MotifAnalyzer‐PDZ: A computational program to investigate the evolution of PDZ‐binding target specificity

Abstract

Recognition of short linear motifs (SLiMs) or peptides by proteins is an important component of many cellular processes. However, due to limited and degenerate binding motifs, prediction of cellular targets is challenging. In addition, many of these interactions are transient and of relatively low affinity. Here, we focus on one of the largest families of SLiM‐binding domains in the human proteome, the PDZ domain. These domains bind the extreme C‐terminus of target proteins, and are involved in many signaling and trafficking pathways. To predict endogenous targets of PDZ domains, we developed MotifAnalyzer‐PDZ, a program that filters and compares all motif‐satisfying sequences in any publicly available proteome. This approach enables us to determine possible PDZ binding targets in humans and other organisms. Using this program, we predicted and biochemically tested novel human PDZ targets by looking for strong sequence conservation in evolution. We also identified three C‐terminal sequences in choanoflagellates that bind a choanoflagellate PDZ domain, the Monsiga brevicollis SHANK1 PDZ domain (mbSHANK1), with endogenously‐relevant affinities, despite a lack of conservation with the targets of a homologous human PDZ domain, SHANK1. All three are predicted to be signaling proteins, with strong sequence homology to cytosolic and receptor tyrosine kinases. Finally, we analyzed and compared the positional amino acid enrichments in PDZ motif‐satisfying sequences from over a dozen organisms. Overall, MotifAnalyzer‐PDZ is a versatile program to investigate potential PDZ interactions. This proof‐of‐concept work is poised to enable similar types of analyses for other SLiM‐binding domains (e.g., MotifAnalyzer‐Kinase). MotifAnalyzer‐PDZ is available at http://motifAnalyzerPDZ.cs.wwu.edu.

1. INTRODUCTION

Protein–protein interactions enable cellular regulation and function. Many protein–protein interactions involve recognition of a small number of residues, called peptides or short linear motifs (SLiMs). These SLiM‐mediated interactions are often transient and less specific than those which involve large binding interfaces.1 SLiM binding motifs are dependent on a minimal number of residues, and can consist of as few as two amino acids.2, 3 Thus, predicting relevant endogenous interactions based on the motif alone is challenging.

The ability to target such interactions, with small molecules, peptidomimetics, or biologics can have profound effects in a number of human diseases.4 Examples of biomedically important SLiM‐dependent interaction domains are: kinase (e.g., cancer), phosphatase (e.g., diabetes, cancer, autism), PDZ (e.g., cystic fibrosis, cancer, pain/addiction), SH2 (e.g., immunodeficiency, X‐linked agammaglobulinemia), SH3 (e.g., kidney disease), PTB (e.g., Alzheimer's disease, diabetes, and coronary heart disease), and WW (e.g., X‐linked intellectual disabilities, cancer) domains.5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16 Therefore, understanding the basic principles of SLiM‐recognition by protein domains promises to provide powerful insight into drug discovery efforts.17, 18, 19, 20, 21

PDZ (PSD‐95/Dlg1/ZO‐1) domains bind the C‐termini of target proteins, and are important regulators of signaling and trafficking pathways in many cell types.2, 3 There are approximately 270 PDZ domains encoded in the human proteome, with relevant proteins consisting of 1–13 PDZ domains. All of the observed PDZ domains to date share a common structural fold and characteristic interaction features (Figure 1a,b).2, 22 Furthermore, PDZ domains are an ancient binding module, and are conserved in bacteria, plants, fungi, and animals, for example, those found in degradation of periplasmic or high‐temperature requirement (Deg/HtrA) proteases (Figure 1c).23, 24, 25, 26, 27 The structures of HtrA PDZ domains are specifically highlighted here because biochemical studies of Deg/HtrA PDZ domains have demonstrated that target selectivity is conserved.24 This is not unique, phage display experiments using PDZ domains from Homo sapiens and Caenorhabditis elegans also revealed that specificities evolved early and are maintained in distantly related organisms.28

Figure 1.

The PDZ domain fold is conserved in evolution. Common structural fold of a PDZ domain, represented by (a) the first PDZ domain crystal structure solved (PDB ID: 1BE9), where peptide residues are labeled in black and the characteristic “GLGF” motif is highlighted (black arrow), (b) a 2D schematic of a PDZ domain, and (c) PDZ domains from various organisms, shown in cartoon (PDZ) and stick (peptide) representation. HtrA homologues (green) from human (PDB ID: 2JOA), bacteria (1SOZ), and plants (4FLN) are shown, as well as a homology model of the Nsa111p PDZ‐1 domain (cyan), the HtrA homologue from yeast (template: 5M3N)

Class I PDZ domains are defined as recognizing the motif S/T‐X‐ϕ_COOH, where ϕ = any hydrophobic residue and X = any amino acid. The most C‐terminal residue is termed P⁰, and adjacent residues are P⁻¹, P⁻², and so on.2, 22 Based on the aforementioned motif, over 2000 unique proteins are predicted to bind the 130 Class I PDZ domains.29 However, work from us and others revealed that recognition of a PDZ domain for each target is dependent on more than the two residues that characterize the motif.28, 29 Characterization of the binding of CFTR Associated Ligand (CAL) PDZ to peptides suggested that the last six C‐terminal residues provide the majority of the free energy of binding, and that these “modulator,” or non‐motif preferences, likely exist broadly across the PDZ family.29

Predicting endogenous PDZ or other SLiM‐binding domain targets is challenging. There are a number of programs and publicly available servers that predict SLiM‐based interactions, including those for PDZ domains, and many are focused on identifying relatively high affinity interactions.30, 31, 32, 33, 34, 35, 36 A number of factors make de novo protein predictions challenging, including gene expression, cellular localization, and critically, the relatively low affinity of many biologically significant interactions. For example, the binding of CFTR to CAL PDZ, an important endogenous interaction, is >400 μM in in vitro binding assays, which is outside the range that would be considered in an affinity‐based prediction or peptide design program.15, 29, 37

We created a pairwise protein comparison algorithm, MotifAnalyzer‐PDZ, that identifies all motif‐satisfying sequences in any proteome and then compares and filters those sequences based on motif and non‐motif residues. The two main benefits of our program are that it allows us to easily (a) track sequence conservation across the evolutionary tree, even in non‐annotated proteins, revealing possible cellular PDZ targets, and (b) assess if certain amino acids are enriched at modulator positions in motif‐satisfying sequences. Although not explicitly tested here, because our program allows a user to easily define specific residues at as many positions (e.g., P⁻¹, P⁻³, etc.) as desired, MotifAnalyzer‐PDZ could easily filter motif‐satisfying sequences into those that also contain experimentally‐determined positive or negative preferences at non‐motif positions for a given protein. Taken together, MotifAnalyzer‐PDZ is a novel program that highlights potential PDZ targets in a way that is unbiased towards binding affinities or protein identity, allowing prediction of previously unidentified interactions.

2. RESULTS AND DISCUSSION

2.1. A program to extract and analyze C‐terminal sequences from any proteome

We developed a program, MotifAnalyzer‐PDZ, to filter and identify C‐terminal sequences in any publicly available proteome from UniProt (Figure 2). MotifAnalyzer‐PDZ allows the command line inputs of: organism names to search on UniProt (if proteome data not supplied directly), number of residues to consider, residue location(s) of interest, and the required amino acids at certain positions (or the motif). The user also specifies which outputs are desired, which may include calculating enrichment values for non‐motif positions in the sequence and/or running MotifMatcher, a subprogram that looks for sequence matches in reference and target proteomes, including the number of user‐defined allowable substitutions (Figure 1).

Figure 2.

Pipeline of the MotifAnalyzer‐PDZ program. Following the initial filtering of motif sequences that occur in the proteome of a given organism, MotifAnalyzer‐PDZ can calculate positional enrichment values for amino acids in non‐motif positions. Optionally invoking MotifMatcher searches for conservation in sequences among various organisms

Briefly, the main Python program file, MotifAnalyzer‐PDZ.py, downloads the specified set of proteomes from UniProt and retrieves the last k residues of each protein sequence. These k‐mers are then filtered to remove redundancy and according to the user‐specified motif, that is, to include the amino acids that are required at up to k positions. The positional enrichment values for the matching k‐mers are calculated, from which heatmaps are generated and output to the user, accompanied with raw data comma separated (CSV) files. Enrichment values (E) for each amino acid are calculated as: E = M/N, where M = the frequency of an amino acid at the specified position in motif‐satisfying sequences and N = the frequency of occurrences of the same amino acid at the specified position in all unique C‐terminal sequences of the proteome. Also available is a graphical user interface (GUI) that launches with MotifMatcher. For detailed information about protein scoring in MotifMatcher and input parameters for MotifAnalyzer‐PDZ, as well as heat map generation and statistical analyses, please see Section 4 and Supporting Information Materials and Methods. MotifAnalyzer‐PDZ is available at http://motifAnalyzerPDZ.cs.wwu.edu.

2.2. Using MotifAnalyzer‐PDZ to predict targets of SHANK homologues

An important feature of MotifAnalyzer‐PDZ is the ability to predict endogenous targets of PDZ domains in any organism. To this end, we used the unicellular eukaryote, Monosiga brevicollis, to test our ability to identify cellular targets of a choanoflagellate PDZ domain that is homologous to the human SH3 and multiple ankyrin repeat domains protein 1 (SHANK1). Choanoflagellates are the closest non‐metazoan ancestor to humans, and M. brevicollis was the first sequenced choanoflagellate genome.38 The SHANK proteins are located in the postsynaptic density of glutamatergic synapses in the central nervous system, and recent studies suggest genetic and functional correlation amongst mutations in SHANK proteins, autism, and cancer.39, 40 Thus, a greater understanding of SHANK1 PDZ evolution and binding may provide important insight into drug targeting in human disease.

Choanoflagellates do not have a central nervous system, but there is a SHANK1 homologue, although its function is unknown. Human SHANK1 is a 2161 residue protein that consists of six ankyrin repeat domains, as well as an SH3, PDZ, and SAM domain. The M. brevicollis homologue, mbSHANK1 (NCBI reference sequence XP_001748613), shares 40% sequence identity to human SHANK1 over 23% of the sequence, including the ankyrin repeats (40% sequence identity), PDZ domain (34% sequence identity), and SH3 domain (33% sequence identity) (Figure 3).41

Figure 3.

SHANK1 is conserved in choanoflagellates. (a) The scaled domain architectures of SHANK homologs from humans (SHANK1) and choanoflagellates (mbSHANK1). (b) Alignment of the SHANK1 and mbSHANK1 PDZ domains reveals strong sequence similarity. Secondary structure elements in SHANK1 PDZ (PDB ID: 3QJN) are highlighted by arrows (β‐strands) or curved lines (α‐helices)

Critically, although SHANK1 is conserved in M. brevicollis, a number of its known human targets are not, suggesting divergence in the function of this protein and signaling pathways between humans and choanoflagellates. Human targets of SHANK1 PDZ include: SAPAP3/GKAP1a/DLG3 (C‐terminal sequence: EAQTRL), β‐PIX (WDETNL), Neuroligin (HSTTRV), mGluR1 (QSSSTL), mGluR5 (QSSSSL), and the L‐type Ca²⁺ channel CAC1C (VYVSSL).42, 43, 44, 45, 46, 47 Another target, α‐latrotoxin (QLVTSL), does contain a homologue in M. brevicollis, the GPS‐containing G‐protein‐coupled receptor (GPCR, GenBank: ACR39371), which is 28% identical over 45% of the sequence. This protein contains C‐terminal sequence YCVTLL, with three substitutions to the human sequence (conserved residues are underlined), suggesting the α‐latrotoxin and ACR39371 proteins are likely related and mbSHANK1 PDZ may bind this protein. Overall, due to this lack of conservation in C‐terminal sequences, it is unclear what the cellular targets are of mbSHANK1 and its overall function in choanoflagellate signaling.

We generated a homology model of M. brevicollis mbSHANK1 in order to investigate whether or not peptide‐binding preferences are likely to be conserved between these two PDZ domains.48, 49 Our model aligned to a structure of human SHANK1:β‐PIX (PDB ID: 3QJN), which was used as the template for homology modeling, with an overall RMSD value of 0.123 Å (231 main chain atoms, Figure S1a). The mbSHANK1 model is consistent with proper protein folding, for example, hydrophobic residues are buried in the core of the protein and polar residues are entirely located on the surface of the protein (Figure S1b,c). Our model confirms that many of the binding site residues are conserved (Figure S2), with the exception of those interacting directly with the P⁻³ peptide position, as discussed below. Thus, we hypothesized that the PDZ domains of SHANK1 and mbSHANK1 likely bind similar sequences, even though the exact cellular targets are different.

Using MotifAnalyzer‐PDZ, we identified a number of M. brevicollis proteins that contain C‐terminal sequences similar to those listed above (Table 1). Of particular interest, and those that we chose to test in vitro were UniProt IDs A9UP44 (QSESRL, which shares four residues with the C‐terminus of mGluR1), and A9UXE1 (QDETAL) and A9V7Z4 (EDETAL), which both share four residues with the C‐terminus of β‐PIX. We chose these three because the human proteins they share the highest degree of sequence identity with are the tyrosine kinases Csk (30% identical over the entire kinase domain), the Met receptor (44% identical to the C‐lobe of the kinase domain), and Abl (35% identical over the entire kinase domain), respectively. This is intriguing because human PDZ domains are known to interact with tyrosine kinases, including the IGF‐1 receptor (targeted by the IIP‐1 PDZ domain), c‐Src (targeted by AF‐6 PDZ domain), and ephrin receptors (targeted by AF‐6 PDZ domain).50, 51, 52 In addition, PTN13 is a human tyrosine phosphatase that contains five PDZ domains and has a homologous protein in M. brevicollis (XP_001750852, 34% identical over 74% of the 2485 amino acid protein). Taken together, this suggests that PDZ domains may also mediate trafficking of tyrosine kinases in choanoflagellates.

Table 1.

Choanoflagellate sequences with strong similarity to human SHANK1 PDZ targets

Human sequence	M. brevicollis sequence	UniProt ID	Number of matching residues
WDETNL	EDETAL	A9V7Z4_MONBE	4
	QDETAL	A9UXE1_MONBE	4
VYVSSL	TTVSSL	A9UX56_MONBE	4
QLVTSL	QLLTSL	A9VC27_MONBE	5
EAQTRL	EAVTAL	A9UV83_MONBE	4
QSSSTL	QSESRL	A9UP44_MONBE	4
	PSSSTA	A9UYJ6_MONBE	4
	CSSSTF	A9V128_MONBE	4
	LSSSTA	A9V8P9_MONBE	4

We determined the binding affinities of recombinant human SHANK1 and mbSHANK1 PDZs with our choanoflagellate, human β‐PIX, and human mGluR1 sequences, using fluorescence polarization. Briefly, PDZ domains were expressed in E. coli cells and purified (see Section 4).53, 54 Fluorescence polarization experiments were performed as previously described.19, 29, 54, 55 Binding affinities are reported in Table 2 and the average binding curves are shown in Figure 4. Overall, all of the M. brevicollis sequences bind mbSHANK1 with biologically relevant affinities (K _i values are <100 μM).55, 56 H. sapiens SHANK1 binds these 3 sequences with slightly better affinities (K _i values are <50 μM), but the preferences are conserved (e.g., order of K _i values are A9V7Z4 < A9UP44 < A9UXE1). Interestingly, mbSHANK1 binds the human β‐PIX sequence 1.5x better than SHANK1, but the mGluR1 sequence >10× worse, with a K _i value >1 mM.

Table 2.

Binding affinities of choanoflagellate sequences for H. sapiens SHANK1 and M. brevicollis mbSHANK1 PDZ domains

		K i (μM)
Peptide	Sequence	SHANK1	mbSHANK1
β‐PIX	NDPAWDETNL	20 ± 4.0	13 ± 4.7
mGluR1	RDYKQSSSTL	90 ± 11	1,020 ± 260
A9UP44	EDTNQSESRL	34 ± 10	44 ± 23
A9UXE1	ANPIQDETAL	30 ± 5.0	72 ± 14
A9V7Z4	GTSLEDETAL	9.8 ± 3.1	39 ± 19

Figure 4.

Binding curves of SHANK1 PDZ and mbSHANK1 PDZ with human and choanoflagellate peptides. Average fluorescence polarization displacement isotherms are shown for SHANK1 PDZ (a) and mbSHANK1 PDZ (b). Titration curves are shown for the following peptides: β‐PIX (circles) and mGluR1 (squares), or choanoflagellate proteins A9UP44 (diamonds), A9UXE1 (triangles), and A9V7Z4 (upside‐down triangles). Error bars indicate standard deviation from the mean for triplicate experiments

Structural analysis of SHANK1 PDZ and our homology model of mbSHANK1 PDZ suggest that the increased affinity of mbSHANK1 for β‐PIX, but dramatically decreased affinity for mGluR1 is perhaps due to amino acid substitutions interacting with the P⁻³ peptide residue (Figure 5a). The P⁻³ residue in β‐PIX is an aspartic acid, which will interact favorably with Arg679 in SHANK1, but be electrostatically repulsed by the nearby Glu703 residue. However, in mbSHANK1, this glutamic acid is replaced with Arg485, a favorable interaction for the P⁻³ Asp. Although SHANK1 Arg679 is not conserved in mbSHANK1 (Ser468), there is an additional arginine residue, Arg466 (Val677 in human SHANK1) that is near enough to interact with the P⁻³ Asp (Figure 5a).

Figure 5.

Structural insights into the binding preferences for the SHANK1 and mbSHANK1 PDZ domains. The differing nearby residues of SHANK1 and a homology model of mbSHANK1 for the P⁻³ peptide position are consistent with measured binding affinities. In all, PDZ domains are shown as gray cartoons, with specific residue side chains of SHANK1 PDZ (green carbons, colored by heteroatom) and mbSHANK1 PDZ (cyan carbons, colored by heteroatom) highlighted as sticks. Bound peptides are shown in stick representation. For all, oxygen is red and nitrogen is blue. (a) The P⁻³ Glu in β‐PIX (yellow carbons, colored by heteroatom) can favorably interact with R679 in SHANK1 PDZ (left panel), but will be electrostatically repulsed by E703 (PDB ID: 3QJN). In contrast, this residue is likely to interact favorably with both R466 and R485 in mbSHANK1 PDZ (right panel, homology model using 3QJN as a template). (b) The P⁻³ Ser in mGluR1 and mGluR5 (blue carbons, colored by heteroatom) may form favorable electrostatic interactions with R679 and E703 in SHANK1 PDZ (light panel). The peptide here is based on an alignment of the Tamalin PDZ domain bound to mGluR5 (PDB ID: 2EGN), RMSD = 0.626 Å over 267 main chain atoms. We hypothesize that R485 may sterically hinder favorable interactions between mGluR1 and mbSHANK1 PDZ (right panel), suggesting why the affinity of mbSHANK1 PDZ for mGluR1 is 10x worse than for SHANK1 PDZ

In order to investigate binding of mGluR1 to these PDZ domains, we aligned a structure of the Tamalin PDZ bound to mGluR5 (sequence: SSSSL, PDB ID: 2EGN) to the human SHANK1:β‐PIX structure (RMSD = 0.626 Å over 267 main chain atoms) (Figure S1d). In this case, whereas human SHANK1 Glu703 can likely form hydrogen bonds with the P⁻³ Ser in mGluR1/5, as well as potentially Arg679, the two arginine residues in mbSHANK1 (Arg466 and Arg485) appear to be improperly positioned to interact with the P⁻³ Ser (Figure 5b). While we cannot make conclusions without experimental data, these structures illuminate subtle binding cleft differences between human SHANK1 and mbSHANK1. Specifically, the only amino acid substitutions that directly interact with the peptide are located in residues near the P⁻³ position. Taken together, we successfully used MotifAnalyzer‐PDZ to predict a number of choanoflagellate sequences that bind mbSHANK1 with biologically‐relevant affinities and which are located on signaling proteins.

2.3. Evolution of PDZ motif‐satisfying sequences

We next hypothesized that analyzing conserved PDZ motif‐satisfying sequences in various organisms may lead to the discovery of previously unrecognized human PDZ targets. We used MotifAnalyzer‐PDZ to look for sequences that are conserved in non‐motif residues, but are not currently known to interact with human PDZ domains. Two examples are shown in Figure 6: dynactin subunit 5 (DCTN5, sequence: LPLTQV) and homeobox C12 (HOXC12, QALSFF). Specifically, the C‐terminal sequence of HOXC12 is 100% conserved in multiple organisms, with a single S‐to‐T substitution at the P⁻² motif position in the jawless vertebrate Callorhinchus milii (the Australian ghostshark). For reference, H. sapiens and C. milii diverged >450 million years ago, providing evidence that conservation of this 6 residue sequence is likely not due to chance.57

Figure 6.

Sequence conservation of predicted human PDZ targets. The evolutionary relationships of nine organisms are highlighted. The sequences of the motif‐satisfying C‐termini of dynactin subunit 5 (DCTN5, A) and homeobox C12 (B) homologs are strongly conserved. Human sequences, as well as those that are 100% conserved are colored blue; orange and gold sequences each have one substitution in a motif position

Previous work using phage display revealed that the NHERF2 PDZ2 (N2P2) and SHANK3 PDZ domains contain strong preferences for a P⁰ Phe, as well as an aromatic residue at P⁻¹ [Trp (SHANK3‐1) or Trp and Tyr (N2P2)], such as those in the HOXC12 sequence listed above.28 The SHANK1 PDZ domain is 84% identical to SHANK3 PDZ. Therefore, we tested a HOXC12 peptide sequence with purified NHERF1 PDZ1 (N1P1), N2P2, and SHANK1 PDZ domains using fluorescence polarization (Figure S3, Table 3). Our results reveal that N2P2 binds the HOXC12 peptide with relatively high affinity (K _i = 7.8 μM) and N1P1 and SHANK1 PDZ domains with moderate affinity (K _i = 140 and 170 μM, respectively) (Figure S3, Table 3). We also included a negative control, mbSHANK1 PDZ from the choanoflagellate Monosiga brevicollis, which does not bind HOXC12 with detectable affinity (K _i > 1,000 μM) (Figure S3, Table 3).

Table 3.

Binding affinities of the Homeobox C12 peptide sequence for PDZ domains

		K i (μM)
Peptide	Sequence	N1P1	N2P2	SHANK1	mbSHANK1
HOXC12	LLREQALSFF	140 ± 50	7.8 ± 0.3	170 ± 10	>1,000
SSRT‐CFTR	SSRTVQDTRL	1.5 ± 0.4	1.3 ± 0.2	ND	ND
β‐PIX	NDPAWDETNL	ND	ND	38 ± 9.0	12 ± 1.6

The homeobox proteins are a large family of transcription factors that play fundamental roles in developing organisms.58, 59 Of the more than 120 homeobox proteins, there are over a dozen classifications.59 Of the 15 proteins in the HOX class of homeodomain proteins, three contain Class I PDZ domain‐targeting sequences: HOXC12, as well as HOXA3 (sequence: PKLTHL) and HOXA4 (PVPSSI). Furthermore, there are known PDZ interactions that occur in the nucleus.60 Our results suggest that, dependent on localization, HOXC12 may very well interact with PDZ domains in the cell, and it would be interesting to test DCTN5 and HOXC12 PDZ interactions in vivo.

2.4. The enrichment of certain amino acids in PDZ motif‐satisfying sequences by organism

Our previous work shows that there are positive and negative preferences that modulate PDZ affinity at non‐motif residues29; thus, we were curious if such preferences could be predicted on a proteome‐wide scale for a given class of PDZ domains. We used MotifAnalyzer‐PDZ to ask if there are over‐ or under‐enriched amino acids at non‐motif positions in PDZ motif‐satisfying sequences. We chose to focus on a limited Class I PDZ binding motif, with P⁰ = I/L/V/F and P⁻² = S/T, in order to reduce the number of sequences and thus, potential noise, in our analyses. Enrichment values were calculated as described earlier. Overall, the calculated enrichment values suggest a stronger degree of conservation than we hypothesized (Figure 7a). Specifically, over‐enrichment of a P⁻¹ His or Arg and P⁻³ Glu (through fungi) appear to be well‐conserved, as well as under‐enrichment of a P⁻³ Trp residue (Figure 7a). These heat maps provide a template for peptide design. For example, a peptide that contains over‐enriched residues for vertebrate organisms would be: TKETHL, KKETHL, or RKETHL. Other substitutions include a P⁻¹ Ser, P⁻³ Cys, or P⁻⁴ Arg or Gln. Examples of peptides with under‐enriched residues include variations of, P⁻¹: G/M/P, P⁻³: A/L/P/W, P⁻⁴: A/G, and P⁻⁵: A/C/F/G, for example, FAWTML or GALTML, among many others. Additional work needs to be done to determine whether or not enrichment of specific amino acids at non‐motif residues is biologically relevant and/or how the amino acids in PDZ domains that interact with these positions evolved.

Figure 7.

Positional enrichment values reveal organismal trends in non‐motif residues. (a) Enrichment values were calculated for the sequences from 14 organisms that follow the Class I PDZ motif, P⁰ = I/L/V/F and P⁻² = S/T. These values are shown as heat maps for the non‐motif positions, P⁻¹, P⁻³, P⁻⁴, and P⁻⁵. The key is to the right and amino acids are grouped by chemical properties and labeled by their one‐letter abbreviations. Despite some obvious trends, for example, P⁻¹ His or P⁻³ Glu or Trp, there is quite a bit of variability, particularly in the P⁻⁴ and P⁻⁵ peptide positions. (b) Normalized enrichment value SequenceLogos for motif‐satisfying sequences in human (left) and choanoflagellate (right) proteomes show variability in the peptide sequence. One was subtracted from each enrichment value and then normalized by non‐motif position (values of 0.95 and 1.05 were disregarded). If the value is positive, the amino acid is considered over‐enriched. If the value is negative, the amino acid is considered under‐enriched. The amino acids are colored by chemical property

We chose to create a modified WebLogo, which we will refer to here as a SequenceLogo, by graphing the enrichment values of all four non‐motif positions for a specified organism on the same axis (Figure 7b). This type of visualization allows us to directly compare amino acid enrichment of individual organisms. For example, our SequenceLogos reveal that human motif‐satisfying sequences are quite enriched in positive residues at the P⁻¹ (H), P⁻⁴ (K/R), and P⁻⁵ (H/K/R) positions, and negative residues at the P⁻³ (D/E) position. In contrast, motif‐satisfying choanoflagellate sequences are more enriched in polar residues (e.g., C, S, Q, N) at these sites. In addition, motif‐satisfying choanoflagellate sequences reveal enrichment of hydrophobic residues (F/I/V) at P⁻¹, which human sequences do not (Figure 7b). Future work will investigate if these differences are broadly reflected in peptide preferences for Class I PDZ domains, or if PDZ preference is uniquely an individual protein characteristic.

2.5. Evolution of PDZ‐binding sequences in known human targets

We next wanted to expand on our SHANK1/mbSHANK1 case study to identify PDZ target sequences that are broadly conserved in multiple organisms. We chose 12 Class I sequences that are known targets of human PDZ domains (Table 4).41, 44, 55, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79 Many of the PDZ domain‐containing proteins listed in Table 4 are homologous to choanoflagellate proteins (e.g., Dlg1, PTPN13, MUPP1, and NHERF1/MAGI1 with XP_001743865.1, XP_001750852.1, XP_001745891.1, and XP_001742647.1, respectively). MotifAnalyzer‐PDZ can be used to investigate the evolution of the these 12 hexameric PDZ target sequences. If we search for 0 or 1 mutation, we do not see many sequences in organisms that diverged from H. sapiens longer than ~450 million years ago (e.g., C. milii) (Tables 5 and S1). However, if we allow for sequences with up to two mutations, we see examples for almost all of our test sequences in 11 of the 14 organisms we searched (Tables 5 and S1), with the exceptions of bacteria (Escherichia coli), slime mold (Dictyostelium discoideum), and yeast (Saccharomyces cerevisiae). Sequence results are shown for those that are related to the human β‐catenin and mGluR1 sequences (Figure 8).

Table 4.

Test sequences and their PDZ binding partners

Protein	Abbreviation	C‐terminal sequence	PDZ binders
Adenomatous polyposis coli	APC	YLVTSV	Dlg1, PTP‐BL61, 62
β‐Catenin		WFDTDL	TIP1, MAGI‐1/2/3, Erbin, NHERF‐1, LIN7, DVL1/3, GRIP1, TIAM2, ZO‐3, SHANK3, SCRIB, PSD95, GRASP55, MAST1, MUPP1, PTP0BL, PARD3, PDZK7, Neurabin‐1, PDZK3, SYNIP, PDZ11, SAP9763
Cystic fibrosis transmembrane conductance regulator	CFTR	VQDTRL	NHERF‐1/2/3, CAL, MAST20555, 64, 65
Glutamate receptor 1	GluR1	LGATGL	Dlg1, PICK1, Syntenin66
Mitogen‐activated protein kinase 12	MAPK12 or ERK3 or SAPK3	SKETPL	SNTA1, LIN7C, SCRIB41
Dual specificity mitogen‐activated protein kinase kinase 2	MEK2	PTRTAV	Dlg1, RGS1267, 68
Metabotropic glutamate receptor 1	mGluR1	QSSSTL	CAL, Tamalin, SHANK1/369, 70, 71
Neuroepithelial cell‐transforming 1	NET1	RKETLV	Dlg1, MAGI‐1, SCRIB72, 73
Neuroligin		HSTTRV	PSD95, DLG‐2/3, MAGI‐1/2/3, SHANK1/3, PICK1, CAL, PDLIM5, SEMCAP3, PDZ‐RGS344
Phosphatidylinositol 3,4,5‐trisphosphate 3‐phosphatase	PTEN	TQITKV	Dlg1, MAGI‐1/2/3, MAST20561, 74, 75
Rho guanine nucleotide exchange factor 26	SGEF or ARHGEF26	GLETNV	Dlg176, 77
Lymphokine‐activated killer T‐cell‐originated protein kinase	TOPK or PBK	ALETDV	Dlg1, SCRIB, Erbin78, 79

Table 5.

Conservation of test sequences in evolution

A. Number of C‐terminal sequences with 0 or 1 mutation in non‐motif residuesa
Organism	YLVTSV	WFDTDL	VQDTRL	SKETPL	PTRTAV	RKETLV	HSTTRV	TQITKV	LGATGL	GLETNV	ALETDV	QSSSTL
E. coli	0	0	0	0	0	0	0	0	0	0	0	0
D. discoideum	0	0	0	0	0	0	0	0	0	0	0	0
S. cerevisiae	0	0	0	0	0	0	0	0	0	0	0	0
A. thaliana	0	0	0	0	0	0	0	0	0	0	0	2
M. brevicollis	0	0	1	0	0	1	0	0	0	0	1	0
N. vectensis	0	0	0	0	0	0	0	0	0	1	0	0
C. elegans	0	1	0	0	0	0	0	0	0	1	1	1
D. melanogaster	0	0	0	0	0	0	0	0	0	0	0	0
C. milii	1	1	0	0	0	0	1	1	0	0	2	0
D. rerio	1	1	1	4	0	1	1	2	0	3	0	3
X. tropicalis	1	1	1	2	1	1	1	1	1	3	3	2
B. taurus	1	0	0	3	1	1	1	1	1	2	1	3
M. musculus	1	1	1	1	1	1	1	1	1	2	0	3
H. sapiens	1	1	1	2	1	2	2	1	1	2	2	2

B. Number of C‐terminal sequences 2 or less mutations in non‐motif residuesb
Organism	YLVTSV	WFDTDL	VQDTRL	SKETPL	PTRTAV	RKETLV	HSTTRV	TQITKV	LGATGL	GLETNV	ALETDV	QSSSTL
E. coli	0	0	0	0	0	0	0	0	1	0	0	0
D. discoideum	0	2	0	2	0	0	0	2	0	0	0	1
S. cerevisiae	0	0	0	1	0	1	0	0	1	0	0	5
A. thaliana	9	4	1	4	4	6	3	1	3	3	2	22
M. brevicollis	1	1	2	0	1	2	1	0	1	3	2	2
N. vectensis	2	2	6	7	2	11	3	3	1	6	9	3
C. elegans	3	1	4	6	4	3	2	4	0	4	2	7
D. melanogaster	3	1	5	4	6	4	0	1	2	1	4	6
C. milii	3	1	3	6	1	11	3	3	2	9	3	7
D. rerio	6	2	9	17	3	12	8	9	3	10	20	15
X. tropicalis	7	4	6	14	6	11	4	5	2	7	8	10
B. taurus	10	2	4	11	4	8	3	6	4	7	9	8
M. musculus	10	4	6	19	4	11	2	4	11	9	8	15
H. sapiens	11	3	6	13	4	11	4	4	5	7	11	6

^aSequence information is in Table S1A.

^bSequence information is in Table S1B.

Figure 8.

Sequence evolution of select human PDZ targets. An evolutionary tree for 14 organisms is shown, including vertebrates, invertebrates, fungi, plants, and prokaryotes, with the conserved sequences of either β‐catenin (a) or mGluR1 (b) that exist in each organism. If an exact match was not found, a sequence was chosen that had the highest pairwise similarity to the target sequence. Conserved (blue) or substituted (orange) residues in each sequence are highlighted and the number of additional sequences in the proteome which also contain 0, 1, or 2 substitutions are indicated to the right of the sequences. Sequence information is in Table S1

These two sequences are highlighted for interesting reasons. Namely, β‐catenin is a promiscuous PDZ target, interacting with at least 27 PDZ domain‐containing proteins, whose homologues have been studied in sponges and choanoflagellates (Table 4).80, 81 While choanoflagellate β‐catenin (NCBI XP_001747419.1) does not contain a PDZ‐targeting C‐terminal sequence, one of the choanoflagellate proteins identified in our results, A9V400_MONBE, is most similar to the human Ras‐associating (RA) domain of RADIL (33% identity over 120 residues), a protein that is involved in cell adhesion similar to β‐catenin.82, 83 This result suggests that a similar PDZ‐mediated mechanism may occur in these organisms, particularly in the multicellular‐like rosette morphology state, although the exact details and protein players likely vary and require additional investigation.84

We also took a closer look at the mGluR1 sequence (QSSSTL), an important glutamate and G‐protein coupled receptor that is involved in neurotransmission (Figure 8). We were surprised to see strong conservation in the mGluR1 C‐terminal sequence in lower organisms, specifically those that lack a nervous system. However, investigation of the five yeast (Saccharomyces cerevisiae) sequences with 0, 1, or 2 substitutions reveals that one is a mitogen‐activated kinase (SMK1_YEAST, sequence: LSSSSL) and another is involved in K⁺ transport (TRK1_YEAST, RSSTTL), again suggesting that although the pathways are highly diverged, PDZ domains may have facilitated signaling and trafficking pathways in similar ways throughout evolution.

2.6. Adapting MotifAnalyzer‐kinase to identify Src and Abl motif‐satisfying sequences

Finally, we adapted MotifAnalyzer to look for targets of additional SLiM‐binding domains that can occur anywhere in the protein sequence. Initially, we chose to investigate PDZ domains for three major reasons: (i) the domain family is large, (ii) individual PDZ domains are important nodes in signaling and trafficking networks, and (iii) PDZ domains recognize the extreme C‐termini of proteins, and isolating the final x residues of proteins in sequenced proteomes was straightforward. However, our program is versatile for other proteins that recognize short linear motifs, for example kinase domains.

Kinase domains typically recognize and phosphorylate serine, threonine, or tyrosine residues in target proteins, and although some kinases are promiscuous, with motifs limited to a small number of amino acid residues, others are more specific, with up to 10 amino acid positions indicating a favorable phosphorylation site.85, 86, 87, 88, 89, 90 For example, if we look for motif‐satisfying sequences that match the Src motif (E/D‐X‐I/L/V‐Y‐E/G‐X‐L/F, where X is any amino acid and the phosphorylated tyrosine is in bold) 89 in the entire human and choanoflagellate proteomes, we find 165 sequences in H. sapiens and 64 in M. brevicollis that contain all motif‐satisfying residues (Table S2). Previous work indicates that the selectivity of choanoflagellate Src and Src‐related kinases are similar to that of their human homologs.91, 92, 93 Many of the proteins with Src motif‐satisfying sequences in H. sapiens are known signaling proteins (e.g., small GTPases, SH3 domain‐containing proteins, Src‐associated proteins, or tyrosine phosphatases), suggesting this is a good approach to identify cellular targets. We can do a similar search for Abl targets, using motif E‐X‐I/L/V‐Y‐A‐X‐P/LF,89 identifying 46 sequences in H. sapiens and 43 in M. brevicollis (Table S2). In addition, as with our PDZ searches, we know the UniProt ID for each protein with a potential Src‐ or Abl‐targeting sequence and can expand our searches to look for sequences with 1, 2, or n number of substitutions (Table S2). This example provides evidence that MotifAnalyzer can be easily adapted for use investigating other domain families.

3. CONCLUSIONS

The prediction of protein–protein interactions dependent on degenerate motifs is challenging. SLiM‐binding domains, such as the PDZ domain, highlight this complexity by interacting specifically with a small number of residues on target proteins, and with relatively weak binding free energies. Motifs consist of required amino acids at two positions, and utilizing just the motif predicts >2000 target proteins for a given PDZ domain, despite work by ourselves and others that suggest the motif alone does not adequately describe interaction networks.28, 29, 94, 95, 96 Therefore, we developed a bioinformatics program to analyze sequence conservation in an unbiased way in order to predict PDZ interactions in humans, as well as distantly related organisms.

As highlighted in our results, we are able to predict re‐wired signaling networks in choanoflagellates by identifying potential protein targets, despite clear divergence from human signaling pathways. In addition, we predict currently unidentified human PDZ targets that bind human PDZ domains with biologically‐relevant affinities, as well as provide a template for the organism‐specific design of peptides that contain either under‐ or over‐enriched amino acids in PDZ‐satisfying sequences.

Broadly, our program MotifAnalyzer‐PDZ allows us to: (i) predict novel human PDZ interactions, (ii) predict PDZ targets in any organism with homologous PDZ domains to human proteins, and (iii) compare PDZ motif‐satisfying sequences broadly across proteomes and organisms. The evolution and expansion of PDZ domains throughout evolution is well‐studied; however, many of the target sequences of PDZ domains are not strongly conserved and less work is focused on how PDZ domains are utilized in different organisms and in these divergent signaling pathways.97, 98, 99

Moving forward, we will continue to adapt MotifAnalyzer to look for motif‐satisfying sequences of various SLiM‐binding domains, expand our evolution‐ and enrichment‐based analyses, and create a web‐based interface. Previous studies on other SLiM‐binding domains, for example SH2 and SH3 domains, suggests that these interactions are also more specific than previously realized.100, 101, 102, 103 MotifAnalyzer is a tool that can be used in silico to predict endogenous targets of these domains and more, for example, tyrosine kinases, using a similar pipeline (Figure 9). Thus, our MotifAnalyzer program can be broadly applied to many important SLiM‐binding domain families.

Figure 9.

Application of MotifAnalyzer‐PDZ to other SLiM‐binding domains. Similar analyses can be applied to other SLiM‐binding domains. The structures of example SLiM‐binding domains are shown as cartoons and rainbow colored; these include: PDZ (PDB ID: 1BE9), kinase (Src: 2SRC and Abl: 1OPL), SH2 (1OPL), SH3 (1OPL), PTB (1WVH), WW (1K9R), and phosphatase (1YGR) domains. The arrows indicate that MotifAnalyzer can be adapted to the motifs for proteins of any of these families and analyses of conserved sequences or positional enrichments performed

4. MATERIALS AND METHODS

4.1. Expression and purification of PDZ domains

Expression and purification of N1P1 and N2P2 was performed as previously described.55 Expression and purification of SHANK1 PDZ (residues 654–762) and mbSHANK1 (residues 442–544, NCBI reference sequence XP_001748613.1) followed a similar protocol. Briefly, His‐tagged versions of the PDZ domains were inserted into the pET28a vector (GenScript) and expressed in Escherichia coli BL21 (DE3) cells. Cells were lysed using sonication and immobilized metal‐affinity chromatography was used to purify proteins from the clarified supernatant. The protein was further purified on a Superdex S75 column, using gel filtration buffer [25 mM Tris pH 8.5, 125 mM NaCl, 10% (w/v) glycerol, 0.5 mM TCEP]. Protein was quantified using the A₂₈₀ and theoretical extinction coefficient values of 8,480 cm⁻¹ M⁻¹ for SHANK1 PDZ and 11,000 cm⁻¹ M⁻¹ for mbSHANK1.

4.2. Binding assays by fluorescence polarization

Fluorescence polarization assays were performed as previously described.29, 55, 104 Replicate experiments were performed to determine the K _D values of SHANK1 PDZ and mbSHANK1 PDZ for the fluorescent peptide, F*‐β‐PIX (sequence: FITC‐NDPAWDETNL). The following K _D values (and reporter peptides) were used for each of the proteins in competition K _i experiments: N1P1 (reporter: F*‐CFTR₆, K _D = 0.486 μM),55 N2P2 (F*‐CFTR₁₀, K _D = 0.232 μM),55 SHANK1 PDZ (F*‐β‐PIX, K _D = 5.1 μM), and mbSHANK1 PDZ (F*‐β‐PIX, K _D = 7.3 μM). The final protein concentration for K _i experiments was equal to 1.4–2*K _D and experiments were performed in triplicate. Binding affinities for K _i experiments were determined using SOLVER, as previously described.29, 55, 104

4.3. Design and development of MotifAnalyzer‐PDZ

The software was designed in a modular, object‐oriented fashion to facilitate ease of use, and also to permit extension and modification of the code by individuals with software development skills. See Supporting Information for detailed information about input parameters to the program.

4.4. Protein scoring

In MotifAnalyzer‐PDZ, protein sequences are filtered according to the motif supplied by the user, and identical sequences are removed. For each sequence in the reference organism proteome, a comparison is made against every other protein from selected proteomes, excluding those of the reference organism. The pairwise comparison score is given by the equations below, where n is motif length, p_{j, i} is the i‐th residue of the j‐th reference sequence, p_{k, i} is the i‐th residue of the k‐th query sequence and S is the scoring function. The results of these pairwise comparisons are saved as an XML file.

$$T=\sum _{i=0}^{n}S\left(p_{j,i},p_{k,i}\right)$$

$$S\left(p_{j,i},p_{k,i}\right)=\left\{\begin{array}{c}1\text{if}{p}_{j,i}=p_{k,i}\\ 0\text{else}\end{array}\right.$$

The information is then reported as the total number of sequences in each score category. Information about statistics is in the Supporting Information.105

4.5. Protein analyses

Sequence alignments were performed using BLASTP or T‐COFFEE and visualized using JalView, with ClustalW coloring.106, 107, 108 Structure figures were rendered in PyMOL, and homology modeling was done using SwissModel.48, 109, 110 Evolutionary trees were generated using PhyloT (phylot.biobyte.de) and visualized using iTOL.111

The heat maps were created with Matplotlib.112 The enrichment values produced by the MotifAnalyzer‐PDZ program were plotted with the imshow() function. The enrichment values were bounded by −0.2 and 2.2. Outlier enrichment values were set to the lower or upper bound. The SequenceLogos show the relative over‐ and under‐enrichment of the amino acids at non‐motif positions with the enrichment values. Any enrichment value greater than 0.95 and less than 1.05 was disregarded when generating the SequenceLogos as these were only slightly over‐ or under‐enriched. One was subtracted from each enrichment value and then normalized by non‐motif position. A positive value indicates an over‐enriched amino acid, and vice versa for a negative value. The new enrichment values were provided to WebLogo3 in the transfac format—the over‐ and under‐enriched SequenceLogos were generated separately and combined in post‐processing.

Supporting information

Appendix S1: Description of supplementary material. The supplemental material includes expanded descriptions of the materials and methods used, specifically of the options included in running the MotifAnalyzer‐PDZ program and how statistics were performed. The name of the file is MotifAnalyzer‐PDZ_supp_submit_withtables.pdf. There are also 3 supplemental figures.

Figure S1

Figure S2

Figure S3

Valgardson J, Cosbey R, Houser P, et al. MotifAnalyzer‐PDZ: A computational program to investigate the evolution of PDZ‐binding target specificity. Protein Science. 2019;28:2127–2143. 10.1002/pro.3741