Summary
Clathrin-mediated endocytosis (CME) underlies intra- and extracellular material trafficking in eukaryotes, and is essential to protein metabolism, intercellular signaling, membrane remodeling and other cell regulatory processes. Although CME is usually driven by F-actin polymerization, membrane invagination can also occur through actin independent mechanisms. Here, we show that viscoelastic protein condensates that form via liquid-liquid phase separation at the sites of endocytosis initiation facilitate actin independent CME. The work required to drive membrane invagination is generated by binding energies of the condensate with the membrane and surrounding cytosol. Our findings expand the repertoire of functions associated with protein condensates that form via liquid-liquid phase separation to include their ability to do work at soft interfaces, thus shaping and organizing cellular matter.
Introduction
Evolution has resulted in numerous innovations by which morphogenesis of organisms occurs within limits imposed by physical and chemical constraints on the underlying biochemical processes (Darwin 1859, Thompson 1917). One such process is clathrin-mediated endocytosis (CME) a fundamental mechanism of cell surface membrane receptor turnover and recycling, nutrient uptake and synaptic vesicle regeneration, among others (Conner and Schmid 2003). The mechanism of membrane invagination in CME has most convincingly been demonstrated to be growth of membrane-bound branched actin, however CME has also been shown to occur under conditions where actin polymerization is absent and the mechanisms by which this happens remain unclear (Aghamohammadzadeh and Ayscough 2009, Li, Shao et al. 2015). Here, we demonstrate that membrane invagination can arise from liquid-liquid phase separation (demixing) of proteins with prion-like domains (PLD) from the cytosol (Fig. 1a). Demixing of these proteins results in formation of a droplet (or condensate), which, by virtue of its viscoelastic properties, binds to and deforms plasma membrane and cytosol. Demonstration that phase separated droplets can perform mechanical work expands the repertoire of known functions of protein condensates to include the ability to do work at the droplet interfaces. Similar mechanisms may govern or contribute to other membrane shaping, invagination and budding processes that are involved in the cellular material uptake, secretion, and cell shape remodeling.
In S. cerevisiae, the dominant mechanism for vesicle generation in CME is branched actin assembly, which is required to compete against intracellular turgor pressure and membrane tension to drive the invagination of the plasma membrane (Carlsson and Bayly 2014, Dmitrieff and Nedelec 2015). If, however, turgor pressure is eliminated, CME can also occur independent of actin polymerization (Aghamohammadzadeh and Ayscough 2009, Li, Shao et al. 2015). Complementary mechanisms have been proposed to explain actin-independent membrane invagination in CME include intrinsic twisting of the membrane by the clathrin matrix, binding of curved BAR (Bin/Amphiphysin/Rvs) domain-containing proteins (Yu and Schulten 2013), protein domain insertion in the membrane bilayer (Ford, Mills et al. 2002), local relief of turgor pressure (Scher-Zagier and Carlsson 2016), lipid modifications and a reorganization of lipid bilayers (Anitei, Stange et al. 2017) or steric repulsion of coat and adaptor proteins due to their crowding (Busch, Houser et al. 2015, Derganc and Copic 2016). Although the possibility of these mechanisms have been demonstrated in vitro, their importance in vivo remain unknown (detailed in Material and Methods) (Boettner, D’Agostino et al. 2009, Carlsson and Bayly 2014, Kukulski, Picco et al. 2016).
We investigated an alternative potential mechanism of CME in a yeast cell mutant model in which turgor pressure is relieved and actin polymerization is specifically inhibited (Fig. 1a, Fig. S1-2). This potential mechanism was suggested to us by the observation that there is a common amino acid sequence pattern called prion-like domains (PLD) found among coat and adapter proteins (Fig. 1a) (Alberti, Halfmann et al. 2009, Malinovska, Kroschwald et al. 2013). Such proteins are known to phase separate in vitro and in cells. Phase separation leads to spherical condensates or droplets that are hundreds of nanometers to micrometers in size with a range of viscoelastic properties (Guilak, Tedrow et al. 2000, Pappu, Wang et al. 2008, Brangwynne, Eckmann et al. 2009, Hyman, Weber et al. 2014, Banjade, Wu et al. 2015, Jiang, Wang et al. 2015, Kroschwald, Maharana et al. 2015, Molliex, Temirov et al. 2015, Nott, Petsalaki et al. 2015, Zhang, Elbaum-Garfinkle et al. 2015). The idea that membranes can be deformed by liquid-liquid phase separation of droplets is supported by in vitro evidence of membrane nanotubes formed by displacement of small polymer droplets contained within giant phospholipid bilayer membrane vesicles (Li, Lipowsky et al. 2011). We postulate that such droplets exist at CME initiation sites and that, owing to their viscoelastic properties and interfacial tension, bind to the plasma membrane adaptors and generate a force that drives invagination of the membrane (Hertz 1882, Johnson 1971, Style, Hyland et al. 2013).
PLD-containing CME proteins accumulate and phase separate at cortical sites
Evidence that a protein droplet (henceforth called the cortical droplet) could form at CME sites include first, electron and light microscopic studies that reveal a region surrounding CME membrane invaginations and mature vesicles of ~200 nm diameter that are devoid of ribosomes (Kukulski, Schorb et al. 2012, Picco, Mund et al. 2015). This “exclusion zone” thus appears to present a physical barrier to large molecular complexes at least as large as ribosomes (> 10 nm) (Kukulski, Schorb et al. 2012). Furthermore, we and others have observed an object at cortical sites of ~200 nm diameter by super-resolution imaging of the endocytic coat protein Sla1 in cells treated with Latrunculin A (Lat A), an inhibitor of actin polymerization. Therefore, the exclusion zone cannot be attributed to F-actin bundles (Fig. 1b, Fig. S3) (Picco, Mund et al. 2015). Our results agree with quantitative immuno-EM data which show that many endocytic coat proteins (including Sla1/2 and Ent1/2) are located in a space of similar dimensions, consistent with a protein droplet that associates with the membrane on cortical sites (Idrissi, Grotsch et al. 2008, Idrissi, Blasco et al. 2012).
The simple alcohol 1,6-hexanediol (HD) has been demonstrated to prevent liquid-liquid phase separation of proteins in vivo and in vitro (Updike, Hachey et al. 2011, Kroschwald, Maharana et al. 2015, Molliex, Temirov et al. 2015, Wheeler, Matheny et al. 2016). CME, as measured by cell uptake of a lipophilic membrane-bound fluorescent dye (FM4-64), was inhibited by HD, whether or not turgor pressure and actin polymerization were present (Fig. 1c, left versus right panels, respectively). Furthermore, an HD dose-response of uptake of the fluorescent dye (Lucifer Yellow) into vacuoles and formation of puncta monitored as Sla1-GFP fluorescence at cortical sites were prevented, but not in cells treated with the related alcohol 1,2,3-hexanetriol that does not disrupt droplets (Fig. 2a, Fig. S4). The other PLD-containing proteins, including Sla2, Ent1, Ent2, Yap1801 and Yap1802, all failed to form puncta in cells treated with HD (Fig. S4). Pulse-chase experiments showed that HD-dependent dissolution of Sla1 puncta was reversible (Fig. S5 and Movie S1). Finally, PLD-containing proteins can also form amyloid aggregates, which can be diagnosed by binding and co-localization of Thioflavin T (ThT) to the aggregates (Khurana, Coleman et al. 2005). We observed no colocalization of ThT with Sla1-mCherry-labelled puncta (Fig. S6).
The PLDs of cortical CME proteins were essential to their localization to cortical sites (Fig. 2b). Furthermore, CME was significantly reduced in cells where the PLDs of Sla1 and Ent1 were deleted and with substitutions of proline for other residues in the Sla1 PLD, which weakens the driving force for phase separation (Fig. 2c, Fig. S7) (Toombs, McCarty et al. 2009, Crick, Ruff et al. 2013). Our results support evidence that there is a functional redundancy among most of the PLD-containing proteins with the two that are more essential, perhaps required for specific functions mediated by other domains within their sequences (Watson, Cope et al. 2001).
The interactions among proteins in liquid-liquid phase separated droplets are expected to be weak and this is assessed by their rapid exchange within and between droplets and their surroundings (Brangwynne, Eckmann et al. 2009, Elbaum-Garfinkle, Kim et al. 2015, Lin, Protter et al. 2015, Feric, Vaidya et al. 2016). In fluorescence recovery after photobleaching (FRAP) experiments we measured equivalent mobile and immobile fractions (0.50 ± 0.02; mean ± sem) for the protein Sla2 and a rapid recovery time (5.96 ± 1.15 seconds; mean ± sem) (Fig. 2d), similar to other protein and nucleic acid droplets including the dense internal fibrillar component of X. laevis nucleoli (Feric, Vaidya et al. 2016). Taken together, these results support the hypothesis that the cortical bodies are phase separated viscoelastic droplets. We next set out to determine the material properties of the cortical droplets and to test our postulate that their binding to the plasma membrane generates the force that drives invagination of the membrane.
Cortical droplets can mechanically deform both cytosol and membrane
We hypothesized that free energy released by cortical droplet phase separation is converted into mechanical work to deform the membrane and the cytosol. This mechanical work is manifested as an inward pressure on the membrane created by expansion of the droplet and the requirement that volume of the droplet is conserved. Phenomena where geometric organization of matter is driven by the balances of opposing forces have been described at subatomic up to stellar scales, examples of which include “fingering instabilities” (Kull 1991, Hester 2008, Xi, Byrnes et al. 2017).
The mechanics of CME can be described by analogy to a soft viscoelastic and sticky balloon bound to a soft elastic sheet (Fig. 4a, Movie S2). If you stuck your finger through the center of the sheet-balloon interface to create an invagination, the surface area of the balloon would have to increase to maintain the volume and density of the balloon constant. Equally, but in an inverse sense, if you were to grasp the sticky surface of the balloon with your hands and pull outwards equally over the surface, except at the elastic sheet-balloon interface, a tiny increase of the surface area would require a compensating adjustment of the shape so that the balloon keeps a constant volume. Since force is being applied outwards everywhere except at the sheet balloon interface, it is here that an invagination of the membrane-balloon interface would compensate for the pressure generated by the outward force on the balloon surface.
In the case of CME, the grasping force is caused by binding of molecules at the cortical droplet-cytoplasm interface. Balance between this binding and elastic deformation energies is achieved when the membrane invaginates. This idea is captured in a simple phenomenological model expressed as the sum of mechanical strain energy (ϕ term) and work (ψ term), respectively;
Here, U is the mean-field energy, δ is the invagination depth of both the membrane and cytosol (which are coupled to each other by virtue of conservation of volume of the droplet) and the exponent ε > 0 reflects the deformation geometry (Material and Methods). Close to equilibrium (as ∂U/∂δ approaches 0) we expect invagination to balance the two contributions so that δ* minimizes energy in (1) resulting in,
Equation (2) shows that the invagination depth d is determined by the ratio ψ/ϕ and the deformation geometry ε. Values of ϕ and ψ can be determined as functions of individual geometries, elasticities, and viscosities of cytosol, droplet and membrane and interfacial tensions among them (Material and Methods). These in turn can be determined by super-resolution imaging (geometries) and elastic and viscous moduli, taken from the literature or determined by active micro-rheology experiments as described next.
We used active rheology to determine the material properties of the cytosol in which cortical droplets are embedded and then, because the droplets are too small to probe directly, we calculated their properties through well-understood relationships between the properties of materials in contact and their resulting geometries, as described below. Specifically, we used optical tweezers to examine the frequency-dependent amplitude and phase responses of polystyrene beads that are embedded in cells (Fig. 3a, Material and Methods). 200 nm diameter polystyrene beads were integrated into cells by osmoporation (Fig. S8) (da Silva Pedrini, Dupont et al. 2014). Measurements of passive diffusion of the beads showed mean square displacements (MSD) close to that of random mechanical noise caused by vibration of the microscope (Fig. S8). Furthermore, we established that the osmoporation procedure did not affect rheological properties of cells by measuring the MSD of expressed viral capsid microNS particles labeled with GFP in untreated or osmoporated cells and showing that their diffusion behaviors were identical (Fig. S9) (Munder, Midtvedt et al. 2016).
For active rheology experiments, we used an acousto-optic device to oscillate the position of the optical trap in the specimen plane at frequencies over four orders of magnitude and measured the displacement of trapped beads from the trap center using back focal plane interferometry (Fig. 3b). We could thus measure the viscoelastic properties of the cytosol surrounding the beads by measuring their phase and amplitude response to the oscillations of the optical tweezers. Then by calculating the power spectrum of unforced fluctuations of the bead we obtained storage (G′) and loss (G″) moduli as a function of frequency (Fig. 3c-d, Fig. S10, Material and Methods) (Fischer, Richardson et al. 2010, Hendricks and Goldman 2017).
In addition to obtaining quantities essential to calculate material properties of the cytoplasm and droplet, active rheology combined with spatiotemporal dynamics of interacting materials can inform of their structures. The mechanical properties of living cells can be compared to that of the popular children’s toy “Silly Putty” (Cross 2012). Like this material, cells and underlying structures show different mechanical properties depending on the rates at which forces are applied to them (Hendricks, Holzbaur et al. 2012, Guo, Ehrlicher et al. 2013, Guo, Ehrlicher et al. 2014). If a force is applied at a low velocity, the cell behaves like a viscous fluid; flowing and taking on whatever shape it is forced into. When a force is applied at higher velocity, however, the material behaves like an elastic object, bouncing back to its original shape. As we discuss below, these behaviors reflect the manner and strengths with which the molecules that make up a material interact with each other and their environment.
In specific terms, the material properties of the yeast cytoplasm and its interactions with the cortical droplet could be interpreted from the complex modulus versus frequency plot as follows (Fig. 3d). The inflection of the G’ modulus at 2 Hz results in similar G’ and G” values at low frequencies, which indicates that the cytosol is more viscous near rest. When deformed by the droplet growth (at a velocity of growth = 2360 ± 120 nm s-1; corresponding to a stress at ~30 ± 2 Hz) the cytosol is more elastic, whereas membrane invagination occurs at a rate at which the cytoplasm is more viscous (a velocity of 7.4 ± 2.5 nm s-1; corresponding to 0.1 ± 0.04 Hz) (Fig. 1b, 3d, Fig. S11). The G′ and G″ we measured are similar to the cytoplasm of adherent mammalian cells and indicate that the beads are confined within a dense network of interacting molecules (Hendricks, Holzbaur et al. 2012, Guo, Ehrlicher et al. 2013, Guo, Ehrlicher et al. 2014).
We could now determine the mechanical properties of the cortical droplet as follows. First, our data are consistent with both cortical droplets and cytosol behaving as predominantly elastic materials (Fig. 3d). Classic Hertz theory relates contact geometries of elastic materials to their mechanical properties. We could thus, use the geometry of the cortical droplets determined in our super-resolution imaging experiments, and the moduli of the cytosol in which they are embedded to estimate the cortical droplet elastic modulus to be 59 Pa (Fig. 1b, 3d, Material and Methods; Eq. 3.7-3.10) (Hertz 1882). These results are consistent with protein condensates that form elastic materials (Reichheld, Muiznieks et al. 2017) and suggest that the cortical droplets have similar material properties as the surrounding cytosol, which has an elastic (or Young’s) modulus of 45 Pa at 1 Hz (Material and Methods). We estimated the average mesh size and permeability of the cortical droplets by probing them with fluorophore-conjugated dextran molecules of 2.4, 5.8, and 10.4 nm in size. We measured FRAP and colocalization of thes dextran molecules with either Sla1-mCherry or Syp1-mCherry puncta (Fig. 3e-f, Fig. S12-13). Both 2.4 nm and 5.8 nm dextran-FITC recovered equally in the droplet and cytosolic zones. In contrast, the 10.4 nm dextran-FITC molecules scarcely permeate the PLD-rich protein network in the droplet whereas they are mobile in the neighbour cytosol. If cortical droplets are dissolved by addition of 1,6-hexanediol, we observe equivalent mobility of 10.4 nm dextran-FITC between cortical sites, labelled with the protein Syp1-mCherry, which is membrane-bound at cortical patches in an HD-resistant manner, and neighboring cytosol (Fig. 3f, Fig. S13). These results are consistent with an exclusion zone for ribosomes as discussed above and with exclusion of dextrans by known protein-RNA phase separated droplets called P granules (Updike, Hachey et al. 2011, Kukulski, Schorb et al. 2012, Wei, Elbaum-Garfinkle et al. 2017).
Cortical droplet binding to cytosol provides the energy to drive membrane invagination
The deformation of the membrane in response to contact with a soft object depends on the geometries and mechanical properties of the object and the vessel it is in (in our case the cytosol of a cell) and the membrane (Fig. 4a). Evidence from electron and super-resolution fluorescence microscopy indicate that the favored geometry of the membrane is flat with invagination centered in the middle of the droplet (Fig. 4a, lower). Such geometries could be explained by a local radial stress-gradient generated by the droplet adhesion to both the membrane and cytosol, or by local binding of adaptor proteins and distinct lipid composition. Simply stated, as the droplet grows the binding to the cytosol draws it inward and the membrane follows, mediated by its own binding to the droplet and the requirement that the volume of the droplet be conserved.
We could now quantify the work performed by the droplet to invaginate the membrane using the storage and loss moduli obtained from the micro-rheology experiments, geometric data obtained from super-resolution imaging and other data available from the literature, to solve the explicit form of the ϕ and ψ terms (mechanical strain and work, respectively) in Equation (1) as functions of membrane and cytosol invagination δ (Material and Methods; Eq. 4.25-4.26). Using the Young-Laplace equation, we first estimated an interfacial tension for the droplet-cytosol interface to be approximately γdc of 7 × 10-5 N•m-1. This estimate is based on the pressure difference across the cytosolic interface and the droplet mean curvature (Material and Methods; Eq. 4.6). Our estimate for the interfacial tension falls within the range of 10-5 N•m-1 to 10-4 N•m-1 that has been reported for other protein droplets, including nucleoli and P granules (Material and Methods; Eq. 4.9) (Brangwynne, Mitchison et al. 2011, Elbaum-Garfinkle, Kim et al. 2015).
Given our estimates of γdc we also determined the work of adhesion that is released when the droplet surfaces are created, as described by the Young-Dupré equation (Fig. 4a, Material and Methods; Eq. 4.11). We calculated an adhesion energy (ψ) of 4.9 × 10-18 J from interactions between the cortical droplet and both the membrane and cytosol (Fig. 4b, Fig. S14, Material and Methods; Eq. 4.26). Our results suggest that the most significant contribution of the mechanical energy comes from the droplet-cytosol interface, where the adhesion energy of 2.9 × 10-18 J is enough to overcome an energy penalty of 2.4 × 10-18 J to deform the membrane and the cytosol. This energy cost includes the elastic, viscous, and interfacial stress penalties (Fig. 4b, Fig. S14, Table S4). We also calculated an average adhesion energy of 1.3 kJ•mol-1 at the droplet-cytosol interface (Material and Methods), which is consistent with the free energies expected of non-covalent interactions (Mahadevi and Sastry 2016).
Our model provides a physical framework to explain how cortical droplets do the mechanical work needed to induce invagination of membranes in actin-independent CME. The interface between droplets, formed, by phase separation of disordered proteins into cortical bodies, and the cytosol- membrane interface deforms the surrounding materials through adhesive interactions. Invagination occurs when ψ dominates ϕ and this is favored within the observed δ interval of 40 nm to 80 nm (Fig. S14). Notably, this predicted δ interval is within the range of plasma membrane invagination of ~70 nm at which point a membrane scission mechanism is activated and vesicle generation is completed (Idrissi, Blasco et al. 2012).
We propose that cortical droplets store and dissipate mechanical energy in the form of surface tension, whereby the composition of the droplets determines their interfacial interactions and provides the energy for adhesion and invagination of membranes. Accordingly, the underlying energy stored within the droplets and the balance of interactions amongst droplet components and solvent governs the nature of the interface. The effective potential energy ψ of droplets, which is equivalent to the total work of adhesion, should be dictated by the density and strengths of physical interactions amongst proteins within the droplet (the droplet cohesion and interfacial tensions). We tested this hypothesis by weakening the favorable free energies of the protein-protein interactions that hold droplet components together using 1,6-hexanediol (HD). These are the interactions that drive the phase separation of cortical droplets, and so would correspond to a decrease of the droplet surface tension (γdc or ψ). Our model predicts that invagination depth should continuously vary with ψ from Equation (2). We titrated HD below the effective concentration that prevents protein phase separation and quantified individual membrane excision events by quantifying uptake of the lipophilic membrane probe FM4-64 into cells by fluorescence microscopy (Fig. 2a, Material and Methods). In Lat A treated GPD1∆ cells, this measures the amount of labeled membrane taken up into cells under the action of cortical droplets alone. By increasing subcritical HD concentration (corresponding to a decrease in ψ), the average fluorescence-labeled membrane per vesicle (a proxy for invagination δ) was continuously reduced over one order of magnitude in the value of γdc (Fig. 4c, Material and Methods; Eq. 2.8). This observation fits with the reduced membrane invagination that we predicted at the outset (i.e., that d scales with the ψ/ϕ ratio) when the droplet cohesion (γdc or ψ) is also reduced (Fig. 4c, Material and Methods; Eq. 4.2).
Discussion
Our results provide a framework for answering many questions regarding CME and other membrane budding processes. Given our observations, how is CME coupled to multiple signaling pathways that integrate to regulate vesicle formation? For instance, the PLD-containing CME proteins we investigated are enriched for multiple phosphorylation sites, which undergo changes in response to activation of a CME-regulating signaling pathway (Kanshin, Bergeron-Sandoval et al. 2015). Since the amount and distribution of charge in disordered regions of proteins regulate their interactions and conformations (Das and Pappu 2013), such post-translational modifications may be important to regulating CME. Our fluorescence microscopy and electron micrographic evidence from the literature suggests that the cortical droplet remains associated temporarily with mature vesicles (Kukulski, Schorb et al. 2012). Does the droplet play any role in trafficking and fusing with, for instance, plasma membrane (protein recycling) or lysosome (protein degradation)? CME underlies several fundamental mechanisms of vesicle trafficking and attendant membrane and vesicle protein cargo transport, including late secretory pathways, endocytosis and neuronal synaptic vesicle recycling. Yeast and human proteins implicated in clathrin-mediated vesicle trafficking are enriched for long disordered protein domains (47/23% of proteins with long consecutive disordered regions of 30 residues and more for humans and yeast, respectively) whereas those involved in two other vesicle trafficking systems are not (COPI: 8/5%; COPII: 8/5%) (Pietrosemoli, Pancsa et al. 2013). These observations argue for investigating the generality and conservation of protein droplet adhesion-driven membrane invagination as the basis of clathrin-mediated vesicle trafficking in the absence of actin polymerization.
It is possible that other liquid-liquid phase separated protein and protein nucleic acid droplets may influence cellular sub-structural dynamics and thus contribute to shaping cell, tissue, and organism morphology (Bergeron-Sandoval, Safaee et al. 2016, Bauerlein, Saha et al. 2017). More broadly, interfacial contact potentials between different biological materials could represent a vastly underestimated source of complex pattern formation in biology, such as has been observed in embryonic tissue layers (Foty, Pfleger et al. 1996) or recently in a model of growing brain convolutions (Tallinen, Chung et al. 2016), in protein stabilization (Gupta, Donlan et al. 2017) and in the ability of clathrin-coated structures to wrap around and pinch collagen fibers (Elkhatib, Bresteau et al. 2017).
Author contributions
L.P.S.B. and S.W.M. designed all of the research and R.V.P. helped in research designing; L.P.S.B. performed biological research; L.P.S.B. and H.K.H. performed micro rheology experiments; L.P.B.S., H.K.H., A.J.E. and A.G.H analyzed micro rheology data; L.P.S.B., A.J.E. and S.W.M. analyzed biological data; L.P.B.S., H.K.H. and P.F. developed physical droplet model; L.P.S.B., R.V.P., and S.W.M. combined physical models with data analysis; all authors wrote the paper.
Acknowledgments
The authors acknowledge support from CIHR grants MOP-GMX-152556 (SWM), the US National Institutes of Health grant R01NS056114 (RVP), the Fonds Québécois de la Recherche sur la Nature et les Technologies (SWM an PF) and the Human Frontier Science Program RGP0034/2017 (SWM and RVP). We thank Simon Alberti for insightful discussions and microNS plasmid, Jackie Vogel for strains, Jacqueline Kowarzyk for technical assistance, Susan Liebman for the Sup35 plasmids, Daniel Zenklussen and Pascal Raymond for help with FRAP experiments.
References
- 1.↵
- 2.↵
- 3.↵
- 4.
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.
- 29.↵
- 30.↵
- 31.
- 32.↵
- 33.
- 34.
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.
- 40.
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.↵
- 66.
- 67.
- 68.↵
- 69.
- 70.
- 71.↵
- 72.↵
- 73.
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.
- 82.↵
- 83.↵
- 84.
- 85.