Anchored and labile pools of Calcineurin are enabled by a disordered AKAP 1 scaffold 2 3

Signalling requires precise spatial and temporal regulation of molecular interactions, which is frequently orchestrated by disordered scaffolding proteins, such as A-kinase anchoring protein 5 (AKAP5). AKAP5 contains multiple Short Linear Motifs (SLiMs) that assemble the necessary components, including the phosphatase Calcineurin, which is anchored via a well-characterised PxIxIT SLiM. Here we show, using a combination of biochemical and biophysical approaches, that Calcineurin also recognises additional lower-affinity SLiMs C-terminal to the PxIxIT motif. Moreover, we demonstrate that the assembly is in reality a complex system in which AKAP SLiMs spanning a wide affinity range act cooperatively to maintain distinct pools of anchored and more loosely held enzyme, analogous to transcription factor search complexes on DNA, and compatible with the requirement for both stable anchoring and responsive downstream signalling. We conclude that the AKAP5 C-terminus is enriched in lower-affinity/mini-SLiMs that cooperate to maintain a structurally disordered but tightly regulated signalosome.


Introduction 47
Many enzymes involved in signal transduction are promiscuous and act in multiple 48 pathways. Scaffolding proteins are therefore central to signalling since, by tethering 49 signalling components and localising them, they enhance specificity and reaction 50 rates by effectively enhancing local concentrations, and thereby achieve strict spatial 51 and temporal control of catalysis 1,2 . A-kinase anchoring proteins (AKAPs) are a 52 family of ca. 14 intrinsically-disordered proteins (IDPs) that were originally 53 characterized by their ability to scaffold Protein Kinase A (PKA) 3,4 , but have since 54 been shown to assemble multiple components into 'signalosomes' that can be 55 targeted to specific receptors or sites 5,6 . Currently there are 14 annotated AKAPs 56 (AKAP1 to AKAP14). 57 the PRIEIT sequence in NFATc1 18,19 . PRIEIT is a canonical Calcineurin recognition 87 motif obeying the consensus sequence PxIxIT, a so-called 'PxIxIT' SLiM. This 88 interaction has a 25 μM affinity, which is within the optimal window for its signalling 89 function 20 . AKAP5 also has a PxIxIT SLiM with which it engages Calcineurin, albeit 90 one with a higher affinity -PIAIIIT -that is suited to anchoring (0.4 μM), while not so 91 tight that downstream signalling is inhibited 21 . In this study, we examined the 92 AKAP5/Calcineurin interaction from the perspective of the IDP scaffold. Using 93 multiple biophysical techniques including NMR spectroscopy, we show that binding 94 is far more extensive than the PxIxIT alone and we identify additional regions that 95 interact with μM -mM affinity. We hypothesise that these additional low-affinity 96 sites help to capture and maintain the high-affinity anchored Calcineurin. Moreover, 97 they provide a separate 'labile but localised' pool of rapidly dissociating and 98 rebinding Calcineurin, which could be intercepted by proteins such as NFAT that 99 compete for Calcineurin in order to enact optimal downstream signalling. 100 101

Results 102
The cAKAP5 scaffold is highly disordered and monomeric 103 AKAP5 can loosely be divided into three regions (Fig. 1a). Here, we have chosen to 104 focus on the C-terminal region (300-427; cAKAP5), which we define as the region of 105 To fully understand the role of disorder in the assembly of signalling 113 complexes by scaffolding proteins, it is important to understand the level and type of 114 disorder present. The term "intrinsic disorder" 28 is used to describe all proteins that 115 populate transient structures through to those that approach true random coils (i.e. 116 those with no structural preferences and uniformly fast backbone dynamics). Two 117 spectroscopic techniques were deployed to assess and delineate regions of nascent 118 structure and to probe the backbone dynamics in cAKAP5: nuclear magnetic 119 resonance (NMR) and circular dichroism (CD). The 15 N-HSQC spectrum (Fig. 1b) 120 displays the low 1 H N chemical-shift dispersion (0.7 ppm) and narrow line widths 121 typical of disordered proteins. A near complete assignment of the backbone H, C, N 122 and C β chemical shifts was possible (Supp. Fig. S2), from which the secondary 123 structure propensity (SSP 29 ) could be calculated (Fig. 1c). SSP scores in cAKAP5 are 124 close to zero with alternating stretches of weak α-or β-propensities. For 125 comparison, SSP scores are ca. +1 or -1 in stable α-and β-structures, respectively. It 126 is interesting to note that the propensities of the Calcineurin-and PKA-binding SLiMs 127 (red and yellow, respectively, in Fig. 1c) reflect the stable structures adopted in the 128 bound states (SSP mean = -0.116 ± 0.038 and +0.079 ± 0.037, i.e. weak β and α, 129 respectively). Backbone dynamics were probed by { 1 H} 15 N heteronuclear NOE 130 (HNOE; Fig. 1c), which reveals motions on a time-scale faster than overall tumbling 131 (HNOE < ca. +0.6). HNOEs were fairly uniform throughout cAKAP (disregarding the 132 highly mobile N-and C-termini). The average value was -0.20 ± 0.29, with no single 133 residue exhibiting a value > +0.1, and all SLiMs falling within one standard deviation 134 of the mean. This is consistent with a highly dynamic polypeptide chain. CD 135 spectroscopy also showed cAKAP5 to be predominantly disordered (Fig. 1d) Fig. S3). 152 Given AKAP5 forms complex assemblies within the cell, we investigated 153 whether full length AKAP5 dimerises in a cellular context (HEK293T and CHO) with a 154 highly sensitive live-cell luminescence assay (NanoBiT®, Promega) that uses a split-155 luciferase reporter engineered to give reduced background. A strong 156 complementation signal was seen between PKA and AKAP5, however, AKAP5 did not 157 give a self-complementation signal in these assays that was significantly different 158 from the HaloTag® negative control (Fig. 1f). This does not preclude AKAP chains 159 being brought together in vivo by interactions with multimeric receptors, e.g. Calcineurin activity, this exposes a second SLiM-binding region, that interacts with 189 πɸLxVP (π = polar, ɸ = hydrophobic) in a Ca 2+ -Calmodulin dependent manner 35 . 190 Although the focus of this study is primarily the dormant PxIxIT-bound 'anchored' 191 complex, we wanted to establish whether Ca 2+ or Ca 2+ -Calmodulin binding alters the 192 interaction with PIAIIIT or the new interactions found above. 193 We first titrated the pre-formed AKAP/Calcineurin 1:1 complex with Ca 2+ , 194 followed by Calmodulin. No changes were seen on addition of Ca 2+ to a large excess 195 (20:1; Fig. 2b). cAKAP/Calcineurin binding is therefore calcium-independent. On 196 Calmodulin but without Calcineurin present showed a similar response, but with a 204 larger mole fraction bound (Supp. Fig. S4). Calmodulin therefore appears to compete 205 with Calcineurin for binding to residues 390-417, a region ca. 50 residues from the 206 canonical PxIxIT site. This region is known to form an amphipathic helix on binding 207 PKA, and contains two clusters of bulky hydrophobic residues separated by a stretch 208 of polar residues, which is a feature of Calmodulin binding 36 , although the sequence 209 was not predicted to be Calmodulin-binding by a database search (Calmodulin Target  210 Database (http://calcium.uhnres.utoron-to.ca/ctdb/)). We could not confirm 211 whether Calmodulin also bound Calcineurin in the canonical manner, but the pattern 212 of shifting and broadening of peaks continued beyond a 1:1 molar ratio of 213 Calmodulin, which is compatible with binding to two sites. 214 215

PxIxIT is not required for binding to non-canonical sites 216
Given the involvement of PxIxIT-flanking and distal regions of cAKAP5 in binding to 217 Calcineurin, we investigated binding to cAKAP5 in which the PIAIIIT site had been 218 deleted, cAKAP5 ΔPIAIIIT . The titration with Calcineurin ( Fig. 3a) was carried out as 219 previously. From the overlay of the intensity ratios (Fig. 3b), it is clear that the 220 immediate footprint around the PIAIIIT site (residues 330-350) is lost, as expected. 221 However, intensities more distal to the PIAIIIT site still showed significant decreases 222 in intensity, in the same regions and even more pronounced than in WT cAKAP. 223 Binding to these sites is thus PIAIIIT-independent, and competitive with PIAIIIT, 224 although it is not possible to say whether they bind the same site on the Calcineurin 225 side. Three regions with pronounced dips in intensity were seen, 361 QFLIS 365 , 226 390 TLLIET 395 and 404 IQLSIEQLVN 413 , which we termed secondary sites 1, 2 and 3, 227 respectively. 228 Inspection of the secondary site sequences revealed that they each contained 229 three or more bulky hydrophobic residues. Given that all known Calcineurin-binding 230 SLiMs contain hydrophobic residues 35,37 , a further AKAP mutant was also tested: 231 AKAP ILVF->SA , in which the PIAIIIT site was present, but all the hydrophobic I, L, V and F 232 residues in the three secondary sites were mutated to S or A to give 361 QASAS 365 , 233 390 TASAET 395 and 404 AQSSAEQSAN 413 . Following reassignment of the chemical shifts, 234 the protein was again titrated with Calcineurin (Fig. 4a). Similar to the WT, large 235 reductions in intensities were observed in and around the PIAIIIT site (Fig. 4b). This 236 was more pronounced in AKAP ILVF->SA indicating higher occupancy of the site by 237 Calcineurin. No dips were seen at secondary sites 2 and 3, (intensities were similar to 238 the unbound AKAP) and site 1 was partially attenuated (overlap with the PxIxIT 239 footprint made the effect less clear). Therefore, binding of Calcineurin to the PIAIIIT 240 or secondary sites appears to be independent but competitive, the secondary sites 241 Corresponding WT data (magenta) reproduced from Fig. 2b. Grey boxed regions 1-3 indicate 'secondary' Cn binding sites identified in Fig. 3b, with the corresponding amino acid sequences shown above. Grey asterisks indicate potential 'mini-SLiMs' (see main text).
both fast and slow kinetics in both the association and dissociation phases. In 258 addition, a fraction of the Calcineurin remained bound at long times. The complexity 259 was reduced somewhat in the curves for cAKAP5 ILVF->SA (Fig. 5b), which showed a 260 clear saturation plateau that was absent in the WT, and complete dissociation. For 261 cAKAP5 ΔPIAIIIT (Fig. 5c), the curves showed a clear plateau and complete dissociation, 262 but weaker overall binding. Also noteworthy was that the binding curves for 263 cAKAP5 ILVF->SA and cAKAP5 ΔPIAIIIT did not sum to reproduce the cAKAP binding curve, 264 i.e. A ≠ B + C, despite their comprising separately the two kinds of binding site 265 present in the WT. This was particularly apparent in the dissociation phase, where 266 the WT showed a sizeable fraction of Calcineurin was essentially permanently 267 bound, while no long-lasting binding of Calcineurin was observed with either mutant. 268 The binding to the PxIxIT and the secondary sites is therefore cooperative. 269 An estimate of the K d could be obtained from fitting the signal (RU) at the 270 end of the association period vs Calcineurin concentration (Fig. 5d) Fig. S5), and to extract k off vs K d distribution plots (Fig. 5e). The 291 complexity seen for Calcineurin binding to cAKAP5 was deconvolved into 3 or 4 292 classes of surface sites: one or possibly two very similar sites with K d 10-100 nM, 293 another with K d 1-10 μM, and a final class with a very slow k off that presumably 294 accounts for the lack of full dissociation observed even at long times (i.e. poor 295 reversibility). The number of surface site classes was reduced in cAKAP5 ILVF->SA to one 296 main site with K d 100 nM-1 μM. The distribution was simpler again for cAKAP5 ΔPIAIIIT 297 which displayed a single class of sites with K d 10-100 μM. These results support a 298 picture in which cAKAP5 acts as a multivalent surface capable of binding one or more 299 molecules of Calcineurin at different sites. There are also two indications that the 300 PIAIIIT and secondary binding sites may bind cooperatively: the K d seems to vary by 301 as much as an order of magnitude dependent on the presence or absence of the 302 other class of site, also, there appears to be a 'long-lived' (slow k off ) bound 303 population that is only significant when both sites are present. This latter 304 phenomenon is reminiscent of the re-capture behaviour seen for bivalent 305 antibodies, which also show anomalously slow dissociation, or ligands rebinding to 306 receptors at cell surfaces 39 . 307 308 Discussion 309 From our experiments, it is clear that the Calcineurin binding site in cAKAP5 is far 310 more extensive than PIAIIIT, involving additional SLiMs and 'mini-SLiMs' in the C-311 terminal region that interact with μM-mM affinity: non-consensus SLiMs containing 312 as few as three bulky hydrophobic residues appear able to bind Calcineurin with low-313 to-mid μM affinity (cAKAP5 ΔPIAIIIT ; Fig. 5d&e), and mini-SLiMs containing only one or 314 two hydrophobic residues produce detectable binding by 15 N-HSQC (Fig. 4b). These 315 additional sites cooperate to increase both the effective affinity, and the binding 316 capacity of the scaffold. The effective affinity is raised through both an increase in 317 the number of productive encounters given the high density of SLiMs, and through 318 rebinding mechanisms that reduce the diffusion of Calcineurin away from the 319 scaffold since the molecule is likely to be intercepted by another binding site rather 320 than diffusing away. This latter point is borne out directly by the lack of complete 321 dissociation of Calcineurin on cAKAP5 seen by SPR (Fig. 5a). The 'labile but localised' 322 pool of rapidly dissociating and rebinding Calcineurin can presumably also be 323 captured by downstream effectors such as NFAT despite their mid μM affinities (Fig.  324   6). 325 These new insights are made possible by the combination of NMR mapping 326 experiments and a non-traditional approach to SPR data fitting that is validated by 327 strategic mutagenesis. The experimental study of molecular recognition by 328 disordered proteins can be very challenging due to the rapid increase in complexity 329 inherent in multivalent binding. Competition, cooperativity and multiple binding 330 modes often limit the study to global behaviour only (in which mechanistic 331 information may be lost), or dissection into simpler pairwise interactions, treating 332 the intervening sequences as inert (which may incur the loss of important 333 information on e.g. cooperativity). SPR data on such systems is often high quality in 334 both signal-to-noise and reproducibility, while being deemed "unfittable" due to the 335 Figure 6: Non-canonical SLiMs could facilitate both anchoring and formation of a labile Calcineurin pool that enables efficient transfer to effectors. cAKAP5 captures and retains Calcineurin using its consensus PIAIIIT (red arrow) and secondary sites (red rectangles), building anchored and signalling pools of bound Calcineurin (blue and green, respectively). Calcineurin binds with lower-affinity to the secondary sites, enabling effective competition by weaker PxIxIT motifs on downstream effectors, e.g. the PRIEIT SLiM in NFAT, and optimal signalling. ill-posed problem of decomposing a distribution of exponentials. However we find 336 that the combination of regularization and Bayesian approaches 38 , designed for the 337 deconvolution of mass transport and surface heterogeneity, is also very effective at 338 decomposing multivalent interactions. The resulting distributions may be 339 subsequently validated by competition or, as we have done here, mutagenesis, 340 alongside mapping by NMR, which is both residue-specific and sensitive to a wide 341 range of interaction affinities 40 . 342 In addition to the secondary sites and mini-SLiMs we establish in this study, 343 further evidence for weak, non-canonical engagement of Calcineurin with 344 hydrophobic residues is seen in the existing PxIxIT peptide/Calcineurin X-ray crystal 345 structures, in which the reverse face of the PxIxIT site provides an additional 346 potential encounter point ( Supplementary Fig. 6a). In the structures, one with PVIVIT 347 20 and one with the AKAP sequence PIAIIIT 21 , the stoichiometry in the asymmetric 348 unit is two Calcineurin heterodimers to one peptide, with the second heterodimer 349 binding to the opposite side of the PxIxIT site, out of register by one residue towards 350 the C-terminus of AKAP5. The peptide thus forms an axis of pseudosymmetry. In 351 both cases, the first heterodimer that engages the consensus elements of the PxIxIT 352 sequence has a larger buried surface area and a greater number of hydrogen bonds. 353 The opposite side displays VxV or AxI, to which a second heterodimer is bound in the 354 crystal, using the same hydrophobic PxIxIT-binding groove, but with significantly 355 lower affinity, since no evidence for a 2:1 stoichiometry was seen in solution by size 356 exclusion chromatography or SEC-MALS 20,21 . However, the presence of hydrophobic 357 residues in the two 'x' positions may serve as an additional capture surface on PxIxIT, 358 even if unstable. Their presence may (partially) account for the fact that in our 359 experiments, mutation of the three distal secondary sites in cAKAP5 ILVF->SA did not 360 restore 'pure' pseudo first-order binding characteristic of simple bimolecular 361 collisions (Fig. 5b & e). Considered as a group, PxIxIT affinities span a wide range, and 362 the fine-tuning of PxIxIT affinities by altering the 'x' positions presumably facilitates 363 evolution into either anchoring or signalling roles. It is interesting to note that during 364 selection for competitive peptide inhibitors of PxIxIT binding there was a clear 365 enrichment for hydrophobic residues (V, I and L) in the two x positions 37 , while 366 many of the low affinity substrates favour charged or polar amino acids in these 367 positions that would be incompatible with the non-polar groove on Calcineurin e.g. 368 PRIEIT in NFATc1 (25 μM; Supplementary Fig. 6b). 369 Our findings also demonstrate that SLiM recognition in cAKAP5 is non-370 exclusive: both Calcineurin and Calmodulin recognise features of the canonical PKA 371 SLiM and therefore all three are in competition for the same site (Fig. 2 & 3). 372 Interactions of one SLiM with many partners have been documented 41 , and due to 373 The scaffolding protein AKAP5 contains a well-characterised PxIxIT-motif Calcineurin 405 binding site. We characterise at least three additional binding sites of lower affinity 406 C-terminal to the PxIxIT site and propose that these additional sites create a 'labile 407 but localised' pool that both supplies Calcineurin to the canonical anchoring site and 408 to downstream effectors such as NFAT. We conclude that in AKAP5, the disordered 409 regions between anchoring SLiMs, rather than being inert tethers, are enriched in 410 lower-affinity/mini-SLiMs that act together to maintain a stable but responsive 411 signalosome. Triple resonance experiments were recorded with 25% nonuniform sampling, using 547 Poisson-gap sampling 54 , and reconstructed using the Cambridge CS package and the 548 CS-IHT algorithm 55 . Assignments were made using CcpNmr Analysis v. 2.4 56 . 549 Chemical shifts were referenced to 2,2-dimethyl-2-silapentane-5-sulfonic acid (DSS). 550 Heteronuclear NOE values were obtained at 600 MHz with either 4 s of 1 H saturation 551 using a 120° pulse train or a 4 s delay prior to the first 15 N pulse 57 . Complexes with 552 Calcineurin were first prepared at the defined stoichiometries in 10 mM Tris/HCl pH 553 7.4, 150 mM NaCl, 2 mM DTT and then dialysed into NMR buffer. Intensity ratios 554 were calculated using fitted peak heights in Analysis. Chemical-shift differences were 555 calculated using Δδ = [(Δδ H ) 2 + (0.15 × Δδ N ) 2 ] 1/2 40 . 556 557

Data availability 623
Chemical-shift assignments for cAKAP5 (see also Supp.