Summary
The migration of immune cells is guided by specific chemical signals, such as chemokine gradients. Their trajectories can also be diverted by physical cues and obstacles imposed by the cellular environment, such as topography, rigidity, adhesion, or hydraulic resistance. On the example of hydraulic resistance, it was shown that neutrophil preferentially follow paths of least resistance, a phenomenon referred to as barotaxis. We here combined quantitative imaging and physical modeling to show that barotaxis results from a force imbalance at the scale of the cell, which is amplified by the acto-myosin intrinsic polarization capacity. Strikingly, we found that macropinocytosis specifically confers to immature dendritic cells a unique capacity to overcome this physical bias by facilitating external fluid transport across the cell, thereby enhancing their space exploration capacity in vivo and promoting their tissue-patrolling function. Conversely, mature dendritic cells, which down-regulate macropinocytosis, were found to be sensitive to hydraulic resistance. Theoretical modeling suggested that barotaxis, which helps them avoid dead-ends, may accelerate their migration to lymph nodes, where they initiate adaptive immune responses. We conclude that the physical properties of the microenvironment of moving cells can introduce biases in their migratory behaviors but that specific active mechanisms such as macropinocytosis have emerged to diminish the influence of these biases, allowing motile cells to reach their final destination and efficiently fulfill their functions.
Introduction
The adaptive immune system of mammals is composed of distinct cell populations that traffic between peripheral tissues and lymphoid organs. Adaptive immunity therefore relies on the ability of these cells to migrate all over the body, where they encounter numerous constraints imposed by the structure of the tissue (Heuze et al., 2013). Two important physical constraints that limit immune cell migration in tissues are geometrical confinement and hydraulic resistance (HR), which is induced by the interaction of a moving cell with the surrounding fluid (Bergert et al., 2015; Prentice-Mott et al., 2013). So far HR has been ignored as it exerts forces on cells that are supposed to be about a hundred fold smaller than the traction forces that cells themselves exert on the substratum (Bergert et al., 2015), suggesting that migrating cells should not be capable of sensing them. Nonetheless, recent findings highlighted that in vitro neutrophil-like cells preferentially migrate in confined environments towards low-hydraulic resistance paths (Prentice-Mott et al., 2013). This phenomenon, referred to as barotaxis, was attributed to the ability of these cells to sense and respond to differences in external hydraulic resistance using an active mechanism that may involve specific receptor(s) and signaling pathway(s). However, such mechanisms have not been identified so far. On the other hand, it was recently shown that cells migrating in confined environments use an adhesion-independent amoeboid-like migration mode that involves forces about a hundred fold smaller than the traction forces exerted by mesenchymal adhesive cells (Bergert et al., 2015). The magnitude of these forces is thus similar to those exerted by hydraulic resistance, suggesting that, in the absence of adhesions, hydraulic resistance could become the dominant resistive force that cells have to fight against to move.
We here combined theoretical physics, micro-fluidics and intravital imaging to investigate the mechanisms of barotaxis and its impact on immune cell behavior in vivo. We show that barotaxis results from a small force imbalance at the scale of the cell, which is amplified by the actomyosin network, while specific receptors and signaling pathways are not required. We further demonstrate that the ability of immature dendritic cells (DCs) to undergo macropinocytosis, i.e. non-specifically ingest extracellular fluid, renders them insensitive to HR and thereby facilitates their capacity to patrol their environment. By contrast, mature DCs are no longer macropinocytic and may benefit from barotaxis to find low-resistance paths while migrating to lymph nodes to initiate the immune response. Thus, specific mechanisms have emerged for DCs to overcome the physical obstacles they are exposed to in vivo and efficiently achieve their immune sentinel function.
Results and discussion
A theoretical framework to predict barotaxis
As mentioned above, the range of forces exerted by non-adherent cells migrating under confinement (such as immune cells) is comparable to the range of HR forces. Therefore, we hypothesized that barotaxis could result from a purely mechanical interaction of the cell migratory apparatus with the surrounding fluid. In other words, cells would preferentially migrate to lower HR paths because these paths oppose the lowest force to their migration. To test this hypothesis, we extended a well-established hydrodynamic physical model that successfully reproduces the fast mode of migration of non-adherent cells in confinement from a minimal set of hypotheses (Callan-Jones and Voituriez, 2013). In this model, cells are described as made of an active poro-elastic material, whose constitutive equations stem from the symmetries and conservation laws of the cortical actomyosin system (Joanny et al., 2007; Juelicher et al., 2007; Kruse et al., 2005). The main parameters of the model are: (1) the amplitude of the contractile stress, which encodes the interaction between myosin-molecular motors and the cortical actin and (2) the cell membrane-fluid permeability, which encodes the forces arising as the extracellular fluid passes through the confined cell. The model harbors a simple motility mechanism in 1d-confined environments that is critically controlled by both the level of contractility and the cell length. Above a critical level of activity, cells exhibit a spontaneous polarization mechanism: random fluctuations of 1d non-polarized cells are amplified, breaking the front-rear symmetry through an accumulation of contractile activity at the back. This drives spontaneous retrograde flows of the actin cortex, resulting in a net thrust force on the channel walls and therefore a polarized motile state. To investigate the mechanism of barotaxis, we extended this model (Callan-Jones and Voituriez, 2013) from a 1d-channel to a three-way bifurcation (See Supplementary Information) exhibiting different values of HR in the 2 upper paths (Fig. 1A). Barotaxis was quantified in these three-way bifurcations as the bias of cell direction toward lower resistance path. We observed that this directionality bias increased progressively with HR asymmetry until it reaches an upper threshold of 0.45 (corresponding to about 70% of cells choosing the low resistance side) (Fig. 1B), indicating that our schematic model is sufficient to recapitulate barotaxis. Therefore, barotaxis can be predicted by the mechanical interaction between the migrating cell and its environment without the need to introduce any receptor or signaling pathway dedicated to pressure sensing.
To experimentally address the predictions inferred from the model, we designed Y-shaped micro-fabricated devices exhibiting different HR and a small cross-section (18µm2) to ensure confinement (Fig. 1C). We verified that HR asymmetry did not induce any asymmetry in terms of channel wall coating or medium filling (Fig. S1). We let neutrophil cell lines (HL60) as well as immature dendritic cells (iDCs) migrate in the different bifurcations and quantified their bias toward the lower resistance side as we did in the cell simulation (Fig. 1D and movie S1). In accordance with the model prediction, we observed a gradual response to HR of both HL60 (Fig. 1E, consistent with published data (Prentice-Mott et al., 2013) and iDCs (Fig. 1F) up to a certain threshold of 0.61 and 0. 54 resp. (corresponding to about 80% of cells choosing the low resistance path). Neutrophil-like cells were slightly more sensitive to low increments in HR (×5 and ×20) as compared to iDCs. Noticeably, a bias of 1, which corresponds to 100% of cells choosing the low resistance path, was never observed (even when cells faced an infinite HR generated by a dead-end (bifurcation DE/1)). In order to test whether cells were responding to an absolute value of HR or a relative one, we modified our dead-end bifurcation (DE/1) to increase the HR of the low resistance side (bifurcation DE/20, Fig. 1C). In these bifurcations, we observed the slightly decreased bias (Fig. 1E-F) predicted by our model (Fig. 1G). These experimental and theoretical results suggest that cells are sensitive to differences in HR and choose the side with a relatively lower resistance, with a sensitivity that only weakly depends on the absolute HR value. Thus, the experimental data obtained with confined HL60 and iDCs agrees with the predictions obtained from the simulations and validate our hypothesis: barotaxis is a mechanical process in which cells choose the path opposing the least resistance to migration.
A small force imbalance amplified by the actomyosin network helps cells choosing low-resistance paths
In order to better understand how cells respond to HR asymmetry and choose a direction, we characterized the shape dynamics and actin distribution of simulated cells (Fig. 2A and movie S2) and compared it to iDCs facing bifurcations (Fig. 2B and movie S2). We focused on symmetric (X1) and fully asymmetric (DE/1) bifurcations. The dynamics predicted by the model can be interpreted as follows: at the bifurcation, cells form two extending arms (one per outlet). The subsystem constituted by the two arms eventually reaches a critical size and its actomyosin network self-polarizes spontaneously towards one of the two available directions through a contractile instability analogous to the spontaneous polarization mechanism observed in 1d channels (Callan-Jones and Voituriez, 2013). This leads to the retraction of the arm with increased actin content and thus the choice of a direction (Fig. 2C). Strikingly, even in asymmetric bifurcations, both arms extend symmetrically, independently of HR values, and only a small difference in extension could be observed prior to the directional choice (Fig. 2C). In the model, this effect results intrinsically from the fact that the choice of direction is made through a dynamic instability of the actomyosin cytoskeleton. When the total spatial extension of the subsystem constituted by the two upper arms is smaller than a critical length, this subsystem is stable and only weakly responds to the hydraulic resistance bias. When reaching the critical length, this upper subsystem becomes unstable, its actomyosin network self-polarizes and dramatically amplifies any infinitesimal difference existing between the extensional speeds of the two arms. In symmetric bifurcations, this difference is exclusively due to fluctuations and the resulting cell direction is thus random. In contrast, in asymmetric bifurcations, the arm extensional speed is on average slightly faster toward the path of least resistance. The amplification of such small difference therefore leads to a significant barotactic bias toward this path (See Supplementary Information). These non-trivial predictions of the model were confirmed experimentally by quantifying arm extension in iDCs. When encountering a bifurcation, cells extended two arms (one in each microchannel path) until one arm retracted, resulting in a choice of direction (Fig. 2B). Strikingly, both arms extended at very similar speeds, even in asymmetric bifurcations (Fig. 2D and S2). This experimental observation strongly suggests that barotaxis is well described by the model: the difference in mechanical load stemming from the difference in the HR of the two paths slightly slows down the arm facing the highest resistance path; this narrow difference is then amplified by the spontaneous polarization of the actomyosin network, leading to a strong bias in the directional choice.
To obtain direct experimental evidence for the role played by the actomyosin system in this amplification mechanism, we quantified the distribution of actin in simulated cells and LifeAct-GFP transgenic iDCs. In both symmetric and asymmetric bifurcations, we observed that retraction of the losing arm was associated to actin accumulation (Fig. 2E and S2). This accumulation was observed earlier when cells were facing asymmetric bifurcation compared to symmetric bifurcations (Fig. 2E, right), consistent with the model’s predictions (Fig. 2E, left).
These findings support the idea that the actomyosin network does indeed spontaneously polarize and directly controls arm retraction and thus the choice of direction. Accordingly, we experimentally found that impairment of actomyosin contractility (in Myosin II knock out (KO) iDCs) decreased the directional bias toward the low resistance path (Fig. 2F), as predicted by the model (Fig. 2G). Our findings thus demonstrate that HR asymmetry strongly biases cell directionality by generating a small force imbalance that is further amplified by the actomyosin cytoskeleton. Therefore, the intrinsic properties of the actomyosin system are sufficient to explain barotaxis, with no requirement for specific pressure sensors.
Macropinocytosis provides immature DCs with a unique capacity to overcome barotaxis
We reasoned that an important parameter for barotaxis should be the permeability of the cell to the surrounding fluid, which is quantified by the cell-fluid resistance parameter ζ. If cells oppose a strong resistance to permeation flows (low permeability, large ζ), the forces experienced by the two cell arms at a bifurcation are controlled by the difference in hydraulic resistance, and could be in principle infinite in the case of a dead-end. On the other hand, if fluid flows freely through the cell (large permeability, low ζ), migration can occur without moving the extracellular fluid and hydraulic resistance is irrelevant. To test this parameter, we quantified barotaxis in our model for different levels of cell permeability (obtained by varying ζ/R). We consistently found that cells endowed with a large effective permeability (ζ/R=10−3) are insensitive to barotaxis (bias of 0), while cells exhibiting a small effective permeability (ζ/R=10−1) are extremely sensitive to HR asymmetry (Fig. 3A). Therefore, our experimental results showing that cells have a strong bias towards open ends, but that the bias never reaches 100% (Fig. 1), suggest that they are slightly permeable but still oppose a resistance to fluid flow.
These observations, together with the predictions of the model, made us wonder whether some specific cells could be permeable enough to exhibit no bias at all and freely explore dead-ends, similarly to the largely permeable simulated cells (ζ/R=10−3, Fig. 3A). Cells can be permeable to fluid by leaving small interstices between the cell body and the channel walls, or by taking up fluid at their front and releasing it at their back. Different mechanisms can be at stake: (1) fluid can enter and move passively out of the cell using membrane channels such as Aquaporins or (2) fluid can be taken up and secreted via vesicles. The latter is particularly efficient in iDCs as they are endowed with the ability to constitutively internalize extracellular fluid through macropinocytosis (Sallusto et al., 1995). This evolutionary conserved mechanism relies on the formation of giant vesicles (> 200 nm) where liquid is internalized in a non-specific manner (Buckley and King, 2017). We thus hypothesized that iDCs performing macropinocytosis may not be barotactic. To test this hypothesis, we manipulated the macropinocytic capacity of iDCs using different strategies. First, we modified the dimensions of Y-shaped microchannels to allow for efficient formation of macropinosomes. Indeed, in small channels as the ones used in Fig. 1 and 2 (< 20 um2-section), F-actin is barely recruited to the cell front (our unpublished data) and therefore macropinosomes do not form, contrary to what had been observed in larger channels (> 30 um2-section (Chabaud et al., 2015)) (Fig. 3B). We thus quantified the bias of HL60 cells and iDCs in symmetric (X1) and asymmetric (DE/1) bifurcations of both small (18 µm2) and large (32 µm2) cross-sections. HL60 cells, which are known as non-macropinocytic cells, exhibited no bias in symmetric bifurcations, but were strongly barotactic when facing dead-ends (Fig. 3C). Yet, we observed a small decrease in barotaxis in large channels compared to small channels. This can probably be explained by the lower level of confinement of HL60 cells in the large channels, which could let fluid flow freely on the sides of the smaller cells. iDCs are bigger than HL60 cells, and are thus well confined both in small and large channels (Fig. 3B). In small channels, iDCs had most of their actin localized at the back, and did not perform any giant vesicle, suggesting they are not macropinocytic in these conditions (Fig. 3B, top). As expected, there were biased toward low resistance path when facing asymmetric bifurcations (Fig. 3D). However, in large channels, we could observe large vesicles covered in actin at the front of the DCs (Fig. 3B, bottom), suggesting they do perform macropinocytosis, and as predicted by the model, they totally lost barotaxis and explored dead-ends equally well as open-ends (Fig. 3D). Consistent with these data, pharmacological inhibition of macropinocytosis using EIPA (also known as Amiloride) or Rottlerin, two specific macropinocytosis inhibitors described in the literature (Koivusalo et al., 2010; Sarkar et al., 2005; West et al., 1989), restored barotaxis in iDCs migrating in large microchannels (Fig. 3E). These inhibitors had no impact on cells migrating in small channels (Fig. 3E). Equivalent results were obtained when genetically compromising macropinocytosis by knocking out (KO) the gene encoding for CD74/Ii, which recruits Myosin II at the front of DCs to promote macropinosome formation (Chabaud et al., 2015): Ii KO iDCs were non macropinocytic and barotactic even in large channels (Fig. 3F). We further confirmed these findings by indirect inhibition of macropinocytosis (Vargas et al., 2016) using the Arp2/3 inhibitor CK666 (Fig. S3). Altogether these results show that, as predicted by our model, permeability to external fluid is a key parameter for barotaxis. They demonstrate that macropinocytosis can efficiently suppress the directional bias introduced by HR, likely by increasing cell permeability to extracellular fluid.
Down regulation of macropinocytosis in mature DCs restores barotaxis
When iDCs are activated by danger-associated signals (such as microbial components), they enter into a “maturation” program that triggers their migration to lymphatic vessels and lymph nodes where they activate T lymphocytes to launch the adaptive immune response. Importantly, macropinocytosis is down regulated during this maturation process (Sallusto et al., 1995). Accordingly, we found that treatment of iDCs with the bacterial wall component lipopolysaccharide (LPS) was sufficient to inhibit macropinocytosis and promote barotaxis even in large channels (Fig. 3G). DCs KO for the protease Cathepsin S (CatS), which remain macropinocytic upon LPS treatment (Chabaud et al., 2015), were not barotactic in large channels (Fig. 3H). This was also confirmed by treating mature DCs (mDCs) with the formin inhibitor Smifh2: as previously shown (Vargas et al., 2016), this small molecule impaired macropinocytosis down-regulation and accordingly, it compromised barotaxis (Fig. S3). Hence, the down-regulation of macropinocytosis that accompanies DC maturation is associated to the acquisition of barotactic properties.
Macropinocytosis increases with hydraulic resistance
Our results showing that macropinocytosis allows iDCs overcoming HR and exploring dead-ends prompted us to closely monitor this process before and after cells chose their direction. For this, we followed in time the distribution of LifeAct-GFP, as we have shown that actin accumulation at the front of iDCs can be used to quantify macropinocytosis (Vargas et al., 2016). Interestingly, we found that in small channels, which diminish the macropinocytic process, the few cells that showed actin accumulation at their front before reaching the bifurcation (Fig. S4) chose open-end at 51% and dead-end at 49%, thus rendering no bias. In contrast, cells showing no actin accumulation at their front before reaching the bifurcation chose open-ends at 78%. Hence, even in small channels, the few cells that displayed characteristics of macropinocytosis (front actin) are insensitive to HR whereas non-macropinocytic cells, including mDCs, strongly undergo barotaxis.
When analyzing the few iDCs that chose dead-ends in these channels, we observed that most of them reached the dead-end, suggesting that they were able to cope with the amount of fluid that was contained in the channel. Remarkably, we noticed that they actively extended and retracted their front and ingested increased amounts of extracellular fluid compared to the iDCs choosing open-ends (Fig. 4A-B). Interestingly, this change of behavior was also observed when comparing iDCs migrating in dead-ends versus open-ends in large channels (Fig. 4A-C). We therefore hypothesized that elevated HR stimulates the macropinocytic process itself, thereby helping cells cope with the external fluid while migrating in high-resistance paths. Signatures of the macropinocytic activity include a decrease in cell speed and an increase in the protrusion/retraction activity in addition to actin accumulation at the cell front (Chabaud et al., 2015; Vargas et al., 2016). Strikingly, all these hallmarks of macropinocytic cells were observed in dead-ends but not in open-ends in small channels, and were enhanced in dead-end large channels compared to open-end counterparts (Fig 4D-F and S4). These observations indicate that iDCs react to changes in HR by increasing their capacity to ingest extracellular fluid, suggesting that macropinocytosis can be modulated by environmental physical cues.
Macropinocytosis favors space exploration while barotaxis enhances directionality in complex environments
The lack of sensitivity of iDCs to HR might improve their capacity to explore tissues in an unbiased way, making them more efficient as immune sentinels. Therefore, macropinocytosis, which is used by iDCs to sample tissues for the presence of danger-associated molecules-for example those belonging to infectious agents-a key step in the initiation of immune responses, could also contribute to the patrolling function of iDCs by allowing them to overcome barotaxis and thus explore spaces inaccessible to other cells such as dead-ends.
In order to investigate the role of macropinocytosis in tissue exploration in vivo, we generated mixed bone marrow chimeric mice containing both macropinocytic (wild-type, WT) and non-macropinocytic (Ii KO, (Chabaud et al., 2015)) iDCs that can be distinguished based on the fluorescent protein (GFP or YFP, resp.) they express. The migration of iDCs in the ear skin was assessed by two-photon imaging, at steady state or after induction of an edema by sub-cutaneous injection of λ-carrageenan (Winter et al., 1962) (Fig. 5A). Space exploration was quantified as the volume explored by Ii WT or Ii KO cells over time (Fig. 5B). At steady state, similar volumes were explored by macropinocytic DCs (WT) and non-macropinocytic DCs (Ii KO). In contrast, macropinocytic DCs (WT) were significantly more exploratory than their non-macropinocytic counterparts (Ii KO) in the presence of an edema (Fig. 5C). Hence, cells permeable to fluid, and thus insensitive to HR, such as macropinocytic immature DCs, have an increased capacity to explore inflamed tissues in which the volume of fluid to displace is higher. This might help iDCs to uphold their sampling function in inflamed tissues.
Conversely, in mature DCs, which are no more exploratory but need to reach lymph nodes rapidly and efficiently, barotaxis might be beneficial. To address this question, we took advantage of our theoretical model by analyzing the stochastic dynamics of simulated cells in mazes whose nodes are interconnected randomly to only three nearest neighbors and whose edges have identical HR (Fig. 5D). The starting position in the maze is at the bottom left corner and the only exit is located at the opposite corner. Simulated cells visit multiple nodes before reaching the exit, and in each of them they encounter a Y-shaped bifurcation. Importantly, at each bifurcation, the path of lowest HR is precisely the shortest path to the exit. We observed two qualitatively different modes of exploration controlled by the ratio ζ/R between cell-fluid resistance and hydraulic resistance (See Supplementary Information). Highly permeable simulated cells (i.e. weakly barotactic, highly macropinocytic, ζ/R=10−3) chose randomly between both outgoing paths at every node. As a result, they tended to explore all accessible nodes in the random maze before exiting, performing a purely diffusive motion through the network (Fig. 5E, left). Impermeable simulated cells (i.e. strongly barotactic, weakly macropinocytic, ζ/R=10−1) were able to follow the path of lowest resistance, and therefore the shortest path to exit among all possible configurations (Fig. 5E, right). Explored area (mean area explored over the total network area) and escaping time (mean time to escape the maze) were quantified for both types of simulated cells (Fig. 5E). This showed that non barotactic cells were the most efficient at fully exploring the network, while barotactic cells, which selected the path with the least resistance at each node, were able to identify the shortest route across the random maze. Similar results were obtained for simulation in mazes presenting two exits (Fig. S5). This capacity to avoid dead-ends or longer paths might help mature dendritic cells to navigate the complex networks such as peripheral lymph vessels. Consistent with this finding, we had reported that CatS KO DCs, which do not down-regulate macropinocytosis as they mature, were defective when migrating from peripheral tissues to lymph nodes (Faure-Andre et al., 2008). Altogether, our experimental and model results suggest that modulation of barotaxis by macropinocytosis in dendritic cells might dose how much their physical environment affects their migration, to adapt their trajectories to their function as immature and mature cells.
In summary, we present here an approach that combines physical modeling and experiments to unravel the mechanisms underlying barotaxis of cells migrating in confinement. Our data show that barotaxis can be simply explained by a small asymmetry in the resisting forces due to HR, amplified by the acto-myosin cytoskeleton. We also found that, in iDCs, macropinocytosis cancels barotaxis, by increasing the permeability of cells to the extracellular fluid. This helps iDCs to explore larger territories than non-macropinocytic cells, which might facilitate their tissue sampling function. Upon microbial sensing, DCs mature and down-regulate macropinocytosis, thus regaining a barotactic behavior. Guidance by differences in HR might help mDCs avoiding dead ends while migrating to lymph nodes. This study therefore proposes a novel and unexpected role for macropinocytosis in modulating to which degree cell trajectories are biased by physical cues. It also suggests that HR might be an important cue for immune cell migration through tissues. How these physical cues interact with the chemical cues formed by chemokines gradients in vivo shall now be addressed.
Author contributions
Conceptualization: HDM, CBM, RA, RV, MP, AMLD. Methodology: HDM, CBM, RA, PB, JFJ, RV, MP, AMLD. Investigation: HDM, RA, ZA, CBM. Formal analysis: HDM, CBM, ZA, MM. Resources: PB. Writing: HDM, CBM, RV, MP, AMLD. Visualization: HDM, CBM, MM. Supervision: RV, MP, AMLD.
Declaration of Interests
The authors declare no competing interests.
Figure legends
Movie S1: Examples of iDCs, migrating in small channels presenting three types of bifurcation: symmetric ((R+ΔR)/R=1, left), weakly asymmetric ((R+ΔR)/R=5, middle) or strongly asymmetric (DE/1, right).
Movie S2: Actin distribution in simulated cell (left) and iDC (right) migrating in symmetric bifurcations.
Movie S3: Examples of iDCs migrating in open-ends or dead-ends, in small or large channels. Green: LifeAct-GFP. Red: Dextran-AlexaFluor 647.
Movie S4: Space exploration by iDCs migrating at steady-state in the ear dermis. Magenta: Ii-WT cells. Cyan: Ii-KO cells. Yellow outline: cumulated space explored.
Movie S5: Space exploration by iDCs migrating after local edema induction in the ear dermis. Magenta: Ii-WT cells. Cyan: Ii-KO cells. Yellow outline: cumulated space explored.
Acknowledgements
We acknowledge the Cell and Tissue Imaging Facility of the Institut Curie (PICT), a member of the France BioImaging national Infrastructure (ANR-10-INSB-04) and the Curie animal facility. We thank C. Hivroz and M. Bretou for insightful comments on the manuscript. We thank Z. Garcia, J. Postat and C. Grandjean from the Bousso lab for technical help. This work was supported by grants from Fondation pour la Recherche Médicale (FRM SPF20140129479) and Association pour la Recherche contre le Cancer (ARC-PDF20140601095) to HDM, the DCBIOL Labex (ANR-10-IDEX-0001-02-PSL and ANR-11-LABX-0043) to A.M.L.-D., as well as the ANR (PhyMax), the Fondation pour la Recherche Médicale and the Institut National du Cancer to A.-M.L.-D., M.P. and R.V.