Correspondence to: Daniele ZINK, Institute of Bioengineering and Nanotechnology, 31 Biopolis Way, The Nanos, Singapore 138669, Singapore. Tel.: +65 6824 7107 Fax: +65 6478 9565 E-mail: firstname.lastname@example.org
The generation of tissue-like structures in vitro is of major interest for various fields of research including in vitro toxicology, regenerative therapies and tissue engineering. Usually 3D matrices are used to engineer tissue-like structures in vitro, and for the generation of kidney tubules, 3D gels are employed. Kidney tubules embedded within 3D gels are difficult to access for manipulations and imaging. Here we show how large and functional human kidney tubules can be generated in vitro on 2D surfaces, without the use of 3D matrices. The mechanism used by human primary renal proximal tubule cells for tubulogenesis on 2D surfaces appears to be distinct from the mechanism employed in 3D gels, and tubulogenesis on 2D surfaces involves interactions between epithelial and mesenchymal cells. The process is induced by transforming growth factor-β1, and enhanced by a 3D substrate architecture. However, after triggering the process, the formation of renal tubules occurs with remarkable independence from the substrate architecture. Human proximal tubules generated on 2D surfaces typically have a length of several millimetres, and are easily accessible for manipulations and imaging, which makes them attractive for basic research and in vitro nephrotoxicology. The experimental system described also allows for in vitro studies on how primary human kidney cells regenerate renal structures after organ disruption. The finding that human kidney cells organize tissue-like structures independently from the substrate architecture has important consequences for kidney tissue engineering, and it will be important, for instance, to inhibit the process of tubulogenesis on 2D surfaces in bioartificial kidneys.
Tubule formation in the developing mammalian kidney is initiated by outgrowth of the ureteric bud from the Wolffian duct (early kidney development is reviewed in [1, 2]). The outgrowing ureteric bud invades the metanephric mesenchyme and undergoes branching morphogenesis, which gives rise to a network of epithelial tubes. This tubular system develops into the collective duct network. Inductive processes lead to the condensation of the metanephric mesenchyme around the tips of the ureteric bud-derived tubules. The condensed mesenchyme polarizes, and the cells transdifferentiate into epithelial cells by a mesenchyme-to-epithelial transition process. In the course of these processes, new epithelial tubules are formed, which give rise to the nephrons.
The formation of kidney tubules in vitro is studied by using 3D gels [3–12]. The most widely used in vitro system consists of Madin–Darby canine kidney (MDCK) cells grown in 3D gels consisting of collagen I and/or matrigel [5–8, 12]. A general drawback of such 3D gel-based systems is that high-resolution imaging of intact functional tubules within gels is difficult. In addition, tubules embedded in 3D gels are difficult to access, and manipulations or applications of drugs cannot be performed in a well-controlled manner. These challenges limit the usefulness of in vitro generated kidney tubules in functional studies and applications. One major area of interest for applications of in vitro generated kidney tubules lies in in vitro nephrotoxicity studies.
The proximal tubule, which has a length of ∼13 mm in the adult human kidney, is one of the major target sites of nephrotoxic agents. Animal- or human-derived proximal tubule cells are widely applied for in vitro nephrotoxicology [13–20]. Monolayer cultures are predominantly used in such studies, which give only limited insights into drug-induced effects on tubular functions and integrity.
Proximal tubule cells are also applied in kidney tissue engineering, particularly for the development of bioartificial kidneys (BAKs) [21–29]. BAKs take advantage of hollow fibre cartridges, where proximal tubule cells grow on the insides of the hollow fibre membranes. Here they should form and maintain a well-differentiated epithelium, in order to mimic the architecture and replace the functions of the proximal renal tubule.
Herein we show that primary human renal proximal tubule cells (HPTCs) do not remain organized as a flat epithelium on 2D surfaces and in tubular substrates, but form large (lengths of up to several millimetres) and functional renal tubules. The process of tubule formation and the properties of the resulting tubules are described. The consequences of the findings for BAK development, as well as potential applications of the in vitro system described here, are discussed.
Materials and methods
Different batches of HPTCs were obtained from ScienCell Research Laboratories (Carlsbad, CA, USA). They were cultured in basal epithelial cell medium supplemented with 2% foetal bovine serum and 1% epithelial cell growth supplement (ScienCell Research Laboratories). In some experiments, transforming growth factor (TGF)-β1 (R&D Systems, Minneapolis, MN, USA) was added at a concentration of 10 ng/ml after monolayer formation. Cells were cultured on uncoated multiwell plates (Nunc, Naperville, IL, USA) or plates coated with human laminin or other extracellular matrix (ECM) coatings as described in  (see also Fig. S1). The seeding density was 5 × 104 cells/cm2, unless otherwise indicated.
The wells of diagnostic printed slides have a diameter of 2 mm and a glass bottom. No ECM coating was applied in these experiments. Cells were seeded at a density of 2.65 × 105 cells/cm2 into glass capillaries (inner diameter = 0.58 mm; Sutter Instrument, Novato, CA, USA), and kept in static culture. Only the ends of the capillaries were examined.
Matrigel (BD, Franklin Lakes, NJ, USA) was applied according to the manufacturer’s instructions.
Tubules were treated (if necessary for the experiment) and fixed (3.7% formaldehyde in phosphate buffered saline, 10 min. at room temperature) after their appearance. Most tubules appeared during week 2 after cell seeding (Fig. S1). Tubules were imaged immediately or during the following days after fixation and were stored in phosphate buffered saline or mounting medium.
Immunostaining, histochemistry and transport assays
Histochemical detection of γ-glutamyl transpeptidase (γGTP) activity, formaldehyde fixation and immunostaining were performed as outlined previously . Transport assays were performed as described in , and more details are provided in the Supporting Information.
Reverse transcription and quantitative real-time PCR
Details of RNA isolation, reverse transcription procedures and primer sequences (Table S1) are provided in the Supporting Information. Quantitative real-time polymerase chain reaction (qRT-PCR) was performed by using the iCycler system and software (BioRad, Hercules, CA, USA). Gene expression levels were calculated relative to the expression levels of the house keeping gene glyceraldehyde 3-phosphate dehydrogenase (GAPDH) using the BioRad software.
Imaging, statistics and software
Imaging was performed with a Zeiss AxioObserver Z1 microscope (Carl Zeiss, Jena, Germany) using the Zeiss AxioVision imaging software. Calculations and statistics (unpaired t-test) were performed with Excel 2003. Figures were arranged with AdobePhotoshop CS3 and ImageJ.
Human renal tubule formation on 2D surfaces
In a previous study, we investigated the impact of ECM coatings on the formation and maintenance of differentiated epithelia by HPTCs in vitro. After the initial formation of well-differentiated epithelia, subsequent disruption of the epithelia was consistently observed. α-smooth muscle actin (SMA)-expressing myofibroblasts, which appeared in increasing amounts during the culture period, seemed to play a crucial role. Herein we analysed these processes in further detail.
In accordance with our previous results, we observed initial formation of a flat and well-differentiated epithelial monolayer when HPTCs were cultured in multiwell plates (Fig. 1A). Subsequently, increasing amounts of α-SMA-expressing myofibroblasts appeared. These myofibroblasts formed large aggregates (Fig. 1B). In the surroundings of such aggregates, the epithelium became reorganized. Firstly, highly coordinated and simultaneously directed movements of large numbers of cells led to retraction of the monolayer on one side of myofibroblast aggregates, leaving behind the largely empty surface of the well (Fig. 1C, E). Subsequently, the monolayer retracted on the other side of the myofibroblast aggregates (Fig. 1D, F). These highly coordinated cell movements led to the formation of stripes of cells, with a length of up to several millimetres or even centimetres (Fig. 1F). The stripes included the myofibroblast aggregates.
The cells that have organized into a stripe then performed additional dynamic reorganizations, which gave rise to tubule formation (Fig. 1G, H). The human renal tubules formed in this way on 2D surfaces were straight and not branched, and typically have a length of several millimetres (Fig. 1G). The tubules always remained attached to myofibroblast aggregates, which could be associated with one (Fig. 1G) or both ends of a tubule, but could also be found at mid-tubular regions. When an end of a tubule was not attached to a myofibroblast aggregate, the tubular epithelium was continuous with the remainder of the monolayer (Fig. 1G, H). The finally formed tubular epithelium and attached epithelia were well differentiated (Figs 1H and 3C, D).
Cell movements involved in monolayer reorganization and tubule formation were not only highly coordinated, but also rapid. Figure 2 shows that condensation and folding up of the cell stripe, which ultimately led to tubule formation, occurred within minutes. We were not able to determine definitively how the tubule was folded from the stripe. This would require extensive live cell tracking and detailed analyses of fast movements of large numbers of cells. Such extensive live cell analyses would be performed in a separate study.
After tubule formation, the structures remained dynamic. Myofibroblast aggregates and attached tubules further rearranged, and eventually large cell aggregates developed, which detached and floated away. Most of the tubules disappeared within a period of three days (Fig. S1D), and the duration for which tubules remained stable was ∼2–9 days.
Tubules have a lumen lined by a differentiated epithelium and display transport functions
Sectioning of the tubules formed by the processes described above confirmed that the tubules enclosed a lumen (Fig. 3A, B). The lumen was lined by a well-differentiated epithelium displaying extensive tight junction formation (Fig. 3C). Usually, some α-SMA-expressing cells were associated with the tubular structures. These could be located on the outside of the tubule or within the lumen. The brush border marker γGTP was expressed within the tubules (Fig. 3D), confirming cell type-specific differentiation and apical–basal polarity of the epithelial cells.
In order to test whether the tubular epithelium displayed typical transport functions, we incubated tubules with the fluorescent organic anions lucifer yellow, rhodamine 123, 5,6-carboxyfluorescein and 4,4-difluoro-5,7-dimethyl-4-bora-3a,4a-diaza-s-indacene-3-propionic acid (BODIPY FL) verapamil. Lucifer yellow and 5,6-carboxyfluorescein are substrates of the p-aminohippurate transport system [31, 32]. Stronger fluorescence was observed within the tubular lumen, as compared to the surrounding medium and the epithelial cells lining the lumen. Thus, these organic anions became enriched in the tubular lumen (Fig. 4), providing evidence for transport across the tubular epithelium. Also, rhodamine 123, which is a substrate of the multidrug resistance-1-encoded P-glycoprotein (P-gp) transport system, became enriched in the tubular lumen (Fig. 4). Rhodamine 123 is actively transported, whereas BODIPY FL verapamil is transported by the P-gp system via electrodiffusive anion transport [33, 34]. BODIPY FL verapamil became enriched within the cells (Fig. 4), suggesting transport of this substrate at the basolateral sites into the cells, but slower or no transport at their apical sites. Whether this reflected impairment of the P-gp mediated electrodiffusive anion transport at the apical sites of the cells was not clear, as the kinetics and exact routes of BODIPY FL verapamil transport in the native human proximal tubule were not characterized. Together, the results showed that various organic anions were transported across the tubular epithelium, and suggested that at least two different major transport pathways were functional.
Tubule formation by HPTCs on 2D surfaces and in 3D gels
In order to address tubule formation by HPTCs in 3D gels, we cultured the cells in matrigel. Tubule formation in matrigel (Fig. 5E–H) involved branching of cells at the initial stages of tubule formation and outgrowth of branched cells. Outgrowing branches then formed tubules, and budding from these tubules could occur, giving rise to branched tubular structures (Fig. 5H). It was important to note that formation of epithelial monolayers and coordinated movements of large numbers of cells were not involved in tubule formation in 3D gels. The resulting human proximal tubules obtained in 3D gels were relatively small, displayed multiple branches (Fig. 5H), could be convoluted (Fig. S2), and were never attached to myofibroblast aggregates, in contrast to the tubules obtained on 2D surfaces.
In order to test whether matrigel had an influence on the process of tubulogenesis, we have investigated tubule formation in matrigel-coated multiwell plates, where HPTCs grew on top of the matrigel coating. Under these conditions, tubule formation occurred in a similar way as observed before on 2D surfaces (Fig. 5A–D). Generally, we found that different ECM coatings consisting of laminin, collagen IV, a mixture of these components or other components could influence the timing of monolayer reorganization , and the extent of myofibroblast aggregation and tubule formation (Fig. S1). However, when tubules were formed on 2D surfaces, they always occurred by the same process as illustrated in Figures 1 and 5A–D, regardless of the ECM coating used (data not shown). Tubulogenesis occurred also on uncoated 2D surfaces (Fig. S1), when HPTCs from a given donor grew and survived under such conditions (growth and survival of HPTCs on uncoated surfaces shows interdonor variability).
Together, the results showed that HPTCs form small and branched tubules in 3D gels by a process of budding and branching morphogenesis, whereas large and straight tubules are formed on 2D surfaces by a process involving large-scale reorganizations of epithelial monolayers and interactions between epithelial cells and myofibroblasts. These findings suggest that the different morphologies of the tubules observed depend on the different processes of tubule formation involved.
Tubule formation on 2D surfaces is enhanced by a 3D architecture
Although renal tubule formation occurred on 2D surfaces, several observations suggested that formation of 3D tissue-like structures by HPTCs was enhanced by a 3D substrate architecture. Most striking was the finding that a closed circle formed by several tubules with attached myofibroblast aggregates could be formed within a well close to its edge, with all tubules displaying a similar distance to the edge (Fig. 6A, B). By contrast to the centre of the well, the edge of the well has a 3D architecture. Another observation suggesting a crucial role for a 3D architecture in enhancing tubule formation was that initial retraction of the epithelial monolayer started in most cases first at the edge of the well (Fig. 6C, D).
In order to determine whether the presence of a 3D edge indeed enhanced the initiation of tubule formation, we seeded cells in parallel on 18 mm cover slips (no 3D edge), 24-well cell culture plates (with 3D edge) and diagnostic printed slides (with 3D edge). Figure 7 shows that initial retraction of the epithelium first occurred at the edges of the wells of the 24-well plates (well diameter = 15 mm) and diagnostic printed slides (well diameter = 2 mm). Subsequently, monolayer reorganization as well as cell aggregate and tubule formation were observed within the wells of these devices. By contrast, the epithelial monolayer was not reorganized on cover slips during the monitoring period of 8 days, and no retraction of the monolayer occurred at the edges of the cover slips (Fig. 7).
The surface material was glass in the case of cover slips and diagnostic printed slides, whereas 24-well plates consisted of tissue culture plastic. Furthermore, these items are different in sizes, and thus the numbers of cells that could be involved in the reorganization processes on the different surfaces were also different. The results suggested that tubule formation was not dependent on the material, surface area and cell numbers involved, but was only dependent on the presence of a 3D edge.
It is currently unclear how HPTCs sense the edge. An important parameter could be lateral attachment of the outermost cells to the rigid wall of the well. In order to find out whether cells might sense the rigid wall, we seeded HPTCs into glass capillaries. In glass capillaries, HPTCs could only attach with their basal sides to the rigid substrate, but not with their lateral sides. Furthermore, it was important to find out whether the process of tubule formation might be inhibited if the cells were already arranged into a tubular architecture. This question is of major practical importance, because HPTCs are applied in BAKs, where they grow within hollow fibre membranes. Tubule formation within such hollow fibre membranes would be a major challenge in BAKs.
Figure 8 shows that HPTCs formed tubules within capillaries. These results demonstrated that tubule formation was not inhibited by arranging HPTCs into a pre-formed tubular architecture. Tubule formation within capillaries was accomplished by the same process as employed on 2D surfaces, involving monolayer formation and subsequent appearance of myofibroblast aggregates. Obviously, this process would lead to hollow fibre clogging in BAKs. The results also revealed that lateral attachment to a rigid substrate was not important for the sensing of a 3D environment. Three-dimensional edges were probably sensed indirectly through the impact of the substrate architecture on surface tension and fluid dynamics of the medium .
TGF-1 induces the initial steps of human renal tubule formation on 2D surfaces
Besides substrate architecture, interactions between epithelial cells and myofibroblasts appeared to be important for tubule formation. Unresolved questions included where the myofibroblasts were derived from, whether their presence was due to contaminations of the epithelial cells and what role they played in tubule formation. Figure 9C shows that α-SMA-expressing cells were not detectable by immunostaining in the initial cultures of HPTCs. The expression levels of α-SMA, as determined by qRT-PCR, were very low in such initial cultures, and were not higher than the expression levels in HeLa cells or human embryonic kidney (HEK) 293 cells (Fig. 9A). HeLa cells are negative for α-SMA , and HeLa as well HEK293 cells are well-established epithelial cell lines free of contaminations with other cell types.
By contrast, large amounts of α-SMA-expressing cells could be detected by immunostaining after maintaining such initially myofibroblast-free HPTC cultures for 4 weeks under in vitro conditions (Fig. 9D). Accordingly, the expression levels, as detected by qRT-PCR, increased (Fig. 9B). As kidney epithelial cells can transdifferentiate into myofibroblasts under in vitro conditions by an epithelial-to-mesenchymal transition process , the most likely explanation for the appearance of myofibroblasts in the initially myofibroblast-free HPTC cultures is transdifferentiation of epithelial cells into myofibroblasts by an epithelial-to-mesenchymal transition process. The data also indicate that the initial cultures of HPTCs were not contaminated with α-SMA-expressing myofibroblasts. As tubulogenesis was initiated in the vicinity of myofibroblast aggregates, we examined which types of myofibroblast-derived signalling molecules were expressed in our in vitro cultures (Fig. 10A). The expression levels of myofibroblast-derived growth factors were determined by qRT-PCR in parallel with the expression levels of the myofibroblast marker α-SMA. In vitro cultures were monitored over a time period 4 weeks. Typically, myofibroblast aggregates appeared after 1–2 weeks of in vitro culture (Fig. S1). Cell loss could occur in the course of reorganization events due to detachment of huge cell aggregates and epithelial sheets.
The results revealed that keratinocyte growth factor and hepatocyte growth factor were not expressed or were expressed at very low levels. In contrast, substantial expressions of TGF-β1, leukaemia inhibitory factor and fibroblast growth factor 2 were observed, along with the expression of α-SMA. In all of these cases, a massive increase in expression levels was observed after 1 week of in vitro culture. This was in accordance with the observation that myofibroblasts appeared after 1–2 weeks of in vitro culture, and the expression levels of α-SMA and TGF-β1 remained high until the second week. The expression levels of leukaemia inhibitory factor and fibroblast growth factor 2 dropped after week 1, suggesting down regulation. The levels of TGF-β1 and α-SMA expression were significantly lower at week 3 (as compared to week 2), and this might reflect down regulation and/or cell loss. The expression levels of TGF-β1 significantly increased again at the end of the monitoring period (week 4).
It has been described before that TGF-β1 triggers the initial steps of tubule formation on 2D surfaces, namely the formation of a condensed stripe or cord of cells from a monolayer . As these previous experiments were performed with rabbit cells, we tested here whether primary human cells reacted in a similar way. Indeed, treatment with TGF-β1 led to rearrangement of the monolayer into a condensed stripe of cells and to the formation of cell aggregates (Fig. 10B–D). This result showed that TFG-β1 induced the initial steps of human renal tubule formation on 2D surfaces.
Herein we showed that HPTCs form large and functional human renal tubules on 2D surfaces. Tubulogenesis on 2D surfaces involved large-scale reorganizations of epithelial sheets around myofibroblast aggregates. Budding and branching morphogenesis, typically occurring in 3D gels [3–12], did not play a role, and these processes are also not involved in proximal tubule formation in vivo[1, 2, 39]. Tubulogenesis on 2D surfaces was induced by TGF-β1, which was probably released in the in vitro system by myofibroblast aggregates. It is well documented that myofibroblasts release TGF-β1, but a possible contribution of epithelial cells to the increased levels of TGF-β1 expression observed must be carefully addressed in future studies.
Increased expression of TGF-β1 and appearance of myofibroblasts are a normal response to kidney injury [41–45], and might reflect in the cell culture system described here a response of HPTCs to organ disruption and artificial in vitro conditions. Remarkably, functional tissue-like structures were reorganized in this in vitro system after the appearance of myofibroblasts and increased TGF-β1 expression. It is thought that myofibroblasts and TGF-β1 have important roles in tissue regeneration after injury [41, 42], apart from their prominent roles in fibrosis, which is difficult to study in vivo. The in vitro system described here allows us to study in detail the roles of TGF-β1, myofibroblasts and other factors in proximal tubule regeneration. It might also be interesting for the screening of anti-fibrotic drugs, which should inhibit accumulation of myofibroblasts and down regulate TGF-β1. Furthermore, this in vitro system allows us to investigate in vitro how tissue-like structures are generated by different interacting cell types in the absence of an organizing matrix.
Our finding that the addition of TGF-β1 induced the initial steps of tubulogenesis on 2D surfaces is in accordance with the results of a previous study performed with rabbit cells . Although the role of TGF-β1 in tubulogenesis on 2D surfaces would be further underlined by showing that anti-TGF-β1 antibodies inhibit tubule formation, we were not able to perform such experiments, simply because the cells consistently died after antibody application (antibodies from three different companies were tested). The reasons for cell death are unclear, but probably the HPTCs used require low amounts of TGF-β1 for survival. Nevertheless, we found that tubule formation on 2D surfaces is inhibited by bone morphogenetic protein 7, which is another growth factor from the TGF-β superfamily that counteracts TGF-β1-induced effects [46, 47]. Due to space limitations, the factors and conditions inhibiting the processes described here are outlined in detail elsewhere (Farah Tasnim, Jackie Y. Ying and Daniele Zink, unpublished data).
Inhibition of tubule formation would be particularly important in BAKs using hollow fibre membranes seeded with proximal tubule cells [21–29]. So far, this problem has not been addressed in studies on BAK development. One article published in 1992 describes tubule formation on 2D surfaces . However, that study, which uses rabbit proximal tubule cells, only outlined that tubule formation on 2D surfaces depends on exogenously added TGF-β1, epidermal growth factor and retinoic acid. Spontaneous occurrence of this phenomenon, which is the major cell-related problem in BAK development in our experience ( and data not shown) and is likely induced by TGF-β1 released by the cells, was not described. This might be one of the reasons why spontaneous tubule formation has not been recognized, so far, as a problem in BAK development.
Proximal tubules generated on 2D surfaces are interesting for a variety of applications. In contrast to renal tubules generated in 3D gels, the tubules generated on 2D surfaces were easily accessible for manipulations, and drugs could be applied in a highly controlled manner. In addition, renal tubules exposed on 2D surfaces could be easily imaged by high-resolution light microscopy. These properties make this in vitro model interesting, for instance, for applications in tubular transport studies and in vitro nephrotoxicology. In vitro generated human proximal tubules might be a physiologically more relevant test system than monolayers of animal or human renal tubule cells, which are currently widely used for in vitro nephrotoxicology [13–20].
The authors thank Dr. Pei-Yong Chow, Dr. Sean Liour, Dr. Karthikeyan Kandasamy, Joscha Muck and Yong Yeow Lee for valuable help and discussions, and for providing materials. This work is supported by the Institute of Bioengineering and Nanotechnology (Biomedical Research Council, Agency for Science, Technology and Research, Singapore).
Conflict of interest
The authors confirm that there are no conflicts of interest.