Phosphorylation triggers presynaptic phase separation of Liprin-α3 to control active zone structure

Liquid-liquid phase separation enables the assembly of membrane-less subcellular compartments, but testing its biological functions has been difficult. The presynaptic active zone, protein machinery in nerve terminals that defines sites for neurotransmitter release, may be organized through phase separation. Here, we discover that the active zone protein Liprin-α3 rapidly and reversibly undergoes phase separation upon phosphorylation by PKC at a single site. RIM and Munc13 are co-recruited to membrane-attached condensates, and phospho-specific antibodies establish Liprin-α3 phosphorylation in vivo. At synapses of newly generated Liprin-α2/α3 double knockout mice, RIM, Munc13 and the pool of releasable vesicles were reduced. Re-expression of Liprin-α3 restored these defects, but mutating the Liprin-α3 phosphorylation site to abolish phase condensation prevented rescue. Finally, PKC activation acutely increased RIM, Munc13 and neurotransmitter release, which depended on the presence of phosphorylatable Liprin-α3. We conclude that Liprin-α3 phosphorylation rapidly triggers presynaptic phase separation to modulate active zone structure and function.


Introduction
We next assessed whether these fluorescent droplets are indeed membrane-free protein dense 120 condensates. We used correlative light-electron microscopy (CLEM) and found that Liprin-α3 121 condensates were electron-dense structures without surrounding lipid bilayers (Figs. 1f, 1g). We 122 conclude that Liprin-α3 rapidly and reversibly forms phase-separated condensates as a function 123 of PLC/PKC signaling. 124 125 PKC phosphorylates Liprin-α3 in vitro and in vivo to trigger condensate formation 126 We hypothesized that PMA triggers PKC activation followed by phosphorylation of Liprin-α3 to 127 induce phase separation. To investigate whether Liprin-α3 is a PKC substrate, we purified GST-128 fusion proteins covering the entire Liprin-α3 protein, and incubated them with 32 P-labelled ATP Extended Data Fig. 2c). Upon PMA addition, and consistent with the PKC consensus sequence, 140 the phospho-S760-Liprin-α3 increased, and disappeared when co-incubated with PKC blockers, 141 while phospho-S764 signals were unchanged. Phospho-S760 Liprin-α3 was not detected in 142 Liprin-α3 knockout neuronal cultures (Fig. 2d), confirming antibody specificity. In vivo, phospho-143 S760 Liprin-α3 was present in the frontal cortex, hippocampus, cerebellum and brain stem with 144 high perinatal levels that gradually decreased over time (Extended Data Fig. 2d). Our data establish that PKC phosphorylates Liprin-α3 at S760. To corroborate that this site 147 mediates phase separation, we generated phospho-dead (S760G, Liprin-α3 SG , using S->G 148 substitution to make it similar to other Liprin-α proteins, Extended Data Fig. 2e) and phospho-149 mimetic (S760E, Liprin-α3 SE ) mutants. Liprin-α3 SG abolished PKC-induced phase separation, 150 and Liprin-α3 SE formed constitutive condensates independent of PKC activation (Figs. 2e, 2f).  Liprin-α3, RIM1α and Munc13-1 are co-recruited into membrane-attached liquid 160 condensates 161 We reasoned that if phase separation of Liprin-α3 controls active zone assembly, active zone 162 proteins must interact with this liquid phase. Co-expression of cerulean-Liprin-α3 with either 163 RIM1α-mVenus or Munc13-1-tdTomato in HEK293T cells resulted in recruitment of each protein 164 into PMA-induced condensates (Extended Data Fig. 3a). Discrete, PMA-insensitive 165 condensates were also observed when RIM1α was expressed alone (Extended Data Fig. 3b), in 166 agreement with its intrinsic ability to phase separate 6 . Munc13-1 did not form droplets on its 167 own, but PMA-dependent membrane recruitment was observed as previously described [29][30][31] . number and size (Figs. 3a-3c). Remarkably, these condensates were not distributed throughout 172 the cytosol, different from Liprin-α3 phase condensates. Instead, they were in the cell periphery 173 in close proximity to the plasma membrane, and the condensates contained all three proteins.

174
To assess whether they were membrane attached, we used CLEM on PMA-treated cells. The 175 fluorescent signals were highly overlapping with large protein densities that were not enclosed 176 by membranes, but instead appeared attached at one side to the plasma membrane (Figs. 3d, 177 3e).

179
We finally used FRAP to assess turnover of Liprin-α3, RIM1α and Munc13-1 in these 180 condensates. All three proteins rapidly recovered when the entire condensate was bleached   If Liprin-α phase separation controls active zone assembly, Liprin-α knockout should impair its 209 structure and function. We generated new knockout mice to simultaneously ablate Liprin-α2 and to previously generated constitutive Liprin-α3 knockout mice (Liprin-α3 -/-) 16 (Fig. 4a). We used 214 cultured hippocampal neurons of Liprin-α2 f/f /Liprin-α3 -/mice infected with lentivirus expressing 215 cre recombinase (to generate KO L23 neurons) and neurons from Liprin-α2 f/f /Liprin-α3 +/mice 216 infected with lentiviruses that express truncated, inactive cre recombinase (to generate 217 control L23 neurons). First, we assessed the composition of synapses by confocal microscopy by 218 measuring protein levels within synapses (Fig. 4b). Liprin-α2 and Liprin-α3 were efficiently 219 removed and the remaining signals are typical for antibody background 19,33 . The levels of RIM, We next asked whether these changes in protein levels occur at the active zone and whether 226 they are present in both excitatory and inhibitory synapses (marked with PSD-95 and Gephyrin, 227 respectively (Extended Data Fig. 7)). At side-view synapses, the peak levels of Munc13-1, RIM,      We reasoned that if Liprin-α3 functions depend on its propensity to phase separate, its ability to 281 rescue should be altered when S760 is mutated to abolish phase separation. We directly 282 compared the ability of wild type Liprin-α3 and Liprin-α3 SG to rescue RRP and RIM ( Fig. 5s-5w).  Liprin-α3 SG expressing neurons ( Fig. 6a-6c), establishing that Liprin-α phosphorylation is 300 important for this enhancement. Similarly, the RRP estimated by hyperosmotic sucrose 301 14 application was increased, but this was significantly tempered when non-phosphorylatable 302 Liprin-α3 was present ( Fig. 6g-6i). It is noteworthy that the RRP enhancement is overestimated 303 because of the robust increase in mEPSC amplitude (Fig. 6f), and as a consequence the 304 impairment in pool enhancement of Liprin-α3 SG is likely underestimated. In summary, these data 305 indicate that PKC phosphorylation and phase separation of Liprin-α3 modulate the RRP.

307
We finally investigated whether Liprin-α3 phase separation controls active zone structure. We  Self-assembly of proteins into liquid phases is a biophysical mechanism used by cells for the 320 formation of membrane-less compartments 1,2 . We investigated molecular pathways that drive       gel electrophoresis and Coomassie staining using increasing BSA concentrations as reference.

492
The following GST-tagged proteins were produced from pGEX-KG2 constructs: pGEX Liprin-α3 Custom antibodies were generated using procedures as described 34 . In brief, phospho-specific 510 Liprin-α3 antibodies were generated using keyhole lympet hemocyanin (KLH) conjugated 511 CKAPKRK(pSer)IKSSIGR or CAPKRKSIKS(pSer)IGRL, for phospho-S760 and phospho-S764, 512 respectively. KLH-conjugated peptides were injected into rabbits whose sera had been pre-513 screened to prevent non-specific antibody signal. Rabbits were given boosters every 2 weeks 514 and bleeds were collected every 3 weeks. Serum that showed the strongest Liprin specificity in 515 western blotting were processed by affinity purification as described 34 . Western blotting 518 Samples were prepared in SDS sample buffer as described 32 , run on SDS-PAGE gels and 519 transferred to nitrocellulose membranes at 4 o C for 6.5 hr in buffer containing (per L) 200 mL 520 methanol, 14 g glycine and 6 g Tris, followed by a 1 h block at room temperature in saline buffer 521 with 10% non-fat milk powder and 5% normal goat serum. Primary antibodies were incubated 522 overnight at 4 o C in saline buffer with 5% milk and 2.5% goat serum, followed by 1 h incubation 523 at room temperature with horseradish peroxidase-conjugated secondary antibodies prior to 524 visualization of the protein bands. Primary antibodies used: rabbit anti Liprin-α1 (A121, 1:500), 525 Liprin-α2 (A13, 1:500), Liprin-α3 (A115, 1:500) and Liprin-α4(A2, 1:500) were gifts from S. 526 Schoch 35 ; rabbit anti phospho-760 Liprin-α3 (generated for this study; A231; 1:1000) and  within ROIs was quantified after local background was subtracted using the "rolling average" 583 ImageJ plugin (diameter = 1.4µm). Data was plotted normalized to the average intensity of the 584 control group (control L23 ) per culture. For co-localization analyses, the "Coloc 2" imageJ plugin 585 was used following default thresholding. For example images, brightness and contrast were 586 linearly adjusted equally between groups and interpolated to meet publication criteria.     respectively. When test assumptions were met, parametric tests (t-test or one-way ANOVA) 692 were used. Otherwise, the non-parametric tests (Mann-Whitney U or Kruskal-Wallis) were used.

693
For paired pulse ratios, a two-way ANOVA was used. Tukey-Kramer or Holm corrections for 694 multiple testing were applied for parametric and non-parametric post-hoc testing. All data were