Arguably, some limitations of existing biomolecule gradient systems lay in the nonphysiological makeup of the substrate on which the gradient-exposed cells are cultured, as well as the fact that many gradients are soluble rather than substratetethered. Improvements of the physiological relevance of micron-scale gradient systems have resulted in approaches to tether gradients, for example via micro-contact printing of discrete cell-guidance patches (51), photo-polymerization of hydrogels containing a microfluidically generated gradient (52,53), or adsorption of laminin gradients onto cell culture plastic (54). However, the generation of in vitro model systems composed of independent overlapping gradients has been challenging. To address this technology gap, we have recently developed a versatile poly(ethylene glycol) (PEG) hydrogel system to capture from solution microfluidically generated gradients of biotinylated and/or Fc-tagged fusion proteins via NeutrAvidin or protein A, respectively, displayed on the gel surface (Figure 3C). The selectivity and orthogonality of the chosen protein binding schemes enabled the independent formation of parallel and orthogonal overlapping gradients of multiple proteins (55). The presentation of overlapping gradients on biomimetic substrates should expand the possibilities for studying a wealth of in vitro biological questions, and potentially enable high-throughput investigation of combinatorial effects of biomolecules on cell fate.

Patterned 3-D niche models

So far, we have focused on 2-D stem cell culture systems. Most stem cell niches, however, are 3-D microenvironments composed of hydrated, crosslinked networks of ECM proteins and sugars (Figure 1A). The three-dimensionality can change cell behaviors and induce striking differences in cell shape, proliferation, migration, and differentiation (4,56,57). In typical 3-D cultures, cells are randomly embedded in scaffolds either formed from naturally derived ECM components such as collagen I, Matrigel [i.e., a gel formed from a complex mixture of laminin, collagen IV, heparan sulfate proteoglycans, and growth factors (58)], or synthetic polymers. However, efforts are underway to recapitulate the heterogeneous 3-D architecture of stem cell niches using microfluidics and micromolding approaches. Micromolding typically involves the replication of gel surfaces against a ‘positive’ micro-structured stamp, and the subsequent layer-by-layer assembly of hydrogels and cells (59,60). Whereas this method can be used to create well-defined but not interconnected cellular structures in one focal plane, compared with microfluidics it is less versatile to generate spatially modular microenvironments. Gillette et al. fabricated microfluidic scaffolds from interpenetrating networks (IPN) of collagen I and either alginate, Matrigel, or fibronectin to pattern cells and collagen I matrices in 3-D (61). ‘Doping’ the IPN with collagen induced the nucleation of the microfluidic collagen I phase at the interface to the IPN, which ensured the proper crosslinking of the two ECM phases and eliminated the risk that contractile forces of seeded cells could destroy the intended geometry. Others have fabricated multilayered hydrogel structures in an array of hexagonal posts that provide support and contain the gels during the injection process by balancing capillary forces and surface tension (62) or using microfluidic devices (63). Notably, microfluidically generated or micromolded 3-D gel structures enable the fabrication of multicellular, tissue-like structures exhibiting important physiological behaviors such as epithelial-to-mesenchymal transitions (60,61) and can be combined with microfluidic devices to generate 3-D gradients (64,65). These are impressive advances, but the spatial resolution of microfluidic patterning is currently limited and this technique does not allow the dynamic modification of the 3-D matrix properties.

An alternative approach addressed these limitations by incorporating photolabile building blocks into a PEG hydrogel (66). The resulting hydrogel could be cleaved partially or completely by exposure to UV light, allowing the rapid prototyping of 3-D patterns at micron resolution using a two-photon laser-scanning microscope. hMSCs responded to locally induced changes in stiffness and availability of photolabile cell adhesion ligands. Natively, hMSCs produce the adhesion protein fibronectin and corresponding integrins. In differentiating cells, the secreted ECM is extensively remodeled by a gradual upregulation of collagen II and glycosaminoglycan (GAG) synthesis. By removing the photolabile fibronectin-derived peptide RGDS after 10 days of culture, this dynamic change in ECM composition could be efficiently mimicked and led to a significant increase of chondrogenic differentiation.

The approaches discussed above represent important advances in the fabrication of patterned 3-D cell culture systems. In combination with advanced, cell-instructive biomaterials (58), these technologies are anticipated to make exciting contributions to the field of stem cell biology. However, 3-D cell culture remains challenging because the sparse and homogeneous distribution of cells within a gel matrix may involve demanding imaging requirements. Another drawback of current 3-D patterning methods is the inefficient alignment of micropatterned protein structures to cells at a single cell level. Several groups have explored dielectrophoresis (DEP) to handle and pattern cells in 3-D (67,68,69). DEP relies on the migration of cells in an externally applied heterogeneous electrical field and hence allows immobilizing cells into a hydrogel after cell patterning. However, the toxicity of some buffers and heat generation remain challenges of DEP for 3-D cell patterning (18,70). Other techniques such as acoustic traps or optical tweezers (71), in combination with appropriate microfluidic systems and biomaterials may be used alternatively to align cells and protein patterns in 3-D.

Microwell arrays as pseudo–3-D niche models

Because of the above challenges in precisely controlling cell and protein distribution in 3-D matrices, as well as the laborious imaging of cells in 3-D, the use of microwell arrays as engineered pseudo–3-D microenvironments has emerged as an alternative strategy to study single stem cells or multicellular colonies (72). Microwell arrays are topographically structured surfaces with hundreds to thousands of miniature cavities, arrayed into a regular grid. A wealth of methodologies has been developed to fabricate microwell arrays with microwell diameters ranging from tens to hundreds of micrometers. Cells are trapped by gravitational sedimentation and hence the number of cells per microwell can be controlled by the cell seeding density and microwell diameter. Due to the stochastic capturing process, the number of single cells per trap follows a Poisson distribution. The maximal frequency of microwells containing only one single cell is ~30–40%. However, microwells containing more than one cell at the onset of an experiment can be efficiently eliminated by retrospective image analysis.

Just as micropatterned ECM adhesion sites allow the control over cell spreading in 2-D (33), the capture of single cells in sufficiently small microwells can allow a quasi–3-D control over cell shape and spreading (73). For example, Kurth et al. recently probed the influence of microwell size on human hematopoietic stem cell (HSC) fate (74). Single HSCs were seeded on fibronectin-coated microwell arrays. Small microwells that nearly encircled single cells led to decreased proliferation and differentiation, quantified by the expression of the HSC marker CD34. Increasing the microwell diameter resulted in faster proliferation and lower expression of CD34, indicating that the adhesive cell-matrix contact area—and consequently the number of engaged integrins—could be involved in regulation of HSC quiescence.

Production of homogeneous stem cell colonies

Microwell arrays also allow the clonal growth of stem cells (such as ESCs) into larger 3-D cell colonies, termed embryoid bodies (EBs). These multicellular spheroids mimic some of the early stages of embryonic development, including an initial formation of the three germ layers. The traditional production of EBs by scrapping ESC monolayer culture results in heterogeneous mixtures of different shapes and diameters of EBs, which negatively affects experimental data and reproducibility (75). The alternative technique, the hanging drop culture, is based on the aggregation of cells at the tip of a single drop hanging on a lid of a Petri dish. Due to a tight control over the initial cell number per drop, this technique produces very homogeneous EBs. However, hanging drop cultures are laborious and practically preclude automation and high-throughput handling. Microwell arrays, due to their defined diameter, allow a simple choice over the initial cell number in a single spheroid and hence afford control over EB size. Microwell arrays fabricated from non-adhesive culture substrates— such as those made of PEG (76,77), agarose (78), or poly-dimethylsiloxane (PDMS) (73,75)—completely restrict cell migration, eliminating merging of multiple spheroids. These reproducibly fabricated and size-controlled EBs exhibit a more homogeneous differentiation pattern than ESC aggregates produced by traditional methods (75). Indeed, the size-controlled EB formation in microwell arrays led to insights on how EB diameter can affect differentiation. For example, it was found that cardiac— but not endothelial—differentiation was primarily induced in larger EBs. In contrast, small EBs resulted in significantly higher endothelial cell differentiation with reduced cardiogenesis (79).

The need to homogenize cell spheroid size is not only crucial in ESC biology, but also for various other stem cells such as neural or mesenchymal stem cells (80,81), as well as in cancer cells (82). Multicellular tumor spheroids, for example, reflect some of the physiological properties of metastasis and can acquire radio- and chemoresistance to apoptosis-inducing drugs, mimicking the resistance found in solid tumors. The ongoing integration of microwell arrays into classical cell culture platforms [such as multi-titer plates (83,84,85)] should make microwell arrays a useful tool for many fundamental studies in stem cell biology as well as applications in drug screening.

Analyses of single stem cell fates

Stem cells are inherently heterogeneous cell populations. For example, although HSCs can be isolated with relatively high purity, they show distinct reconstitution patterns in single-cell transplantation assays (86), the gold standard to prove stem cell function. A similar heterogeneity can be found in ESCs with regard to the expression levels of the pluripotency marker Nanog. Five to twenty percent of cells in a typical ESC population express low levels of Nanog even under self-renewing, leukemia inhibitory factor (LIF)–containing conditions (87). Recent experiments indicate that this heterogeneity is caused by an oscillating expression of a set of synergistically and antagonistically acting genes and could therefore be considered an integral part of pluri-and multipotency for both embryonic and adult stem cells (87,88,89).

To investigate these often low frequency events, single stem cell fates have been traditionally analyzed in standard multititer cell culture plates, such as the 96-well plate (90). These unicellular cultures allow cells to be analyzed at the single-cell level and followed over time as single cell–derived clones, but they require disproportionately high amounts of cell medium and are highly inefficient. Microwell arrays are well-suited to analyze large populations of single cells at the single-cell level. Of course, similar single-cell analyses can also be conducted on micro-contact printed substrates (11) but the topography of microwells extends single-cell analyses to non-adherent cells with clinical relevance such as neurospheres or hematopoeticstem cells (81,84).

A pioneering high-throughput single-cell microwell study was conducted by Chin et al. on rat adult hypocampal progenitor. Using time-lapse microscopy, a large number of quiescent or slowly dividing cells were identified alongside a small fraction (3–4%) of highly proliferating cells. As a consequence of differences in proliferation capacities, 62% of all the cells at day 4 of culture were derived from only 23% of the initial population (91). Our group expanded on these studies to investigate the clonal growth of single mouse neural stem cells into multicellular neurospheres on microwell arrays fabricated from PEG hydrogels. The hydrogel proved to be a superior cell culture substrate in terms of cell survival compared with tissue culture plastic, presumably due to its biomimetic properties. The microwell-confined culture revealed a subpopulation of slowly dividing cells that would be masked by the merging of spheroids in conventional cultures (81) (Figure 4A).

Figure 4. PEG Microwell arrays for high-throughput study of stem cell heterogeneity and fate. (Click to enlarge)

Another stem cell population increasingly investigated using microwell arrays are HSCs (Figure 4B) (74,83). Confirming work by others (92), Lutolf and Blau et al. demonstrated that cell cycle times of individual long-term repopulating stem cells were significantly prolonged and showed a wider distribution compared with multi-potent progenitor cells. Expectedly, the cultured cells did not show any repopulation potential when transplanted into irradiated mice (83). To further elucidate if the addition of exogenous factors can induce self-renewal divisions in culture, 14 putative niche factors—either provided as soluble proteins or tethered to the bottom of individual microwells—were screened for their effect on single-cell proliferation in vitro using microwell arrays cast in 96-well plates. To probe gel-immobilized stem cell regulatory proteins, micro-molding and micro-contact printing were combined and specific protein immobilization was achieved by functionalizing the regulatory proteins with a heterofunctional PEG linker covalently bound to the hydrogel. Site-specific immobilization of protein A extended the method to pattern any Fc-tagged transmembrane proteins. Strikingly, 40% of the tested 14 growth factors yielded significant changes in single-cell proliferation kinetics compared with the basal medium. For example, Wnt3a significantly slowed down proliferation, whereas some factors, such as thrombopoietin, led to accelerated cell division rates. A third class of growth factors, including Shh or immobilized N-cadherin, yielded a significantly higher frequency of single cells producing only three progeny within a week, suggesting that the founder cell had undergone an asymmetric division (Figure 4C). Importantly, the slow or asynchronous in vitro proliferation behavior correlated with the in vivo blood repopulation capacity (83).

Microwell arrays provide a powerful tool for the culture of cell spheroids, to assess the spatial and dimensional effects of geometric patterns at a single cell level, as well as the tracking of clonal growth of populations of single cells. However, it should be mentioned that microwell arrays are passive structures, which restricts the possibilities to manipulate trapped cells. Therefore, integration of microwell arrays into microfluidic devices (93) or the development of releasable microwells (94) may provide new opportunities to handle and analyze cells cultured in microwells.

Microfluidic approaches to track single stem cell development in vitro

Results of time-lapse experiments on single stem cells cultured in microwells (e.g., Reference 83) suggest that HSCs can undergo self-renewal divisions in vitro, and that cell fate choices are under the control of niche factors. However, the distinction between symmetric and asymmetric divisions (Figure 1B) cannot be made a priori using microwell cultures alone. The understanding of these particular fates in stem cells is of utmost importance for clinical applications of many stem cell types (e.g., HSC expansion for transplantations to treat blood cancers). Indeed, in many mammalian tissues it is not yet known whether homeostasis is maintained by asymmetric divisions or by a ‘population-based’ strategy that uses symmetric divisions to balance stem cells and differentiated progeny (2). Because rare adult stem cell divisions can hardly be imaged in vivo (95) (e.g., HSCs that are buried in a poorly accessible bone marrow), there is a considerable demand for in vitro platforms to address this question in live cells.

State-of-the-art in vitro approaches to assess single-cell fate changes—in particular, the symmetry of division— include the manual tracking and analyses of dividing cells in conventional cultures. Although this approach is technically highly demanding, Schroeder and colleagues succeeded in visualizing blood generation from hemogenic epithelium and in tracking the progeny of single hematopoietic progenitor cells (Figure 5A) (96,97). To achieve this goal, these researchers imaged single cells seeded in standard culture plates at a frequency of 2–3 min and developed an advanced cell tracking software to cope with the tremendous data volumes. Although it should be possible to increase the efficiency of data acquisition in these studies using, for example, microwell arrays, the rate-limiting step of this approach is the downstream semi-manual image analysis. Notably, there are a few ongoing efforts addressing the image analysis challenge in microwells. For example, Kachouie et al. developed an algorithm to detect cells and PEG microwells in fluorescence images, taking advantage of the autofluorescence of the PEG hydrogel (98). A software that can recognize PDMS microwells in bright-field images, and that detects and clusters cells based on fluorescence images was also reported (99). However, because both algorithms depend on fluorescent images for cell recognition, and because frequent fluorescent imaging can lead to phototoxic effects in the imaged cells, fluorescence-based cell recognition may be problematic to reliably track cell divisions requiring continuous, long-term observation of primary stem cells (96).

Figure 5. Tracking the genealogies of single stem cells. (Click to enlarge)

Rowat et al. reported on a microfluidic device overcoming the difficulties associated with continuous imaging. Their chip allowed the tracking of multiple yeast lineages in parallel by trapping single cells and constraining them to grow in channels for as many as seven generations (100). Aligning all of the single-cell progeny in lines facilitates image analysis because it reduces the complexity of cell recognition from a 3-D to a 1-D problem. Whereas cell tracking on normal culture substrates requires finding a cell in both space and time, cell recognition in microfluidic devices can be reduced to detecting the time-dependent occupancy of predefined microfluidic traps; it should therefore be more amenable to complete automation.

Despite these exciting advances, the question of divisional symmetry remains poorly addressed. We believe that microfluidic technology has the potential to aid in revealing the molecular mechanisms that govern the fate of individual stem cells—in particular, the symmetry of stem cell divisions. Importantly, any imaging-based analysis of single-cell behaviors would need to be complemented by further assays to characterize cell fate, such as by differential phenotype or gene expression pattern. This could be done by physically separating daughter cells in a predictable manner and removing them from their microfluidic environment after separation. Along those lines, Faley et al. designed a microfluidic device to study the behavior of human HSCs and chronic myeloid leukemia cells (101). Their microfluidic device consisted of an array of hydrodynamic traps that allowed the anecdotal observation of cell divisions using stained mitochondria. However, the parallel arrangement and the size of the hydrodynamic trap were not designed to reliably separate stem cells upon division. To accomplish this more effectively, we adapted a microfluidic perfusion chip having a consecutive number of serial hydrodynamic traps (102,103), small enough to host only one cell and reliable enough to capture daughter cells after division (102) (Figure 5B). By optimizing the hydrodynamic conditions of this trap and the perfusion of the chip, we reached a trapping efficiency of up to 97% and showed viability of nearly 95% of the trapped cells in long-term in vitro experiments. These prerequisites enabled us to visualize cell divisions of non-adherent leukemia cells under perfusion. Because each single cell trap was small enough to host only one single cell, one of the daughter cells generated upon division remained in the trap whereas the other one was transported into the next free trap by perfusion (Figure 5C). The automated physical separation of daughter cells via microfluidics, combined with additional (on-chip) fate analyses including PCR (28), could provide the foundation to rationally assess the regulatory mechanisms that govern single stem cell proliferation and lineage commitment.

Enhancing existing stem cell culture platforms

To date, microscale technologies for stem cell biology have primarily been based on classical cell culture substrates, in particular, tissue culture plastic and glass. However, it has been increasingly acknowledged that the biophysical characteristics of a cell culture substrate can significantly influence stem cell behavior. A seminal study by Engler et al. demonstrated the influence of substrate stiffness on mesenchymal stem cell differentiation (104). Soft hydrogels mimicking the elastic properties of the brain led to the preferred differentiation of hMSCs into neurons; stiffer matrices were myogenic and relatively rigid substrates (as found in the collagenous bone) induced osteogenesis. We expect that the awareness of cell sensitivity to biophysical factors will result in the increased use of soft biomaterials as cell culture substrates in microfabricated platforms (9,81,83), and will also initiate the increased development of approaches to micropattern such materials using micro-contact printing (83), photolithography (66) or microfluidics (53,55). This convergence of existing technologies could generate truly unique microenvironments for cell fate manipulation. To this end, the integration of advanced hydrogel chemistries, micromolding, and micro-contact printing into a single platform has been demonstrated (83). This engineering approach yielded PEG microwell arrays displaying tethered proteins on a soft, tissue-like substrate. This enabled the effect of tethered factors on single non-adherent hematopoietic stem cells to be investigated, which would not have been possible on unstructured substrates due to the extensive migration of HSCs on flat substrates.

The same properties that allow microwells to extend the application of protein patterns to non-adherent cell types may also be used for testing the effect of protein combinations on stem cell differentiation. Combinatorial growth factor and ECM microarrays typically bear the risk of differential adhesion preferences of cells to different substrates, potentially affecting the outcome of an experiment (11). A micro-contact printing technology that combines microarrays and microwell fabrication with soft materials possessing tunable biophysical properties should enable the screening of different substrate properties and protein combinations with well-controlled cell numbers and thereby open up new possibilities to define functional components of stem cell niches.

Recent efforts to develop 3-D microarrays—where cells and test compounds are embedded into a hydrogel—indicate the need to expand screening possibilities to the third dimension (105). Apart from robotic spotting, other techniques might be well-suited for this purpose. Because the encapsulation of cells into a hydrogel eliminates the need to fix the hydrogel samples statically onto a substrate, encapsulated cells could also be cultured in suspension. One powerful approach could be the microfluidic fabrication of hydrogel microbeads containing cells and test compounds. Microfluidic-based droplet generation is based on the injection of an aqueous solution (the discontinuous phase) into an immiscible carrier fluid (the continuous phase), typically oil, inside a microfluidic chip. The immiscibility of the two phases leads to the well-controlled emulsification where the two phases meet and to the generation of up to 10,000 microdroplets per second with a very low size distribution (<2% variation) (106). Due to the very high throughput of microdroplet generation and the engineering of sophisticated on-chip droplet handling methods, microdroplets in combination with a bar-coding system represent a powerful alternative to conventional screening platforms. They were already successfully used in crystallization or cytotoxicity screenings (27,107) (Figure 6). However, microdroplet-based systems are currently limited to nonadherent cells because they lack a substrate for cells to adhere. The in-droplet gelation of a hydrogel would therefore not only extend the possibilities of microdroplets-based screenings to adherent cells, but also open the door for the high-throughput and consistent fabrication of microtissues (108).

Figure 6. High-throughput screening using microdroplets. (Click to enlarge)

In conclusion, the above stem cell culture systems, built at the interface of microfabrication and biomaterials technology, could greatly contribute in identifying the role of specific niche components and the niche architecture in regulating stem cell fate, including (symmetry of) cell division, self-renewal, and differentiation. These approaches currently represent highly simplified models of the in vivo niche, but they allow deconstructing the in vivo complexity and reconstructing it in a well-defined fashion (i.e., from the bottom up). By analyzing the dynamic responses of stem cells to these artificial niches, we should expect advances in the generation of adequate numbers of stem cells and the ability to control their directed differentiation in order to maximize their utility for cell-based therapeutics, as well as drug screening applications.