Externally applied forces as inputs that regulate cellular responses

The most obvious developmental process in which forces play a role is morphogenesis. This requires cells and tissues to undergo a number of crucial deformations, such as twisting, bending and stretching against their environment. These deformations are made possible by a number of forces that are produced by cells directly or indirectly. One such example is cardiac looping during heart development, in which cells bend, twist and stretch the straight heart tube while pressure gradients caused by global contraction asymmetrically increase tube stiffness, prompting the tube to bend (Voronov et al., 2004). Though a great deal of information has been gleaned from this and other animal models of development, researchers have begun developing reductionist techniques – particularly using 2D cultures – to understand how stem cells themselves respond to these inputs and cause subsequent and specific cellular responses or outputs. Here, we consider three main types of externally applied force: twisting, cyclic stretching or strain, and shear – discussing how they can be induced (physiologically and experimentally) and measured, as well as their functional consequences.

Twisting can be both induced and measured by magnetic twisting cytometry (MTC, Fig. 2A), a torque-based deformation method whereby a magnetic field is applied to a bead attached to cell(s) (Wang et al., 2002). The torque experienced by cell(s) can induce changes in behavior, similar to those experienced during development. For example, embedding of paramagnetic beads into embryoid body (EB) cultures and long-term application of force can push embryonic stem cells (ESCs) into contractile fates such as cardiomyocytes or smooth muscle cells (Geuss et al., 2014). Recently, ESCs have been allowed to internalize the paramagnetic particles, and, upon long-term stimulation, this was shown to drive the formation of EBs and subsequent differentiation towards the mesodermal cardiac pathway (Du et al., 2017). Although MTC can be used in 3D systems, it has more commonly been applied in planar culture to observe the response of a cell to a cyclic twisting force. In mouse ESCs, chronic application of MTC induced cell spreading and decreased expression of pluripotency genes, e.g. Oct3 and Oct4 (Chowdhury et al., 2010).

As well as its use in applying long-term external force stimulation, MTC has also been used to help measure intrinsic properties, e.g. cell stiffness, via a short-term application of force and subsequent measurement of the deformation of a cell from that force (Wang et al., 2002). Mouse ESCs, when placed on substrates with different stiffness, do not themselves stiffen (Poh et al., 2010), i.e. their deformation does not change over time when a force is briefly applied to the paramagnetic particle. On the other hand, short-term application of force to adult human mesenchymal stem cells (MSCs) shows that they do exhibit a stiffening response when cultured on hydrogels of different stiffness, and this response drives differentiation into osteoblasts versus adipocytes independent of induction method (Ahn et al., 2014).

Cyclic stretching or strain is a second force ‘input’ to which stem cells respond. Cells are seeded on a deformable membrane and are subjected to periodic strain to mimic the intermittent stretching (Fig. 2B) that occurs in vivo, e.g. in the beating heart or blood vessels. Stretch can occur in uni-, bi- or equibi-axial (where cells are stretched in a manner such that they are confined to regions of homogenous strain in both) directions. Unlike MTC and depending on the stem cell type and niche context, cyclic stretching can modulate the balance of self-renewal and differentiation. For example, cyclic strain induces increased proliferation for MSCs but reduces it for adipose-derived stem cells (ASCs) (Lee et al., 2007; Song et al., 2007). As with other actively applied forces, cyclic strain also induces ESC differentiation into cardiomyocytes and vascular smooth muscle cells, improves maturity and aligns cells along the direction of stretch (Gwak et al., 2008; Heo and Lee, 2011). This behavior mirrors in vivo behavior, where cyclic strain induced by blood flow contributes to cell alignment along the direction of stretch (Sinha et al., 2016). One problem with this ‘input’ method is that there are no community-wide standards for stretch duration, intensity or direction of stretching, and we do not know how best to set the parameters to recapitulate in vivo conditions; uni-axial stretch is most common and may imitate some vascular conditions, but no clear consensus has emerged.

Such standards are more straightforward for the third type of input discussed here: fluid shear forces applied by fluid flow (Fig. 2C). Hemodynamic forces are essential during development; reductions of these forces induced via deletion of cardiac-specific genes result in embryonic death (Culver and Dickinson, 2010). In development after flow is established, additional transcriptional regulation of many vasoactive endothelial genes, e.g. via KLF2 (Lee et al., 2006) and ephrin B2 (Masumura et al., 2009) among others, occurs. As the embryo develops, higher hemodynamic forces are correlated with further maturation and gene expression (Culver and Dickinson, 2010; Lee et al., 2006). Although significant exploration of the consequences of shear on cell fate has occurred in animal models, equally important analyses have been conducted in vitro. For example, shear forces produced by a pulsatile flow bioreactor upregulated endothelial and downregulated smooth muscle cell markers in MSCs (Dong et al., 2009). Recent data in cultured human induced pluripotent stem cells also indicate that initial specification with differentiation media containing VEGF (i.e. inducing endothelial fate) is augmented by shear stress (Sivarapatna et al., 2015). Stem cell models have also proven useful when investigating how shear stress is transduced to regulate transcription, with receptors including FLK1 (Wolfe and Ahsan, 2013) and primary cilia (Hierck et al., 2008) having been implicated. Although further discussion of the interplay of developmental shear stress with other signals is beyond the scope of this Review (for further details see Freund et al., 2012), this is an area in which significant efforts are aimed at recapitulating these combinations in vitro.

Understanding these external forces, how to measure them and what their influences are is crucially important, but – as alluded to above – one problem that has plagued mechanobiology has been the limited information in the literature about the amount, timing and location of the forces required by cells to activate appropriate signaling pathways. Some progress has been made with force-sensitive signaling probes, including fluorescent-based tension sensors (Grashoff et al., 2010; Morimatsu et al., 2013) or rupturable gauges (Wang and Ha, 2013), to quantify the amount of force required to activate a biophysically induced signaling cascade. Yet determining the sensitivity of individual stem cell types and lineages to force-activated signaling remains a major challenge for the field. Further challenging these efforts is the likelihood that the method of force application and its directionality (e.g. compressive pushing inward, tensile pulling outward or shear acting on a surface) may influence the response, further complicating our understanding of exactly how a stem cell acts to integrate signals.

Biomaterials mimic passive (and active) cues that regulate stem cell responses

The previous section focused on stem cell responses from applied forces, but there are numerous situations where a stem cell simply responds to its surroundings as it pulls on and ‘feels’ it. Thus, we will next discuss techniques developed over the past decade to model physical properties of the environment, as well as the outputs regulated by these properties (Fig. 1, right). A crucial development in the late 1990s from Pelham and Wang was the adaptation of polyacrylamide gels, conventionally used for electrophoresis, as a cell culture substrate (Fig. 2D). Using this system, they noted that fibroblasts and endothelial cells spread in a stiffness-dependent manner – stiffer substrates induce greater spreading (Pelham and Wang, 1997). Subsequent work with bone marrow-derived MSCs showed that matrix stiffness could activate serum-based responses and drive cells towards neurogenic, myogenic or osteogenic lineages when cultured on soft neural-like, firm muscle-like and stiffer bone-like polyacrylamide hydrogels, respectively (Engler et al., 2006), i.e. they express lineage-specific transcription factors. At the time, stiffness-mediated signaling was mainly thought to be regulated by intracellularly generated forces from non-muscle myosins, as increasing contractile forces were required to deform increasingly stiff matrices to the same extent (Legant et al., 2009). Subsequently, stiffness regulation has been observed in ESCs similar to that for MSCs (Evans et al., 2009), where stiffness directed the expression of transcription factors consistent with a single cell fate. Additionally, adipose-derived MSCs that migrated to matrices of muscle stiffness underwent myogenesis and increased fusion rate into myotubes compared with cells on other stiffness. Last, unipotent progenitor cells in muscle cultured on muscle stiffness self-renew, whereas they do not on rigid plastic dishes (Gilbert et al., 2010).

As well as stiffness, substrate geometry – which defines cell area (McBeath et al., 2004) and shape (Kilian et al., 2010) – can modulate stem cell differentiation; small and large microprinted ‘islands’ or low and high membrane curvature encourage adipogenic or osteogenic differentiation of MSCs, respectively (Liu et al., 2006) (Fig. 2E). Similar to MSCs, epidermal stem cells cultured on small micro-printed islands underwent differentiation, whereas cells on large islands did not (Totaro et al., 2017). In either case for cell shape and area, intracellular forces are integrated by the cell to create these responses, even though the forces are in static equilibrium and not dynamically changing. As stem cells rarely exist in isolation, how this equilibrium is maintained across cell pairs has also been explored (Maruthamuthu et al., 2011), revealing that cells will additionally balance forces across their cell-cell junctions.

Taken together, these data reinforce the concept that multiple physical inputs can be modulated by other cues, as implied in Fig. 1. However, what complicates this calculation is its non-linearity; the intracellular forces generated from contracting against ECM of a particular stiffness, or of a specific composition, can be permissive or not for differentiation when presented individually. In combination, however, they are not additive (Kourouklis et al., 2016; Rowlands et al., 2008). Even within a particular cue, e.g. ECM composition, hundreds of unique combinations do not appear additive from their individual counterparts (Flaim et al., 2005; Kourouklis et al., 2016). Thus, one needs to appreciate the complexity of these cues and their influence on force, but even that may not predict how they will be integrated by a stem cell.

Just as cells do not exist in isolation, environmental cues that induce intracellular forces are not static; niche properties change both with space and time (Fig. 2F), and this could affect how a stem cell integrates cues and makes decisions. In culture, stiffness has typically been presented as a single value, but it changes in vivo during development and often with disease. One question that arises is at what point is stem cell commitment no longer able to respond to the physical attributes of the niche? Temporal gradients can be induced via sequential ECM crosslinking using biomaterials with single or multiple crosslinking methods (Guvendiren and Burdick, 2012; Young and Engler, 2011). Conversely, materials can have crosslinks degraded to soften ECM or induce stress relaxation (Chaudhuri et al., 2016; Kloxin et al., 2009). Interestingly in all cases, the consequence of the change in stiffness varies according to when it is induced. Thus, for example, during MSC differentiation, changes applied in the first week appear reversible, but those made in subsequent weeks are not (Guvendiren and Burdick, 2012; Young and Engler, 2011). Similarly with neural stem cells, there is a small temporal window in which ECM stiffness maximally affects neurogenic commitment; altering stiffness signaling in this window dramatically impacts neurogenesis (Rammensee et al., 2017), whereas changes at other times have more minor effects.

Tissue stiffness also typically contains spatial gradients, which can be accomplished by changing gel thickness, crosslink density or micropost length in a spatially dependent manner (Hadden et al., 2017; Tse and Engler, 2011; Zaari et al., 2004). Stiffness can vary by six orders of magnitude in vivo (Discher et al., 2009) and gradients within tissues can vary by up to three orders of magnitude (Vincent et al., 2013). Most committed cells migrate preferentially to stiffer regions via unbalanced forces created by these gradients, i.e. ‘durotaxis’ (Lo et al., 2000; Vincent et al., 2013). MSCs also exhibit these unbalanced forces but are even more sensitive to gradient slope and range, exhibiting directional migration even when the gradient is at or below natural physiological variation in tissues (Tse and Engler, 2011; Vincent and Engler, 2013). As these cells are uncommitted, it is perhaps surprising that they would exhibit such preference, which also biases their differentiation towards more contractile lineages, e.g. bone. Interestingly, some report similar ‘memory’ of their previous niche post-durotactic migration, e.g. MSCs that migrated from soft to stiff regions still expressed neuronal markers (Tse and Engler, 2011), which suggests that as with temporal changes, spatial variation can impact stem cell plasticity.

Both ESCs and MSCs use cell-based forces to contract against their niche, but important differences occur between them that give rise to different sensitivities. Common methods to measure these deformations include micropillars (Tan et al., 2003). In the case where posts are fabricated to different heights while maintaining a planar surface, larger posts are more deformable and produce more bending compared with shorter posts (Fig. 2G), creating a gradient similar to those in solid surfaces. MSCs cultured on short rigid posts are well spread with highly organized actin fibers compared with cells on longer more-deformable posts. Furthermore, differential post deformability altered differentiation, as osteogenic fates were favored on short posts and adipogenic fates were favored on long posts (Fu et al., 2010). However, rather than determining cell fate, decreasing pillar length induced increased human ESCs contractility, which makes for a more rigid substrate surface and maintains pluripotency in isolated cells. Furthermore, removal of cell-cell contacts (i.e. only single cells) muted cell-micropost interactions and indicated synergy between the stimuli; thus, the percent of pluripotent cells was reduced regardless of substrate stiffness when cell-cell contact is present (Sun et al., 2012b). These data are consistent with the interpretation that cell-cell and cell-matrix forces balance to maintain cell fate. In an expansion of this application, magnetic wires have been added to these posts to actively stimulate cells via torque induced by a magnetic field moving the micropillars (Sniadecki et al., 2007). Post-stimulation, intracellular forces rearrange and change magnitude (Khademolhosseini et al., 2016; Lin et al., 2012), suggesting the stem cells responses, e.g. fate, could be driven by these externally applied local deformations. Together, these data – particularly the difference in response between MSCs and ESCs – suggest that not only is the cell response to force context specific, it is also stem cell type specific.

Actomyosin-based regulation of cell fate

Actin and myosin II form the classic cytoplasmic contractile apparatus found in muscle and, more generically, in all adherent cells (Fig. 3A): actin forms thin filaments upon which myosin thick filaments bind and slide to create a contractile force. In the fields of stem cell mechanobiology and biomechanics, there has been particular focus on understanding how non-muscle myosin II (NMMII) enables progenitor cells to ‘feel’ the stiffness, porosity and topography of their environment, and develop intracellular tension (Engler et al., 2006; Kilian et al., 2010; McBeath et al., 2004; McMurray et al., 2015; Wang et al., 2002). These different passive properties of the niche may seem disparate, but inhibition of NMMII actin-binding activity (Guilak et al., 2009) or depolymerization of the structure entirely (Arnsdorf et al., 2009) renders stem cells unable to respond to any of these material effects. Chronic NMMII inhibition can also reduce contractility-related apoptosis in stem cells in culture (Chen et al., 2010; Watanabe et al., 2007). Direct application of external force is also crucial for myosin II-mediated cell responses (Riveline et al., 2001). Although there are direct mechanical connections between the actomyosin contractile apparatus and focal adhesions (Zamir and Geiger, 2001), the intra-cellular forces are distinct, spatially decoupled from each other (Plotnikov et al., 2012) (i.e. one adhesion can be actively transducing a contraction whereas an adjacent one does not), and insensitive to other stimuli that affect maturation of focal adhesion structures (Stricker et al., 2013). Although this and other evidence strongly suggests a role for actomyosin contractions in transducing biophysical signals, the specific molecular details in stem cells regarding how NMMII could convert these contractions into biochemical signal remain unclear. More generally, however, one potential actomyosin transduction candidate is serum response factor (SRF), which is a transcriptional regulator controlled by actomyosin dynamics (Connelly et al., 2010).

‘Molecular strain gauges’ and mechanically activated transcription factors

Focal adhesions are a cluster of proteins physically bound to one another that connect extracellular matrix receptors with the cytoskeleton of the cell, which contains its actomyosin contractile structures. Within these adhesions, transcription factor-binding sites have been identified that, when exposed, could act as molecular switches (Fig. 3B, star) that convert a stem cell into a mature cell based on the passive forces that the cell can exert on its environment. One of the best-studied examples of a focal adhesion tension sensor is vinculin (Holle et al., 2013). In human MSCs, vinculin can act as a strain gauge and control stiffness-mediated differentiation; too little or too much force transduced across the protein can lead to cryptic kinase-binding sites remaining buried or completely unfolding, respectively. Thus, vinculin-mediated signal transduction via kinases is force sensitive. For example, MSCs lacking vinculin remain adherent to their ECM and contractile but fail to differentiate into muscle in stiffness-based differentiation assays due to a lack of mitogen-activated protein kinase (MAPK) binding to a cryptic site on vinculin. Stem cells rescued with mutant forms of vinculin (headless, tailless or lacking the MAPK-binding site) also failed to differentiate into muscle. These data are consistent with observations in other cell types in which partial length vinculin was insufficient for many cell functions (Dumbauld et al., 2013). Beyond vinculin and MAPK signaling, other focal adhesion-based, strain-sensitive regulators that employ a force-dependent, conformational change-based mechanism, i.e. molecular strain gauge, have been identified by high-throughput screening to detect proteins that when deleted would prevent stiffness-based induction of differentiation (Holle et al., 2016). These sensors often will have different set points to regulate differentiation to other lineages. For example, sorbin and SH3 domain-containing protein 1 (SORBS1) was found to regulate stiffness-sensitive osteogenesis in MSCs in the same manner that vinculin governs myogenesis, albeit with a higher stiffness set point – i.e. greater force is required to activate SORBS1 than vinculin. This type of mechanism offers great precision; the sensor is tuned to unfold or change conformation only over a narrow range of tensions. Thus, its ‘set point’ is exceedingly sensitive.

Force-sensitive transcription factor localization

Strain-sensitive proteins in focal adhesions may change their confirmation, but additional force-sensitive switches, including some transcription factors, appear to shuttle information to the nucleus in a similar manner. For example, YAP/TAZ is a robust signaling complex that translocates from the cytoplasm to the nucleus as the niche becomes stiffer or stem cell shape changes for MSCs among other stem cell types (Dupont et al., 2011). Translocation can also occur via changes in cell density, shear stress, in 2D or 3D, and via stretch (Panciera et al., 2017). This regulation of cell fate occurs independently of the Hippo/LATS cascade and upstream of Notch signaling to regulate somatic cell stemness (Totaro et al., 2017). Conversely, TWIST1 acts as a mechanomediator that induces epithelial-to-mesenchymal transition (EMT) on stiff matrices by translocating to the nucleus. (Wei et al., 2015). This pathway appears to function independent of the classic YAP/TAZ ‘switch’ identified in MSCs in the past decade. In both instances, the exact mechanisms are not entirely clear, but they may involve changes in the phosphorylation of the factor or its chaperones (Shamir et al., 2014; Wei et al., 2015). However, it is important to note that initial stiffness-sensitive transcription factor nuclear translocation may not be sufficient for complete lineage specification; MSCs, which express stiffness-sensitive RUNX2 (Engler et al., 2006), require cell-cell contact for later osteogenesis markers (e.g. alkaline phosphatase and calcium deposition) to be induced. These data demonstrate the interplay that exists between force-sensitive differentiation mechanisms and other inputs (Mao et al., 2016). Regardless, transcription factor localization appears switch-like, centered around a particular set point as with focal adhesion-based sensors, i.e. mechanism in Fig. 3B.