Identifying Stereochemical Preferences that Modulate Affinity and Selectivity among Five PDZ Domains that Bind CFTR: Comparative Crystallographic and Sequence Analyses

Summary

PDZ domain interactions are involved in signaling and trafficking pathways that coordinate crucial cellular processes. Alignment-based PDZ binding motifs identify the few most favorable residues at certain positions along the peptide backbone. However, sequences that bind the CAL (CFTR-Associated Ligand) PDZ domain reveal only a degenerate motif that overpredicts the true number of high affinity interactors. Here, we combine extended peptide-array motif analysis with biochemical techniques to show that non-motif ‘modulator’ residues influence CAL binding. The crystallographic structures of 13 new CAL:peptide complexes reveal defined, but accommodating stereochemical environments at non-motif positions, which are reflected in modulator preferences uncovered by multi-sequence substitutional arrays. These preferences facilitate the identification of new high-affinity CAL binding sequences and differentially affect CAL and NHERF PDZ binding. As a result, they also help determine the specificity of a PDZ domain network that regulates the trafficking of CFTR at the apical membrane.

Introduction

If each residue is uniquely specified, a decapeptide can encode more than 10¹³ distinct sequences, providing a versatile and specific mechanism to encode protein:protein interactions. Indeed, domains recognizing short linear motif (SLiM) peptides are found throughout the human genome, including the abundant family of PDZ domains first recognized in the proteins PSD-95, Dlg, and ZO-1 (Davey et al., 2012; Harris and Lim, 2001; Lee and Zheng, 2010; Nourry et al., 2003). PDZ target interactions are governed by a variety of constraints, including local concentration, multidentate binding interfaces, and interactions with additional proteins (Luck et al, 2012). Nevertheless, the recognition of a generally C-terminal peptide by the PDZ binding cleft is the core of the interaction, conferring critical target specificity. As a result, the underlying sequence:affinity relationships are essential to understand connectivity in PDZ-mediated protein networks.

To identify the binding preferences of individual domains, early studies compared the sequences of known binding partners and highlighted the importance of the amino acids at the P⁰ (extreme C-terminus) and P⁻² positions (e.g., Songyang et al., 1997). High-throughput screens of phage-display and peptide-array libraries have since revealed more complex ‘motifs’ involving varying combinations of up to seven C-terminal residues, consistent with stereochemical interactions observed in individual complexes (Doyle et al., 1996; Laura et al., 2002; Schultz et al., 1998; Skelton et al., 2003; Stiffler et al., 2007; Tonikian et al., 2008). Furthermore, although few motifs have been investigated beyond the P⁻⁶ position, biochemical and structural studies have identified affinity contributions and stereochemical contacts extending as far as the P⁻¹⁰ position (reviewed in Luck et al., 2012).

Even with extended motifs, PDZ binding motifs on average constrain fewer than four of the potential interacting residues and often accommodate similar residues at a given position (Davey et al., 2012). As a result, PDZ domains are frequently promiscuous, binding multiple target proteins and sharing targets with other PDZ domains. This is illustrated by a set of PDZ proteins that regulate the intracellular trafficking and localization of the cystic fibrosis transmembrane conductance regulator (CFTR): the CFTR-Associated Ligand (CAL) and the NHE3 Regulatory Factor proteins NHERF1 and NHERF2 (Cheng et al., 2002; Guerra et al., 2005; Wolde et al., 2007). Consistent with shared target specificity, sequence alignments revealed overlapping motifs at the P⁰ and P⁻² positions, and sequence optimization in the four C-terminal residues led to only 10-fold selectivity for the CAL PDZ (CALP) domain (Vouilleme et al., 2010). Using a peptide-array approach to extend the sequence iteratively towards the N-terminus, we engineered a CALP inhibitor (iCAL36; sequence ANSRWPTSII) with 170-fold selectivity (Cushing et al., 2010; Vouilleme et al., 2010).

Here we characterize the interaction of the CAL PDZ domain with these N-terminal (‘upstream’, i.e., P⁻⁴ to P⁻⁹) binding determinants and with the P⁻¹ and P⁻³ side chains. Although not captured by sequence alignment techniques, they act as binding ‘modulators’ for each domain, exerting a modest influence individually, but a powerful impact collectively. Crystallographic analysis of 13 new CALP:peptide complexes spanning a wide range of affinities reveals an accommodating pharmacophore binding model. In-depth stereochemical and substitutional array analysis provides clues to the identification of preferences that remain hidden from alignment motifs. These modulatory effects facilitate the identification of high-affinity binding sequences and contribute to inhibitor specificity in the context of CFTR trafficking.

Results

Sequence alignments identify only core binding motifs for CAL and NHERF

Based on the demonstrated contributions of upstream residues to the selectivity of our engineered iCAL36 peptide, we attempted to identify extended binding motifs for the CAL and NHERF PDZ domains. We performed WebLogo analysis of all 10 C-terminal positions in peptides identified by screening a 6223HumLib array (Cushing et al., 2008) with CALP or with the NHERF1 and NHERF2 PDZ1 and PDZ2 domains (N1P1, N1P2, N2P2, and N2P2, respectively). However, the results (Figures 1A and S1A) did not highlight any significant upstream preferences outside the previously examined core residues P⁰ to P⁻³ (Roberts et al., 2012; Vouilleme et al., 2010).

Figure 1. Sequence determinants of CAL PDZ binding and CAL/NHERF specificity.

(A) WebLogo analysis (Crooks et al., 2004) of the top 100 sequences that bound CALP in a 6223HumLib peptide array experiment reveals clear preferences at the P⁰ and P⁻² positions, similar to the short, degenerate binding motifs seen with the NHERF PDZ domains (Figure S1A). The C-terminal residue (P⁰) position is labeled 0; adjacent residues are −1, −2, etc.

(B) For each PDZ domain, free energies of binding (kcal/mol) were calculated from the binding affinities determined by FP for 10 different iCAL36/CFTR chimeric peptide sequences (Table 1). The number of intersection points in pairwise free-energy plots (Figure S2) were tabulated to reveal rank-order exchanges as a function of sequence.

Weak secondary CALP preferences are observed at a few positions (e.g., S>E>V at P⁻³; Figure 1A). To test the robustness of such signatures, we calculated motifs using a varying number of input sequences, and observed that both the rank order and stereochemical nature of the residues changed. For example, at P⁻³, the top 10 CALP-binding sequences yielded the pattern E>S>T, whereas the top 50 sequences yielded S>I>Q (Figure S1B). Furthermore, while the top 10 CALP-binding sequences show modest signals, e.g., for Arg at P⁻¹ or Ser at P⁻⁷, these grow progressively weaker or shift with the addition of more sequences (Figures 1A and S1B). To evaluate the possibility of correlated binding requirements across multiple ligand positions, we reprobed our top 100 sequences using the MUltiple Specificity Identifier (MUSI) program, which uses both single and multiple position weight matrices to determine specificity patterns (Kim et al., 2011). Overall, the results again confirm the highly degenerate CALP binding motif originally seen at the P⁰ and P⁻² positions (Figure S1C).

Similar patterns are observed for the NHERF domains (Figure S1A). In some cases, secondary preferences seen at the P⁻³ position (e.g., S>E>T for N1P2 and E>S>D for N2P2) are slightly stronger than those observed for CALP. At least in the case of the N2P2 domain, the presence of acidic residues is broadly consistent with a P⁻³ preference for Asp reported in earlier phage-display studies (Tonikian et al., 2008). However, as for CALP, no strong signatures were observed upstream of P⁻³.

Chimeric peptides reveal positive and negative affinity modulators

Although our experience suggested that residues outside the P⁰/P⁻² motif positions significantly enhanced binding selectivity of iCAL36 for CAL vs. NHERF domains, we had not focused on CALP affinity per se, which changed relatively little in the later stages of iCAL36 sequence engineering (Vouilleme et al., 2010). Given the lack of upstream binding motifs, we decided to explore the free-energy contributions of non-motif residues, using nested sets of chimeric peptides based on reference sequences with both high (iCAL36; K_i = 23 μM) and low (CFTR; K_i = 420 μM) affinity for the CALP domain (Amacher et al., 2013). Using fluorescence polarization (FP) competition experiments, we determined K_i values for each of the chimeric peptides binding to CALP (Table 1). CALP has modest preferences for an ^-9ANSR^-6 sequence over ^-9TEEE^-6 (iCAL36_VQDTRL vs. CFTR), Trp over Val at P⁻⁵ (iCAL36_QDTRL vs. iCAL36_VQDTRL), and Gln over Pro at P⁻⁴ (iCAL36 vs. iCAL36_Q^-4). A positive CALP preference for a P⁻¹ Arg is also suggested by the 5-fold higher affinity of the iCAL36_TRL sequenced compared to iCAL36_L and the approximate equivalence of the Ser/Thr motif at P⁻² (Vouilleme et al., 2010). However, the most significant effect is seen at P⁻³. Replacing ^-5WPT^-3 with ^-5VQD^-3 in iCAL36_VQD^-3 dramatically reduces CALP affinity (K_i > 1 mM). Since neither the P⁻⁵ nor P⁻⁴ substitutions account for such a large effect, we synthesized a CFTR peptide with a single substitution at P⁻³. CFTR_T^-3 (TEEEVQTTRL; K_i = 18 μM) binds 23-fold more tightly than the native CFTR sequence (Table 1).

Table 1.

FP binding affinities of CALP and NHERF PDZ domains for iCAL36/CFTR chimeric peptides.

		Ki (μM)
Peptidea	Sequenceb	CALP	CALP E317A	N1P1	N1P2	N2P1	N2P2
iCAL36TRL	ANSRWPTTRL	4.5 ± 0.8	2.2 ± 1.2	8.0 ± 1.4	110 ± 10	8.7 ± 3.1	16.5 ± 3.9
iCAL36Q-4	ANSRWQTSII	14.8 ± 3.5	NDd	280 ± 100	>1000	300 ± 100	450 ± 150
CFTRT-3	TEEEVQTTRL	18 ± 5.5	15.1 ± 2.0	0.76 ± 0.51	4.6 ± 1.6	1.3 ± 0.4	0.45 ± 0.25
iCAL36c	ANSRWPTSII	22.6 ± 8.0	15.1 ± 2.1	>1000	>1000	>1000	>1000
iCAL36Lc	ANSRWPTSIL	23.6 ± 2.2	NDd	230 ± 100	>1000	76.8 ± 1.6	530 ± 120
iCAL36QDTRL	ANSRWQDTRL	56.2 ± 5.6	16.2 ± 3.5	1.5 ± 0.2	6.7 ± 2.4	3.1 ± 1.4	1.1 ± 0.6
iCAL36VQDTRL	ANSRVQDTRL	230 ± 30	140 ± 40	2.5 ± 0.3	6.5 ± 2.3	3.8 ± 1.4	0.94 ± 0.42
CFTRc	TEEEVQDTRL	420 ± 80	64.7 ± 1.8	0.47 ± 0.2	1.8 ± 0.5	0.83 ± 0.33	0.07 ± 0.05
CFTRIc	TEEEVQDTRI	490 ± 120	NDd	10.7 ± 2.8	39.8 ± 1.0	19.1 ± 2.9	2.5 ± 0.43
iCAL36VQD-3	ANSRVQDSII	>1000	770 ± 50	>1000	>1000	>1000	420 ± 120

^aSubstitutions noted in subscript; the final substituted position is C-terminal unless numbered.

^bBold indicates residue(s) from the CFTR sequence; non-bold indicates residues from iCAL36.

^cValues previously reported in Amacher et al., 2013.

^dND indicates that value was not measured.

To test the hypothesis that non-motif residues also impact NHERF1 and NHERF2 affinities, we measured the binding constants of the chimeric peptides for each of the four PDZ domains (N1P1, N1P2, N2P2, and N2P2, respectively) in FP displacement assays (Table 1). The NHERF PDZ domains bind the CFTR and iCAL36 peptides with high affinity for CFTR (K_i <2 μM) and low affinity for iCAL36 (K_i >1 mM) (Amacher et al., 2013). Among these domains, substitutions at P⁻¹ (iCAL36_TRL vs. iCAL36_L), P⁻³ (CFTR_T^-3), P⁻⁴ (iCAL36_Q^-4), and P⁻⁶-P⁻⁹ (CFTR vs. iCAL36_VQDTRL) each affect individual K_i values.

Finally, to explore the potential of these non-motif differences to contribute to PDZ inhibitor selectivity, we developed an approach to identify substitutions that switched rank order of affinity. We first plotted the binding free energies for the series of peptides along a vertical axis for each domain. When placed side-by-side for a given pair of PDZ domains, with straight line connections between the values for each peptide, rank-order switches appear as points of intersection (Figure S2). Compared to each NHERF PDZ domain, the CALP affinities exhibit 19–25 pairwise affinity crossovers (Figure 1B) that presumably account for the engineered selectivity of iCAL36. Thus, although the CAL and NHERF binding motifs fail to highlight preferences at P⁻¹, and P⁻³ through P⁻⁹, these residues are capable of ‘modulating’ both the affinity and selectivity of peptides for these PDZ domains. Furthermore, despite high sequence identity (58–72%) between the NHERF PDZ domains, pairwise comparisons reveal 2–6 crossovers (Figure 1B), suggesting potential non-motif contributions to selectivity even in this tightly related cluster.

All six C-terminal positions form side chain-dependent interactions

Despite the absence of motif signatures at four of the six residues from P⁰ to P⁻⁵, each residue directly contacts the surface of the CALP binding site, as illustrated in Figure 2A for one of the two protomers in our previously published CALP:iCAL36 structure (Amacher et al., 2013). Corresponding contact surfaces on the peptide are highlighted by comparison of the solvent-accessible surface of the bound peptide in the absence (Figure 2C) and presence of the CALP domain (Figure 2D).

Figure 2. Six peptide positions interact with the CALP binding cleft.

(A) The van der Waals surface of the CALP protein is shown together with the bound iCAL36 peptide (stick figure), with individual residues labeled and colored by position. The CALP surface (cyan) is colored at contact sites according to the closest peptide residue: P⁰ Ile (red), P⁻¹ Ile (purple), P⁻² Ser (gray), P⁻³ Thr (orange), P⁻⁴ Pro (yellow), P⁻⁵ Trp (blue), P⁻⁶ Arg (green), P⁻⁷ Ser (black), P⁻⁸ Asn (pink), and P⁻⁹ Ala (forest green).

(B) The peptide side-chain interaction index (I_PDZ) is shown as a function of residue position for the peptides interacting with the A- (gray), and B- (black) CALP protomers of the asymmetric unit.

(C–D) The van der Waals surface of the bound conformation of the iCAL36 peptide is shown in the same perspective as in panel A (left) and following rotation by 135° around the vertical axis (right). The surface is colored by residue as in (A). To illustrate the surface contacted by CALP, individual atoms with I_PDZ values > 0.5 are colored white in (D), involving residues P⁰-P⁻⁵.

The non-motif contacts also represent potentially significant stereochemical constraints. The fraction of solvent-accessible surface area (SASA) buried serves as an interaction index (I_PDZ) that gauges the extent of side-chain engagement at each position along the backbone (Figure 2B). At the non-motif position P⁻⁵, I_PDZ values (0.87 and 0.63 for the A and B protomers, respectively) are comparable to those for the motif residue P⁻² (0.73 and 0.64). The other side chains exhibit I_PDZ values ranging from 0.26 to 0.60, and even for residues with lower side-chain I_PDZ values, several side-chain atoms are occluded at least 50% upon CALP docking (atoms colored white in Figure 2D).

Of course, contact mapping cannot by itself distinguish between favorable and unfavorable binding relationships. It also does not capture long-range (e.g., electrostatic) interactions or shifts in the free energy of the unbound state. Nevertheless, the observed steric interactions are consistent with evidence that non-motif residues can impact CALP affinity.

Defined CALP binding sites interact with each residue position

Given the characteristically shallow peptide-binding pocket of the CALP domain, we wished to determine whether our substituted peptides adopt significantly variable conformations, which could complicate stereochemical analysis and obscure modulator preferences in sequence alignments. To address this point, we crystallized and solved the structures of CALP bound to four chimeric iCAL36/CFTR peptides, selected to cover a range of affinities: iCAL36_TRL (K_i = 4.5 μM), iCAL36_Q^-4 (K_i = 14.8 μM), iCAL36_QDTRL (K_i = 56.2 μM), and iCAL36_VQD^-3 (K_i >1 mM),. To further explore sequence and conformational space, we also crystallized CALP bound to two additional low-affinity binding partners with unrelated sequences. HPV E6 viral oncoprotein C-terminal peptides have been studied in complex with other PDZ domains by NMR and X-ray crystallography (Charbonnier et al., 2011; Liu et al., 2007; Mischo et al., 2013; Zhang et al., 2007). Here, we generated CALP complexes with the decameric E6 C-terminal peptides from HPV16 (SSRTRRETQL) and HPV18 (RLQRRRETQV) (Jeong et al., 2007; Pim et al., 2012), with K_i values of 340 ± 70 μM and 490 ± 20 μM, respectively.

Structure determination and refinement protocols were designed to avoid phase bias (Table S1; Figures S3–4). Final data and refinement statistics are shown in Table 2. In general, the iCAL-based structures are very similar. When the PDZ backbone atoms of each structure are aligned to the CALP:iCAL36 B-protomer, they exhibit an average RMSD value of 0.34 ± 0.09 Å, with a maximum of only 0.52 Å (B-protomer of CALP:iCAL36_TRL). Visually, the superposition of these CALP domains (Figure 3A) confirms that the framework of the binding cleft is largely independent of ligand sequence.

Table 2.

Data collection and refinement statistics for CALP co-crystals with iCAL36/CFTR chimeric peptides.

	iCAL36Q-4	iCAL36TRL	iCAL36VQD-3	iCAL36QDTRL	iCAL36QDTRL CALP-E317A	HPV16 E6	HPV18 E6
Data Collection
Space Group (Number)	P 21 21 21 (19)	P 1 (1)	P 21 (4)	P 63 2 2 (182)	P 63 2 2 (182)	P 21 (4)	P 21 (4)
Unit cell dimensions
a,b,c (Å)	36.0,47.7,98	30.8,50.2,55.2	36.3,48.8,54.9	61.6,61.6,97.6	61.8,61.8,97.4	33.1,48.4,52	33.2,48.0,52.9
α,β,γ (°)	90,90,90	68.8,75.8,87.9	90,92.8,90	90,90,120	90,90,120	90,101.6,90	90,102.0,90
Resolutiona (Å)	19.9–1.48 (1.58–1.48)	19.3–1.75 (1.79–1.7)	19.6–1.90 (2.03–1.9)	19.7–1.50 (1.58–1.5)	19.1–1.80 (1.85–1.8)	19.4–1.80 (1.90–1.8)	19.3–1.34 (1.42–1.34)
Rsymb (%)	5.9 (45.0)	4.5 (37.2)	6.5 (36.0)	4.9 (61.5)	11.0 (61.7)	5.9 (52.3)	3.8 (26.5)
I/σI	34.60 (6.93)	15.48 (2.60)	15.58 (3.59)	49.46 (6.32)	26.01 (6.53)	20.37 (3.16)	36.69 (7.53)
Completeness (%)	99.7 (99.2)	94.8 (91.1)	99.6 (99.5)	99.9 (100.0)	99.9 (100.0)	99.4 (98.1)	94.2 (77.8)
Refinement
Total # of reflections	28,816	30,585	15,162	18,179	10,780	14,953	34,482
Reflections in test set	1,460	1,493	763	920	560	758	1,729
Rworkd/Rfreee	18.1/20.9	17.2 /22.1	17.9/22.7	19.7/22.1	18.1/22.1	18.3/22.3	17.2/18.1
Number of atoms:
Protein	1467	2927	1426	774f	772 f	1431	1507
Water	180	267	141	69	69	147	258
Ramachandran plotg (%)	96.8/3.2/0/0	95.6/4.4/0/0	98.0/2.0/0/0	97.3/2.7/0/0	100/0/0/0	96.6/3.4/0/0	97.4/2.6/0/0
Bav (Å2)
Protein	13.27	18.96	22.38	19.36	16.32	23.40	9.86
Solvent	24.62	28.73	31.27	30.03	27.43	31.13	23.10
Bond length RMSD (Å)	0.006	0.006	0.007	0.007	0.008	0.007	0.007
Bond angle RMSD (°)	1.078	1.006	1.097	1.169	1.203	1.064	1.167

^aValues in parentheses are for data in the highest-resolution shell.

^bR_sym = Σ_hΣ_i |I(h) − I_i(h)| / Σ_hΣ_i I_i(h), where I_i(h) and I(h) values are the i-th and mean measurements of the intensity of reflection h.

^cSigAno = |F(+) − F(−)| / σ

^dR_work = Σ ||F_obs|_h − |F_calc||_h / Σ |F_obs|_h, h ∈ {working set}

^eR_free is calculated as R_work for the reflections h ∈ {test set}

^fincluding sulfocysteine at position 319 (see Figure S4)

^gCore/allowed/generously allowed/disallowed

Figure 3. Peptide residues lie in distinct CALP binding sites.

(A) C_α traces are shown for CALP protomers from each of four different space-groups (Table S1), following superposition of main-chain atoms on the B-protomer of CALP:iCAL36. Traces are colored by space group: P2₁2₁2₁ (black), P1 (green), P2₁ (cyan), and P6₃22 (pink). Every 10^th residue is indicated with a black circle and residue number. The tight clustering reveals no gross conformational changes in the PDZ domain based on lattice packing, peptide sequence, or affinity. See also Figure S3.

(B) Following superposition of PDZ main-chain atoms, chimeric iCAL36/CFTR peptides (stick figures) bind CALP (cartoon, gray) in similar conformations, despite sequence and affinity differences (Table 1). Characteristic PDZ hydrogen bonds are shown as dashed lines, including carboxylate interactions with residues Gly³⁰⁰ and Leu²⁹⁹ and a P⁻² interaction with His³⁴⁹ (stick figures). iCAL36 (green carbons) and the highest (iCAL36_TRL; cyan) and lowest (iCAL36_VQD^-3; orange) affinity peptides are highlighted for reference. Other peptides (gray) include iCAL36_QDTRL, iCAL36_Q^-4, iCAL36L, HPV16 E6, and HPV18 E6. See also Figure S8A–G.

(C) Following superposition of PDZ main-chain atoms, bound peptide conformations are shown for C-terminal peptides derived from the HPV16 (black) and HPV18 (pink) E6 proteins together with iCAL36 (green). Binding sites are shown schematically for each position (S⁰, S⁻¹, etc.). Non-carbon atoms are colored by element (red = O, blue = N)

This domain superposition also reveals strong conformational similarity of the peptides. We first compared the complexes formed by derivatives of the iCAL sequence, including the chimeric peptides described above, as well as earlier structures of iCAL36 and iCAL36_L (Amacher et al., 2013). Despite widely varying binding affinities, all peptide co-complexes reflect canonical Class I PDZ domain-target interactions: the P⁰ residue with the carboxylate-binding loop, the P⁻² residue with His³⁴⁹, and conserved β-sheet interactions (Figure 3B). Pairwise comparisons of the peptides reveal C_α variations less than 1.1 Å for the P⁰ to P⁻³ positions, and up to 4.4 Å for the P⁻⁴ and P⁻⁵ positions. These latter variations may partially reflect conformational changes due to differential crystal packing (Figure S3), but the side chains at each position still contact relatively similar surfaces on CALP. Thus, even the substitution of half of the residues in the peptide has only a limited effect on the stereochemical environment experienced by the side chain at a given position along the backbone.

Comparison of iCAL36 with the HPV E6 peptides also revealed similar peptide backbone conformations, despite very low sequence similarity (Figure 3C). Thus, in all of our structures, the side chains at a given peptide position are oriented similarly, contacting a common CALP local environment. As a result, we can extend previous models (e.g., Kang et al., 2003) to define a pharmacophore model in which positions P⁰ to P⁻⁵ each interact with cognate sites S⁰ to S⁻⁵ (Figure 3C). According to this model, each CALP site exhibits a characteristic set of stereochemical constraints that favor the binding of a corresponding subset of complementary amino acid(s). If the constraints are stringent, they may be present in enough sequences to appear as part of a motif. If they are more accommodating, they may not, but may still contribute to affinity as modulators.

Positional SubAna analysis confirms multiple modulatory residues

As a further test of the idea that each peptide side chain contacts a defined stereochemical environment relatively independent of sequence context, we performed substitutional analysis (SubAna) using peptide arrays derived from a variety of starting sequences. Each SubAna array contains a single peptide sequence with all 20 natural L-amino acids individually substituted at each position along the chain (Boisguerin et al., 2007). We chose to use SubAnas based on 10 starting sequences that encompassed a number of different CALP-binding attributes (sequences described in detail in the supplemental information, Figure S5).

Except for the artificial iCAL36 sequence, these peptides are derived from endogenous proteins, include known interacting partners (Cheng et al., 2002; Wente et al., 2005), cover a wide range of affinities for CALP (Cushing et al., 2008), and provide a variety of binding motif residue combinations (S/T at the P⁻² position, I/L/V at the P⁰ position) for comparison. Furthermore, while the amount of protein bound is measured at a single concentration under non-equilibrium conditions, we can qualitatively compare trends across arrays (Boisguerin et al., 2011). In order to compare positional preferences across the full set of peptides, we separated each SubAna array into six strips, corresponding to the residues from P⁰ to P⁻⁵, and then aligned all of the strips for P⁰, all of the strips for P⁻¹, etc. The resulting collage (Figure 4) shows the variability of side-chain preference(s) at a given position. Consistent selective preferences at the P⁰ and P⁻² positions reflect their status as motif residues, largely independent of peptide context.

Figure 4. Position-specific peptide arrays reveal modulator preferences for CALP.

SubAna arrays of 10 peptides are clustered by residue position, with each row labeled according to the background sequence of the source array (see key for sequence information). The WT residues are listed to the right of each row. Residues in the initial sequence are substituted by all 20 L-amino acids, as indicated at top (see text). The motif positions P⁰ and P⁻² reveal clear preferences for only a few residues. Positive modulator preferences are seen as consistently stronger binding than WT following substitution of e.g., Arg at P⁻¹ or Trp at P⁻⁵, independent of the background peptide sequence. Negative preferences are also seen as consistently weaker binding compared to WT, e.g., following substitution of Gly at P⁻¹ or Asp at P⁻³. See also Figure S5.

At the other positions, a more heterogeneous picture emerges, consistent with the lack of motif constraints observed previously (Figure 1B). A residue may be favored for certain sequences, and disfavored for others. For example, a P⁻⁴ Phe residue appears to be strongly favored for CFTR (row 7) and other sequences, while disfavored for VIPR2 (row 9). However, despite this variability, certain trends are observed that apply to most or all sequences. For example, our positional arrays confirm positive preferences for a P⁻¹ Arg and for a P⁻⁵ Trp residue. They also reveal negative preferences, including P⁻³ Asp (Figure 4), consistent with the affinity difference observed between CFTR and CFTR_T^-3 (Table 1). CALP also appears to disfavor a Gly or Pro at P⁻¹ and Pro at P⁻³.

Single substitutions at P⁻⁵ modulate binding affinity

To investigate the stereochemistry of preferential accommodation of a modulator residue, we focused on the impact of single residue substitutions in iCAL36 at the P⁻⁵ position. Competition FP experiments confirm a range of affinities for P⁻⁵-substituted peptides (Table 3). The Leu- and Tyr-substituted peptides bind CALP with K_i ≤100 μM, whereas the Ala- and His-substituted peptides each show ≥10-fold losses in affinity. A P⁻⁵ His side chain results in the weakest affinity of the set, at 450 μM, comparable to that of CALP:CFTR binding.

Table 3.

CALP binding affinities for P⁻⁵-substituted iCAL36 peptides.

		Ki (ìM)	Contact Surface Areac (Å2)
Peptide	Sequencea	CALP	Peptide	P−5 Position
iCAL36b	ANSRWPTSII	22.6 ± 8.0	533.4	109.5
L-iCAL36	ANSRLPTSII	68 ± 20	492.2	73.9
Y-iCAL36	ANSRYPTSII	85.3 ± 1.5	518.2	99.6
F-iCAL36	ANSRFPTSII	110 ± 20	506.9	86.6
V-iCAL36	ANSRVPTSII	190 ± 40	485.8	66.4
A-iCAL36	ANSRAPTSII	275 ± 80	461.9	42.5
H-iCAL36	ANSRHPTSII	450 ± 10	505.5	79.2

^aSubstitutions at the P⁻⁵ position are indicated in bold.

^bValues previously reported in Amacher et al., 2013.

^cContact surface area for each P⁻⁵ position includes all atoms of the residue.

All 6 peptides were crystallized in complex with CALP. Structures were determined using the methods described above for the chimeric peptide complexes (Table S1 and supplemental methods), with data and refinement statistics reported in Table 4. For structural comparisons, we aligned the main-chain atoms of the P⁻⁵-substituted peptide:CALP complexes, using CALP:iCAL36 as a reference. Since the superposition of chimeric peptides had shown slight variations at the C-terminal core (positions P⁰ to P⁻³), we first compared the positions of the C_α atoms in these residues. In this region, the RMSD for the P⁻⁵-substituted peptides is less than the coordinate error (Figure 5A). There is a larger degree of variability at the P⁻⁴ and P⁻⁵ positions (≤3 Å), but the side chains remain clustered, again suggesting that modulations in binding affinity at the P⁻⁵ residue are due to local binding effects, not to large-scale alterations of residue interactions (Figure 5A).

Table 4.

Data collection and refinement statistics for CALP co-crystals with P⁻⁵-substituted iCAL36 peptides.

	A-iCAL36	L-iCAL36	V-iCAL36	H-iCAL36	F-iCAL36	Y-iCAL36
Data Collection
Space Group	P 21 21 21	P 21 21 21	P 21 21 21	P 21 21 21	P 21 21 21	P 21 21 21
Unit cell dimensions
a,b,c (Å)	36.7,47.8,97.6	36.2,47.6,97.3	36.7,47.9,97.8	36.2,47.8,98.1	36.2,48.1,97.6	36.2,47.8,97.5
α,β, γ (°)	90,90,90	90,90,90	90,90,90	90,90,90	90,90,90	90,90,90
Resolutiona (Å)	18.5–1.14 (1.19–1.14)	34.0–1.2 (1.24–1.2)	19.7–1.46 (1.54–1.46)	19.5–1.47 (1.58–1.47)	19.6–1.2 (1.26–1.2)	18.5–1.09 (1.15–1.09)
Rsymb (%)	11.6 (39.3)	6.3 (51.5)	6.8 (50.4)	9.3 (50.6)	8.4 (47.1)	7.5 (34.5)
I/σI	10.14 (3.99)	20.98 (4.24)	16.71 (3.05)	21.17 (6.39)	13.28 (3.86)	15.57 (4.98)
Completeness (%)	99.4 (99.9)	99.9 (99.9)	98.1 (88.7)	98.3 (97.2)	98.3 (95.7)	98.2 (92.4)
Refinement
Total # of reflections	63,011	53,398	30,171	28,702	53,211	70,084
Reflections in the test set	3,144	2,674	1,516	1,446	2,658	3,525
Rworkd/Rfreee	18.5/19.5	18.9/20.1	17.9/19.5	18.0/20.7	18.5/18.8	18.2/18.8
Number of atoms:
Protein	1502	1483	1472	1428	1522	1520
Water	236	230	180	170	227	220
Ramachandran plotf (%)	94.8/3.4/0/0	96.6/3.4/0/0	97.3/2.7/0/0	96.6/3.4/0/0	97.4/2.6/0/0	97.3/2.7/0/0
Bav (Å2)
Protein	12.35	12.27	14.85	9.77	12.20	11.95
Solvent	25.17	23.39	27.11	21.72	24.24	23.52
Bond length RMSD (Å)	0.005	0.006	0.006	0.006	0.006	0.007
Bond angle RMSD (°)	1.108	1.102	1. 043	1.063	1.107	1.073

^aValues in parentheses are for data in the highest-resolution shell.

^bR_sym = Σ_hΣ_i |I(h) − I_i(h)| / Σ_hΣ_i I_i(h), where I_i(h) and I(h) values are the i-th and mean measurements of the intensity of reflection h.

^cSigAno = |F(+) − F(−)| / σ

^dR_work = Σ ||F_obs|_h − |F_calc|| _h / Σ |F_obs| _h, h∈ {working set}

^eR_free is calculated as R_work for the reflections h∈ {test set}

^fCore/allowed/generously allowed/disallowed

Figure 5. Stereochemical analysis of modulator preferences.

(A) The clustering of CALP-bound P⁻⁵-substituted iCAL36 peptides (stick figures) is shown following main-chain superposition of the CALP B-protomers. Peptides are colored by sequence: iCAL36 (W; green carbons), F- (red), Y- (black), H- (yellow), A- (cyan), L- (orange), and V- (purple). RMSD offsets at the P⁰-P⁻³ residues were within coordinate error, while modest variations were observed at the P⁻⁴ and P⁻⁵ positions (black arrow). The largest difference is 3.2 Å, between the C_α of the L- and C_α of the A-substituted peptides.

(B) Each substituted P⁻⁵ side chain (colored stick figure) is shown, as docked against its CALP binding interface (surface, cartoon and stick figure). The residues that contact the P⁻⁵ residue (His309, Gly310, Val311) are labeled in the CALP:F-iCAL36 structure. Each panel is labeled with the P⁻⁵ residue identity. See also Figure S6.

(C) The electrostatic potential surface of WT CALP (left) reveals a highly polarized distribution on either side of the peptide binding cleft. The electrostatic potential surface of CALP-E317A (right) shows a reduced, although still negative, interaction surface for the P⁻³ Asp residue of iCAL36_QDTRL (insets). See also Figure S7.

(D) Electrostatic interactions are shown for P⁻¹ Arg (left) and P⁻³ Asp (right) residues of the iCAL36_QDTRL peptide (stick figure, yellow carbons) bound to CALP (gray C_α trace). At the top, nearby charged CALP residues are shown (stick figures, green carbons). The P⁻¹ Arg forms a hydrogen bond (dashed line) with Gln³¹⁴. The closets charged residue for the P⁻³ Asp is Glu³¹⁷ (d > 5 Å). In the lower panels, the corresponding electrostatic potential surfaces are shown (rendered at 10 k_BT/e) underscoring the negative electrostatic environment of both the P⁻¹ and the P⁻³ side chains.

As seen for the chimeric peptides (Figure 3C), all P⁻⁵ residues interact with S⁻⁵, comprised of the same His³⁰⁹-Gly³¹⁰-Val³¹¹ hydrophobic ledge that interacts with the P⁻⁵ Trp in iCAL36 (Amacher et al., 2013). However, despite this shared environment (Figure 5B; Figure S6), there is no obvious correlation of the observed range of affinities with stereochemical parameters, including electrostatic interactions or buried surface area metrics. For example, although the highest affinity peptide (Trp) is associated with the largest contact surface area, the second highest-affinity peptide (Leu) has a smaller contact surface than the significantly weaker His, Phe, or Tyr peptides (Table 3). However, more sophisticated computational approaches have shown promise in capturing CALP affinity determinants: the K* algorithm used energy minimization of conformational ensembles to estimate partition functions for the free and bound partners. Interestingly, 8 of the top 11 K*-designed sequences contained a P⁻¹ Arg, including kCAL01 (WQVTRV), which has the highest affinity observed to date for a hexameric sequence (Roberts et al., 2012).

Electrostatic potential contributes to modulator preferences at P⁻¹ and P⁻³

Our SubAna comparisons (Figure 4) also revealed prominent modulatory preferences at P⁻¹ (favoring Arg) and P⁻³ (disfavoring Asp), both of which involve charged residues. As a result, we investigated the electrostatic environment of the corresponding binding sites. Rendered at 10 k_BT/e, the CALP van der Waals surface potential reveals distinct positive and negative surfaces on opposite sides of the peptide-binding cleft (Figure 5C, Figure S7A). In the CALP:iCAL36_QDTRL structure, both the P⁻¹ Arg and the P⁻³ Asp residues point towards a ridge of negative potential, comprised of the β2, β3, α1, and β4 secondary structure elements, as well as connecting loops. This electrostatic environment could thus account for both of these modulator preferences (Figure 5D).

Specifically, the P⁻¹ Arg is hydrogen-bonded with Gln³²², facilitating a favorable interaction between the positively charged arginine guanidino group and an area of negative potential (Figure 5D). In contrast, the negatively charged P⁻³ Asp side-chain carboxylate is positioned directly adjacent to the negative potential surface, creating a Coulombic repulsion, even though the nearest charged residue is >5 Å away. In order to test the role of this side chain in determining modulator preferences, we expressed and purified a CALP-E317A mutant. Unfortunately, a crystal structure of the mutant (Table 2, Figure S7B–C and supplemental information available online) showed that the region of negative electrostatic potential was only modestly attenuated relative to wild-type (Figure 5C, insets). Correspondingly small (<7-fold) effects were observed on peptide binding affinities (Table 1).

However, a clearer picture is provided by analysis of the NHERF PDZ domains, which do not share the charge distribution of CALP. For example, in N1P1, the residue corresponding to CALP-Glu³¹⁷ is Leu⁴¹, a nonpolar residue. Consistently, all four NHERF1/2 PDZ domains make favorable hydrogen bonds with the P⁻³ Asp, which is in a positive environment (Figure S7D–G). Indeed, in contrast to the dramatic increase in binding affinity for CALP upon substitution of the P⁻³ Asp to Thr, CFTR_T^-3 affinity was ~2–6-fold weaker than that of CFTR for each of the NHERF1/2 PDZ domains (Table 1), confirming their modest preference for the P⁻³ Asp.

Using modulator preferences to identify new high affinity sequences for CAL

Lastly, we wished to evaluate the ability of modulator preferences to improve our classification of peptide affinities for the CALP domain. The human deca-C-terminome, the set of C-terminal decapeptides for all 84,888 protein isoforms listed in the UniProt database (The UniProt Consortium, 2012), contains a total of 57,632 unique sequences. 2,197 of these sequences conform to the CALP binding motif (Figure 6A). However, CALP only weakly binds some of the peptide sequences that satisfy its motif, including iCAL36_VQD^-3 and the CFTR C-terminus (Table 1), and others previously tested (Cushing et al., 2008; Roberts et al., 2012).

Figure 6. Impact of binding motif and modulator residues on PDZ binding specificity.

(A) Evaluation of the number of unique C-terminal decamers in the human proteome that satisfy the CALP binding motif (red residues in example sequences listed at right).

(B) A schematic pharmacophore model of the series of side chain-binding sites along the CAL PDZ cleft is shown, some of which are identified as binding motif residues by high-throughput analysis (Figure 1A). Positive (green) and negative (red) preferences are shown at non-motif sites, together with the frequencies of these residues in the pool of potential CALP binders. FP displacement titrations confirm the expected affinities for representative peptides that contain positive (left panel, green) and negative (right, red) modulator residues, suggesting that modulators can significantly restrict the number of binding partners by adjusting the overall affinity balance of the PDZ:target interaction. See also Figure S8H.

To detect candidate high affinity sequences, we applied the positive modulator preferences at P⁻¹ and P⁻⁵. Of the 2197 motif sequences, 141 and 36 contain an Arg and a Trp side chain, respectively (Figure 6B). Only four sequences contain both of our positive modulator preferences. Although they may not be physiological partners, all have CALP affinities better than or comparable to previously reported endogenous CALP ligands (Cushing et al., 2008): insulin-like growth factor-binding protein 3 receptor (TM219_HUMAN; RESHWSRTRL; K_i = 29 ± 8 μM), semaphorin 4G isoform 3 (SEM4G_HUMAN; PHSPWSFSRV; K_i = 46 ± 14 μM), histone H4 transcription factor (HINFP_HUMAN; CGCSWFATRV; K_i = 46 ± 24 μM), and proteasome subunit beta type-9 (B4DZW2_HUMAN; GFETWEGSRL; K_i = 120 ± 60 μM). Moreover, the proteasomal sequence contains a P⁻³ Gly residue (underlined), which has a negative influence in the positional arrays (Figure 4) and thus may account for its somewhat lower affinity. Likewise, endogenous targets containing the strongest negative modulator (Asp at P⁻³) are among the weakest CALP interactors (Figure 6B). Taken together, our data confirm the value of modulator preferences in assessing the affinity of candidate CALP-binding sequences.

Discussion

As illustrated by the weak CALP binding of C-terminal peptides from the known endogenous interactors CFTR and β-catenin (Cushing et al., 2008), there is abundant evidence that SLiM peptide affinity is not always the sole determinant of in vivo interactions, which also depend on molecular and cellular context (Luck et al., 2012). Nevertheless, PDZ:peptide binding remains a key element of specificity. Thus, deciphering the relationship between peptide sequence and PDZ affinity is a core requirement for understanding the networks of interactions formed by these ubiquitous protein-protein interaction domains. Over the past 15 years, studies have identified binding motifs and contacts from the C-terminus to the P⁻¹⁰ position (Appleton et al., 2006; Luck et al., 2012; e.g., Skelton et al., 2003; Smith and Kortemme, 2010; Thomas et al., 2008).

However, recent evidence suggests that PDZ sequence:affinity relationships can involve either more or fewer residues than indicated by motifs defined by alignments of high-affinity sequences. For example, while the Lin-7 homolog A (Lin7A) PDZ domain exhibits a highly selective motif with up to 6 determinants (Tonikian et al., 2008), fewer than four of these motif constraints are satisfied in any single Lin7A target sequence (The UniProt Consortium, 2012), and no annotated human C-terminal sequence matches the complete Lin7A PDZ binding motif. In contrast, although the binding motif for the CAL PDZ domain is highly degenerate (Figure 1B, Figure S1A), many motif-compatible sequences actually have negligible affinity for the domain (Roberts et al., 2012).

Overall, it appears that PDZ domains achieve sufficient binding affinity by harnessing strong free energy contributions at a few anchor positions, supplemented by additional stereochemical interactions that may involve either additional motif constraints or non-motif ‘modulator’ residues. Structural comparisons suggest that such modulator preferences reflect relatively loose pharmacophore sites, each of which engages the side chain at a particular position (Figure 3C). In fact, although only a limited number of other PDZ domains have been co-crystallized with more than one peptide ligand (e.g., Babault et al., 2011; Grembecka et al., 2006; Kang et al., 2003; Nomme et al., 2011), as for CALP, superpositions in these cases also suggest that each domain imposes a characteristic binding conformation on its ligands, and can thus define a set of domain-specific sequence preferences (Figure 3B, Figure S8).

Whether these preferences are readily captured in alignment motifs depends on a number of factors, including the stringency of the preference itself, as well as the nature of the experiment, limitations of the technique (e.g., number of sequences tested), and the baseline affinity of the domain for the peptide backbone. In many cases, preferences are subtle and lead only to marginal enrichment relative to random sequences. For example, our P⁻⁵ residue analysis suggests that all 7 substituted amino acids (Trp/Leu/Tyr/Phe/Val/Ala/His) could have been found in peptides that were included in the binding motif analysis, as the substituted iCAL peptides all bind CALP with affinity greater than that of CFTR, a top 100 sequence hit. This variability may reflect the attenuated specificity of surface-exposed stereochemical interactions (e.g., Serrano et al., 1990). Whatever the source of accommodation, alignments are unlikely to achieve sufficient enrichment to create an obvious motif constraint. Negative contributions are particularly difficult to detect robustly by alignment, since the expected observation frequency of a “neutral” residue is its already low rate of natural occurrence, associated with a correspondingly high level of random variability.

Our data (Figure 4) suggest that multi-sequence substitution analysis represents an effective extension of mutagenetic approaches (e.g., Wiedemann et al., 2004) to highlight preferences that are insufficiently robust or sequence-independent to establish clear motif signatures. SubAna arrays can also readily detect negative contributions (e.g., Asp at P⁻³) that have important implications for the biological specificity of PDZ interactions. The P⁻³ Asp side chain at the C-terminus of CFTR significantly limits its affinity for CAL, which targets CFTR to the lysosome, but not for NHERF1 and NHERF2, which functionally stabilize CFTR at the apical membrane (Table 1). Thus, the CFTR sequence at this position specifically favors interactions with PDZ proteins that promote endocytic recycling over degradation (Cheng et al., 2002).

This example underscores the potential for modulator residues to differentially affect interactions with PDZ domains that share overlapping motifs. Recent work by a number of groups suggests that PDZ interactions are a finely tuned product of evolution operating both on the PDZ domains (Ernst et al., 2009; Kaneko et al., 2010; McLaughlin et al., 2012) and on the target C-terminal sequences themselves (Kim et al., 2012). If degenerate two- or three-residue motifs were sufficient to determine target affinity for PDZ domains such as CAL or NHERF, such evolution would be limited by a very small set of combinatorial possibilities, as would the engineering of highly selective sequences. While individual modulator residues may contribute only modestly to target affinity, collectively they have the potential to significantly differentiate PDZ specificity profiles. As a result, characterizing these previously cryptic binding determinants may be critical both for a fuller understanding of PDZ domain interaction networks and for the development of specific peptide inhibitors of individual nodes.

Experimental Procedures

Protein Expression and Purification

The CALP E317A mutant was generated using the Quikchange Lightning (Agilent Technologies) protocol. The WT and mutant CALP (Amacher et al., 2011) and NHERF PDZ domains (Cushing et al., 2008) were expressed and purified as previously described. For fluorescence polarization (FP) experiments, proteins were dialyzed overnight into storage buffer (150 mM NaCl; 0.02% [w/v] sodium azide, 25 mM sodium phosphate pH 8.0, 0.2 mM tris[2-carboxyethyl]phosphine [TCEP], 5% [v/v] glycerol). For CALP crystallization, the protein was stored in GF2 buffer (25 mM Tris, 150 mM sodium chloride, 0.02% [w/v] sodium azide, 0.1 mM TCEP, 0.1 mM ATP, 5% [v/v] glycerol) (Amacher et al., 2011), prior to dialysis into crystallization buffer. Peptides for co-crystallization and FP experiments were synthesized and purified by the Tufts Core Peptide Facility.

Crystallization and structure determination

Detailed methods for crystallization and determination of each co-crystal structure are described in supplemental information (Table S1). Complexes were crystallized, data collected, and structures determined as previously described (Amacher et al., 2011; Amacher et al., 2013). Coordinates and structure factors have been deposited in the Protein Data Bank, with accession codes 4JOE, 4JOF, 4JOG, 4JOH, 4JOJ, 4JOK, 4JOP, 4JOR, 4K6Y, 4K72, 4K75, 4K76, and 4K78 (Table S1).

Structure analysis

Electrostatic potential surface calculations and structure alignments were performed in PYMOL (DeLano, 2008). The contact surface of each peptide residue was visualized using CHIMERA (Pettersen et al., 2004). SASA, including contact, was calculated using AREAIMOL (Lee and Richards, 1971), part of the CCP4 suite of programs (Collaborative Computational Project, 1994). Contact surface area for each peptide residue atom was calculated as [SASA]_COMPLEX – [SASA]_ALONE. The Interaction Index (I_PDZ) is defined as the fractional change in solvent accessible surface area upon ligand binding, using the equation:

$${\mathrm{I}}_{\mathit{PDZ}}=1-\frac{{\mathit{SASA}}_{\mathit{complex}}}{{\mathit{SASA}}_{\mathit{free}}}$$

Fluorescence Polarization (FP)

FP assays were performed as previously described (Cushing et al., 2008; Vouilleme et al., 2010). Triplicate experiments were performed, using separate protein and reporter stock solutions in independent peptide titrations. The K_i values were determined using a SOLVER-based least-squares fitting algorithm in EXCEL, as described previously (Amacher et al., 2013; Cushing et al., 2008; Vouilleme et al., 2010). The following reporter peptides were used: F*-iCAL36 (F*-ANSRWPTSII) for CALP and E317A CALP; F*-CFTR6 (F*-VQDTRL) for N1P1 and N1P2; and F*-CFTR10 (F*-TEEEVQDTRL) for N2P1 and N2P2. F* corresponds to a fluorescein group linked to the peptide N-terminus via an amino-hexanoic acid linker. The concentrations used for each PDZ domain were ~1.5 × K_d, where the K_d values of each PDZ domain for their respective reporter peptides were: CALP (K_d=0.97 μM), E317A CALP (0.61 μM), N1P1 (0.37 μM), N1P2 (1.08 μM), N2P1 (0.32 μM), and N2P2 (0.23 μM).

Peptide Arrays

The full set of 6223HumLib sequences is described elsewhere (Roberts et al., 2012; supplemental material). Position-specific SubAna arrays were generated by juxtaposing data for a given peptide position from a series of SubAna arrays based on distinct peptide sequences. Individual SubAna arrays were synthesized and binding analyzed as previously described (Boisguerin et al., 2007; Vouilleme et al., 2010).

Proteome analysis of sequences

The Python scripting language (van Rossum, 2009) was used to map all occurrences of the CALP binding motif and the frequency distribution of individual modulator preferences in the set of all decameric C-terminal peptides present in the FASTA file provided by UniProt Knowledgebase release 2012_10 (The UniProt Consortium, 2012).