High-throughput competitive fluorescence polarization assay reveals functional redundancy in the S100 protein family

Abstract

The calcium-binding, vertebrate-specific S100 protein family consists of 20 paralogs in humans (referred as the S100ome), with several clinically important members. To explore their protein–protein interactions (PPIs) quantitatively, we have chosen an unbiased, high-throughput, competitive fluorescence polarization (FP) assay that revealed a partial functional redundancy when the complete S100ome (n = 20) was tested against numerous model partners (n = 13). Based on their specificity, the S100ome can be grouped into two distinct classes: promiscuous and orphan. In the first group, members bound to several ligands (> 4–5) with comparable high affinity, while in the second one, the paralogs bound only one partner weakly, or no ligand was identified. Our results demonstrate that FP assays are highly suitable for quantitative interaction profiling of selected protein families. Moreover, we provide evidence that PPI-based phenotypic characterization can complement or even exceed the information obtained from the sequence-based phylogenetic analysis of the S100ome, an evolutionary young protein family.

Abbreviations

CapZ

F-actin-capping protein subunit alpha-2

C-ERMAD

ezrin C-terminal domain

FOP

FGFR1 oncogene partner

FOR20

FOP-related protein of 20 kDa

FP

fluorescence polarization

GST

glutathione S-transferase

HTP

high-throughput

ITC

isothermal titration calorimetry

MK2

MAPK-activated protein kinase 2

MNK1

MAPK interacting serine/threonine kinase 1

NCX1

sodium/calcium exchanger 1

NMIIA

nonmuscle myosin IIA

PPI

protein–protein interaction

RSK1

ribosomal-S6-kinase 1

SIP

Siah-interacting protein

TCEP

Tris(2-carboxyethyl)phosphine hydrochloride

TEV

tobacco etch virus

TRPM4

transient receptor potential cation channel subfamily M member 4

UPGMA

unweighted pair-group method with arithmetic average

Introduction

Biochemical characterization of protein–protein interactions (PPIs) is a challenging field in molecular life sciences, which is usually limited to the determination of steady-state dissociation constants [1]. The accurate determination of thermodynamic parameters of molecular interactions is performed by fast, but superficial, high-throughput (HTP) methods. In the literature, several HTP approaches are applied such as coimmunoprecipitation [2], yeast two-hybrid and spot assays [3], pull-down assay [4], holdup assay [5], and direct fluorescence polarization/anisotropy [6]. In direct fluorescence polarization (FP) experiments, a fluorescent probe (usually a labeled peptide) is titrated with a globular partner. Their association is monitored by the polarization of the emitted light of the fluorophore (Fig. 1A). In a modified FP experiment called competitive assay, both the probe and partner concentration are fixed, and the reaction mixture is titrated with an unlabeled competitor molecule (peptide or protein). Depolarization of the emitted light is indicative of the competition between the probe and the competitor in binding to the partner (Fig. 1B,C). While direct FP can be perturbed by the presence of the fluorescent dye, the competitive assay is unbiased and therefore more suitable for accurate HTP measurements of dissociation constants [7, 8].

Figure 1

The theory of fluorescence polarization assays. (A) Fluorescence polarization/anisotropy experiments can be performed with direct and competitive titrations. In direct assay (direct titration, (d)), the concentration of the protein of interest is increased in the presence of tracer amount of labeled peptide. Upon complex formation, the hydrodynamic radius of the tracer increases causing slower rotation and therefore lower depolarization of the emitted light. In the direct assay, one can measure the minimal and maximal polarization values, a dissociation constant, and importantly, an optimal concentration can be easily determined for competitive assays, which is usually the concentration corresponding to 60–80% saturation. (B) In a competitive assay (competitive titration, (c)), the concentration of the protein of interest is set to this concentration and one can titrate the reaction mixture with a competitor. The competition results in increased level of free labeled peptide and consequently high depolarization of the emitted light. (C) Competitive FP is not affected by the presence of a labeling group in the peptide (unbiased) and has a high dynamic range (approximately two orders of magnitudes around the dissociation constant of the probe). At high concentrations, it can be also used to determine the stoichiometry of the interaction for strong interactions. K_d(d) and K_d(c) correspond to direct and competitive dissociation constants, respectively. The red and blue graphs (on panel C) correspond to two scenarios, in which the competitive K_d (K_d(c)) is 10-fold higher and 10-fold lower than the direct K_d (K_d(d)), respectively.

S100 proteins belong to the superfamily of EF-hand containing calcium-binding proteins. They appeared in early vertebrates and consist of 20 core paralogs in the human proteome [9]. S100s are associated with several disease conditions, such as cardiomyopathies, cancer, and inflammatory and neurodegenerative diseases, in which their overexpression can be observed in the affected cells [10-12]. Due to this reason, they are emerging biomarkers and also promising therapeutic targets [13]. Despite their growing importance, the literature still lacks their comprehensive and systematic analysis, which would be essential for developing rational strategies for drug development. Similar to calmodulin, they can interact with protein or peptide targets in a calcium-dependent manner [14]. They are generally considered as relatively low specificity proteins, with dozens of interaction partners, among them they are unable to maintain high selectivity [15]. In this study, we determined the interaction profile of the full human S100 family (termed here as the S100ome) against a set of diverse known S100 partners (and some of their paralogs) systematically, including kinases such as RSK1 [16] and its paralogs MK2 and MNK1; cytoskeletal elements such as CapZ [17] (commonly known as TRTK12), NMIIA [18], ezrin [19], FOR20 and its paralog FOP [20]; membrane proteins such as NCX1 [15] and TRPM4 [21]; and other signaling proteins such as the tumor suppressor p53 [22-24], SIP [15], and MDM4 [23].

Results

Mapping the S100ome with FP measurements

The interactions between S100 homodimers and their selected labeled peptide partners were studied first by direct FP assay (Figs S1–S13). We have found that all reasonable S100 interactions gave an experimental window of 50–200 mP (polarization). If significant binding was detected (K_d < 200 µm) between a labeled peptide and an S100 protein, a subsequent competitive FP assay was performed. In cases, where no labeled peptide was available (e.g., when globular protein domains were used as competitors), we used noncognate tracers against all possible S100 proteins. Additionally, we tested the possible binding between these competitors and the noncognate probes in direct FP experiments to eliminate the possibility of rebinding (Fig. S14). This way, we tested 180 unique direct and 150 unique competitive interactions and found 89 and 66 significant interactions, respectively (Table 1, Figs S1–S13).

Table 1. Quantitative characterization of interactions between S100 proteins and their selected partners by FP (N.D., not determined; E.A., experimental artifact; (d), direct titration; (c), competitive titration).

	Kd (μm)
	fp53 (d)	p53 (c)	fNMIIA (d)	NMIIA (c)	fCapZ (d)	CapZ (c)	fNCX1 (d)	NCX1 (c)	fSIP (d)	SIP (c)	FOR20 (c)
S100A1	41 ± 11	> 200	0.033 ± 0.0062	0.004 ± 0.001	8.9 ± 2.0	5.2 ± 1.7	2.6 ± 0.14	2.6 ± 0.56	> 200	N.D.	0.50 ± 0.25
S100A2	2.9 ± 0.08	5.2 ± 0.39	0.057 ± 0.0038	0.013 ± 0.0022	> 200	N.D.	5.5 ± 0.37	> 200	102 ± 40	> 200	0.27 ± 0.03
S100A3	7.6 ± 0.44	7.7 ± 1.9	5.5 ± 1.0	E.A.	> 200	N.D.	25 ± 2.5	> 200	> 200	N.D.	29 ± 5.6
S100A4	0.85 ± 0.04	2.1 ± 0.23	0.026 ± 0.0065	0.0088 ± 0.0025	> 200	N.D.	15 ± 0.62	> 200	42 ± 5.2	11 ± 2.7	2.0 ± 0.20
S100A5	26 ± 2.4	67 ± 23	5.8 ± 0.67	2.9 ± 0.51	4.7 ± 0.28	5.4 ± 0.84	4.6 ± 0.22	5.2 ± 1.1	42 ± 3.2	40 ± 7.8	1.9 ± 0.53
S100A6	0.68 ± 0.020	2.2 ± 0.090	0.58 ± 0.07	0.21 ± 0.04	> 200	N.D.	27 ± 1.8	> 200	8.7 ± 0.48	20 ± 1.1	0.007 ± 0.0018
S100A7	> 200	N.D.	13.1 ± 1.4	E.A.	> 200	N.D.	30 ± 2.1	E.A.	> 200	N.D.	5.9 ± 0.54
S100A8	> 200	N.D.	> 200	N.D.	> 200	N.D.	56 ± 5.0	E.A.	> 200	N.D.	> 200
S100A9	> 200	N.D.	> 200	N.D.	> 200	N.D.	20 ± 3.2	> 200	> 200	N.D.	> 200
S100A10	48 ± 2.9	> 200	> 200	N.D.	> 200	N.D.	53 ± 6.6	> 200	> 200	N.D.	> 200
S100A11	10 ± 0.72	12 ± 2.4	49 ± 14	E.A.	> 200	N.D.	52 ± 2.9	E.A.	> 200	N.D.	> 200
S100A12	76 ± 11	N.D.	> 200	N.D.	> 200	N.D.	170 ± 63	E.A.	> 200	N.D.	> 200
S100A13	35 ± 10	> 200	> 200	N.D.	> 200	N.D.	87 ± 14	E.A.	> 200	N.D.	> 200
S100A14	> 200	N.D.	98 ± 127	E.A.	> 200	N.D.	71 ± 7.5	E.A.	> 200	N.D.	63 ± 30
S100A15	> 200	N.D.	17 ± 1.5	E.A.	> 200	N.D.	28 ± 3.4	E.A.	> 200	N.D.	17 ± 2.1
S100A16	79 ± 8.9	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200
S100B	33 ± 3.2	> 200	4.7 ± 0.25	2.9 ± 0.31	2.3 ± 0.13	1.8 ± 0.23	7.7 ± 0.95	> 200	> 200	N.D.	0.25 ± 0.050
S100G	> 200	N.D.	21 ± 1.9	E.A.	> 200	N.D.	24 ± 1.7	E.A.	> 200	N.D.	3.8 ± 1.1
S100P	0.17 ± 0.01	0.54 ± 0.040	0.99 ± 0.11	0.14 ± 0.03	11 ± 1.4	4.1 ± 0.73	4.0 ± 0.23	> 200	38 ± 3.2	> 200	1.2 ± 0.13
S100Z	> 200	N.D.	> 200	N.D.	> 200	N.D.	11 ± 0.45	E.A.	> 200	N.D.	> 200

	Kd (μm)
	fTRPM4 (d)	TRPM4 (c)	fMDM4 (d)	MDM4 (c)	fC-ERMAD (d)	C-ERMAD (c)	fRSK1 (d)	RSK1-CTKD (c)	MK2 (c)	MNK1 (c)	FOP (c)
S100A1	0.38 ± 0.061	> 200	91 ± 14	35 ± 10	> 200	N.D.	4.8 ± 0.72	> 200	15 ± 4.6	> 200	0.35 ± 0.14
S100A2	1.0 ± 0.062	1.1 ± 0.23	49 ± 10	53 ± 16	15 ± 0.62	5.4 ± 0.57	5.1 ± 0.53	> 200	4.5 ± 1.0	18 ± 4.2	0.081 ± 0.019
S100A3	0.91 ± 0.052	3.4 ± 0.45	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	> 200	0.27 ± 0.13
S100A4	5.9 ± 0.36	35 ± 113	> 200	N.D.	11 ± 0.73	6.1 ± 0.80	8.5 ± 0.89	> 200	> 200	24 ± 3.5	0.048 ± 0.010
S100A5	0.60 ± 0.051	2.3 ± 0.71	61 ± 5.4	65 ± 7.5	> 200	N.D.	> 200	N.D.	> 200	> 200	0.56 ± 0.20
S100A6	1.4 ± 0.081	8.4 ± 2.0	> 200	N.D.	> 200	N.D.	8.7 ± 0.81	> 200	> 200	6.7 ± 0.53	0.0056 ± 0.0050
S100A7	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	E.A.	E.A.	> 200
S100A8	3.8 ± 0.57	13 ± 4.4	> 200	N.D.	> 200	N.D.	> 200	N.D.	N.D.	> 200	> 200
S100A9	21 ± 7.3	> 200	> 200	N.D.	> 200	N.D.	> 200	N.D.	N.D.	> 200	> 200
S100A10	17 ± 1.9	125 ± 25	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	> 200	> 200
S100A11	2.4 ± 0.12	106 ± 11	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	> 200	> 200
S100A12	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	N.D.	> 200	> 200
S100A13	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	N.D.	N.D.	> 200
S100A14	12 ± 5.6	> 200	> 200	N.D.	> 200	N.D.	> 200	N.D.	E.A.	E.A.	> 200
S100A15	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	E.A.	E.A.	> 200
S100A16	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	N.D.	> 200	> 200
S100B	16 ± 25	> 200	0.20 ± 0.04	0.15 ± 0.04	> 200	N.D.	2.8 ± 0.47	1.2 ± 0.86	3.2 ± 0.73	> 200	1.2 ± 0.53
S100G	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	E.A.	E.A.	> 200
S100P	0.93 ± 0.24	> 200	> 200	N.D.	> 200	N.D.	4.5 ± 0.47	> 200	2.7 ± 0.47	2.5 ± 0.38	0.066 ± 0.0049
S100Z	> 200	N.D.	> 200	N.D.	> 200	N.D.	> 200	N.D.	N.D.	N.D.	> 200

Compared to the direct assay, competitive FP provides unbiased (or more specific) affinities, unaffected by the chemical labeling, making it a better tool to measure protein–protein interactions (Fig. 2A,B). Nevertheless, there are some pitfalls (Fig. 2), which should be taken into consideration while analyzing competitive data. First of all, the experimental window of the competitive measurement should be the same as the experimental window of the direct measurement (Fig. 2C). Studying large biomolecules (e.g., globular proteins) in a competitive experiment often results in an increased base polarization (P_min) due to the change in biophysical properties of the reaction mixture (e.g., change in viscosity). Moreover, during a competition experiment, it is possible that the competitor can interact with the probe itself, which can also cause an increase in the base polarization (Fig. S14). In rare cases, saturation polarization can be also altered (e.g., due to oligomerization at higher concentration). Additionally, experimental artifacts of unknown origin can be observed occasionally (Fig. 2D). Here, a sharp decline can be detected during the titration, which results in an IC50 value smaller than the fixed receptor concentration. This observed substoichiometric complex formation should be handled with extra care as it is likely due to unexpected biophysical phenomena, such as protein aggregation. To standardize and automatize data handling and to eliminate subjective factors, we developed a Python-based universal program, called ProFit, for fitting all direct and competitive experimental data (freely available at https://github.com/GoglG/ProFit).

Figure 2

Possible outcomes of a competitive experiment. (A) In ‘perfect experiments’, the experimental window is stable and the dissociation constants match between the (cognate) probe and the competitor. (B) As often occurs, fluorescent labeling can alter the binding affinity, resulting in false-positive interaction partners in direct FP experiments. In other cases, the effect is softer and it only causes a dimming effect on the biochemical constant. (C) The reliable experimental window can be different in a competitive experiment. If the change is not extreme, the competitive K_d can be considered (with caution) as the relevant biochemical constant. (D) In some cases, a rapid decline can be observed in the polarization. In this case, the experimentally determined IC50 value should not be used as a dissociation constant. This phenomenon can be explained by a competitor-induced biophysical transition, for example, aggregation or precipitation. In this final case, it is very important to redetermine the concentrations of the receptor and the competitor and to repeat the experiment at different receptor concentrations to properly discriminate the stoichiometric molar ratio from the observed IC50 value.

Validation with ITC measurements

The biochemically described S100 binding motifs, found in the literature, show an extremely low sequence similarity [15, 23] (Fig. 3A). Mostly, linear segments are recognized by the human S100ome; however, no consensus S100 binding sequence can be defined [15]. In general, hydrophobic residues are preferred, but additional basic residues can also be favored in some instances. Moreover, S100 proteins can form two types of complexes (Fig. 3B). Earlier studies showed that a symmetric S100 dimer can recognize two identical binding motifs (1 S100 dimer binds 2 partners, N = 1), symmetrically [17, 25, 26]. In recent studies, however, several asymmetric complexes (1 S100 dimer binds 1 partner, N = 0.5) were also described [18, 27, 28]. In those cases, an S100 dimer captures a single partner at the two binding sites. As the binding affinity highly depends on the stoichiometry of the interaction, we selected a set of significant, peptide-based interactions for isothermal titration calorimetric (ITC) measurements. This way, we validated the interactions that were originally detected by the FP assay and determined the binding stoichiometry in all instances.

Figure 3

The S100 protein family. (A) Multiple short linear motifs are recognized by S100 family members; however, no consensus-binding motif can be defined for the protein family, as indicated here by the sequence alignment of several S100 binding motifs by ClustalW algorithm (UniProt accession codes: CapZ: P52907, MDM4: O15151, NCX1: P32418, C-ERMAD: P15311, FOP: O95684, FOR20: Q96NB1, TRPM4: Q8TD43, NMIIA: P35579, p53: P04637, SIP: Q9HB71, RSK1: Q15418). Though it is noteworthy that hydrophobic residues (green) are preferred, basic residues are also welcome in some cases. (B) S100 proteins act as dimers and are capable of interacting in two distinct ways with other proteins. On the left, a symmetric complex is shown, where one S100 dimer interacts with two peptides (S100B-CapZ, PDB code: 1MWN, [17]). By contrast, a single interacting partner can bind to one S100 dimer asymmetrically, as it is shown on the right side (S100A4-NMIIA, PDB code: 3ZWH, [18]). pymol (Schrödinger, New York, NY, USA) was used for visualization and for preparing molecular images.

All determined K_d values correlated well with the data provided by the orthogonal FP measurements (Table 2, Fig. S15). Symmetric interactions were found with CapZ, NCX1, SIP, TRPM4, and MDM4. In cases of CapZ and MDM4, the experimental data were fitted by a two binding site model indicating slightly different affinities and a complex relationship between the S100 monomers. In contrast, asymmetrical interactions were detected with p53, RSK1, C-ERMAD, NMIIA, and FOP. These findings confirmed the expected binding stoichiometry in all cases and clarified the binding mode of TRPM4 and FOP. We hypothesize that the binding mode of close paralogs should be identical (symmetric or asymmetric); therefore, asymmetric binding was assumed for MNK1, MK2 (based on RSK1), and FOR20 (based on FOP). We performed these ITC measurements in parallel with the FP experiments, and based on the refined stoichiometry, monomer or dimer S100 concentrations were used during the FP data evaluation.

Table 2. Quantitative characterization of interactions between S100 proteins and their selected partners by ITC.

Measurement	T (K)	N (mol peptide·mol^−1 S100 subunit)	Reference N (mol peptide·mol^−1 S100 subunit)	Kd (μm)	ΔH (kJ·mol^−1)	−TΔS (kJ·mol^−1)
S100A6-FOP	310	0.44 ± 0.002	Previously unknown	0.088 ± 0.0073	−73 ± 0.58	31
S100B-CapZ	310	0.44 ± 0.002	1 [17]	3.9 ± 0.39	−15 ± 0.37	−17
		0.343 ± 0.002		0.94 ± 0.03	−3.3 ± 0.67	−33
S100A1-NCX1	310	1.1 ± 0.013	1 [15]a	6.4 ± 0.58	−35 ± 0.80	4.0
S100A5-TRPM4	310	0.89 ± 0.0037	Previously unknown	1.4 ± 0.089	−24 ± 0.21	−10
C-ERMAD-S100A4	310	0.52 ± 0.015	0.47 [19]	18 ± 2.7	−49 ± 3.0	22
S100A5-SIP	310	0.92 ± 0.24	1 [26]a	21 ± 20	−4.1 ± 2.5	−24
S100A1-NMIIA	298	0.49 ± 0.0017	0.5 [18]a	0.009 ± 0.005	36 ± 0.34	−82
S100B-MDM4	310	0.52 ± 0.039	1 [23]	0.71 ± 0.036	−186 ± 142	150
		0.53 ± 0.038		0.63 ± 0.081	111 ± 148	−148

• ^a These interactions were measured with a different S100 paralog.

Specificity map of the S100ome

The 20 S100 paralogs, whose interactions were studied here, represent almost the complete human S100ome [13]. It is a chordata-specific, evolutionary young protein family, and despite the fact that they exhibit moderate sequence similarity, they are structurally very similar owing to their small size (~ 100 residues) and conserved fold (including two consecutive EF-hand motifs) (Fig. 4A). Due to this reason, their phylogenetic analysis generally does not lead to unambiguous results [29, 30]. Applying different parameters during the analyses resulted in varied grouping of the human S100ome; moreover, only a few clades received statistical supports (see our analyses in Fig. 4B). Because of these ambiguities of the phylogenetic analyses, a phenotypic screening and analysis could provide a more reliable grouping and could reveal functional similarities among the paralogs of the protein family of interest beside the sequence-based genealogies. For such purpose, we decided to create a robust phenogram [31], representing the functional relationships within the human S100ome, using hierarchical clustering (UPGMA) [32]. This analysis separated the S100ome into two groups, in which the first group contains S100 proteins generally lacking significant interactions (termed here as ‘orphan’ S100 proteins) and the second group comprises generally good binders (termed here as ‘promiscuous’ S100 proteins) (Fig. 5). While promiscuous S100 proteins showed significant binding to at least a few (4–5) of the tested interaction partners, orphan S100 proteins showed either no sign of partner binding or a weak binding to a single partner.

Figure 4

Phylogenetic map of the S100 protein family. (A) S100 proteins are small (~ 100 amino acids long) EF-hand proteins, sharing high sequence identity (identical residues with S100A1 are shown in gray). The sequences were aligned by clustalw algorithm (UniProt accession codes: S100A1: P23297, S100A2: P29034, S100A3: P33764, S100A4: P26447, S100A5: P33763, S100A6: P06703, S100A7: P31151, S100A8: P05109, S100A9: P06702, S100A10: P60903, S100A11: P31949, S100A12: P80511, S100A13: Q99584, S100A14: Q9HCY8, S100A15: Q86SG5, S100A16: Q96FQ6, S100B: P04271, S100G: P29377, S100P: P25815, and S100Z: Q8WXG8). The fold consists of two consecutive EF-hand motifs, connected by a hinge region. The calcium ions are coordinated by several residues between helices I–II and III–IV (- highlighted with asterisks). (B) The affinity profile of S100ome is clustered by the phylogeny of the different S100 paralogs. For the phylogenetic analyses, the human S100 paralog sequences were aligned by clustalw, muscle, and prank algorithms with default parameters (see Materials and methods section). The evolutionary histories were inferred by maximum-likelihood method with 10 runs, using ProtGamma and LG as substitution model and substitution matrix, respectively. The supports of the branches were tested by bootstrap analysis (1000 replicates) shown as % (values below 60% are not shown). The analyses were conducted by raxml gui. It is shown that S100A2/A3/A4/A5/A6 and S100A13/A14 can be considered as monophyletic groups, supported by high values (S100A7/A15 are almost identical paralogs). Nevertheless, the phylogenetic analyses do not provide unambiguous genealogy for the rest of the S100 proteins.

Figure 5

The phenogenetic map of the S100 protein family. The determined dissociation constants are depicted as a heatmap representing the specificity map of the S100ome. Hierarchical clustering, based on functional relationships, divided the S100ome into two different groups, one of them consists of low(er) specificity and/or more promiscuous S100 proteins (‘promiscuous’), while the other one contains high(er) specificity and/or less promiscuous members of the family (‘orphan’). White and gray fields indicate nondetermined interactions and cases with experimental artifacts, respectively. Stoichiometry (2 mol peptides/2 mol S100 subunit, N = 1; 1 mol peptides/2 mol S100 subunit, N = 0.5) is also shown for all ligands at the end of each rows.

Discussion

Competitive FP as a potent tool to measure high-throughput macromolecular interactions

Although numerous HTP, semiquantitative approaches are available and many low-throughput but highly accurate methods exist to measure PPIs, reliable and quantitative HTP methods are scarce in the literature. On the one hand, direct FP assay can be performed in large scale in multiwell plates, which makes it an ideal method for rapid interaction screening; however, it has the serious limitation of chemical labeling that can perturb the binding measurement. Competitive FP, on the other hand, shares the same properties but without any possible interference from the labeling dye. Moreover, it provides comparative results to other, orthogonal, usually low-throughput, label-free biochemical assays, such as ITC or SPR measurements [33]. In the present work, we applied this robust HTP method to characterize the specificity map of the S100 interactome. We used minimal S100 binding segments (with a few exceptions) due to technical and biochemical reasons. On the one hand, this is a limitation, because the presented affinities might be different in full-length proteins, but on the other hand, the generated data set remained as comparative as possible between different S100 proteins. One should note that binding stoichiometry and oligomerization of the investigate proteins, in this case, S100s, could affect the FP measurements and data evaluation. As both properties could affect the binding ratio and the affinity, orthogonal measurements are important as part of the validation process [34]. In summary, competitive FP assay is robust and has HTP; thus, it is a valuable tool for screening macromolecular interactions involving linear peptide motifs, RNA/DNA oligonucleotides, or fluorescent small molecules [35, 36].

Functional redundancy within the S100 family and possible functions of the orphan group

S100 proteins are usually considered as ‘sticky’, relatively low specificity proteins [15], which is also supported by several studies covering nearly all S100 proteins and only one or few S100 targets [20, 37-41]. Usually, the tested S100 proteins only covered the closest relatives (e.g., S100A2, S100A4, S100A6, S100B, S100P), and the results often showed redundant bindings [19, 27, 40, 42, 43]. In one study, close to the full S100ome was tested against a simple peptide (derived from CapZ), highlighting binding promiscuity for a subset of S100 isoforms [38]. However, no systematic study has been performed to make a specificity map involving the complete S100ome against multiple S100 partners. Based on functional clustering, we have revealed here that the S100ome can be separated into two groups, which is comparable with previous findings based on few partners. The minor group of eight members includes promiscuous paralogs, which clearly suggests functional redundancy, at least in vitro. However, this does not mean that they do not have specific interactions (e.g., RSK1 is highly specific partner of S100B) In contrast, the major group consists of 12 members without a clear binding preference. The function of this orphan group on the molecular level is less defined. All the dimeric S100 proteins (with the exception of S100A10) are calcium sensors; however, if they have no additional interaction partners, which is difficult to prove, they could simply act as calcium buffers (like S100Z) contributing to calcium homeostasis of the cells [44]. Alternatively, and more likely, they can have highly specific, yet undiscovered, interaction partners. In this case, the orphan designation is only temporary and reflects a limitation of our analysis. For example, S100A10, the only S100 protein without a functional EF-hand motif, can mediate a very high affinity and rather specific interaction with annexin A2 [42]. It is still possible that there is functional redundancy within the orphan group, but our knowledge about S100 interaction partners is more limited in this group compared to the promiscuous group as no known interaction partners are available. Moreover, the present study covered only S100 homodimers (and the S100G monomer), although some S100 proteins can form heterodimers [45]. As an example, the S100A8/A9 (both coming from the orphan group) can form a functional heterodimer with known interaction partners [46].

Based on an interaction specificity map of the S100 proteins, we propose here that a more widespread functional redundancy exists in the family than previously thought. Our results provide thermodynamic evidence for possible complex formation with minimal binding segments, and further studies are needed to see whether individual complexes can be indeed formed in a particular cell type and to test that functional redundancy also exists in vivo. Regarding the possible biological relevance, the hereby defined functional redundancy can act as a potential compensatory mechanism under pathological conditions, in which the expression patterns and levels of multiple S100 proteins are altered [12].

Function-based examination of relationships within the S100ome complements phylogenetic analysis

The phylogenetic analyses of the human S100ome resulted in rather ambiguous genealogies, likely due to the young age of the protein family (Fig. 4B). Nevertheless, the clade including S100A2, S100A3, S100A4, S100A5, and S100A6 was supported with high statistical values in all analyses (Fig. 4B) similarly as it had also been found by others [29, 30]. Our functional analysis has revealed that all members of this clade belong to the same subset of the promiscuous group, with a greatly similar functional profile. However, the phylogeny of the rest of the S100ome is supported with lower statistical values. Therefore, we suggest that in such scenarios, function-based phenotypic clustering can complement or even exceed the information obtained from pure sequence-based phylogenetic analysis [47]. In our case, the S100 family can be divided, relatively unambiguously, into two bigger clusters (Fig. 5), thus giving a more robust classification. Mapping the specificity and clustering of the S100ome contribute to the better understanding of this vertebrate-specific Ca2+-binding protein family. An implication of the functional redundancy defined hereby is a possibility that a function-based combinatorial theranostic strategy may be more effective than detecting individual proteins of the S100 family.

Materials and methods

Expression and purification of S100 proteins

Protein preparations were done as described previously [48]. Briefly, the cDNAs of S100 proteins were cloned into a modified pET15b expression vector. All protein constructs were expressed in Escherichia coli BL21 (DE3) cells (Novagen, Kenilworth, NJ, USA) with a tobacco etch virus (TEV)-cleavable N-terminal His₆-tag, and purified by Ni²⁺ affinity chromatography. The His₆-tag was cleaved by TEV protease, which was followed by either hydrophobic interaction chromatography, ion-exchange chromatography, or size-exclusion chromatography with applying standard conditions [48]. The quality of the recombinant proteins was checked by SDS/PAGE analysis.

Expression and purification of kinases

The kinase domains, RSK1-CTKD (411–735), MK2 (1–400), and MNK1 (1–465), were cloned into a variant pGEX expression vector. The kinase domains were expressed in E. coli ROSETTA (DE3) cells (Novagen) with TEV-cleavable N-terminal GST and a noncleavable C-terminal His₆-tag. The recombinant proteins were purified using Ni²⁺ and GST affinity purification. The quality of the kinase domains was checked by SDS/PAGE analysis. FP measurements were performed without cleavage of the GST-tag.

Expression and purification of recombinant peptides

The peptides FOR20 (1–48), FOP (1–48), p53 (1–60; 17–53), NMIIA (1894–1937), C-ERMAD (516–560 and 516–586), and RSK1 (696–735 and 689–735) were expressed in E. coli BL21 (DE3) cells (Novagen) with TEV-cleavable N-terminal GST-tag, and purified by GST affinity chromatography. The tag was cleaved by TEV protease. After cleavage, the TEV protease and GST-tag were eliminated by heat denaturation and centrifugation. The supernatant was purified by RP-HPLC using a Jupiter 300 Å C₅ column (Phenomenex, Torrance, CA, USA). The quality of the expressed peptides was checked by mass spectrometry (Bruker Daltonics, Billerica, MA, USA).

Peptide synthesis

The CapZ (265–276), NCX1 (254–265), SIP (188–202), TRPM4 (129–147), and MDM4 (25–43) peptides were chemically synthesized using solid-phase peptide synthesis with a PS3 peptide synthesizer (Protein Technologies, Tucson, AZ, USA) with Fmoc/tBu strategy in the case of 5(6)-carboxyfluorescein-labeled and 5(6)-carboxyfluorescein-unlabeled version. Peptides were purified by RP-HPLC using a Jupiter 300 Å C₁₈ column (Phenomenex). The quality of the peptides was monitored by HPLC-MS (Agilent, Santa Clara, CA, USA).

Determination of concentrations

Concentrations of peptides and proteins were determined by UV spectrophotometry using the absorbance of Tyr and Trp residues. In the absence of these aromatic residues, the concentrations were calculated by using the absorbance of the compound on 205 and 214 nm [49, 50].

Fluorescent labeling

Chemically synthesized peptides (CapZ, NCX1, SIP, TRPM4, and MDM4) were labeled with 5(6)-carboxyfluorescein (Sigma-Aldrich, St. Louis, MO, USA) at the N terminus at the end of the synthesis. The recombinant peptides (p53, NMIIA, and RSK1) were labeled with fluorescein isothiocyanate (Sigma-Aldrich) at an N-terminal Cys residue using the protocol described previously [48]. C-ERMAD was labeled by Alexa Fluor 568 C₅ maleimide (Molecular Probes, Eugene, OR, USA) [19]. The excess labeling agent was eliminated by using HiTrap Desalting column (GE Healthcare, Little Chalfont, UK). The labeled peptides were further purified and separated from the unlabeled peptides by RP-HPLC using a Jupiter 300 Å C₅ column (Phenomenex). The concentration of fluorescent peptides and the efficiency of labeling were determined by measuring the absorbance of the fluorescent dye and the peptides.

FP measurements

Fluorescence polarization was measured with a Synergy H4 plate reader (BioTek Instruments, Winooski, VT, USA) by using 485 ± 20 nm and 528 ± 20 nm, and 530 ± 25 nm and 590 ± 35 nm band-pass filters (for excitation and emission, respectively) in cases of fluorescein-based (former) and Alexa Fluor 568-based (latter) measurements. In direct FP measurements, a dilution series of the S100 protein was prepared in 96-well plates (96-well skirted pcr plate, 4ti-0740; 4titude, Wotton, UK) in a buffer that contained 150 mm NaCl, 20 mm HEPES pH 7.5, 1 mm CaCl₂, 0.5 mm TCEP, 0,01% Tween-20, and 50-nM fluorescent-labeled peptide (probe). The volume of the dilution series was 50 µL, which was later divided into three technical replicates of 15 µL during transferring to 384-well microplates (low binding microplate, 384 well, E18063G5; Greiner Bio-One, Kremsmünster, Austria). In total, the polarization of the probe was measured at eight different S100 concentrations (whereas one contains no S100 protein and corresponds to the free peptide). In competitive FP measurements, the same buffer was supplemented with S100 proteins to achieve a complex formation of 60–80%, based on the titration. Then, this mixture was used for creating a dilution series of the competitor (e.g., unlabeled peptide or purified protein) and the measurement was carried out identically as in the direct experiment. Competitive FP measurement was executed if the fitted K_d value originated from the direct FP titration was below 200 µm. Table 3 shows the peptides used for direct and competitive FP measurements. The typical experimental window of an S100 interaction was found to be around 100 mP (polarization). However, some direct titration caused marginally small change in the polarization signal (10–30 mP) that we decided not to analyze further.

Table 3. Peptides used in this study.

Name	Region	Sequence	Modification
fp53	p53 (17–56)	GSCETFSDLWKLLPENNVLSPLPSQAMDDLMLSPDDIEQWFTE	Fluorescein isothiocyanate
fRSK1	RSK1 (689–735)	GSCQDLQLVKGAMAATYSALNSSKPTPQLKPIESSILAQRRVRKLPSTTL	Fluorescein isothiocyanate
fNMIIA	NMIIA (1894–1937)	CRKLQRELEDATETADAMNREVSSLKNKLRRGDLPFVVPRRMARK	Fluorescein isothiocyanate
fMDM4	MDM4 (25–43)	NQVRPKLPLLKILHAAGAQ	N-terminal carboxyfluorescein
fCapZ	CapZ (265–276)	TRTKIDWNKILS	N-terminal carboxyfluorescein
fNCX1	NCX1 (254–265)	RRLLFYKYVYKR	N-terminal carboxyfluorescein
fTRPM4	TRPM4 (129–147)	VLQTWLQDLLRRGLVRAAQ	N-terminal carboxyfluorescein
fSIP	SIP (188–202)	SEGLMNVLKKIYEDG	N-terminal carboxyfluorescein
fC-ERMAD	C-ERMAD (516–560)	GSCKRITEAEKNERVQRQLLTLSSELSQARDENKRTHNDIIHNENMRQG	Alexa Fluor 568 C5 maleimide
p53	p53 (1–60)	GSMEEPQSDPSVEPPLSQETFSDLWKLLPENNVLSPLPSQAMDDLMLSPDDIEQWFTEDPGP	None (free N and C terminus)
NMIIA	NMIIA (1894–1937)	YRKLQRELEDATETADAMNREVSSLKNKLRRGDLPFVVPRRMARK	None (free N and C terminus)
MDM4	MDM4 (25–43)	NQVRPKLPLLKILHAAGAQ	None (free N and C terminus)
CapZ	CapZ (265–276)	TRTKIDWNKILS	None (free N and C terminus)
NCX1	NCX1 (254–265)	RRLLFYKYVYKR	None (free N and C terminus)
TRPM4	TRPM4 (129–147)	VLQTWLQDLLRRGLVRAAQ	None (free N and C terminus)
SIP	SIP (188–202)	SEGLMNVLKKIYEDG	None (free N and C terminus)
C-ERMAD	C-ERMAD (516–586)	GSCKRITEAEKNERVQRQLLTLSSELSQARDENKRTHNDIIHNENMRQGRDKYKTLRQIRQGNTKQRIDEFEAL	None (free N and C terminus)
FOR20	FOR20 (1–48)	GSMATVAELKAVLKDTLEKKGVLGHLKARIRAEVFNALDDDREPRPSLSH	None (free N and C terminus)
FOP	FOP (1–48)	GSYAATAAAVVAEEDTELRDLLVQTLENSGVLNRIKAELRAAVFLALEEQ	None (free N and C terminus)

Fitting of FP data

The K_d of the direct and competitive FP experiment was obtained by fitting the measured data with quadratic and competitive equation, respectively [7]. For automatic fitting, we used an in-house developed, Python-based program, called ProFit, which is freely available from GitHub. The program is capable to process multiple experimental data at once, evaluate direct competitive experimental data series pairs, and estimate the variance of the deduced parameters (e.g., dissociation constants) through a Monte Carlo approach. It produces ready-to-use figures for publications, as well as a report sheet for evaluation.

ITC measurements

Titrations were carried out either at 310 or at 298 K in a buffer containing 150 mm NaCl, 20 mm HEPES pH 7.5, 1 mm CaCl₂, 0.5 mm TCEP, using a MicroCal PEAQ-ITC instrument (Malvern Panalytical, Malvern, UK). The acquired data were fitted by PEAQ-ITC analysis software using the model ‘One Set of Sites’ for most of the experiments; however for S100B-CapZ and S100B-MDM4, this model provided unsatisfactory fits and the model ‘Two Sets of Sites’ were applied instead. Note that we used the minimal interacting region (696–735) of RSK1 instead of the larger construct (689–735), which was used in the direct FP assay.

Bioinformatics analysis

For the phylogenetic analysis, the human S100 protein sequences were aligned using clustalw [29] (gap open penalty 10 and gap extension penalty 0.1 for pairwise alignment; gap open penalty 10 and gap extension penalty 0.2 for multiple sequence alignment, BLOSUM weight matrix), muscle [51], and prank [51] algorithms. Gaps were replaced by ambiguous residues (question marks) before the beginning and after the end of each sequence in the raw sequence alignment to avoid the overinterpretation of the highly variant tail extensions in the further analysis. Phylogeny was conducted with raxml gui [52]. Evolutionary history was inferred using maximum-likelihood algorithm with ProtGamma and LG as substitution model and substitution matrix, respectively [53], with 10 runs and 1000 bootstrap replicates. For the mapping of functional relationships and clustering, the dendrogram from the S100ome data set was constructed using the unweighted pair-group method with arithmetic average (UPGMA) method [32] based on the Euclidean distance using the past software [54].

Acknowledgements

We would like to thank Dániel Knapp from the Department of Plant Anatomy, ELTE, for the help in bioinformatics analyses. We would also thank Gitta Schlosser and Dávid Papp from the Department of Analytical Chemistry, ELTE, for the help in mass spectrometry. This work was supported by the National Research Development and Innovation Fund of Hungary (K119359 to LN K120391 to JK). MAS and GG were supported through the New National Excellence Program of the Hungarian Ministry of Human Capacities (ÚNKP-18-2 and ÚNKP-18-3, respectively). We also acknowledge the FIEK_16-1-2016-0005 and VEKOP-2.3.3-15-2016-00011 grants and Project No. 2018-1.2.1-NKP-2018-00005 implemented with the support provided from the National Research, Development and Innovation Fund of Hungary, financed under the FIEK_16, VEKOP-2.3.3-15-2016, and 2018-1.2.1-NKP funding schemes, respectively. The work was completed in the ELTE Institutional Excellence Program supported by the National Research, Development and Innovation Office (NKFIH-1157-8/2019-DT). GG was supported by the Post-doctorants en France program of the Fondation ARC.

Conflict of interest

The authors declare no conflict of interest.

Author contributions

MAS carried out the experiments, analyzed the experimental data, and wrote the paper. GG and LN supervised the research, analyzed the data, and wrote the paper. PE, JK, GMK, ÁLP, and AR contributed by carrying out protein and peptide expression, ITC experiments, and bioinformatics analyses.