Interaction of the tetratricopeptide repeat domain of aryl hydrocarbon receptor–interacting protein–like 1 with the regulatory Pγ subunit of phosphodiesterase 6

Phosphodiesterase-6 (PDE6) is key to both phototransduction and health of rods and cones. Proper folding of PDE6 relies on the chaperone activity of aryl hydrocarbon receptor–interacting protein–like 1 (AIPL1), and mutations in both PDE6 and AIPL1 can cause a severe form of blindness. Although AIPL1 and PDE6 are known to interact via the FK506-binding protein domain of AIPL1, the contribution of the tetratricopeptide repeat (TPR) domain of AIPL1 to its chaperone function is poorly understood. Here, we demonstrate that AIPL1–TPR interacts specifically with the regulatory Pγ subunit of PDE6. Use of NMR chemical shift perturbation (CSP) mapping technique revealed the interface between the C-terminal portion of Pγ and AIPL1–TPR. Our solution of the crystal structure of the AIPL1–TPR domain provided additional information, which together with the CSP data enabled us to generate a model of this interface. Biochemical analysis of chimeric AIPL1–AIP proteins supported this model and also revealed a correlation between the affinity of AIPL1–TPR for Pγ and the ability of Pγ to potentiate the chaperone activity of AIPL1. Based on these results, we present a model of the larger AIPL1–PDE6 complex. This supports the importance of simultaneous interactions of AIPL1–FK506–binding protein with the prenyl moieties of PDE6 and AIPL1–TPR with the Pγ subunit during the folding and/or assembly of PDE6. This study sheds new light on the versatility of TPR domains in protein folding by describing a novel TPR-protein binding partner, Pγ, and revealing that this subunit imparts AIPL1 selectivity for its client.

Cyclic nucleotide phosphodiesterases of the sixth family (PDE6) 2 are the key effectors in the visual transduction cascade in rod and cone photoreceptors (1,2). In the dark, activity of the PDE6 catalytic dimers, PDE6AB heterodimer in rods and PDE6C homodimer in cones, is blocked by two inhibitory ␥-subunits (P␥), allowing for a depolarizing "dark" current through cGMP-gated channels in the plasma membrane. Photoexcitation leads to G protein (transducin)-mediated activation of PDE6 and a drop in cytoplasmic cGMP causing the channels to close and the plasma membrane to hyperpolarize (1). The role of PDE6 in phototransduction is intrinsically linked to its importance to the health and survival of photoreceptor cells. Loss of function of PDE6 leads to elevation of intracellular cGMP and excessive influx of Ca 2ϩ , thereby triggering photoreceptor cell death (3)(4)(5)(6)(7)(8).
Loss of function of PDE6 can result from mutations in PDE6 causing retinitis pigmentosa, achromatopsia, or cone-rod dystrophy (6,7,(9)(10)(11)(12). Another major reason for PDE6 deficits is malfunction of its specialized retina-specific chaperone AIPL1 (aryl hydrocarbon receptor (AhR)-interacting protein-like 1). AIPL1 was named based on the high sequence identity (49%) and similar domain organization with the ubiquitously expressed AIP (AhR-interacting protein) (13). AIP and AIPL1 both contain an N-terminal FK506-binding protein (FKBP) domain and a C-terminal tetratricopeptide repeat (TPR) domain with three tetratricopeptide repeats. The AIPL1 gene encoding AIPL1 has been linked to Leber congenital amaurosis, a severe early-onset blindness (13)(14)(15)(16). Knockout of AIPL1 in mice revealed selective and drastic destabilization of PDE6, resulting in retinal degeneration (17,18). This finding suggested that AIPL1 may function as PDE6 chaperone. Such a role of AIPL1 was ultimately established with heterologous expression of active cone PDE6C in the presence of AIPL1 in HEK293 cells (12).
Among the critical interactions that underlie the chaperone activity of AIPL1 is the direct binding of the C-terminal prenyl modifications of the PDE6 catalytic subunits by the AIPL1 FKBP domain (19). The crystal structures of the AIPL1-FKBP with bound farnesyl or geranylgeranyl ligands revealed its unique structural features enabling prenyl moiety binding and unmasked pathogenic mechanisms of the AIPL1 variants in Previously we demonstrated that AIPL1 interacts with P␥, and P␥ acts as a potent co-chaperone of AIPL1 in folding/assembly of PDE6 (12,24). To identify the interface of this interaction, we utilized an NMR approach by examining chemical shift perturbations (CSPs) of individual 15 N-labeled FKBP and TPR domains of AIPL1 in the presence of P␥ fragments (28,29). The N termini of the P␥ subunits are not conserved among the rod and cone counterparts. Two segments of P␥, including the central polycationic region P␥24 -45 and the C-terminal region P␥46 -87, have been commonly recognized as functionally important parts of the protein contributing to P␥ interactions with PDE6 catalytic subunits and transducin-␣ (30 -33). We first assessed potential effects of synthetic peptides P␥24 -46 and P␥46 -87 on the 15 N/ 1 H HSQC NMR spectra of 15 N-AIPL1 FKBP domain (aa 1-161). Neither P␥24 -46 nor P␥46 -87 induced CSPs in the spectrum of 15 N-AIPL1-FKBP (Fig. S1). However, P␥46 -87, but not P␥24 -46, caused signif-icant CSPs in backbone amide peaks of the AIPL1-TPR domain (aa 171-316), suggesting that P␥46 -87 binds to the TPR domain ( Fig. 1A and Fig. S2). Our recently assigned NMR backbone resonances of the AIPL1-TPR domain (29) allowed us to identify the residues on this protein as being a part of the P␥46 -87-binding site as clearly labeled for the significantly perturbed peaks (Fig. 1A). In addition, selective and substantial CSPs in the presence of P␥46 -87 were detected in the NMR spectrum of 13 C-methyl (IVL)-labeled AIPL1-TPR domain (Fig. 1B). Next, we determined whether binding of P␥46 -87 to AIPL1-TPR influences its thermostability using dynamic light scattering (DLS) technique. P␥46 -87 increased the T onset of AIPL1-TPR by ϳ2°C ( Fig. 2A), and it had an even stronger effect (ϳ4 -5°C shift) on the thermostability of the full-length AIPL1 (Fig. 2B). Thus, the C-terminal region of P␥ binds to the TPR domain of AIPL1, thereby increasing thermostability of the protein.

Analysis of AIPL1-TPR interaction with the C terminus of P␥ by BLI
The kinetics and affinity of binding for AIPL1-TPR and P␥46 -87 were quantitated using BLI, with N-terminally biotinylated P␥46 -87 attached to a streptavidin biosensor (Fig. 3, A  and B). The association and dissociation kinetics for the interaction were consistent with a 1:1 binding model, with an average association constant (k a ) of 0.8 ϫ 10 5 M Ϫ1 s Ϫ1 and an average dissociation constant (k d ) of 0.38 s Ϫ1 , yielding a K D ϭ k d /k a of 4.8 M (Fig. 3A). The K D calculated based on steady-state analysis of the interaction between AIPL1-TPR and P␥46 -87 was comparable (3.0 M; Fig. 3B) and also similar to the K D value of 2.5 M previously reported for the binding of the fulllength AIPL1 and P␥ (24). This confirms that the interface is largely confined to the TPR domain and the C-terminal 46 -87 residues of P␥. Interestingly, the highly homologous TPR domain of AIP failed to appreciably bind to P␥46 -87 in the BLI assay ( Fig. S3), highlighting the specificity of AIPL1-TPR-P␥ interaction. To narrow down the P␥ C-terminal site for binding to AIPL1-TPR, we conducted the BLI assay using N-terminally biotinylated P␥63-87. The association data for up to 15 M of AIPL1-TPR were fitting well to a 1:1 binding model with the average k a of 1.6 Ϯ 0.2 ϫ 10 5 M Ϫ1 s Ϫ1 (Fig. 3C). A K D of 3.1 M was estimated from the kinetics data using the average k d of 0.49 Ϯ 0.02 s Ϫ1 from the dissociation phase of the assay, whereas a K D of 4.6 M was derived from the steady-state analysis (Fig. 3D). These results indicate that the 25 C-terminal residues of P␥ encompass most if not all of the contacts with AIPL1-TPR.

Mapping the interface between the P␥ C terminus and AIPL1-TPR by NMR
To determine the minimal region of the C-terminal fragment of P␥ required for binding to AIPL1-TPR, as well as how various P␥ peptides perturb the residues or binding site(s) on AIPL1-TPR, we performed further NMR binding experiments by titrating P␥ C-terminal fragments with different lengths including P␥63-87, P␥70 -87, P␥74 -87, and P␥63-83 into AIPL1-TPR. Clearly, P␥63-87 binds to AIPL1-TPR as shown by both 15 N and 13 C HSQC spectra (Fig. 4), and P␥63-87 bind-

Interactions of AIPL1 with PDE6
ing perturbs the same set of residues on AIPL1-TPR to a similar degree as observed for P␥46 -87 binding ( Figs. 1 and 4). This result indicates that P␥46 -87 and P␥63-87 binds to the same site on AIPL1-TPR, and P␥ residues 46 -62 are not required for binding to AIPL1-TPR. This result is also consistent with the similar binding affinity detected by BLI for P␥46 -87 and P␥63-87.
NMR titration of P␥70 -87 showed that P␥70 -87 binds to the same site on AIPL1-TPR as P␥63-87 or P␥46 -87 because the same set of residues on AIPL1-TPR are affected upon adding P␥70 -87 (Fig. S4). However, higher P␥70 -87 concentration (P␥:AIPL1-TPR ratio, 1.6:1) is needed to detect a similar degree of perturbations of these affected residues. During the titration experiments, we noted that P␥70 -87 slowly forms a A, overlay of 15 N/ 1 H HSQC spectra. B, overlay of 13 C/ 1 H HSQC spectra. The significantly perturbed backbone amide and methyl peaks upon P␥ peptide binding are labeled using the WT protein sequence numbering. In these experiments, the TPR and P␥ peptide concentrations used were 157 and 166 M, respectively.

Interactions of AIPL1 with PDE6
gel-like solution, indicating reduced solubility. This may have affected the quantification of the peptide and be responsible for the increase in peptide concentration needed to see the similar effect.
NMR titration of P␥74 -87 showed that this peptide binds to AIPL1-TPR very weakly and caused very small CSPs of AIPL1-TPR even at the higher P␥:AIPL1-TPR ratio of 4.8:1 (762 M P␥74 -87 and 160 M AIPL1-TPR). A comparison of P␥70 -87 and P␥74 -87 binding to AIPL1-TPR suggests that P␥ residues 70 -73 are required for the interaction.
NMR titration of P␥63-83 revealed that this peptide binds to AIPL1-TPR because significant CSPs are observed in both 15 N and 13 C HSQC spectra (Fig. S5), but this peptide binds to AIPL1-TPR weaker than P␥63-87 because higher P␥63-83 concentration is needed to induce similar degrees of CSPs ( Fig.  4 and Fig. S5). Interestingly, comparing the shift perturbations caused by the binding P␥63-87 (Fig. 4) and P␥63-83 (Fig. S5), it is clear that the residues located near the N-terminal of AIPL1-TPR including the backbone amides of Ala-177, Gly-183, Glu-184, Asn-186, and Arg-187 (Fig. S5A) and the side chains of Val-178␥1, Val-180␥1, and Leu-232␦1 (Fig. S5B) are much less affected by the binding of P␥63-83. These results indicate that P␥ C-terminal residues 84 -87 bind near these residues. Thus, these NMR titration experiments established that P␥70 -87 is the required minimal sequence for binding to AIPL1-TPR. Furthermore, our NMR results implicate two short stretches of P␥, P␥70 -73 and P␥84 -87, as being critical in P␥ binding to AIPL1-TPR.

Crystal structure of AIPL1-TPR
Given the unique and critical role of the AIPL1 TPR domain in chaperoning PDE6, we sought to solve its crystal structure. The use of His 6 -tagged AIPL1(171-316) led to well-diffracting crystals. All residues of AIPL1-TPR, except for the N-terminal nine residues and the C-terminal two residues were well-resolved in electron density (Table S1) (PDB code 6PX0). The crystal structure revealed a typical TPR domain fold with three TPR repeats (Fig. 5A). TPR1, TPR2, and TPR3 are comprised of antiparallel ␣-helices 1-2, 3-4, and 5-6, respectively, followed by ␣-helix 7. A regular repeat of ␣-helices generates a righthanded helical array and creates a groove that often serves as a binding surface for binding partners of TPR domain proteins (Fig. 5B). AIPL1-TPR is structurally very similar to the TPR domain of AIP, and these two structures are superimposable with a root-mean-square deviation of 1.2 Å for 128 C␣ atoms (Fig. 5B) (34). In contrast to the AIP-TPR structure, electron density for the nine N-terminal residues of AIPL1-TPR preceding helix ␣1 is missing. However, the N-terminal helix of AIPL1-TPR may have been affected by the crystal packing. Furthermore, the NMR analysis indicated that residues 176 -190 form an ␣-helix in solution (29). Thus, the N-terminal helix of AIPL1-TPR is longer than it is observed in the structure. One

Interactions of AIPL1 with PDE6
interesting difference between the structures of the TPR domains of AIPL1 and AIP appears to be a surface charge distribution on the ligand-binding groove of the proteins. The overall surface is positively charged in both TPR domains, but in AIPL1-TPR the positive charge is more dispersed compared with a relatively focal positive pocket in AIP-TPR (Fig. 5, C and D).

Modeling the complex of AIPL1-TPR with P␥63-87
P␥63-87 contains two short ␣-helices composed of residues 69 -74 (␣1) and 78 -84 (␣2), as shown by the crystal structure of P␥46 -87 in complex with G-protein transducin-␣ and RGS9 (PDB code 1FQJ) (33). These helices match well with the helical propensities of the full-length P␥ alone determined earlier by NMR (35). Therefore, we have docked P␥63-87 into the crystal structure of AIPL1-TPR with the CNS program (36) by using the NMR restraints derived from the surface-exposed and chemical shift-perturbed residues of AIPL1-TPR upon binding of P␥63-87 as revealed from the NMR binding experiments. During docking, the P␥ ␣1 and ␣2 secondary structures are maintained, but the relative orientation of these two ␣-helices as well as the P␥ loops are kept fully flexible. Moreover, the ␣2 helix and the C-terminal four residues of P␥ are kept parallel to the N-terminal ␣1 and ␣3 helices of AIPL1-TPR because this binding mode is most consistent with the NMR binding data where deletion of the C-terminal four residues of P␥ significantly reduced CSPs of the residues localized to the N-terminal ␣1 and ␣3 helices of AIPL1-TPR (Fig. S5). The docked and energy-minimized model of AIPL1-TPR in complex with P␥63-87 is shown in Fig. 6 where the significantly perturbed and most severely perturbed residues of AIPL1-TPR upon binding to P␥63-87 are colored in red and magenta, respectively.

Interactions of AIPL1 with PDE6
␣2 helix is also present, and the bending angle between the P␥ C-terminal ␣1 and ␣2 helical segments is similar to that in our model. However, instead of an ␣2 helix as modeled, residues 69 -74 in the PDE/P␥ structures assume a coil-like conformation. This is not surprising because this helix is very short and furthermore contains a Pro at the N-terminal side (Pro-69).
AIPL1-TPR has been previously shown to bind the C terminus of HSP90 (23). The model of the complex of AIPL1-TPR with P␥63-87 indicated overlapping P␥-and HSP90-binding sites (Figs. 5B and 6A). We confirmed this by comparing the patterns of the CSPs of the perturbed residues upon binding the C-terminal HSP peptide (TSRMEEVD) and P␥63-87 to AIPL1-TPR ( Fig. S6 and Fig. 4)

Probing the AIPL1-TPR-P␥63-87 interface with chimeric AIPL1-AIP proteins
To identify residues of AIPL1-TPR that are likely to contribute to binding of P␥63-87, we first conducted conservation analysis of its ligand binding surface using ConSurf (38). This analysis revealed that our model of the complex placed P␥63-87, in particularly its ␣2 and the C terminus, over a highly conserved patch of AIPL1-TPR (Fig. 6B). Moreover, many of the TPR residues with largest CSPs caused by binding of P␥63-87 are highly conserved. However, nearly all of these residues were not AIPL1-specific but were identical or homologous to residues at equivalent positions in AIP. Furthermore, the conservation analysis of the ligand-binding surface of AIP-TPR revealed a pattern of conservation similar to that in AIPL1-TPR (Fig. 6, B and C). Thus, the selectivity of AIPL1 for P␥63-87 is likely defined by the overall conformation of the binding surface with contribution from multiple rather than few residues. Therefore, we utilized a chimera AIPL1-AIP TPR approach to test the significance of various segments of AIPL1-TPR for the interaction with P␥63-87 and for the chaperone activity of AIPL1. Initially, two chimeric TPR domains, TPR-C1 and TPR-C2, were constructed as shown in Fig. 7A, and examined for thermostability and the ability to bind P␥63-87 (Table  S2 and Fig. 7B). Both chimeras were thermally stable, indicating proper overall folding. Chimera TPR-C1 containing ␣1-␣3 ␣helices of AIPL1-TPR and ␣4 -␣7 of AIP-TPR retained higher affinity for P␥63-87 compared with the reciprocal chimera TPR-C2 ( Fig. 7B and Table S2). To test the ability of the C1 and C2 TPR domains to support the chaperone activity of AIPL1, Figure 6. A, structural model of the human AIPL1 TPR domain in complex with P␥63-87. The P␥63-87 peptide was docked into the crystal structure (aa 180 -314) of human AIPL1 TPR domain using the NMR restraints obtained from the analysis of P␥ peptide binding to the protein. The P␥ ␣1 and ␣2 secondary helix structures were derived from the PDB structure (PDB code 1FQJ). The P␥ peptide is shown in yellow. The red and magenta ribbons indicate the TPR residues that were significantly perturbed (with ⌬␦ ppm Ն 0.053 ppm) and most severely perturbed (with ⌬␦ ppm Ն 0.10 ppm) by the binding of the P␥ peptide, respectively. B and C, surface representations for AIPL1-TPR (B) and AIP-TPR (C) colored by residue conservations scores derived from ConSurf analysis of a sample of 87 AIPL1-TPR orthologs (B) and 89 AIP-TPR orthologs (C). The sequences for the analysis with maximal identity of 97% and minimal identity of 60% were collected from the UniProt database. The color scale is as follows: 9, magenta, conserved; and 1, cyan, variable. P␥63-87 (B) and the C terminus of HSP90 (C) are shown as yellow tubes.

Interactions of AIPL1 with PDE6
the full-length AIPL1 proteins (AIPL1-C1 and AIPL1-C2) with the corresponding chimeric TPR domains have been generated for co-expression with PDE6C in the absence or presence of P␥ in HEK293T cell (Fig. 7). AIPL1-C1 induced ϳ40% of basal PDE6C activity observed with the WT AIPL1 (Fig. 7C) and ϳ27% of PDE6C activity compared with WT AIPL1 when the co-expression system included P␥ (Fig. 7D). Accordingly, in the presence of P␥, folding of PDE6C was increased ϳ105-fold by AIPL1, and ϳ67-fold by AIPL1-C1 (Fig. 7, C and D). Thus, the reduction in affinity for P␥ in TPR-C1 was accompanied by the reduction in the P␥ potentiation of the chaperone activity of AIPL1-C1. In contrast, AIPL1-C2 failed to chaperone PDE6C regardless of the presence of P␥. In agreement with NMR data and the model, the results with TPR-C1, TPR-C2, and AIPL1-C1 and AIPL1-C2 are consistent with the ␣1 and ␣3 being the most critical P␥ binding and chaperone activity determinants. TPR-C3 and TPR-C4 were constructed to further probe the role of ␣1 and ␣3, respectively (Table S2 and Fig. 7A). TPR-C3 was significantly more impaired in P␥63-87 binding than TPR-C4 (Fig. 7B). However, the basal chaperone activity of AIPL1-C3 was ϳ60% of that for WT AIPL1, which was significantly higher than the basal chaperone capacity of AIPL1-C4 (ϳ30%) (Fig. 7C). Reflecting a greater decrease in affinity for P␥63-87 in TPR-C3 compared with TPR-C4, the P␥ potentiation of the chaperone activity of AIPL1-C3 (ϳ40-fold) was weaker than that of AIPL1-C4 (ϳ70-fold) (Fig. 7, C and D).

Modeling of AIPL1 and its complex with PDE6
A starting model of AIPL1(1-316) was generated using Modeler and the crystal structures of AIPL1-FKBP (PDB code 5U9A), FKBP52 (PDB code 1P5Q), and AIPL1-TPR domain (PDB code 6PX0), which was extended by nine N-terminal residues and two C-terminal residues according to the crystal structure of AIP-TPR (PDB code 4AIF) (20,34). The model of AIPL1(1-316) thus obtained featured relatively compact conformation with the FKBP domain and the ␣3 helix of its insert region closely opposing the ␣1 of the TPR domain (Fig. 8A). To probe the fitness of the model, we examined conformation of AIPL1(1-316) in solution by analyzing it with SEC-SAXS. The theoretical SAXS profile for the initial AIPL1(1-316) model fit the experimental SEC-SAXS data with a 2 value of 1.67, indicating a general agreement of the model with the protein structure in solution (Fig. 8D).
Next, we sampled potential conformations of AIPL1(1-316) that the protein may adopt in complex with PDE6 using molecular dynamics (MD) simulations. In these MD simulations, in addition to the model of apo AIPL1(1-316), we used a complex of AIPL1(1-316) with P␥63-87 modeled according to the AIPL1-TPR-P␥63-87 model. Nine 230 -280 -ns MD simulations of apo AIPL1(1-316) (MD1-MD9) revealed two general types of trajectories. The first trajectory, observed in MD 1, 2, 3, 6, and 8, showed the FKBP domain rotation against the TPR domain with the ␣3 helix of the insert region moving toward the TPR domain and occupying its ligand-binding surface (Fig. 8B). These trajectories produce closed conformation of AIPL1. The second trajectory type featured the movement of insert ␣3 helix toward the TPR loop ␣1-␣2, where it stayed near the bottom of the TPR domain (MD 4, 5, 7, and 9) (Fig. 8C). This is an open conformation of AIPL1 that allows unrestricted access to the ligand-binding surface of AIPL1-TPR ( Fig. 8C and Fig. S7A). Theoretical SAXS profiles of all conformations of apo AIPL1(1-316) generated in the MD analysis were compared with the experimental SEC-SAXS data, and the top models from trajectories 5 and 6 featured robust 2 values of 1.25 and 1.27, respectively (Fig. 8D) (37). Such superimposition produced no sterical clashes between PDE6 and the AIPL1-TPR domain. Next, we used AIPL1-TPR to superimpose models of AIPL1(1-316) from MD10-MD18 simulations that had better fits ( 2 values) to the experimental SAXS data. Interestingly, models from the MD trajectories of the open conformation type produced little to no clashes with PDE6, whereas models from MD trajectories of the alternative type severely clashed with the PDE6 catalytic domains. A representative model of the AIPL1-PDE6 complex is shown in Fig. 9. The farnesyl moiety in the FKBP-binding pocket of AIPL1 is in favorable proximity to Gly-822, the last C-terminal residue of PDE6A resolved in the structure, whereas farnesylation occurs at PDE6A-Cys-856. Thus, the model is consistent with simultaneous binding of AIPL1-FKBP to prenylated C termini of PDE6 and AIPL1-TPR to the P␥ C-terminal region.

Role of the insert region ␣3 helix of AIPL1-FKBP
MD simulations indicated a potentially critical role of the insert region ␣3 helix in protein dynamics of AIPL1. To probe the significance of this region, we replaced the insert region ␣3 helix with a small flexible loop comprised of five Gly residues. This mutant AIPL1 protein (AIPL1-⌬␣3) was thermally stable, and it bound the farnesyl moiety and P␥46 -87 comparably to WT AIPL1 (Fig. S9, A and B). However, AIPL1-⌬␣3 completely failed to chaperone PDE6C in the HEK293T system. To determine the molecular mechanism, underlying this failure we conducted MD simulations of the AIPL1-⌬␣3 model. In three of four simulations, the FKBP and TPR domains completely separated and moved independently of each other (Fig. S9, C and   D). Such separation of the domains would interfere with the ability of AIPL1 to simultaneously bind the prenylated C termini and P␥ subunit of PDE6.

Discussion
The P␥ subunit of PDE6 is essential for the folding and/or assembly of the latter; in the absence of P␥, virtually no functional PDE6 is formed in mouse photoreceptors, and this leads to rapid retinal degeneration (39). Moreover, in mice bearing a mutation that deletes just two C-terminal Ile residues of P␥, levels of the catalytic subunits of PDE6 are reduced in the retina (40). Recently, it was discovered that P␥ interacts with AIPL1, an obligate chaperone of PDE6; subsequently, it was hypothesized that this interaction markedly potentiates expression of functional PDE6C in a heterologous system and in photoreceptor cells in vivo (12,24). Our study validates this hypothesis and reveals the molecular underpinning of the important biological effects of P␥. Specifically, it demonstrates that the main interaction interface between P␥ and AIPL1 is confined to the P␥ C terminus, P␥63-87, and the TPR domain of AIPL1. Our model of the AIPL1-TPR-P␥63-87 complex, generated based on the newly solved crystal structure of AIPL1-TPR and on NMR analysis and binding of various P␥ C-terminal peptides to AIPL1-TPR, is supported by our data for the interaction of P␥63-87 with chimeric AIPL1-AIP TPR domains. In addition, these data established a correlation between the ability of P␥ to elevate expression of functional PDE6C and the affinity of chimeric TPR domains for P␥63-87. Thus, the biological effects of P␥ as a co-chaperone in folding of PDE6 are mediated by the interaction of P␥ with AIPL1-TPR. Our study also illuminates the puzzling question of why AIPL1 is specific for PDE6. One aspect of the puzzle was that although several essential photoreceptor proteins are prenylated and AIPL1-FKBP appears to bind prenyl moieties irrespective of the sequence of the prenylated peptide or protein (20), only PDE6 is affected significantly in the absence of AIPL1 (17). A second was that although ubiquitous TPR domain proteins, such as AIP, commonly bind HSP90 to assist protein folding, the AIPL1-TPR-HSP90 interaction would not explain the specificity of AIPL1 (25,41). Our study sheds a new light on the versatility of TPR domains in protein folding by identifying P␥ as a novel TPR proteinbinding partner that imparts the selectivity of AIPL1 for its client PDE6. Another notable feature is that in the closed conformation, the ␣3-helix of the FKBP insert region occludes the ligand-binding surface of the TPR domain, whereas in the open conformation this surface was freely accessible to binding partners, such as HSP90 and P␥. Likewise, the prenyl-binding pocket in AIPL1- The AIPL1 FKBP and TPR domains are rendered in brown and orange, respectively. The farnesyl moiety is shown as yellow spheres, and the C-terminal residue present in PDE6 structure is shown as spheres colored by atom type. For clarity, the complex is shown only for the PDEA subunit.

Interactions of AIPL1 with PDE6
FKBP is more accessible in the open conformation. Notably, the closed autoinhibited conformation of AIPL1 is conceptually similar to that of GPSM2-LGN, where the ␣-helices of GoLoco-GPR motifs occupy the binding surface for TPR domain partners (42). Binding of G␣ subunits to GoLoco-GPR releases LGN from autoinhibition and facilitates interaction of the LGN-TPR domain with NuMA (nuclear mitotic apparatus protein), leading to formation of the signaling complex required for asymmetric cell division (42). Another important example of TPR domain-mediated autoinhibition is PP5 (protein phosphatase 5). In this case, the TPR domain engages with and restricts access to the catalytic site (40). This autoinhibition is released by the binding of HSP90 and fatty acids, resulting in the stimulation of phosphatase activity. By analogy to LGN and PP5, the AIPL1 TPR domain partner, HSP90 or P␥, might shift the equilibrium toward the open conformation, thereby promoting formation of the chaperone-PDE6 complex and enabling the expression of functional PDE6.
Interestingly, when the open conformation of AIPL1 was used in modeling the complex with holoPDE6, few or no clashes of AIPL1 with the PDE6 catalytic subunits were observed. Furthermore, in the resulting models of the AIPL1-PDE6 complex, the prenyl moiety in the FKBP domain binding pocket was in close proximity to the presumed prenylated C-terminal Cys of the PDE6 catalytic subunit. Thus, our models of the AIPL1-PDE6 complex are consistent with simultaneous binding of the prenylated C termini of the PDE6 catalytic subunits and P␥ subunits to the AIPL1 FKBP and TPR domains, respectively. The MD simulations suggested that the ␣3-helix of the FKBP insert region plays an important role in the dynamics and interdomain interactions of AIPL1 and demonstrated that deletion of this region (AIPL1-⌬␣3) leads to physical separation of the FKBP and TPR domains. The finding that AIPL1-⌬␣3 fully retained its capacity for binding of the prenyl moieties and P␥46 -87 yet lost the ability to chaperone PDE6 shows that specific orientations of AIPL1 domains are required for it to simultaneously bind the two ligands.
Our data suggest that the HSP90 C terminus and P␥63-87 bind to AIPL1-TPR in a mutually exclusive fashion. We hypothesize the following sequence of interactions during the folding of PDE6. Binding of HSP90 to AIPL1-TPR relieves the autoinhibition in the closed state of AIPL1, facilitating binding of AIPL1-FKBP to the post-translationally prenylated C termini of nascent PDE6 catalytic subunits. Thus, the PDE6 catalytic dimer is recruited to the chaperone complex consisting of AIPL1 and HSP90. In the initial PDE6 -AIPL1-HSP90 complex, HSP90 may assist primarily in folding of the PDE6 catalytic domains because functional expression of PDE6 GAF domains in the absence of AIPL1 has been reported (43,44). During this stage, a small fraction of PDE6 can become catalytically active, and a P␥-binding site on PDE6 may be partially formed. This model is supported by the basal P␥-sensitive PDE6 activity resulting from co-expression of the enzyme with AIPL1 in the absence of P␥ (12). During the next step, HSP90 dissociates from PDE6 or is actively outcompeted by P␥, and P␥ becomes an adaptor between AIPL1-TPR and the PDE6 catalytic domains. In the complex AIPL1-P␥-PDE6cat (or AIPL1-P␥-(PDE6cat) 2 -P␥-AIPL1, given PDE6 is a catalytic dimer), the affinity of AIPL1 for P␥-PDE6 is greater than that for PDE6 during the initial stage. This complex may allow AIPL1-TPR to direct the P␥ C terminus to the PDE6 catalytic pocket and/or induce the latter to adopt its native conformation and achieve its full catalytic power.
A sequential model of client protein interaction with ubiquitous, relatively nonselective chaperones such as HSP90 followed by assembly with specialized co-chaperones might be more broadly applicable in the context of protein folding. Future structural studies of the AIPL1-PDE6 complex are expected to reveal mechanistic details of the final steps in PDE6 folding that are critical for visual perception.

Plasmids/cloning
DNA sequence encoding human AIPL1-TPR (residues 171-316) was PCR-amplified and cloned as described earlier (29). DNA sequences that encode the full-length human AIP and AIP-TPR (residues 172-317) were PCR-amplified from cDNA isolated from HEK293T cells and cloned into the pET15b vector using NdeI/XhoI and Nco1/Nde1 sites, respectively. Chimeric TPR-C1 in which residues 244 -328 of AIPL1 were replaced with residues 245-330 of AIP, was generated in a twostep PCR procedure: (i) a DNA sequence encoding AIP 245-330 was amplified using forward hybrid AIPL1-AIP primers and a reverse primer for AIP with a Nde1 site; and (ii) this PCR product was then used as a reverse primer in a PCR with the AIPL1-TPR template and a forward primer for WT AIPL1-TPR with a NcoI site and His 6 tag. A similar two-step PCR protocol was used to generate TPR-C2, TPR-C3, and TPR-C4 in which residues 162-243, 162-192, and 218 -247 of AIPL1 were replaced with residues 163-244, 163-193, and 219 -248 of AIP, respectively. The PCR product was then cloned into the pET15b vector using NcoI/NdeI site. The AIPL1-⌬␣3 construct was generated by replacing the residues 111-132 of AIPL1 WT with five glycine residues in a two-step PCR procedure. In the first step, the AIPL1(1-110)-Gly-Gly-Gly-Gly-Gly(111-132) sequence was amplified with a forward primer with Nco1 site, His 6 tag and a reverse primer corresponding to AIPL1(100 -110)-Gly-Gly-Gly-Gly-Gly(111-132). This PCR product was used as a forward primer in the second PCR amplification with a primer containing the stop codon and Nde1 site. This resulting PCR product was then cloned into the pET15b vector using Nco1/NdeI sites.
The pcDNA3.1 vector for expression of the full-length mouse AIPL1, FLAG tag full-length PDE6, and enhanced GFPtagged P␥ was described previously (12). Full-length chimeras AIPL1-C1, AIPL1-C2, AIPL1-C3, and AIPL1-C1 in pcDNA3.1 vector were generated similarly to the construction of chimeric TPR domains in pET15b. AIPL1-C1 in which residues 244 -328 of AIPL1 were replaced with residues 245-330 of AIP, was generated in a two-step PCR procedure: (i) a DNA sequence encoding AIP 245-330 was amplified using forward hybrid AIPL1-AIP primers and a reverse primer for AIP containing an hemagglutinin tag and XbaI site; and (ii) this PCR product was then used as a reverse primer in a PCR with the AIPL1 template and a forward primer for WT AIPL1 with HindIII site and Kojak

Dynamic light scattering
DLS was used to examine the degree of polydispersity and thermostability of AIPL1-TPR, AIPL1-⌬␣3, and TPR chimeras. Purified proteins were concentrated up to 3 mg/ml and used for light-scattering experiments. For P␥-bound AIPL1-TPR samples P␥ peptide was added in 2.5-fold excess and incubated for 5 h and centrifuged at 45,000 for 1 h. DLS data for thermostability measurements were collected while heating the samples from 20 to 80°C at 1°C/min. Onset of protein thermal unfolding (T m-onset ) was determined by the sudden increase in hydrodynamic radius during the temperature ramp. Analyses of thermostability of AIPL1, AIPL1-TPR, AIPL1-⌬␣3 with and without P␥ peptide and TPR chimeras were performed using a DynaPro Nanostar instrument (Wyatt), and the data were analyzed using the Dynamics 7.1.7 software.

Biolayer interferometry binding assay
An Octet RED96 system and streptavidin-coated biosensors (FortéBio, Menlo Park, CA) were used to measure association and dissociation kinetics for P␥ C-terminal peptide (P␥46 -87 and P␥63-87) with AIPL1, AIPL1-⌬␣3, AIPL1-TPR domain, TPR-C1, TPR-C2, TPR-C3, and TPR-C4. Binding studies were performed in 20 mM Tris, 250 mM NaCl, 2.5% glycerol, 1 mM tris(2-carboxyethyl)phosphine, 0.5 mg/ml BSA (pH 7.5). All steps were performed at 26°C, with biosensors stirred into 0.2 ml of sample in each well at 1000 rpm and at a data acquisition rate of 5.0 Hz. N-terminally biotinylated P␥46 -87and P␥63-87 peptides were loaded onto streptavidin sensors at a concentration of 5 g/ml for 60 -90 s. Data for association and dissociation phases of the assay were collected as shown in Fig.  3 (A and C). To correct for baseline drift and nonspecific binding, reference sensors with bound P␥ peptide were used in the BLI assays without AIPL1-TPR proteins. Kinetic data fitting was performed using FortéBio Data Analysis software 10.0. For each concentration of AIPL1-TPR, dissociation rate constant (k d ) values were calculated from the corresponding dissociation phases of the curves. These k d values were used to calculate the association rate constant (k a ) values from the association phases for each concentration according to the equation: k a ϭ k obs -k d /[AIPL1]. The average k a and k d were calculated as means of the individual k a and k d values for all curves. K D was calculated as mean k d /mean k a . Steady-state data fitting was performed using the GraphPad Prism 8 software with the equation for one site-specific binding.

Crystallization and crystal structure determination
Purified AIPL1-TPR was concentrated to 8 -12 mg/ml. Crystals of AIPL1-TPR were grown by hanging-drop vapor diffusion at 18°C, against well buffer containing 100 mM Tris, 5-15% PEG8000 (pH 7.5-8.5) using a TTP LabTech Mosquito crystallization robot (TTP LabTech). The crystals grew over periods of 1-3 weeks. Crystals were cryoprotected using 20% glycerol in mother liquor. Data for the crystals were collected remotely from the University of Iowa Protein Crystallography Facility using the 4.2.2 Beamline at the Advanced Light Source (Berkeley, CA). The crystals were exposed to the beam for 0.1 s at a detector distance of 200 mm with a wavelength of 1 Å at 0.2°o scillation per frame, and data were collected across a 180°rotation. The data sets were indexed and integrated using XDS (45) and scaled using the SCALA (3.3.22) (46). PDB code 4AIF was used as search model, and molecular replacement was done using Phaser (47). Structure was refined using REFMAC (48) and Phenix (49), and manual models were built using Coot (50). All figures of structures and alignments were generated using the PyMOL Molecular Graphics System (version 1.8, Schrödinger, LLC). The atomic coordinates of AIPL1-TPR have been deposited in the Protein Data Bank (PDB code 6PX0).

Small-angle X-ray scattering
SAXS data were collected at the BioCAT (Beamline 18-ID; Advanced Photon Source, Argonne National Laboratory) using an in-line SEC-SAXS configuration (51). Concentrated protein samples of hAIPL1 1-316 (10 mg/ml) were loaded on a Superdex 75 10/300 GL gel-filtration column (GE Healthcare) equilibrated against gel filtration buffer containing 50 mM Tris, 100 mM NaCl, 2.5% glycerol, and 6 mm DTT. To separate any aggregates or oligomer states, the samples were subjected to scattering analysis in-line with size-exclusion chromatography. Scattering data were collected every 2 s using a 0.5-s exposure on a Pilatus 3 ϫ 1 M pixel detector (DECTRIS) covering a q range of 0.0040 Ͻ q Ͻ 0.388 Å Ϫ1 (q ϭ 4/ sin , where 2 is the scattering angle). For each protein, the buffer scattering before and after the eluted peak was recorded and used for background correction. The final protein scattering curves were obtained by

Cell culture and PDE6 activity assay
HEK293T cells were cultured and maintained in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum (Gibco). The cells were transfected with human PDE6C alone (2 g) or co-transfected with mouse P␥, mouse AIPL1 (1 g each), or both P␥ and AIPL1 (1 g each) plasmids using FuGENE 6 (Promega) according to the manufacturer's instructions.
For the PDE6C activity assay, PDE6C, AIPL1, or AIPL1 and P␥ co-transfected HEK293T cell lysates were prepared in hypotonic buffer (5 mM Tris-HCl, pH 7.5, 1 mM MgCl 2, buffer A) and centrifuged at 125,000 ϫ g for 30 min at 4°C in a Beckman Optima TLX ultracentrifuge. The supernatants thus obtained were used for all PDE6C activity measurements. cGMP hydrolysis was measured in cell lysates obtained from HEK293T cells 48 h post-transfection. The samples were treated with 0.1 mg/ml tosylphenylalanyl chloromethyl ketone-trypsin (Sigma) on ice for 10 min to selectively degrade P␥, after which trypsin was inhibited with the addition of 10-fold excess of soybean trypsin inhibitor (Sigma) and incubation for 5 min at 25°C. Cell lysates (protein concentration, 5-10 mg/ml) were diluted 4 -600-fold into 40 l (final volume) of a buffer containing 20 mM Tris-HCl (pH 7.5), 120 mM NaCl, 2 mM MgSO 4 , 1 mM 2-mercaptoethanol, 0.1 unit bacterial alkaline phosphatase, 10 M [ 3 H]cGMP (100,000 cpm) (PerkinElmer Life Sciences) for 15 min at 37°C. The reaction was stopped by the addition of AG1-X2 cation exchange resin (0.5 ml of 20% bed volume suspension). The samples were incubated for 6 min at 25°C with occasional mixing and spun at 10,000 ϫ g for 3 min. 0.25 ml of the supernatant was removed for counting in a scintillation counter.

NMR spectroscopy
For the studies of P␥ binding to AIPL1-TPR, NMR spectra were acquired on a Bruker Neo 600-MHz NMR spectrometer equipped with a sensitive QCI-P cryoprobe at 25°C using uniformly 15 N-labeled and selectively 13 C-methyl-labeled (for IVL residues) AIPL1-TPR at a protein concentration of ϳ150 M in a buffer containing 20 mM sodium phosphate (pH 7.5), 400 mM NaCl, and 6 mM DTT in 90% H 2 O, 10% D 2 O. For the studies of P␥ binding to AIPL1-FKBP, NMR spectra were acquired on a Bruker Avance II 800 MHz NMR spectrometer equipped with a sensitive TCI cryoprobe at 25°C using uniformly 15 N-labeled and selectively 13 C-methyl-labeled (for IVL residues) AIPL1-FKBP or uniformly 15 N, 13 C-labeled AIPL1-FKBP at a protein concentration of ϳ150 M in a buffer containing 25 mM sodium phosphate (pH 7.5) and 8 mM DTT in 90% H 2 O, 10% D 2 O. P␥ binding to AIPL1-TPR or AIPL1-FKBP was analyzed by acquiring 15 N/ 1 H and 13 C/ 1 H HSQC NMR spectra. The 1 H chemical shifts were referenced to 2,2-dimethyl-2-silapentane-5-sulfonate. The collected NMR data were processed using NMRPipe (54) and analyzed using NMRView (55). The model of AIPL1-TPR-P␥63-87 complex was calculated using the CNS program (36) with the NMR-derived restraints from the analyses of the P␥ binding data.

Modeling and MD simulations
The AIPL1-TPR crystal structure was first extended by nine helical N-terminal residues and two C-terminal residues according to the structure of AIP-TPR (PDB code 4AIF) using YASARA Structure 18.2.7. Extended AIPL1-TPR structure was used together with the structures of AIPL1-FKBP (PDB code 5U9A) and FKBP52 (PDB code 1P5Q:B) as input into Modeler (56) to produce the initial model of AIPL1(1-316).
MD simulations were performed using YASARA Structure 18.2.7 and the md_runfast macro. The simulations were run using the AMBER14 force field in water at a temperature of 298 K, pH of 7.4, and NaCl concentration of 0.9%. The particle mesh Ewald summation was used to compute long-range coulombic interactions with a periodic cell boundary and a cutoff of 8 Å. The MD simulations were analyzed using the md_analyze macro in YASARA and PyMOL programs. The Crysol program (57) was used to generate and compare fits of theoretical SAXS profiles of the models from MD trajectories (4 models per 1 ns of MD simulation) to experimental SAXS data ( 2 values). Models of AIPL1(1-316) from MD simulations with low 2 values were superimposed with the model of the AIPL1-TPR-P␥63-87 complex and subsequently overlaid with the structure of PDE6 (PDB code 6MZB) by aligning the ␣2-helix residues of P␥, P␥78 -83.