Determinants of affinity, specificity, and phase separation in a supramodule from Post-synaptic density protein 95

Summary The post-synaptic density (PSD) is a phase-separated membraneless compartment of proteins including PSD-95 that undergoes morphological alteration in response to synaptic activity. Here, we investigated the interactome of a three-domain supramodule, PDZ3-SH3-GK (PSG) from PSD-95 using bioinformatics to identify potential binding partners, and biophysical methods to characterize the interaction with peptides from these proteins. PSG and the single PDZ3 domain bound similar peptides, but with different specificity. Furthermore, we found that the protein ADGRB1 formed liquid droplets with the PSG supramodule, extending the model for PSD formation. Moreover, certain mutations, introduced outside of the binding pocket in PDZ3, increased the affinity and specificity of the interaction and the size of liquid droplets. Other mutations within the ligand binding pocket lead to a new binding motif specificity. Our results show how the context in terms of supertertiary structure modulates affinity, specificity, and phase separation, and how these properties can evolve by point mutation.


INTRODUCTION
PSD is a membraneless compartment with a high density of proteins located between the postsynaptic plasma membrane and the cytoplasm of dendritic spines. Dendritic spines of excitatory synapses are plastic, dynamic, and change morphology in response to synaptic strength (Nishiyama and Yasuda, 2015), which is essential for synaptic plasticity, learning, and memory (Lamprecht and LeDoux, 2004). The PSD contains hundreds of different proteins with varying levels of abundance that form a disc-shaped interconnected network (Cohen et al., 1977). Proteins are added and removed in response to synaptic strength, thus constantly remodeling the PSD, but the overall structure of the PSD is retained by a matrix of nearly immobile scaffold proteins (Blanpied et al., 2008).
PSD-95 is the most abundant scaffold protein in the PSD (Cheng et al., 2006) and is required to sustain the molecular organization of the compartment (Chen et al., 2011). PSD-95 has several binding partners in the PSD: CRIPT (Niethammer et al., 1998), SynGap (Zeng et al., 2016), NMDA receptors (Niethammer et al., 1996), cell adhesion proteins (Irie et al., 1997), and other scaffold proteins such as GKAP , emphasizing its importance in a structural organization. PSD-95 is one of four paralogs in humans together with PSD-93, SAP102, and SAP97. They are members of the membrane-associated guanylate kinase family that all share three types of domains: PSD-95, Discs-large, Zona occludens 1 (PDZ), Src homology 3 (SH3), and guanylate kinase (GK) domains, respectively ( Figure 1A). The supertertiary structure formed by the three domains (PDZ, SH3, and GK) is denoted the PSG supramodule. Its function depends on the interplay between the three domains and is not recapitulated by the properties of the isolated protein domains (Feng and Zhang, 2009;Laursen et al., 2020b;Rademacher et al., 2019).
The PSG forms liquid droplets in a complex with SynGap, which binds via its C-terminal to the PDZ3 domain of PSG (Zeng et al., 2016). Based on this observation Zhang and co-workers suggested that liquid-liquid phase separation (LLPS) underlies the formation of the PSD (Zeng et al., 2016). To prove that the PSD forms by LLPS, it was reconstituted by using four scaffold proteins from PSD: PSD-95, GKAP, Shank, and Homer . Indeed, the reconstituted PSD-like assemblies recapitulated many properties of PSD: they could cluster receptors, selectively concentrate enzymes, promote actin bundle formation and expel PDZ is a small (around 100 amino acid residues) protein interaction domain belonging to one of the largest domain families in the human proteome with around 268 members (Christensen et al., 2019). PDZ domains usually bind to the C-terminal (continuous epitope) of other proteins but occasionally also to internal motifs (Mu et al., 2014) and phosphoinositides (Wawrzyniak et al., 2013). PDZ domains are characterized by the C-terminal recognition motif they bind to. Up to 16 distinct specificity classes have been identified in both humans and the nematode Caenorhabditis elegans, suggesting that these classes are conserved across bilaterian animals (Tonikian et al., 2008). However, the three most common classes are categorized based on the amino acid at position P -2 ( Figure 2). The classes have the following PDZ binding motifs (PBMs), where 4 denotes a hydrophobic residue and X any residue: type I (T/S-X-4), type II (4-X-4), and type III (D/E-X-4). PDZ domains are often present in multidomain proteins. Whereas isolated PDZ domains have been thoroughly studied, much less is known about their functional properties and binding specificity in the context of the supertertiary structure, which entails both inter-domain interactions and their dynamics (Tompa, 2012).
In this work, we address general principles that determine the binding affinity and specificity of the PSG supramodule in comparison with PDZ3 alone, by screening 74 putative PDZ binding motifs from proteins present in the PSD ( Figure 1A). In general, both PSG and PDZ3 were found to bind with high affinity to peptides with type I PBM, but with different rankings in specificity for the peptides. In particular, PSG has high affinity and specificity to peptides derived from SynGap and ADGRB1, in comparison with PDZ3. Interestingly, ADGRB1, similarly to SynGap, has a high content of coiled coil (CC) and a type I PBM in its C-terminus, and interaction with PSG resulted in LLPS. We also investigated six designed mutational variants of PSG to shed light on the plasticity of the specificity in the PSG supramodule and its capacity to evolve new functions.

Library design and construction
We designed a human peptide library containing 74 C-terminal 15 amino acid (AA) peptides (Data S1) from proteins with a potential to bind PDZ domains and which are found in the post-synaptic density (PSD), dendrite, PSD membrane or dendritic spine, as reported in Uniprot (www.uniprot.org). Thus, all proteins with hydrophobic (I, V, L, and F) or cysteine residues at the C-terminal position P 0 were chosen for the analysis ( Figure 1A). Two out of the 74 peptides were previously well characterized for both single domain PDZ3 and the three-domain PSG supramodule: CRIPT, the proposed native ligand of PDZ3, which binds with low mM affinity both to PDZ3 and PSG (Laursen et al., 2020a), and the a 1 isoform of SynGap (referred to as SynGap), which binds selectively to PSG as interactions outside of the binding pocket in PDZ3 are required for high affinity (Zeng et al., 2016). The PDZ3 interactome has been comprehensively characterized. It shows type I PBM specificity (Niethammer et al., 1998;Saro et al., 2007;Toto et al., 2016), and requires only six amino acids for high-affinity interaction (Saro et al., 2007;Toto et al., 2016), as interactions outside of the binding pocket do not contribute significantly to the affinity. Too few studies have been carried out with PSG to show any trend regarding structural details of its high-affinity interaction (Laursen et al., 2020a;Figure 2. Continued for the PSG variant, log (K i PSGvariant /K i PSG ) < 0, and blue lower affinity for the PSG variant, log (K i PSGvariant /K i PSG ) > 0. As a general trend, pixels in columns with PSG G322A , PSG G335A , and PSG G322A G335A are red as these proteins have higher affinity than PSG for the peptides. On the other hand, pixels in columns with PSG F325A , PSG I327V , and PSG F325A I327V are blue or white; thus, these proteins have generally lower or equal affinity for the peptides, as compared with PSG. Based on the heatmap, it is possible to identify shifts in binding specificity. For example, the INSY2-derived peptide reported a shift, as pixels for PSG F325A , PSG I327V , and PSG F325A I327V are red, whereas pixels for PSG G335A and PSG G322A G335A are blue. A row with only white pixels means that the peptide does not bind to any of the proteins. (B) K i values were normalized to the K i of PSG:CRIPT, and the logarithm of this value is shown in the heatmap. Thus, 0 (white) represents an affinity similar to that of the wild-type PSG:CRIPT interaction. Color code: dark blue (low affinity/no binding) and red (higher affinity than wild-type PSG:CRIPT interaction). Row with only blue pixels means that the peptide does not bind to any of the proteins. K i values were calculated from IC 50 values determined from fluorescence polarization-monitored displacement experiments performed at room temperature in 50-mM Tris, 100-mM NaCl, 1-mM TCEP, and 0.1% Tween 20 (see Data S1 for numbers). iScience Article Zeng et al., 2016). To investigate structural determinants of the affinity and specificity, six mutational variants of PSG were included in the study: PSG G322A , PSG G335A , PSG G322A G335A , PSG F325A , PSG I327V , and PSG F325A I327V. The mutations were chosen as they were previously shown to have a significant effect by decreasing (F325A and I327V) or increasing (G322A and G335A) the binding affinity of the native ligand CRIPT .

Design of fluorescence polarization assay
The characterization of 592 protein:peptide (8 3 74) interactions required a semi-high throughput experimental set-up that can measure weak interactions (high mM) (Lea and Simeonov, 2011;Nikolovska-Coleska et al., 2004). We used a fluorescence polarization (FP) assay for the library screen ( Figure 1B), in which a labeled probe peptide was displaced from PDZ3 or PSG by unlabeled peptides. A C-terminal peptide (15 AA, GKKVLDTKNYKQTSV) derived from CRIPT was N-terminally labeled with FITC, thus acting as a probe peptide in the FP assay to characterize new as well as previously known PSG:peptide interactions. The peptide derived from CRIPT was selected as the probe peptide as it has a high affinity for wild-type PSG and was previously shown to bind all PSG variants used in the present study   (Laursen et al., 2020a. The affinities were generally 2-to 4-fold lower than those previously measured by kinetic stopped flow experiments, which can be explained by a different assay set-up with regard to temperature and a different N-terminal modification of CRIPT. The FP assay has several advantages in comparison with stopped flow experiments in terms of high throughput and the possibility to measure low-affinity interactions (K d $10-100 mM). However, the accuracy and precision are limited by the size ratio of the protein:probe, and by the affinity of the probe in displacement screening experiments (see below). For example, a better signal-to-noise ratio was obtained for the larger PSG:probe complex than for the smaller PDZ3:probe. For the low-affinity variant PSG F325A I327V , protein solubility hampered complete saturation of FITC-15 AA CRIPT, resulting in a larger variation in the resulting K d from four independent experiments ( Figure S1H).
To determine the affinity of all C-terminal peptides derived from proteins expressed in the PSD, competitive displacement experiments using non-labeled peptide were performed ( Figure 1B). Based on the saturation experiments with FITC-15 AA CRIPT ( Figure S1), the optimal protein concentration for FP-monitored displacement experiments was chosen for each protein. A good assay window is obtained by a concentration of protein that gives 50-80% saturation of the FP probe (Moerke, 2009). Hence, the concentrations for each protein variant were as follows: PDZ3 (1.3-6 mM), PSG (2-3 mM), PSG G322A (0.5 mM), PSG G335A (0.6 mM), PSG G322A G335A (0.3 mM), PSG F325A (3.8-7 mM), PSG I327V (6-12 mM), and PSG F325A I327V (10-14 mM). We used a concentration of PSG F325A I327V that gave a signal below 50% signal, to reduce the amount of protein required for each experiment.

Global mapping of ligand affinity and specificity in the PSG protein-protein interaction matrix
Each protein variant was incubated with 5 nM of FITC-15 AA CRIPT, which was then displaced by increasing peptide concentration (2-fold dilution series) of unlabelled 15 AA CRIPT ( Figure 1E), 15 AA SynGap (Figure 1F) or one of the other 72 peptides in a 96-well plate format. All protein variants were screened in parallel by using the same peptide concentration series for all eight protein variants, thus allowing for a direct comparison of specificity (relative affinity) for the 74 peptides across the variants. The affinity (K i ) was determined from each protein:peptide IC 50 value using the equation described by Nikolovska-Coleska et al.
( Figure 1D) (Nikolovska-Coleska et al., 2004). The FP-determined binding affinities (K i ) of all 74 C-terminal peptides derived from proteins in PSD with the respective PSG variant are summarized in Data S1 and visualized in a heatmap ( Figure 2). Binding is displayed by a colored pixel representing the affinity between PSG variant:peptide as compared with wild-type PSG:peptide (where white is the same affinity) (Figure 2A). iScience Article specificity of the PSG variants for each single peptide, does not show whether any of the peptides have a higher affinity for PSG than the native ligand CRIPT. Therefore, in a second heatmap, all PSG variant:peptide interactions were normalized against one interaction, that for PSG: 15 AA CRIPT ( Figure 2B). Most interactions were found to have lower affinity than native PSG: 15 AA CRIPT, as shown by the overrepresentation of blue color in the heatmap. Three 15 AA peptides derived from CRIPT, FRPD4, and SynGap bound to PSG G322A , PSG G335A , PSG G322A G335A with higher affinity than 15 AA CRIPT to wildtype PSG ( Figure 2B). One single peptide stands out, SynGap, which bound to PSG and even PSG F325A with high affinity. 15 AA SynGap showed a significantly tighter binding to the PSG supramodule than to PDZ3, likely owing to interactions outside of the binding pocket, as previously reported (Zeng et al., 2016). Furthermore, the mutations in PSG had only little effect on the binding affinity of 15 AA SynGap to PSG. In contrast, PSG G322A G335A displays an approximately 8-fold increase in binding affinity to 15 AA CRIPT compared with PSG. On the other hand, the low-affinity PSG variants, PSG I327V and PSG F325A I327V , displayed lower affinity than wild-type PSG for most peptides with few exceptions: INSY2, CNIH2, RHOA, and CCG5 ( Figure 2A).

PDZ3 and PSG have different specificity profiles
Comparison of the binding profiles for PDZ3 and PSG shows similarities and differences. Generally, PDZ3 and PSG bind with high specificity and affinity to all ligands with type I PBM, but we observed differences in the rank of peptides ( Figure 2 and Data S1). PDZ3 displayed binding to 10 peptides with K i below 75 mM listed in order of decreasing affinity: FRPD4, CRIPT, CCG8, SynGap, NETO1, TANC1, RGS14, NLGNX, BCR, NLGN1, and NETO1 (PDZ3). PSG was found to bind eight peptides with K i below 75 mM: SynGap, FRPD4, CRIPT, TANC1, VSTM5, BCR, CCG8, and NLGN1 (PSG).

Mutational paths to high and low affinity and new specificity
The 268 PDZ domains present in our proteome display a range of overlapping specificities. One key question concerning the evolution of new protein function is how many mutations are necessary to switch specificity in a protein-protein interaction. For example, the affinity for unlabelled 15 AA CRIPT obtained in the displacement experiments spanned from 0.22 to 70 mM among the PSG variants, whereas that of 15 AA SynGap varied between 0.24 and 19 mM, suggesting slight switches in specificity ( Figure 2). To further investigate possible specificity switches among the six designed PSG variants (G322A, G335A, G322A G335A, F325, I327, and F325A I327A) we selected eight peptides based on the large-scale screen for more detailed studies: CRIPT, FRPD4, TANC1, NLGNX, BCR, CCG8, INSY2, and SynGap. Seven of the peptides contained a type I motif and one (INSY2) a type II motif. In order to increase the accuracy and precision of the affinity constants for these eight peptides, we determined association (k on ) and dissociation (k off ) rate constants iScience Article using stopped flow spectroscopy. K d is then calculated as the ratio k off /k on (Figure 3 and Data S2). FP was employed for INSY2 when stopped-flow experiments could not be conducted owing to a low-affinity interaction and ITC was used for SynGap ( Figure S2) owing to complex binding kinetics. The residues G322 and G335 are situated outside of the peptide-binding pocket ( Figure 3A) whereas F325 and I327 make direct contact with the bound peptide ( Figure 3E).
The two single mutations G322A and G335A both contributed to increasing the affinity, mainly in an additive fashion, as shown by the low coupling free energy (0 R DDDG c R À0. In other words, the first mutation attenuates the effect of the second mutation. Of all type II peptides in the screen, only INSY2 was identified as a binder, even though two peptides from other proteins, CUX1 and CUX2, had a large hydrophobic group at the P -2 position. CUX1 and CUX2 have a W-X-F binding motif, but large hydrophobic residues both at positions P 0 and P -2 are rare for canonical binding of C-terminal peptides to PDZ domains (Tonikian et al., 2008). PDZ domains that bind to internal peptides can bind to peptides with Phe at the P 0 position, possibly because they have a more plastic binding pocket (Christensen et al., 2019;Penkert et al., 2004).

Correlation between phase separation and affinity
It was previously shown that PSG and a CC fragment of SynGap, denoted ''full-length'' (FL) CC SynGap form LLPS (Zeng et al., 2016). Generally, condensates resulting from LLPS are important for membraneless spatial organization (Li et al., 2012), biomolecular reaction centers, signaling hubs, and support architecture inside the cell (Zhang et al., 2020). PSG variants display a range of affinities toward peptide ligands including a 10-fold difference for 15 AA SynGap (Data S2, Figures 3 and S2). To explore how affinity modulates LLPS, the PSG variants were examined using different assays including sedimentation and turbidity experiments (Figures 4 and S3). First, we mixed PSG WT or PSG G322A G335A with FL CC SynGap and observed the formation of LLPS under the light microscope (Figure 4A). Liquid droplets evolved from a spherical shape to a more elongated shape over time. The elongated shape was most profound for the PSG G322A G335A :SynGap complex, probably owing to a lower LLPS threshold for PSG G322A G335A as compared with PSG WT, which in turn depends on the affinity for SynGap. Therefore, it seems like droplets start to merge into elongated shapes owing to the crowded environment (30 min, PSG G322A G335A :SynGap). Next, we performed a sedimentation assay ( Figure 4B) in which the fraction of LLPS was quantified by SDS-PAGE ( Figures 4C and S3). Indeed, higher concentrations of PSG variant and FL CC SynGap resulted in more LLPS. We observed a discrepancy for PSG F325A in the sedimentation assay, in which a larger fraction of the protein is observed in the pellet fraction than expected from the turbidity assay and affinity ( Figure S3C). A likely explanation is that this variant is unstable over time and precipitates at high protein concentrations.
Finally, in an attempt to assess the kinetics of LLPS we mixed PSG variants with different concentrations of FL CC SynGap and followed the increase in absorbance at 350 nm over time ( Figures 4D and S3). At this wavelength, an increase in absorbance results from light scattering, or turbidity, which can be anticipated to correlate with LLPS. The kinetics were multiphasic including a lag phase showing that the LLPS formation process is complex, as can be expected. Nevertheless, the kinetic transients show that the maximum rate of LLPS formation, occurring between the lag phase and the stationary phase, increases with FL CC SynGap iScience Article concentration ( Figures 4D and S3). At higher FL CC SynGap concentrations, and depending on the PSG variant, the initial binding event(s) occurred too fast to be captured in our experiments as shown by the high absorbance at time zero in the time courses. The observed maximum rate appeared to increase linearly with FL CC SynGap concentration, consistent with a molecular association event ( Figure S3F). A comparison of the maximum rate of LLPS between the PSG variants at 40 mM FL CC SynGap shows a clear correlation where the higher-affinity variants PSG G322A , PSG G335A , and PSG FG332A G335A displayed a higher maximum rate of LLPS ( Figure 4E).
In conclusion, the data on LLPS (Figures 4 and S3) and affinity of SynGap for PSG variants ( Figure S2) show that LLPS formation correlates with affinity ( Figure 4E). This conclusion is supported by light microscopy-monitored liquid droplet formation at different time points for PSG:FL CC SynGap and PSG G332A G335A :FL CC SynGap ( Figure 4A).

ADGRB1 shows LLPS with PSG
It was previously reported that the C-terminal peptide of SynGap induced dimerization of PSG (Zeng et al., 2016). Such dimerization could act as a proxy for phase separation. Therefore, in an attempt to identify new proteins that phase separate with PSD-95, 10 high-affinity ligands identified on the screen were examined for their potential to induce dimerization of PSG ( Figure 4F). Thus, PSG was mixed with each of the ten different 15 AA peptides and with the cross-linker disuccinimidyl glutarate (DSG) for 30 min (Zeng et al., 2016). The amount of crosslinked PSG was quantified by SDS-PAGE. However, all peptides except SynGap (positive control) showed the same amount of PSG cross-linking as afforded by CRIPT (negative control) or water ( Figure 4F), suggesting that dimerization is a unique property induced by SynGap . Whereas dimerization is not required for LLPS (Mitrea and Kriwacki, 2016), it will increase multivalent interactions, which seems to be important for a low LLPS threshold Zeng et al., 2018).
To further investigate LLPS involving putative native ligands for PSD-95 we employed bioinformatics to find proteins with potential for phase separation. LLPS is associated with a high propensity of forming a CC (Ford and Fioriti, 2020;Zhang et al., 2020) as observed for a sequence stretch of 60 amino acids in SynGap. We calculated the CC propensity for all PSD proteins included in the study using the COILS software (Lupas et al., 1991) ( Figure S4). CRIPT was found to have a low CC propensity, which is in line with the finding that CRIPT does not form LLPS with PSG (Zeng et al., 2016). Out of the 74 PSD proteins on the screen, 9 proteins were found to have a high propensity for CC formation ( Figure S4), but only three of them showed an affinity corresponding to a K d < 100 mM, even to the high-affinity PSG G322A G335A variant (Figure 2, Data S1). Myotubularin-related protein 2 (MTMR2) that interacts with PSD-95 to maintain excitatory synapses by modulating endosomal traffic (Lee et al., 2010) and Adhesion G protein-coupled receptor B1 (ADGRB1) were chosen for further analysis, as they combine a high propensity of forming a CC with K i values from the FP-based assay of 95 and 39 mM, respectively, to PSG G322A G335A . Indeed, ADGRB1, but not MTMR2, showed LLPS in initial test studies.
ADGRB1 is a seven transmembrane helix protein highly enriched in the postsynaptic membrane with an N-terminal extracellular and a C-terminal intracellular portion ( Figure 5A). The protein is suggested to Higher concentration of ADGRB1 clearly correlates with higher inclusion of PSD-95 in the pellets representing LLPS, whereas a correlation of lower significance was observed with Homer, Shank, and GKAP. Data were quantified by Lab software from BioRad. The significance (*p < 0.05; **p < 0.01; ***p < 0.001) was evaluated by two-way ANOVA with Tukey test using Prism 9.0 software (GraphPad).

OPEN ACCESS
iScience 25, 105069, October 21, 2022 11 iScience Article regulate Rho signaling through a PDZ:ADGRB1 interaction (Stephenson et al., 2013). The C-terminal part (residues 1,471-1,584) of ADGRB1 is anchored into the PSD, thus, PSD-95 PSG and the C-terminus of ADGRB1 is spatially located such that a physical interaction in the postsynaptic neuron is possible. We therefore subjected this interaction to further experiments. A direct FP-monitored binding experiment with a FITC-14 AA ADGRB1 peptide confirmed the binding to PSG and PSG G322A G335A ( Figure 5B). Furthermore, we observed weaker binding of the ADGRB1 peptide to the single domain PDZ3 constructs, PDZ3, PDZ3 G322A , and PDZ3 G335A ( Figure S5), showing that binding is more specific for the PSG supramodule.
Next, FL CC ADGRB1 (residues 1,471-1,584) was expressed and purified. The far-UV circular dichroism spectrum of FL CC ADGRB1 displayed a minimum shift toward 200 nm rather than the 208 nm expected for an alpha helix, and the second expected minimum at 222 nm was not present ( Figure 5C). Thus, FL CC ADGRB1 has low secondary structure content, as also predicted by the Alpha-Fold 3D model of ADGRB1 (Jumper et al., 2021;Varadi et al., 2022) and suggested from a previous study with the C-terminal part of ADGRB1 (Weng et al., 2019). The binding of FL CC ADGRB1 to PSG and PSG G322A G335A was confirmed with isothermal titration calorimetry ( Figures 5D and 5E).
We then performed the DSG-based crosslinking assay to characterize if ADGRB1 induces dimerization of PSG, like SynGap (Zeng et al., 2016). However, this appears to not be the case, as the gel bands with crosslinked PSG are weak for ADGRB1 in comparison with SynGap, and more similar to the controls, water, and CRIPT ( Figure 5G). PSD-95 has been reported to undergo LLPS with both SynGap and Stg, however in different molar ratio, 2:3 (Zeng et al., 2016) and 1:1 , respectively. Thus, based on the DSG assay and ITC we expected a 1:1 ratio for the ADGRB1:PSG complex. To further address the stoichiometry, we attempted SEC-MALS experiments with PSG and FL CC ADGRB1. However, it was not possible to perform experiments with the protein complex likely owing to phase separation or aggregation at the desired concentration range.
To explore whether FL CC ADGRB1 could phase separate with PSG, like SynGap, we performed light microscopy experiments. Interestingly, the mixing of PSG and FL CC ADGRB1 resulted in the appearance of many small liquid droplets ( Figure 5H). Over time, the small droplets assembled into larger ones. After 7 min, an excess of 15 AA CRIPT was added, resulting in the disappearance of the liquid droplets within seconds. The most likely explanation is that the PSG:FL CC ADGRB1 complex was disrupted by competition, where 15 AA CRIPT replaced FL CC ADGRB1 in the complex. This experiment showed that PSG and FL CC ADGRB1 form reversible LLPS, similarly to the interaction between PSG and FL CC SynGap (Zeng et al., 2016).
In an attempt to quantify the LLPS formation, PSG G322A G335A was mixed with FL CC ADGRB1 and turbidity was monitored over time, at increasing concentrations of ADGRB1 ( Figure 5F) or NaCl ( Figure S6B). The turbidity assay was used to monitor the phase separation of the PSG G322A G335A :FL CC ADGRB1 complex by titrating increasing concentration of FL CC ADGRB1 (0-60 mM) into a PSG G322A G335A solution (28 mM, 100-mM NaCl) and monitoring the light scattering at 350 nm over 10 min. The maximum absorbance is reported against increasing FL CC ADGRB1 or NaCl concentrations. Indeed, PSG G322A G335A :FL CC ADGRB1 LLPS is concentration-( Figure 5F) and salt-dependent ( Figure S6B), similarly as found for SynGap ( Figures 4C, 4E, and S6A) and as previously reported for other protein complexes that induce LLPS (Bai et al., 2021;Zeng et al., 2016). The hypersensitivity to the salt concentration of protein complexes undergoing phase separation has been suggested as a mechanism to regulate protein clustering in neurons by activity-induced ion influx from the synapse (Bai et al., 2021).
Finally, we determined the interaction between ADGRB1 and other PSD proteins with regard to phase separation by using the sedimentation assay. The PSD contains several scaffold proteins and the four most abundant ones are PSD-95, GKAP, Homer, and Shank. These proteins can all undergo LLPS and incorporate other PSD proteins such as SynGap and NR2B . Indeed, we found that ADGRB1 could undergo LLPS and be incorporated in a reconstituted PSD, similarly to SynGap ( Figure 5I).

Multivalency decreases the threshold for phase separation
All three PDZ domains from PSD-95 (PDZ1, PDZ2, and PDZ3) have an overlapping interactome and bind to proteins with type I PBM (Lim et al., 2002). We therefore investigated whether PDZ1 and PDZ2 could also interact with ADGRB1, and compared the affinity with that of SynGap ( Figure S7) iScience Article ADGRB1 ( Figure S7B) and 14 AA ADGRB1 ( Figure S7C) to a PDZ1-2 tandem construct was determined and showed a 3-fold higher affinity in comparison with PSG WT (Figures 5B and 5D), suggesting that the ADGRB1 interaction is not specific for PDZ3 in the PSG supramodule. To further characterize the PDZ1-2:ADGRB1 complex we investigated whether PDZ1-2 could induce phase separation with FL CC ADGRB1. Interestingly, phase separation with FL CC ADGRB1 was not observed for PDZ1-2 under the same conditions as used for PSG ( Figure S7D). Most known PSD-95 complexes have a 1:1 stoichiometry, whereas that between SynGap and PSD-95 has a 3:2 stoichiometry (Zeng et al., 2016). Thus, we performed SEC and SEC-MALS with FL CC ADGRB1 and PDZ1-2 and found that data were most consistent with a 1:1 complex ( Figures S7E and S7F). To corroborate our result, we reconstituted the PSD both with PSD-95 or PSG and PDZ1-2 ( Figures 5I and S8). PSG can form LLPS in the absence of ADGRB1 and SynGap in a multivalent complex with GKAP, Shank, and Homer (4xPSD) even though its K d for GKAP  is 10 times higher than the final concentration of proteins in the reconstituted PSD ( Figure S8C). PDZ1-2 is only enriched in the PSD condensates ( Figure S8B and S8C) at a final concentration of FL CC ADGRB1 ten times above K d ( Figure S7B). In conclusion, SynGap is specific for PSG and induces LLPS, whereas ADGRB1 binds to both PSG and PDZ1-2 with similar affinity, but the threshold for LLPS is significantly lower for PSG than for PDZ1-2 ( Figure S7G).

PSG and ADGRB1 form LLPS in living cells
Next, we tested if PSG and FL CC ADGRB1 display co-localization and undergo a phase transition in HeLa cells ( Figure 6). As negative control experiments, EGFP-FL CC SynGap ( Figure 6A), mCherry-PSG (Figure 6B), and EGFP-FL CC ADGRB1 ( Figure 6C) were individually expressed in HeLa cells, resulting in a uniform distribution of EGFP and mCherry signal. However, when EGFP-FL CC SynGap and mCherry-PSG were co-expressed, many bright puncta with both EGFP and mCherry were observed ( Figure 6D), recapitulating previous experiments and suggesting that SynGap and PSG were enriched in these puncta. Thus, this experiment serves as a positive control and is consistent with the formation of a condensed liquid phase in living cells, as previously demonstrated (Zeng et al., 2016). When EGFP-FL CC ADGRB1 and mCherry-PSG ( Figure 6E) or mCherry-PDZ1-2 ( Figure 6F) were co-expressed in HeLa cells many bright puncta with co-localization of both GFP and mCherry signal were observed. These data suggest that ADGRB1, similarly to SynGap ( Figure 6D), can co-localize and bind to PSD-95. Like for SynGap, this colocalization and the LLPS directly observed in the light microscope ( Figure 5H) indicate that ADGRB1 could form a condensate with PSD-95 in the cell.

Single domains versus supertertiary structure -similar but different
Interaction between two proteins is often mediated via a folded protein domain in one protein and a short linear binding motif in a disordered region of the other one. PDZ domains and their binding to C-terminal sequences provide a typical example of such types of interaction. The majority of all large-scale screens used to identify potential protein-protein interactions use isolated protein domains. However, it is clear that supertertiary structure, i.e., structure and dynamics involving several interacting domains (Tompa, 2012), affects the function of multidomain proteins. We have previously shown that supertertiary structure of the PSG supramodule affects allostery in the PDZ3 domain  and folding of the supramodule (Laursen et al., 2021). In the present study, we found that supertertiary structure also influences the interactome of PDZ3, such that distinct specificity profiles are observed for PDZ3 and PSG. More specifically, whereas both PDZ3 and PSG prefer binding of peptides with a type I PBM, they differ in specificity, i.e., the ranking of peptide ligands based on affinity. In particular, SynGap and ADGRB1 bind with significantly higher affinity to the PSG supramodule than to single domain PDZ3. Furthermore, LLPS involving SynGap and ADGRB1 requires the supramodule PSG. Previous experimental and computational studies, reviewed by Feng and Zhang (2009)  iScience Article (Laursen et al., 2021;Vishwanath et al., 2018), allostery (Kirubakaran et al., 2016;Laursen et al., 2020b), binding affinity (Vistrup-Parry et al., 2021;Zeng et al., 2016), LLPS (Vistrup-Parry et al., 2021;Zeng et al., 2016), and the effect of phosphorylation (Vistrup-Parry et al., 2021) can be affected, when comparing single domain with multiple domain proteins. Here we observe that mutations in PDZ3, as part of PSG, can significantly alter the affinity of CRIPT, whereas the same mutations have a minor effect on the binding of SynGap, likely because noncanonical interactions contribute to the affinity. We also show that FL CC ADGRB1 and FL CC SynGap can undergo LLPS with PSG, but LLPS cannot be induced with PDZ3 alone as multiple interactions are essential. These examples underscore the importance of studying protein interactions in the context of their supertertiary structure. The study of multidomain proteins that induce LLPS will further allow us to reveal the function of allosteric signaling. Decades of allosteric network analysis in PDZ3 indicate that the network is dependent on the context (Gautier et al., 2018), as different networks have been reported in PDZ3 (Gianni et al., 2011), PDZ3Da 3 (Faure et al., 2022;McLaughlin et al., 2012), and PSG . This sensitivity suggests that the allosteric signaling in protein domains will be affected by the crowded milieu in the cell and by LLPS (Kuznetsova et al., 2015), which is further discussed in a recent review by Nussinov et al. (2022).

Change of specificity in PDZ domains
Emergence of new protein-protein interactions frequently occurs during evolution, and the large number of PDZ domains in our proteome demonstrates the versatility of this domain as an interaction module. It is of interest to understand how many mutations are necessary to change specificity in protein interactions. We applied double mutant cycle analysis to measure coupling free energy (DDDG c ), and thereby identify residues that are allosterically coupled (Horovitz and Fersht, 1990). We found a high coupling free energy between F325 and I327 upon binding to INSY2, paralleled by a switch in specificity, from a 67-fold preference for CRIPT over INSY2 (for wild-type PSG), to no discrimination between the peptides (for the double mutant PSG F325A I327V ). The structural basis for this shift from type I toward type II specificity is likely that F325A and I327V make space for the binding of peptides with a large hydrophobic residue at the P -2 position in the binding motif. In a previous study, two other mutations in the binding pocket of PDZ3, G330T and H372A, switched specificity from type I to a type II motif (F-X-4) (Raman et al., 2016). The authors showed that the G330T mutation increased the conformational plasticity of the binding pocket resulting in the conditionally neutral variant PDZ3 G330T with similar affinity for type I and type II PBMs, whereas the H372A mutation switched the specificity. Similarly, the present mutations of residues F325 and I327 in the peptide-binding pocket of PDZ3 in PSG can be described as conditionally neutral or class bridging, as the mutations preserve the original function (binding CRIPT), but also open up for a new function by binding to a type II motif peptide. Obviously, different mutational routes with few mutations can switch specificity from type I to type II in PDZ domains.

Multivalent ligands induce phase separation of scaffold proteins
PDZ3 did not show high specificity and affinity for any particular C-terminal peptide, suggesting that the specificity of PDZ3 is plastic as long as the peptide has a type I PBM. On the other hand, a high-affinity interaction was reported between PSG and SynGap, suggesting that context, i.e., interactions outside of the canonical binding site, shapes domain-peptide interactions (Li et al., 2012), which is important for PDZ domains in scaffold proteins to gain high affinity (Erlendsson et al., 2019) and specificity in a cellular context (Zeng et al., 2016). Comparison of CRIPT and SynGap reveals that both ligands have type I PBM, but only SynGap can be described as a multivalent ligand as interactions outside of the binding pocket contribute significantly to the high affinity of SynGap (Zeng et al., 2016). Multivalency is an important factor for stabilization of LLPS, an ability that monovalent ligands like CRIPT lack. Indeed, monovalent interactions will destabilize LLPS (Ruff et al., 2021) as we demonstrated by the addition of CRIPT to liquid droplets formed by PSG and ADGRB1. A computational study reporting the effect of ligand type on phase separation revealed that LLPS can be induced under the saturation limit of the scaffold protein if a multivalent ligand is present (Ruff et al., 2021). In our study, we observed such behavior for the multivalent ligand FL CC ADGRB1. These results suggest that the multiple interactions formed between PSG and FL CC ADGRB1 are important to boost affinity and achieve a low LLPS threshold. The conclusion is further supported by all the experimental studies with SynGap and PSD-95 (Zeng et al., 2016 and a recent study, where LLPS with tri-Stg was observed for FL PSD-95 at 36 mM but only at a higher concentration (256 mM) for PDZ1-2 . All together the experimental and computational studies show that scaffold proteins can induce LLPS at a concentration threshold lower than its saturation limit in the presence of a multivalent ligand, an important finding for LLPS in vivo, when the endogenous concentration of scaffold proteins can be below the saturation limit (Ruff et al., 2021).

OPEN ACCESS
iScience 25, 105069, October 21, 2022 15 iScience Article Phase transition of the PSG:ADGRB1 complex and its implication for anchoring of PSD to the membrane bilayer We identified ADGRB1 as a potential ligand for PSD-95. Decreased levels of ADGRB1 are reported in patients diagnosed with Parkinson's disease (Choi et al., 2018;Zhong et al., 2020). Overexpression of ADGRB1 resulted in decreased levels of nuclear condensations associated with cell death and Parkinson's disease, suggesting that ADGRB1 can prohibit cell death in the neurodegenerative pathway (a potential therapeutic target) of Parkinson's disease (Choi et al., 2018). In a previous study, ADGRB1 was suggested to prevent polyubiquitination and degradation of PSD-95 through an interaction with the ubiquitin ligase MDM2 (Zhu et al., 2015). An ADGRB1 knockout mouse showed a thin PSD and decreased level of PSD-95, which could be rescued by PSD-95 gene therapy, suggesting synaptic plasticity and a reversible modulation of PSD-95 and PSD through ADGRB1. The indirect interaction between PSD-95 and ADGRB1 was demonstrated by a C-terminal truncation of the PBM (QTEV) in ADGRB1. The truncation decreased the PSD-95:ADGRB1 interaction, but preserved the level of PSD-95 suggesting that modulation of PSD-95 by ADGRB1 is through MDM2 and not through a direct interaction. In the present study, we demonstrate a direct interaction between ADGRB1 and PSD-95, which could even promote LLPS, suggesting a new potential mechanism for PSD to anchor to the transmembrane bilayer. It is clear that multiple interactions are required for full affinity between PSD-95 and ADGRB1, as the coiled-coil domain binds in synergy with the canonical PDZ-binding motif. LLPS is dynamic and reversible, but specific mutations and post-translational modifications will hamper the reversibility and induce protein aggregation, which is a common tell-tale of neurodegenerative diseases. New drug development for neurodegenerative diseases aims to maintain LLPS by suppressing mutations and post-translational modifications that cause aggregation of proteins . Further studies are required to establish the interplay between PSD-95, ADGRB1, and MDM2 and their function in LLPS and protein aggregation.
LLPS currently emerges as a very important property of proteins underlying several functions in the living cell (Alberti and Hyman, 2021). Phase separation facilitates spontaneous and rapid compartmentalization, such as the LLPS induced by the PSD-95:SynGap complex, which is associated with the formation of the PSD (Zeng et al., 2016). LLPS in PSD formation is supported by experiments where the PSD was reconstructed and the concentration threshold for phase separation of PSD-95 and SynGap was reduced by the addition of three other PSD-95 associated scaffolding proteins: GKAP, Shank and Homer . We speculate whether phase separation by the PSG:ADGRB1 complex is associated with PSD anchoring to the membrane bilayer. It is known that PSD-95 can anchor to the membrane bilayer by binding via a second supramodule of PSD-95, the PDZ1-2 tandem, to a PBM from the intracellular C-terminal part of NMDA  or AMPA receptors (Matt et al., 2019). Our current data suggest the possibility of another anchoring point for PSD-95 with the membrane bilayer through the C-terminal part of ADGRB1.
LLPS is concentration-dependent; therefore, liquid droplets of PSG:FL CC ADGRB1 will appear only at a certain concentration threshold. Reduced ADGRB1 expression could decrease the level of ADGRB1 below the LLPS threshold for the PSG:ADGRB1 complex. Zeng et al. reported the first phase transition with PSG through interaction with SynGap (Zeng et al., 2016). They suggested that the dosage sensitivity of SynGap for synaptic functions is a general mechanism for synaptic proteins that undergo LLPS. Whereas both PSG and PDZ1-2 bind to ADGRB1, phase separation seems to be induced by the PSG supramodule, as suggested by its lower LLPS concentration threshold, which is below K d for the PSG:ADGRB1 interaction, upon PSD reconstruction. Thus, multivalent interactions involving an extended interaction interface between PSG and the coiled-coil of ADGRB1 or SynGap are essential for LLPS, consistent with previous findings for protein:protein interactions to induce phase separation (Banjade and Rosen, 2014;Li et al., 2012;Zeng et al., 2016).
In summary, through our screen of proteins expressed in the PSD, we found that ADGRB1, like SynGap, binds specifically to the multidomain protein PSD-95. Furthermore, supertertiary structure and multivalent interactions alter affinity, specificity, and phase separation of PSD-95 interactions. Our data also suggest a putative role for ADGRB1 in the regulation of synaptic plasticity by anchoring PSD-95 to the postsynaptic membrane bilayer.

Limitations of the study
Co-localization experiments in HeLa cells were performed by overexpression of GFP or mCherry-tagged proteins owing to limitations in the detection of endogenous proteins in cells and the unavailability of validated antibodies against the respective proteins. LLPS is observed and described in an artificial system iScience Article in vitro, where we only use the C-terminal part of ADGRB1 or SynGap. The common experimental approaches to study LLPS, which are applied in this study, are limited to soluble proteins, therefore it is not possible with the chosen methods to study LLPS with FL ADGRB1 as it is a membrane-attached protein.
Affinity measurements are also performed in in vitro systems. Whereas absolute affinities might be different in vivo, it is usually assumed that differences in affinity are similar in vitro and in vivo such that, for example, wild-type proteins and mutational variants can be compared and the results can be extrapolated to an in vivo situation.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We thank Eva Andersson for assistance with sample preparation, Prof. Mingjie Zhang (School of Life Sciences, Southern University of Science and Technology (SUSTech), Shenzhen, China) for plasmids encoding FL CC SynGap, GKAP, Homer and Shank, and Dean Derbyshire from ProLinC (Linkö ping University) for assistance and use of SEC-MALS instrument. The project has received funding from the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement No 67341 (to P.J. and S.Ø.) and the Swedish Research Council (grant no. 2020-04395) (to P.J.).

DECLARATION OF INTERESTS
The authors declare no competing interests.

Materials availability
All unique constructs generated in this study are available from the lead contact.

Data and code availability
All data reported in this paper will be shared by the lead contact upon request. This paper does not report an original code.

METHOD DETAILS
Identification of putative PDZ-binding proteins in the PSD, peptide library design and peptide synthesis High coiled coil content and affinity are properties reported to be important for SynGap to undergo LLPS with PSD-95. Coiled coil percentage was calculated for all proteins in the library using online available software and compared with the K i value reported from binding experiments ( Figure 2). We picked MTMR2 and ADGRB1 for further analysis and expressed their FL CC domain, as they had a high coiled coil percentage and a decent affinity for PDZ3.
We searched manually in Uniprot for proteins predicted to be located in proximity to PSD-95 in PSD. In more detail, we used term-search in Uniprot using dendrite, post synaptic density, post synaptic density membrane and dendrite spine, respectively. It is well-known that a hydrophobic amino acid at the C-terminus (denoted P 0 ) is essential for binding to PDZ3 in PSD-95. Thus, 74 out of 80 C-terminal peptides from proteins with Leu, Ile, Cys, Phe or Val at their C-terminal position (denoted P 0 according to established PDZ nomenclature) were included in the final peptide library. We limited the study to the main isoform of proteins in proximity to PSD-95 to reduce the size of the peptide library, and splice variants as well as peptides that proved difficult to handle were therefore excluded. Peptides were designed with a free C-terminal, 14 native amino acids, a Tyr at P -14 for concentration determination by absorbance, and with acetylated N-terminal.
All peptides for fluorescence polarization experiments were synthesized by parallel 96 format peptide synthesis using Intavis multipep RSi. 10 mg per well of Fmoc-preloaded resins were used. Fmoc-amino acids (Fmoc-AA) 0.3 M in dimethylformamide (DMF) containing 0.3 M OxymaPure were used. Three couplings using 100 mL Fmoc-AA and 10 mL DIC (3 M in DMF) were employed with coupling times of 5, 15 and 60 min. Removal of Fmoc was performed by washing twice with 120 mL of 25% piperidin in DMF for 2 and 8 min. Trifluoroacetic acid (TFA) cleavage was done by 92% TFA containing 4% triisopropylsilane, 1% thioanisol, and 3% H 2 O for 2 hours. A total volume of 1 mL TFA was added to each well. TFA was reduced in volume to approximately 150 mL by N 2 flow followed by precipitation by the addition of diethylether. The peptides were transferred to Waters solvinert plates and washed thoroughly five times with diethylether. After washing, the peptides were dried and then dissolved in neat DMSO. Peptides were analyzed by UPLC-MS and showed an average purity of 85%.
Several of the peptides were re-ordered from GL Biochem (see Data S1). 15 AA peptides for stopped flow analysis with dansylated N-terminal and 15 AA CRIPT, 14 AA ADGRB1 and 18 AA MTMR2 with FITC labelled N-terminal for fluorescence polarization experiments were ordered from Genecust.

Protein expression and purification
PDZ3, PSG, PSG G322A , PSG F325A , PSG I327V , PSG G335A , PSG G322A G335A , PSG F325A I327V and PDZ1-2 were expressed and purified as previously described . Plasmids (pETMG3C) encoding GB1-His 6 tag full length coiled coil (FL CC) SynGap, MTMR2 or ADGRB1 were transformed into Escherichia coli BL21(DE3) pLys cells (Invitrogen). Cells were first grown in LB medium at 37 C in a rotary shaker. Overexpression of proteins was induced with 1 mM isopropyl-b-D-1-thiogalactopyranoside at OD 600 of 0.6-0.8, and the cells were incubated 4 hours at 30 C (FL CC ADGRB1 and FL CC MTMR2) or overnight at 18 C for FL CC SynGap. Proteins were purified from the soluble fraction on a Nickel Sepharose Fast Flow column ll OPEN ACCESS