Enhanced Differentiation of Human Embryonic Stem Cells Toward Definitive Endoderm on Ultrahigh Aspect Ratio Nanopillars

Differentiation of human embryonic stem cells is widely studied as a potential unlimited source for cell replacement therapy to treat degenerative diseases such as diabetes. The directed differentiation of human embryonic stem cells relies mainly on soluble factors. Although, some studies have highlighted that the properties of the physical environment, such as substrate stiffness, affect cellular behavior. Here, mass‐produced, injection molded polycarbonate nanopillars are presented, where the surface mechanical properties, i.e., stiffness, can be controlled by the geometric design of the ultrahigh aspect ratio nanopillars (stiffness can be reduced by 25.0003). It is found that tall nanopillars, yielding softer surfaces, significantly enhance the induction of definitive endoderm cells from pluripotent human embryonic stem cells, resulting in more consistent differentiation of a pure population compared to planar control. By contrast, further differentiation toward the pancreatic endoderm is less successful on “soft” pillars when compared to “stiff” pillars or control, indicating differential cues during the different stages of differentiation. To accompany the mechanical properties of the nanopillars, the concept of surface shear modulus is introduced to describe the characteristics of engineered elastic surfaces through micro or nanopatterning. This provides a framework whereby comparisons can be drawn between such materials and bulk elastomeric materials.


Introduction
Human embryonic stem (hES) cells have the potential to generate all cell types in the body, and it is suggested that they could provide an unlimited source for cell replacement therapy FULL PAPER 816 wileyonlinelibrary.com ©2015 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim micro and nanotopographies for neurons [ 22 ] and mesenchymal stem cells. [ 23 ] Taking the mechanical properties of the niche or biomaterials into account may help obtaining effi cient, fully defi ned and xeno-free differentiation protocols to obtain mature functional cells of the endoderm lineage.
The mechanical properties of a surface can be readily modulated using hydrogels, which was originally reported by Pelham and Wang. [ 24 ] They demonstrated the importance of surface rigidity to control epithelial cell migration and spreading. Furthermore, Engler et al. demonstrated how surface stiffness could be linked to the path of differentiation of mesenchymal stem cells. [ 11 ] Trappmann et al. showed that substrates having similar mechanical properties but which varied in porosity showed differences in the ability to direct the differentiation of epidermal stem cells. [ 25 ] Thus, despite the ease of regulating the mechanical properties of gels, the mechanical changes can be associated to other physical differences. In an alternative approach, Chen and co-workers used rubber-like material polydimethylsiloxane (PDMS) micropillars to regulate the effective mechanical properties of substrates. [ 26,27 ] By adjusting the height and diameter of the pillars, it is possible to vary the apparent elasticity in shear sensed by the cells according to the following equation [ 28 ] 3 64 where k is the corresponding spring constant, D denotes the diameter, L is the length, and E is the Young's modulus of the material. In PDMS, assuming a modulus of 2.5 MPa, [ 26 ] a pillar with a diameter of 2 µm gives spring constant from 6454 nN µm −1 (stiffer surface) to 3 nN µm −1 (softer surface) by changing the height from 0.97 to 12.9 µm, respectively. Fu et al. demonstrated that a large spring constant (effectively stiffer material) led to osteogenic differentiation whereas the low spring constant (softer material) led to adipogenic differentiation of human mesenchymal stem cells, [ 26 ] which was in line with the results demonstrated by Engler et al. [ 11 ] Yet, to make an appropriate comparison between the two different model systems, it is necessary to fi nd a common descriptor. Work carried out using, e.g., acrylamide gels or PDMS is typically characterized by the mechanical properties of the substrate using the Young's Modulus, E . Whereas the PDMS micropillar system [ 26,29 ] is characterized by the spring constant, k , of the microfabricated pillars. Thus we propose the following approach. We consider the force applied by the cells to be in-plane of the substrate and thus are better characterized by the effective shear modulus, G, of the substrate. For a bulk system the shear modulus is given by where ν is the Poisson ratio of the material. On the other hand, a material with an array of micro-or nanopillars has an effective shear modulus which can be described by where E is the Young's modulus of the bulk material from which the pillars are made and f is the surface coverage of the pillars. A full derivation of these equations is available in the Supporting Information. Equipped with these relationships it is now possible to make a more direct comparison between the two different model systems which is listed in Table 1 . While PDMS used in research is better defi ned than many hydrogels and is in general biocompatible and can be replaced with FDA (Food and Drug Administration) approved silicones, PDMS or corresponding silicone devices are not amenable for large volume production. By contrast, traditional engineering hard plastics such as polycarbonate (PC), polymethylmethacrylate, and polystyrene are suitable for high volume replication by, for example, injection molding. The challenge, however, is that the Young's modulus of PC is 2-3 GPa as compared to 2-4 MPa for PDMS. In previous work, we have demonstrated how it is possible to manufacture ultrahigh aspect ratio nanopillars in PC by injection molding. Here we fabricated arrays of 100 and 150 nm diameter PC pillars with heights between 500 and 2000 nm. [ 30 ] The corresponding spring constant for such nanopillars range from 5 to 1400 nN µm −1 -similar to that of the PDMS work by Fu et al. [ 26 ] Moreover the softest pillars have an effective shear modulus similar to that of a commonly used acrylamide hydrogel. In the present work we demonstrate that these polycarbonate injection molded nanopillar surfaces can Fu et al. [26] 2000 6100 26 11.4 Fu et al. [26] 2000 970 6454 452 Engler et al. [11] Acrylamide ( E = 40 kPa) ---13.7 wileyonlinelibrary.com ©2015 The Authors. Published by WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim be used to improve differentiation from embryonic stem cells to DE or PE cells. The reported data indicate that this could be a commercially viable way to improve differentiation yield.

Fabrication and Mechanical Measurements of Nanopillar Substrate
Arrays of polymeric nanopillars were fabricated on a 25 mm × 25 mm sample by injection molding as previously described. [ 30 ] The dimensions of the nanopillars and their equivalent mechanical properties are provided in Table 1 together with commonly used model systems from the literature.
The dimensions were characterized by scanning electron microscopy (SEM) (diameter) and atomic force microscopy (height). The corresponding spring constant, of the nanopillars, given in Table 1 , was calculated using Equation ( 1) and the Young's modulus of the PC (2.35 GPa provided at www.bayer.com ). Oblique view SEM images of the different nanopillars can be seen in Figure 1 . Given that the bulk shear modulus of polycarbonate is 650 times higher than that of PDMS, yet it is possible to reduce its effective value by up to 25 000 times through the fabrication of nanopillars.

DE Differentiation on Nanopillars
In order to see possible effects of nanopillars on DE differentiation, the following DE differentiation protocol was employed. Samples coated with fi bronectin were seeded with pluripotent stem cells (Oct3/4 expressing cell, [ 31 ] expanded for 4 d, and subsequently sequentially exposed to Wnt3a and Activin A [ 3 ] to differentiate them toward DE [ 32 ] ( Figure 2 ). The level of differentiation was followed by immunofl uorescent staining for Oct3/4 (hES cell marker) and Sox17 (DE marker) respectively. The total number of cell nuclei was determined by staining the cells with a DNA-binding fl uorescent dye (DAPI). Image analysis was carried out to determine the ratio of Oct3/4 and Sox17 positive cells.
At the point of seeding, all cells were Oct3/4 positive and were uniformly seeded across the entire sample. However, it was clear from samples fi xed one day after seeding (day 1) that the areas with nanopillars exhibited a reduced adhesion of the cells, especially the soft and intermediate nanopillars ( Figure 3 ). Higher magnifi cation images are available in Figure S1 (Supporting Information). Moreover, cells attached to these had a distinctly different morphology from cells on the control surface (this is the fl at area surrounding the nanopillared regions). They exhibited an elongated and aligned morphology with respect to the symmetry of the nanopillar pattern ( Figure 3 ). After 4 d in culture, the hES cells had proliferated extensively, and the cell layer was confl uent on the control surface. However, on the soft and intermediate nanopillars, cells only attached at the edge of the nanopillar area and in a few small tight clusters on the tall nanopillars ( Figure 3 ). This indicated that the soft and intermediate nanopillars arrays did not favor hES (Oct3/positive) cell attachment or proliferation. Subsequently, the DE differentiation was initiated by priming the cells with Wnt3a. One day after Wnt3a priming (day 5) the fi rst Sox17 positive cells started to appear at the border of the confl uent cell layer on the soft nanopillars. Such early Sox17 induction was not observed in the fl at control sample ( Figure 3 and Figure S2, Supporting Information). The DE differentiation protocol was continued with the addition of Activin A for three days. After just one day with Activin A treatment (day 6), the previously empty space on the soft nanopillars was occupied by Sox17 positive cells ( Figure 3 ).
To quantify cell numbers on each pillar array, we used the CellProfi ler [ 33 ] software suite to measure the expression of both markers in each cell. Over 90% of the cells were Oct3/4 positive at day 1 (data not shown), but after 8 d of expansion and differentiation, cells on the softest pillars exhibited a far greater level of Sox17 positive cells as compared to both the control surface (≈50%) and the other nanopillar surfaces (<75%) ( Figure S4, Supporting Information). This combined with images acquired at the interface between the fl at region and the nanopillars ( Figure S2  Already on day 6 nearly 50% of the cells are positive for Sox17 on the soft nanopillars as compared to 0% for the fl at control (Figure 4 ). After 8 d, the conversion is nearly 100% on the softest substrate whereas we only observe ≈50% conversion on the control substrate. From the data, it is clear that the induction of Sox17 positive cells is more effi cient on the nanopillar substrates as compared to the fl at substrate.

Selectivity of a Heterogeneous Cell Population
In order to test if DE cells selectively attach to the nanopillars, a 1:1 mixture of ES cells and predifferentiated DE cells were seeded on the pillar arrays, and fi xed the following day. There appeared to be no signifi cant difference in the number of cells attached on nanopillars and the control surface. However, the ratio between DE and hES cells was higher on the softest nanopillars as compared to the control ( Figure 5 ). It appeared that both Oc3/4 and Sox17 positive cells show tendencies to cluster to the respective cell type on the pillars but not on the fl at surface ( Figure 4 ). This indicates that the DE cells favor the pillars through either an initial attachment preference for DE cells or a subsequent migration during the 24 h incubation.

Pancreatic Endoderm Differentiation
The next differentiation step to PE was also investigated on the nanopillars. To allow an equal starting point for all surfaces, DE cells were produced using a highly effective differentiation protocol [ 9 ] in a cell culture fl ask and subsequently seeded on the nanopillar substrates. DE cells were differentiated to PE, using a 2 week differentiation protocol with bFGF. [ 9 ] Subsequently the differentiated samples were fi xed and stained for the PE markers Pdx1 and Nkx6.1. [ 34 ] Pdx1 and Nkx6.1 positive cells appeared in tight clusters with multiple cell layers ( Figure 6 ). These clusters were signifi cantly smaller on   (Figure 6 ). In contrast to DE differentiation, there appeared to be no correlation between the pillar elasticity and the PE induction ( Figure 6 ). The percentage of Pdx1 positive cells was generally lower on the nanopillars compared to the fl at control surface ( Figure 6 ). This indicated that the PE differentiation required stiffer substrates such as the fl at control surface.

Discussion
Several studies have demonstrated that cellular behavior, including cell differentiation, is affected by the mechanical properties of the environment. [ 11,[17][18][19][20][21][22][23] However, these studies were performed with mesenchymal and adult stem cells. Here we have demonstrated that nanopillars greatly infl uence the Adv. Funct. Mater. 2016, 26, 815-823 www.afm-journal.de www.MaterialsViews.com On day 5, hES cells were located at the edge of the nanopillar area whereas the middle of the nanopillar area was almost empty with only a few clusters of hES cells (Figure 3 ). The previously empty space was occupied by Sox17 positive cells after only one day of Activin A treatment (day 6 from differentiation start). Such a fast repopulation of the nanopillar areas suggests that Sox17 cells migrated toward the nanopillars while the hES cells (Oct3/4 positive cells) were mainly restricted to the rigid control surface. Furthermore, the total cell number on soft nanopillars increased with time while the total number of cells decreased on the surrounding fl at surface (Figure 4 ), which indicates a directed movement of cells from the fl at surface toward the pillar region or cell proliferation. DE cells may have a slightly higher affi nity to the pillar substrates compared to hES cells ( Figure 5 ). Selective cell attachment using nanotopography has previous been reported between different cell types, [ 35,36 ] but to our knowledge it has not been observed between hES cells and hES cell derived DE cells. The selective affi nity alone however cannot explain the high fraction of Sox17 on the soft nanopillars ( Figure 3 ) because the ratio of hES to DE cells is ≈1:20 after the differentiation on soft pillars and only 1:1.29 using predifferentiated cells ( Figure 5 ). Taken together, a reasonable line of events ( Figure 7 ) is that hES cells preferentially grow on fl at surfaces (Figures 3 and 4 ) leaving soft pillar surfaces almost empty. DE cells preferentially grow on softer nanopillars, and a potential mechanism to move into the area is by migration through durotaxis. The latter suggestion is strengthened by the observation that only the really high soft pillars had DE cell accumulation while the shorter harder pillars had an intermediate effect on DE cell accumulation. This suggests that the soft high pillars work as enrichment sites for DE cells. In contrast to the DE cells, PE cells preferred the harder surface indicating a selective advantage of DE cells on the soft pillars.
The observation that the differentiation was initiated very early on the soft nanopillars ( Figure 3 ) indicates that the pillars may also directly and actively improve the outcome of Sox17 positive cells. Sox17 positive cells appeared already after the Wnt3a priming (day 5) on the soft nanopillars and such early induction was not seen on the fl at control surface ( Figure 3 ). Targeting the Wnt signaling using Wnt3a has previously been reported to increase the effi ciency of mesendoderm specifi cation, the transient stage prior the DE stage. [ 3,7,37 ] However, Sox17 is not expressed in mesendoderm but fi rst later in the DE stage proper. [ 3,37 ] This indicated that the soft nanopillars together with the Wnt3a provided suffi cient signals to differentiate the hES cells all the way to the DE stage and to not stop at the transient mesendoderm stage. One possible mechanism could involve the TGF-β pathway. The intracellular proteins YAP (Yes-associated protein) and TAZ (transcriptional coactivator with PDZ-binding motif) are involved in mechanotransduction [ 38 ] and these proteins are situated at the center of the TGF-β signaling pathway. [ 39 ] The TGF-β signaling pathway activated though Nodal/Activin A during the DE formation. [ 40 ] Thus, the soft nanopillars may provide mechanical signals to activate the TGF-β signaling pathway.
The migration capabilities of DE cells are consistent with the literature describing the development of an embryo and in vitro DE differentiation. In the early development of an embryo the DE is generated by epiblast cells undergoing EMT (epithelial to mesenchymal transition). The cells delaminate from the epiblast epithelial layer, break through the embryonic basement membrane and migrate through the primitive streak, where they will go through a mesendoderm transition state and subsequently become DE cells. [ 2,41 ] EMT has also been reported in the in vitro differentiation of hES cells to DE. [ 37 ] Our observations suggest a directed migration of DE cells, which is also observed in vivo during gastrulation. In vivo the migration through the primitive streak is guided by cues from the extracellular matrix and soluble factors. [ 42 ] Adv. Funct. Mater. 2016, 26, 815-823 www.afm-journal.de www.MaterialsViews.com In contrast to the DE differentiation the PE differentiation was repressed by the nanopillars (Figure 6 ). This indicated that the DE and PE differentiation had different mechanical requirements of the environment. In published stem cell differentiation protocols, soluble molecules are added in different concentrations at different time points to mimic the stepwise embryonic development. [ 43 ] Likewise, the physical microenvironment changes during embryonic development, [ 44 ] and hereby the different differentiation steps may require different physical environments. Further the expression of the mechanical cell receptors integrins changes during stem cell differentiation. [ 45 ] Together with our results this suggests that the mechanical requirement for the environment changes throughout the different differentiation steps.
Our results demonstrate that nanopillars with shear modulus constant can be used as model substrates with different stiffness. Previous studies using hydrogels with different stiffness have demonstrated that cellular behavior including stem cell differentiation is infl uenced by the mechanical properties of the cell culture substrate. [ 11 ] However, other papers have suggested that the mechanical properties of the gel are not the only factors regulating fate. [ 25 ] Using a similar approach as Chen and co-workers with pillars to regulate the effective mechanical properties of substrates, [ 26,27 ] we were able to regulate the apparent stiffness of the PC substrate. We manufactured the PC nanopillars sample with injection molding, which is a highly reproducible and high content production method. Furthermore, the nanopillar arrays provide a fully defi ned synthetic surface for differentiation, which can be used for GMP (good manufacturing practices) compliant production of cells. Our fi ndings suggest that DE differentiation can be signifi cantly increased using soft nanopillars. Thus, taking the mechanical properties of the surface into account can improve the differentiation of hES cells. These fi ndings can potentially also be used to develop a surface which can either purify DE cells or improve DE differentiation. One suggestion is to have areas of blank surfaces, which allow hES cells to attach and proliferate, and surrounding surfaces with nanopillars, where the DE cells migrate into.

Conclusion
Using mass produced, nanoengineered substrates we have been able to affect the differentiation of hES cells to DE and PE. More importantly we have observed that by providing a softer substrate interface, DE differentiation rate and effi ciency were signifi cantly higher than on the fl at control surface. However, the underlying mechanism for the surface stiffness effect on the in Adv. Funct. Mater. 2016, 26, 815-823 www.afm-journal.de www.MaterialsViews.com

Experimental Section
Fabrication of the Nanopillars : The fabrication of the UHAR (ultrahigh aspect ratio) nanopillar samples has previously been described. [ 30 ] In brief, nanofabricated master inlays were made by a combination of electron beam lithography and reactive ion etching. The inlays were then used in an injection molding process whereby 100-1000 s of substrates have been made for the experiments. Macrolon OD2015 from Bayer was used was feedstock. Prior to cell seeding the samples were briefl y oxygen plasma treated, 20 s at 40 W, to improve adhesion of the cells. The samples were placed in six-well plates (Nunc) and sterilized with 70% ethanol for 30 min and washed three times with phosphate buffered saline (PBS) (Gibco). Prior cell seeding, the samples were coated with 50 µg mL −1 fi bronectin (Sigma) in PBS for 30 min at 37 °C in the cell incubator. The fi bronectin solution was removed immediately before the seeding of cells.
Cell Culturing and Differentiation : The hES cells (SA121, Takara Bio Europe AB) were cultured in the feeder-free and defi ned DEF-CS500 system according to instruction from the supplier (Takara Bio Europe AB). The hES cells were cultured and differentiated in a humidifi ed incubator with 5% CO 2 .
For the differentiation toward defi nitive endoderm on the nanopillars, the following differentiation protocol was used. Cells were dissociated into single cell suspension with TrypLE Select (Invitrogen) and seeded at 40 000 cells cm −2 in DEF medium with 5 × 10 −6 M RockI (Sigma-Aldrich). Cells were kept undifferentiated for 4 d in DEF medium. The differentiation was initiated on day 4 by rinsing with RPMI16040 + Glutamax medium (RPMI -Roswell Park Memorial Institute) (Gibco) and the addition of 100 ng mL −1 recombinant mouse Wnta3 (R&D Systems) in basal medium, containing RPMI medium with 2% B27 minus insulin (Gibco) and 0.1% Pen Strep (Invitrogen). On day 5, the cells were rinsed with RPMI medium and 10 ng mL −1 Activin A (Peprotech) in basal media was added. Fresh basal media with 10 ng mL −1 Activin A was added on day 6 and 7 and the differentiated was terminated on day 8. For the differentiation toward pancreatic endoderm the following protocol was used. The cells were fi rst differentiated to defi nitive endoderm in fi bronectin coated cell culture fl ask using a highly effective patented protocol (WO 2012175633 A1). Subsequently, the cells was rinsed with PBS and dissociated to single cell suspensions with TrypLE Select at room temperature. Seeding of the cells was performed with 200.000 cells cm −2 in basal media with 100 ng mL −1 Activin A and 5 × 10 −6 M RockI. The cells was differentiated toward pancreatic endoderm with a 13 d published protocol with RPMI medium containing 64 ng mL −1 FGF2 (Peprotech), 12% knockout serum replacement (Gibco) and 0.1% pen strep. [ 9 ] Media was changed daily.
The attachment assay with heterogeneous cell population containing undifferentiated hES cells and DE cells was carried as follow. DE cells were derived with the patented differentiation protocol described above. DE and undifferentiated hES cells was mixed in the 50:50 ratio and seeded at 200.000 cells cm −2 in basal media with 100 ng mL −1 Activin A and 5 × 10 −6 M RockI. The cells were incubated in the cell incubator overnight.
Images were acquired using an Olympus CX41 upright microscope equipped with a prior motorized stage and 10× or 4× objective, operated by ImageProPlus (Media Cybernetics, UK). A total of 1026 images were processed, creating a dataset containing over 400 000 cells. The nuclear size and shape, along with intensity of DAPI, OCT4, and SOX17 markers within each nucleus, was measured using CellProfi ler software suite (V2.1, Broad Institute, Harvard). Processing the full dataset took ≈6 h on an Intel Core i7 2600 CPU (central processing unit) @ 2.4 GHz with 16 Gb DDR2 RAM (random access memory) -this included various image measurements which were not used in the fi nal analysis. The data were exported for processing and plotted using MATLAB.
Statistical Analysis : All values were presented as mean ± S.E.M. Data are presented for experiments performed at least as three independent experiments. For statistical analysis, ANOVA (analysis of variance) was used and a p -value of 0.05 was considered statistically signifi cant.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.