Contact inhibition of locomotion is a complex process that involves many different molecular mechanisms. Each of the four distinct steps of CIL requires changes to the cytoskeleton driven by a variety of molecular components [36, 48, 49]. The following part of this review will break down the process of CIL into these four stages and highlight the key components involved in driving each step.
A contact is formed between the cells
Formation of a cell–cell adhesion complex
It has long been established that the formation of a physical contact between colliding partners is a requirement for CIL and no changes occur in the lamellae prior to this event . The fact that an adhesive contact must be forming between colliding cell partners was further evident by the observation that tension is generated in the lamellae across a contact [25, 28, 50]. After Abercrombie’s discovery of CIL in fibroblasts, work was done to elucidate the nature of these adhesions using the microscopy techniques available at the time. Heaysman and Pegrum coupled the behaviour of the adhesions to the different stages of CIL in fibroblasts . They noted that cell–cell adhesions formed between colliding cells soon after a collision and speculated that the abrupt separation of the cells was due to the loss of these adhesions. Interestingly cell–cell adhesions were not observed when fibroblasts collided with sarcoma cells , where normal CIL behaviour is known to be lost . Although the exact nature of these adhesions was speculated upon , the limitations of the microscopy and molecular biology techniques available prevented the identification of the molecular components involved. It was not until decades later that the nature of these adhesions could begin to be elucidated. One potentially surprising aspect of the cell–cell adhesions identified in CIL is that they do not all belong to the same family of adhesion complexes. This suggests that CIL may be driven through a variety of different mechanisms. We will discuss some of the adhesion molecules involved in CIL.
Cadherins The first family of cell–cell adhesion molecules to be identified in CIL were the cadherins . Cadherins are a family of transmembrane glycoproteins that facilitate calcium-dependent cell–cell adhesions. They form adherens junctions between neighbouring cells and tightly regulate the actin cytoskeleton . Their importance in CIL was first identified in L-cell lines where it was demonstrated that the presence of E-cadherin, the cadherin predominantly expressed in epithelial cells, caused paralysis of the lamellae upon a collision . Furthermore, E-cadherin has since been identified as the adhesion molecule required to inhibit the protrusive activity and migration of confluent epithelial cells  and its disruption has been associated with the loss of this behaviour in carcinoma cells . N-cadherin, the cadherin first discovered in the neural plate, is required for CIL in a variety of cell types [14, 30, 56]. In myoblasts and glial cells it is required for the cessation of migration and paralysis of lamellae upon a collision [14, 56]. In addition N-cadherin and cadherin-11 are essential for CIL between neural crest cells where their loss inhibits the migration of the neural crest in vivo [30, 57]. In vitro cultures of neural crest cells show normal CIL behaviour, where colliding cells form a contact, collapse protrusions and cease migration before repolarising and migrating away from each other. When either N-cadherin or cadherin-11 is inhibited the colliding neural crest no longer show normal CIL behaviour, instead they continue migrating in the direction of contact and no longer repolarise away from the contact. In addition, there is an increase in protrusive activity at the contact, indicating that the normal paralysis of lamellae is lost. Interestingly, blocking N-cadherin junctions in Schwann cells seems to promote a CIL like process, where the cells pull away from each other after coming into contact .
Eph-ephrin Another group of proteins that are known to mediate cell–cell interactions during CIL are the Eph receptors. These are a group of tyrosine kinase receptors that bind transmembrane ephrin ligands from the neighbouring cell and couple the cells upon cell–cell contact. The binding of the ligand by the receptor triggers bidirectional signalling cascades in both the ligand-expressing and the receptor-expressing cells . Eph/ephrins are expressed in all germ layers. They are essential for many aspects of development including vascular and skeleton morphogenesis, boundary formation and axon guidance (as reviewed in ) and their dysregulation is associated with disease . Interestingly Eph-ephrin mediated cell–cell interactions are often, but not always, associated with a repulsive response in the coupled cells causing the cells to retract upon contact in a process similar to CIL [62–64]. EphA signalling can facilitate CIL in prostate cancer cells by promoting a repulsive behaviour between cells [8, 35]; whereas, EphB signalling suppresses CIL and increases membrane ruffling at the site of contact by promoting cell–cell attraction [16, 64]. Interestingly, this difference in behaviour controlled by a shift in the balance of activities of EphA to EphB, is strikingly similar to the cadherin switch from E- to N- that dictates whether neural crest cells undergo CIL or not . Both EphA and EphB are required for CIL in Cajal–Retzius neurons and to drive their proper dispersion . EphB signalling gives rise to CIL in a carcinoma cell line and can induce high levels of CIL behaviour, which can override chemotactic cues . Whether the full spectrum of cell–cell adhesion complexes that contribute to CIL have been identified is unknown. During CIL of haemocytes in Drosophila [6, 17, 48] zyxin has been shown to localise at the cell–cell contact ; however, the molecular nature of the cell adhesion molecule at the contact remains unknown. The engagement of this unidentified cell–cell adhesion is essential for CIL through its ability to couple the cytoskeletons in the colliding partners, allowing tension to be built up in their lamellae prior to separation .
Protrusive activity is inhibited at the site of contact
Regulation of small GTPase activity
The distinct steps of CIL are each driven by cytoskeleton rearrangements and dynamics that in turn are controlled by the activity of Rho family GTPases . RhoA and Rac1 are the best understood members of the RhoGTPases. The canonical understanding is that RhoA generates contraction through the regulation of actomyosin and activation of ROCK , while Rac1 drives the formation of lamellipodia  through the mediation of actin polymerisation. Here we highlight the RhoGTPases identified at the contact during CIL.
One distinct feature of CIL is the paralysis of membrane ruffling and inhibition of protrusive activity at the leading edge upon a collision [25, 28, 30, 48, 68]. In a free migrating cell Rac1 is active in the leading edge. This drives actin polymerisation and subsequently protrusion formation at this site . Upon a collision a switch in the activity of the RhoGTPases occurs at the contact site, whereby RhoA is activated and Rac1 is inhibited, driving the paralysis in the membrane and loss of protrusions (Figs. 1c, 2) [5, 30, 69]. In neural crest cells, this switch is dependent upon the activation of the non-canonical Wnt-planar cell polarity (PCP) pathway (Fig. 2c) [5, 69, 70]. Upon a collision many PCP elements, including Dishevelled, Prickle1 and Strabismus, are recruited to the receptor Frizzled7 at the cell–cell contact where their presence is required to drive CIL [5, 10]. The activation of the PCP pathway results in the activation of RhoA, which drives the contraction of the lamellae in a manner dependent on ROCK activity. If ROCK activity is blocked the protrusions fail to collapse at the contact and normal CIL behaviour is lost [5, 10, 69]. In addition, Rac1 activity is inhibited at the contact site, resulting in collapse of the protrusions [29, 70]. This loss of Rac1 activity could in part be due to the antagonistic behaviour that is known to occur between RhoA and Rac1, where the activation of one results in the inhibition of the other . The requirement of RhoA/ROCK activity at the contact site in CIL has also been further established in chick embryonic heart fibroblast where their absence prevents the cells from undergoing CIL, instead they continue migrating in their given direction upon contact as there is no paralysis of membrane ruffles and protrusions . Furthermore, the perturbation of Rac1 in NIH3T3 fibroblasts, either through the use of dominant active Rac1, dominant negative Rac1 or an increase in RhoA activity, results in the loss of CIL when they confront chick heart embryonic fibroblasts . As well as its inhibition downstream of PCP signalling, the inhibition of Rac1 is also driven by the formation of N-cadherin junctions at the contact in the neural crest (Fig. 2b). Blocking N-cadherin, either by antisense morpholino or blocking antibodies, results in a loss of CIL due to an increase in Rac1 activity at the contact driving protrusions at this site . In addition, the overexpression of E-cadherin in the neural crest also results in an increase in Rac1 activity at the contact . Furthermore, these E-cadherin overexpressing cells no longer undergo CIL.
The precise mechanism by which N-cadherin leads to the activation of RhoA and inhibition of Rac1 remains unknown although there are many possibilities (Fig. 2a, b). One possibility is through p120-catenin, which binds to N-cadherin and regulates its turnover . Cytosolic p120-catenin can enhance protrusion formation through the activation of Rac [74, 75]. Interestingly, when it is sequestered to the cell–cell adhesion complex it can no longer promote the activation of Rac and protrusions are inhibited . During CIL N-cadherin could be sequestering p120-catenin preventing it from activating Rac at the contact. Furthermore, the elevation of Rac1 at the contact in neural crest cells overexpressing E-cadherin appears to be dependent on its interaction with p120-catenin and when this interaction is blocked Rac1 activity is once again reduced at the contact . This suggests the ability to prevent p120-catenin from activating Rac1 is specific to the way it is sequestered by N-cadherin. p120-catenin has also been implicated in modulating RhoGTPase activity downstream of Wnt signalling [76–78]. It is also possible that p120-catenin may be modulating the activity of Rho and Rac at the contact after activation of the PCP pathway. The RhoGTPase switch that occurs at the contact upon a collision could also be mediated by the inhibition of the GEF-Trio at this site. Trio can activate Rac1 and modulate the activity of RhoA. It localises to the cell–cell contact in the neural crest in vivo, downstream of the polarity protein Par3, where its inhibition appears to be required for CIL . Furthermore, there is evidence that Trio is recruited downstream of cadherin-11 (Fig. 2a) and its inhibition could provide a mechanism for RhoA activation and Rac1 inhibition upon a collision . It is likely the cadherins recruit Par3 to the contact where it inhibits Trio, resulting in the inhibition of Rac1. An additional mechanism driving the RhoGTPase switch is through the interaction between the nucleotide diphosphate kinase–nm23, and the GEF-Tiam1 that activates Rac1 (Fig. 2b). Nm23 has been identified at the cell–cell contact site in glial cells undergoing CIL where it is localised to N-cadherin . At the cell–cell contact nm23 associates with Tiam1 and inactivates it resulting in the inhibition of Rac1 at this site. EphA/ephrinA signalling leads to RhoA/ROCK activation at the contact (Fig. 2d) , via the GEF-Vav2, which is recruited to EphA when it is activated upon binding ephrinA . Furthermore, it has recently been discovered that Rac1 activity in the overlapping protrusions of colliding fibroblasts is regulated by the GAP srGAP2 . It appears that slit-robo signalling is activated in overlapping protrusions during a collision resulting in the activation of srGAP2 and the localised regulation of Rac1 activity . This localised signalling event is required to prevent the cells continued migration and drive the repolarisation of the cells. Each of these different mechanisms regulating small GTPases during CIL could happen in distinct cells or in the same cell. If they occur in the same cell the net balance of all these molecular interactions will determine the final outcome and if a cell undergoes a CIL response.
Microtubules upon a collision
In addition to their role in regulating the actin cytoskeleton, RhoGTPases also play an essential role in the regulation of microtubules. Microtubules are stabilised in the leading edge where they are important for maintaining the polarity of a cell and driving directional migration [81, 82]. Stabilised microtubules promote membrane ruffling and the formation of lamellipodia , whilst inhibiting contractility through the down regulation of stress fibre and focal adhesion formation . Furthermore, microtubules help maintain cell–cell adhesion complexes . In haemocytes, microtubule bundles are observed in the leading edge where they stabilise the protrusion . When two haemocytes collide the microtubule bundles align across the two colliding cells , this coincides with a deceleration of the cells during CIL . It is likely the alignment of microtubule bundles in colliding haemocytes plays a role in the inhibition of the forward movement of the cells, potentially by generating a physical barrier that prevents the cells’ continued migration. If the microtubules cannot be stabilised then polarity is lost in the haemocytes and they no longer undergo CIL . It is possible that the initial coupling of microtubules in colliding cells promotes the formation of the cell–cell adhesion complex that is required to drive CIL.
The cells repolarise and new protrusions form away from the contact
Rac1 activity away from the contact
Another key feature of CIL is the repolarisation of the cells away from the contact after a collision (Fig. 1d). The repolarisation of colliding cells requires a switch in front–rear polarity. In order for this switch to occur not only does RhoA have to be elevated and Rac1 inhibited at the contact, as discussed above, but a new leading edge must form away from the contact. The formation of a new leading edge is dependent on the interplay between adhesions, RhoGTPases and the cytoskeleton. This requires the increase in Rac1 activity away from the contact driving the formation of lamellipodia in this region [33, 85]. During collisions of neural crest cells the switch in the localisation of Rac1 activity has been visualised . In a free migrating cell Rac1 is activated in the leading edge of the cell. Upon a collision Rac1 is inhibited at the contact and subsequently becomes active away from the contact . An elegant experiment in the neural crest recently showed the importance of Rac1 activity in the leading edge after a collision. Cells overexpressing E-cadherin, where Rac1 activity is increased near the contact, do not separate after colliding. However, the activation of photoactivatable Rac1 in the free edge of a cell is sufficient to promote the separation of the cells . This is of particular interest as it suggests the repolarisation of the cells away from the contact is enough to drive separation of the cell even when Rac1 activity is elevated at the contact due to the presence of E-cadherin.
In addition to a switch in Rac1 activity, a switch in the dynamics of microtubules is also required to drive the repolarisation of cells after a collision [36, 37, 86]. Microtubules are stabilised in the leading edge of a cell where they are required to reinforce its polarity [81, 82]. Upon a collision there is a change in the dynamic behaviour of the microtubules at the site of contact, with an increase in the frequency of catastrophe events and rates of shrinkage and growth . This increase in dynamic behaviour at the contact is required for CIL [6, 8, 36, 37]. In the neural crest the dynamic behaviour of microtubules seems to be dependent upon the cell polarity protein—Par3 . Par3 localises to the cell–cell contact where it promotes microtubule catastrophe through the inhibition of the GEF-Trio and subsequent inhibition of Rac1. In haemocytes microtubule bundles align between colliding cells upon a collision and their subsequent collapse is required for a normal CIL response . In addition to an increase in their dynamics at the contact site, microtubules also become stabilised away from the contact further driving the repolarisation of the cell .
The cells separate and migrate away from each other
Tension build-up across the contact
The driving force behind the cells’ separation after a collision is still not fully understood (Fig. 3). It has long been established that there is a build-up in tension across the contacting lamellae [25, 28, 48, 50]; however, how this tension builds up and whether this tension alone is sufficient to tear apart the contacting cells remains unknown. In haemocytes a sudden retraction of lamellae is observed as the cell–cell adhesion complex is broken and the tension across the complex is released . There is much speculation as to what triggers separation and we shall discuss the possibilities.
One possible event that could be triggering the separation of cells after a collision is the disassembly or internalisation of the cell–cell adhesion complex (Fig. 3a). This would uncouple the cells and release the tension across the contact causing the cells to come apart. An alternative possibility is that tension is built up to such a degree across the contact that it forces the cell–cell adhesion apart. This tension could be generated through various means. The activation of RhoA and subsequently ROCK at the contact upon a collision [5, 16, 69] was believed to trigger actomyosin contraction. Actomyosin contraction in the contacting lamellae would result in tension being generated across the contact (Fig. 3b). Myosin II coated stress fibres align between colliding haemocytes and mutants that are lacking in myosin II show a reduction in lamellae tension in the contacting lamellae . It has been hypothesised that myosin-driven contraction of these stress fibres could be sufficient to drive the separation of the cells. Interestingly, however, there is evidence that RhoA/ROCK activation at the contact site does not act through actomyosin contraction as normal CIL behaviour can still occur when myosin contraction is blocked through the use of blebbistatin . It appears instead that RhoA/ROCK activity acts through the mediation of microtubule dynamics . Upon a collision an increase in microtubule dynamics and catastrophe events is required for CIL [6, 8, 36, 37]. Thus, a microtubule catastrophe event could trigger the separation of the cells after a collision by causing a sudden increase in tension across the contact that may be sufficient to force the contact apart (Fig. 3c).
The coupling of the actin cytoskeletons in colliding cells can also generate tension by linking the actin retrograde flow in the lamellae of both cells via cell–cell adhesions across the contact. In a mechanism similar to integrin, the cell–cell adhesions act as a clutch by anchoring the cytoskeleton to a point of resistance [48, 49, 87]. This causes a deceleration of the continuous actin retrograde flow and results in a build-up of tension across the cell–cell contacts and in the lamellae, as actin retrograde flow continues to generate a force that is pulling the cells away from each other. This actin retrograde flow alone could generate enough tension across the cell–cell contacts that a point is eventually reached where the force is too great and the cell–cell adhesion is pulled apart (Fig. 3d).
In addition, the repolarisation of the cells as a whole is necessary for the separation of the cells after a collision (Fig. 3e) [33, 86]. The neural crest cell–cell adhesion complexes remain intact when protrusions are inhibited from forming away from the contact due to physical constraint . This suggests the cells need to pull apart from each other in order for the cell–cell adhesions to be lost. Furthermore, stimulating protrusion formation through the use of a photoactivatable Rac1 in the free edge of cells overexpressing E-cadherin, which do not separate upon a collision, is sufficient to drive the separation of these cells . This indicates that neural crest cells start migrating away from each other prior to the loss of the cell–cell adhesions and this pulling apart is necessary and sufficient to drive the breakdown of these adhesions.
It appears that a variety of mechanisms (Fig. 3) may be stimulating tension generation across the contact and the disassembly of cell–cell adhesions. Each event alone may not be sufficient to drive the separation of the cells, but together they generate enough force and possibly stimulate a signalling event that results in the disassembly of cell–cell adhesions and the subsequent separation of the cells. It is unclear how cell dependent the precise mechanism of separation is, or whether it is conserved across different cell types. A more thorough examination of this event is required to fully understand what drives the separation of cells after a collision.
Cell–matrix adhesions play a core role in cell migration and therefore are central to CIL. Cell–matrix adhesions form a transmembrane complex that crosslinks the extracellular matrix to the intracellular cytoskeleton via integrins and adapter proteins. This generates a physical connection linking the external environment to the cytoskeleton and results in force generation and cytoskeletal rearrangements. In addition, this link can also induce internal signalling that can be stimulated by the external environment. The behaviour of cell–matrix adhesions during CIL was first speculated upon by Abercrombie , although their behaviour and importance during this process is still not fully understood. Cell–matrix adhesions were first characterised during CIL by Harris where he observed a detachment of cell–matrix adhesions in the lamellae upon a collision. This would lead to the cell–cell contact coming under tension once these adhesions to the substrate were lost . The cells would subsequently separate after the complete loss of cell–matrix adhesions . Interestingly however, Abercrombie noted a conflicting observation using interference reflection microscopy , a method that assumes strong cell–matrix adhesions occur where the cell membrane is at its closest to the substrate . Using this imaging technique to infer where cell–matrix adhesions are, Abercrombie concluded that adhesions to the substrate actually persist during a collision even when the lamellae contract . These apparent contradictory results have not being revisited in the 40 years since these observations, and it is still unknown what happens to the cell–matrix adhesions upon collision and if they play a role in driving separation.
Integrin signalling has been identified in myoblasts where ectopic expression of either α5 integrin, β1 integrin or downstream effectors of integrin—such as paxillin and FAK—results in a paralysis of membrane ruffling and lamellae activity upon a collision . There is further evidence of cell–matrix adhesions during CIL in the neural crest. Syndecan-4, a transmembrane heparan sulphate proteoglycan that can crosslink the extracellular matrix to actin via the adapter protein α-actinin  and stimulate focal adhesion formation , is essential for the directional migration of the neural crest in vivo . In addition, the loss of syndecan-4 results in a loss of CIL with protrusions no longer inhibited towards the contact, as in the case in control cell, due to a huge increase in Rac1 activity across the whole cell periphery. This suggests the presence of syndecan-4 inhibits Rac1 activity at the contact, although where syndecan-4 is localised in the neural crest or how it inhibits Rac1 activity has not yet been identified. In fibroblasts, however, there is evidence that syndecan-4 regulates Rac1 activity through the mediation of PKCα, which plays a role in localising Rac1 activity to the leading edge . Integrin-based cell–matrix adhesions have been visualised in the neural crest [10, 33]. Interestingly, they show a distinct difference in morphology in the free edge versus the site adjacent to the contact. Large elongated adhesions are observed in the free edge, whereas the adhesions near the contact are much smaller and rounded in shape. Interestingly, these small adhesions near the contact become enlarged when E-cadherin is overexpressed . Whether this enlargement is a contributing factor or just a consequence of the loss of CIL in E-cadherin overexpressing cells is unknown.
Cell–matrix adhesions are important mediators of actin retrograde flow rates . The engagement of these adhesions slow actin retrograde flow by generating friction between the actin network and the substrate, consequently generating traction . Changes in actin retrograde flow during CIL have recently been visualised in haemocytes in vivo . It is possible these changes are not solely due to the engagement of the cell–cell adhesion complex, as discussed above, but also driven by changes in cell–matrix adhesion behaviour. It would be of interest for cell–matrix adhesions to be imaged in this in vivo model so their dynamics during CIL can be understood.