists of a single ventromedial lobe overlying the gallbladder and two dorsolateral lobes flanking the caudal foregut laterally. This architecture is clearly seen in pig (Bradley, 1908) and rodent embryos (Nettelblad, 1954; Kogure et al. 1999; Wong et al. 2012). Although separate dorsolateral liver lobes are not recognizable in human embryos, these parts of the embryonic liver are known as its ‘wings’ (Lewis, 1912).

Most textbooks show the right hepatocardiac channel, from which the portion of the inferior caval vein between the outflow of the hepatic veins and the entrance into the right atrium forms, to originate from the right vitelline vein solely. However, early studies with detailed reconstructions already described the right umbilical and vitelline veins as having a common entrance into the right horn of the venous sinus (Ingalls, 1908). Others described the vessels as the ‘vitello‐umbilical trunks’ (Girgis, 1926; Nettelblad, 1954). Of note, this common distal portion is already identifiable at CS10, or 2 Carnegie Stages earlier than the development of the common cardinal veins in man (Fig. 3). The hepatocardiac channels are longer in pig than in the human embryo and, hence, easier to identify. After CS12, the right‐sided hepatocardiac channel and sinus horn expand at the expense of the left counterparts (Fig. 7D) following the rightward shift of the sinuatrial junction. The asymmetric development of the hepatocardiac channels, therefore, most likely has a hemodynamic origin.

Our reconstructions show that the hepatoblasts do not fragment the vitelline and umbilical veins into sinusoids during or upon their incorporation into the liver, as assumed by the vestigial theory (Lewis, 1904, 1912; Wake & Sato, 2015). Instead, after being engulfed by hepatoblasts, the vessels remained identifiable as large conduits with smooth transitions between the extra‐ and intrahepatic trajectories for ~ 2 Carnegie stages (~ 5 days in man). Moreover, the diameter of the right vitelline vein along the dorsal edge of the liver increased until the main left–right anastomoses (portal sinus and venous duct) had sufficiently developed (cf. figs 4 and 5 in Bradley, 1908). The subsequent disruption of the large right‐sided intrahepatic portion of the vitelline vein coincided with the outgrowth of portal branches from its upstream (caudal) portion and the adjacent portal sinus. Simultaneously, portal branches began to appear from the intrahepatic part of the left umbilical vein, with Cantlie's line forming in between. Downstream in the liver, a network of draining veins developed in retrograde direction and with a pronounced left–right asymmetry from the hepatocardiac channels.

Virtually all textbooks show three intervitelline anastomoses. Only the dorsal and intrahepatic anastomoses are important as the ventral anastomosis is only small and temporary. The dorsal anastomosis accounts for the passage of the portal vein across the duodenum. We consider, in agreement with others (Ingalls, 1908; Lassau & Bastian, 1983), the portal sinus as an intrahepatic left–right anastomosis that lies ventral to the foregut. However, this anastomosis is not intervitelline, as the left vitelline vein was already regressing when this left–right shunt developed. The portal sinus becomes the transverse portion of the left portal vein, while the part of the umbilical vein that remains patent postnatally becomes the umbilical portion of the left portal vein.

The origin of the more upstream (extrahepatic) portion of the portal vein is poorly documented, but this study (Fig. 8) as well as our earlier study (Soffers et al. 2015) clearly document it to be a small left‐sided vein coming from the developing midgut mesentery and draining into the left vitelline vein. The main trunks of the vitelline veins themselves follow the stalk of the yolk sac and anastomose at several places (cf. plate 3 in Ingalls, 1908).

Whereas the incorporation of the vitelline veins into the dorsolateral lobes or wings of the liver was similar in human, pig and mouse embryos at CS12, that of the umbilical veins was not. In all species, hepatoblasts had enclosed the upstream (caudal) portion of the umbilical vein into the ventromedial liver lobe between late CS13 and early CS14. In human embryos, a remnant of the umbilical vein in the lateral body wall, which is sometimes shown in reconstructions (Ingalls, 1908) and often in textbooks, is in fact rarely seen. In pig embryos, remnants in the lateral body wall were seen, which suggests that the incorporation took somewhat longer, whereas in mouse embryos the enclosure on the left side preceded that on the right side by a half day (~ 1 Carnegie stage), so that both vessels could be observed to flank the gall bladder (Fig. 6F). In all species, however, the main events occurred during CS13 and CS14 and the changes in vascular topography were similar.

The downstream (cranial) portion of both umbilical veins ended initially in the hepatocardiac channels, that is, at the junction of the dorsolateral and ventromedial lobes (Fig. S3). As soon as the umbilical veins had been incorporated in the liver, this connection was lost, so that the veins ended in the hilum (Fig. 6F). While the right umbilical vein regressed, the left one formed the portal sinus as a connection with the right‐sided vitelline vein (Ingalls, 1908; Lassau & Bastian, 1983; Mavrides et al. 2001) and the venous duct as a connection with the right hepatocardiac channel. Both connections reflect the hemodynamic conditions in the liver with nearly all blood entering the liver on its left side and leaving the liver on its right side (Ingalls, 1908; Lassau & Bastian, 1983). In fact, a left‐sided venous duct appears to require a persisting left hepatocardiac channel (Yoshinaga & Kodama, 1997).

Except for its portion near and in the liver hilum, the umbilical vein is, like the venous duct, a branchless vessel in the neonate. This characteristic has not yet been explained, but our observations show that the ventral portion of the transverse septum gradually transforms into the relatively thin falciform ligament because the liver expands and the septum does not, and that the venous duct is initially an short intrahepatic vessel that gradually acquires a location in the lesser omentum. This outcome indicates that the extensive and continuing growth of the liver in all directions confers the parts of the umbilical vein and venous duct that were initially located in the transverse septum and lesser omentum, respectively, to the fissures of the respective vessels on the visceral surface of the liver. These findings explain how the portion of the umbilical vein that produces the intrahepatic portal branches during CS14 and CS15 transforms into the (peri‐)hilar region of the liver.

The rare presence of the gallbladder on the right side of the falciform ligament without situs inversus is known since its first description as ‘left‐sided gallbladder’ (Hochstetter, 1886). However, this condition appears to represent persistence of the right instead of the left umbilical vein (Nagai et al. 1997; Shindoh et al. 2012), so that a “right‐sided falciform ligament” is a more appropriate description. In agreement, our findings (Fig. 6F in particular) show that both umbilical veins enclose the gall bladder before joining the portal sinus. Apparently, the right umbilical vein can replace the left umbilical vein in rare instances.

It is generally accepted that the ventral pancreas rotates to become located on the left side of the descending part of the duodenum before fusing with the dorsal pancreas (Odgers, 1930; Adda et al. 1984; Park et al. 1992). However, our reconstructions and measurements unambiguously demonstrate that rotation is secondary to a marked ventrocaudal growth of the duodenum and its dorsal mesentery, while the position of the gall bladder and ventral pancreas remains more or less fixed due to their intimate connection with the liver via the transverse septum (ventral mesentery).

The vitelline and umbilical veins share a common entrance into the venous sinus, become engulfed by hepatoblasts, but remain at least temporally identifiable as large veins inside the liver. Since their fate can be traced, we summarize these events as unfolding according to the ‘lineage theory’. The subsequent asymmetric development of the afferent and efferent intrahepatic vessels probably finds its origin in the repositioning of the sinuatrial junction and accounts for the development of the portal sinus and venous duct as large left‐right intrahepatic anastomoses as formulated in the ‘hemodynamic theory’. Portal branches start to develop from the intrahepatic portion of the large veins only after the establishment of these left‐right shunts. We found no evidence in favor of the ‘vestigial theory’, which postulates fragmentation of the vitelline veins into a sinusoidal network before forming portal and hepatic veins.

The authors declare that they have no competing interests.

Fig. S1. Quality check of image transformation from Amira 3D to Cinema 4D and explanation of performed measurements.

Fig. S2. Three‐dimensional (3D) PDF's of human hepatic venous development.

Fig. S3. 3D PDF's of porcine hepatic venous development.

Fig. S4. Correlation between Carnegie Stages and embryonic ages.

We thank Drs Maurice van den Hoff (AMC) and Marco de Ruiter (LUMC) for giving us access to their institutional series of human embryos. Additional embryos came from the Virtual Human Embryo project of Dr John Cork (Cell Biology & Anatomy, LSU Health Sciences Center, New Orleans; http://virtualhumanembryo.lsuhsc.edu), who made digitized sections available to us. Special thank goes to Els Terwindt (Maastricht University) for her technical assistance. The financial support of ‘Stichting Rijp’ is gratefully acknowledged.