Promiscuity Mapping of the S100 Protein Family Using a High-Throughput Local Surface Mimetic Holdup Assay

S100 proteins are small, typically homodimeric, vertebrate-specic EF-hand proteins that establish Ca 2+ dependent protein-protein interactions in the intra- and extracellular environment and are overexpressed in various pathologies. There are about 20 distinct human S100 proteins with numerous potential partner proteins. Here, we used a quantitative holdup assay to measure anity proles of most members of the S100 protein family against a library of chemically synthetized foldamers. The proles allowed us to quantitatively map the binding promiscuity of each member towards the foldamer library. Since the library was designed to systematically contain most binary natural amino acid side chain combinations, the data also provide insight into the promiscuity of each S100 protein towards all potential naturally occurring S100 partners in the human proteome. Such information will be precious for future drug design to interfere with S100 related pathologies.


Introduction
The vertebrate-speci c calcium-binding S100 protein family (termed here as the S100ome) belongs to the superfamily of the EF-hand containing proteins and consists of at least 20 core members of small (10 kDa), usually homodimeric proteins that play role in cellular regulation both intra-and extracellularly via protein-protein interactions (PPIs) in a Ca 2+ -dependent manner 1,2 . Under physiological conditions, their expression pattern is tissue-speci c and they are present usually in low concentrations 3 . However, their expression level and pattern can be altered under pathological conditions, leading to severe consequences 4 . Speci cally, elevated cellular concentrations of certain S100 proteins were observed in cancer, cardiomyopathies, in ammatory and neurodegenerative diseases 4,5 , pointing to them as potential biomarkers and/or therapeutic targets of these diseases 6 . Development of selective inhibitors have great pharmaceutical potential, but it is still challenging due to the structural similarity within the S100 family. Thus, comprehensive, and accurate mapping of the speci c S100 interactome is required for such purpose 7 . Although numerous S100 binding partners are known, they are rather restricted to a small subset of the protein family (e.g. S100B, S100A4) 1 . Therefore, a family-wide systematic screening is in need to map the speci city and a nity pro les within the entire S100 family and to identify new binding partners.
Experimental characterization of protein surfaces having shallow binding clefts is a great challenge in drug discovery; however, tools of fragment-based approaches have become e cient techniques toward identi cation of small-molecule drug candidates 8 . Mapping the binding surface of proteins can be performed with short recognition elements (i.e., small patches of the binding interface) displaying reduced structural complexity [9][10][11] . Local surface mimetic (LSM) foldamers having water-stable H14 helical conformation proved to be e cient probes for screening these shallow binding clefts of different protein targets 12 . In a high-throughput (HTP) experimental setup, the binding properties of the S100ome can be characterized comprehensively by using this LSM foldamer library. The library members are able to point two proteinogenic side chains at i and i + 3 positions toward the surface and thereby mimic the small pieces of the complementary binding interface for the tested protein. A 256-member foldamer library ( Fig  S13) was designed to cover all the possible combination pairs of 16 different proteinogenic side chains 13 . Regardless of their secondary structure, LSMs can mimic binary surface fragments with their 5 Å distances between the proteinogenic side chains resulting small surface patches rather than secondary structure mimetics. (Fig. 1A). The bulky cyclic amino acids in the helix (trans-2-aminocyclohexancarboxylic acid) shield the proteinogenic side chains introduced by β 3 -amino acid derivatives from the solvent molecules 12 .
The members of the S100 family are often regarded as rather unspeci c, promiscuous proteins 15 . Based on our recent study, the S100ome can be divided into two groups, according to binding preference against several natural S100 partners 7 . The partner preferences give a good approximation for the classi cation of S100 member with multiple partners; nevertheless, the speci city and a nity pro le of S100 proteins without a clear binding preference (orphan) are still unknown. Here we reasoned that the binding surface of the S100ome could be mapped extensively by the application of the foldamer-based library containing most natural side chain combination, which cover the general side chain preference of the S100ome by mimicking the complementary binding surface of interacting partners (Fig. 1A). In this study, we thoroughly investigated the general and unique characteristics of the binding surface of the S100ome by determining the binding a nities of the diverse H14 LSM foldamer library towards the S100 proteins in a HTP holdup (HU) assay 13,16,17 . Our experimental results revealed the binding preferences of not only S100 proteins with multiple known interactions but also S100 members lacking known interaction partners (orphans) in living organisms.

Results
Screening the binding a nities of the S100ome against the LSM library by a HTP HU assay We screened the binding a nities of the S100ome towards the LSM library by using a HTP HU assay ( Fig.   1B), in which the 256-member LSM library was divided into four sub-libraries (each containing 64 individual foldamer fragments) ( Fig S13). S100 proteins were immobilized on Co 2+ -resin through their Nterminal His 6 -tag and incubated with the foldamer sublibraries. Experimental conditions were set so that each S100 protein was in equimolar amount (64 µM) with the global concentration of the foldamer sublibrary (containing the 64 foldamer fragments in 1 µM), thus all foldameric fragments had the opportunity independently to bind to the protein target, as described previously 13 . After the co-incubation, the unbound foldamers (the ow-through fraction) were separated from the protein-foldamer complexes (resin-bound fraction). Samples were analyzed on LC-MS system, and library members were characterized quantitatively in all samples by their area under the curve (AUC) in the total ion chromatograms ( Fig S17). The AUC value of the appropriate foldameric element in the ow-through fraction was compared to a control sample (comprising all the components of the assay except the immobilized S100 protein) prepared under the same conditions. In this way, we quanti ed the fraction of each foldamer that was speci cally retained on the resin containing immobilized S100 protein. This approach allowed us to determine a bound fraction values (F B ).
We used this approach to map the binding a nities of the complete S100ome and determined the F B constants of 5120 interactions (20 S100 proteins versus the 256-member foldamer library), depicted as heat maps ( Fig. 2A, Fig S1). The binding patterns of the S100 proteins for LSMs were found to be highly diverse. Some S100 members (e.g. S100A16, S100G) displayed only weak interactions (F B < 0.2) toward the foldamers, while other family members (e.g. S100A2, S100A6) showed high propensity to bind LSM probes ( Fig S1). The pair of identical hydrophobic binding pockets in S100 homodimers 2 , created by the Ca 2+ -induced conformational changes, could be recognized by highly hydrophobic side chains with limited selectivity and the LSM library generally displayed enrichment for residues Trp, Phe, Ile and Leu. Beside the most favored hydrophobic side chains, which can often be observed in systematic libraries 18 , foldamers containing basic and polar residues were also enriched on the protein binding sites in some cases, providing useful information to increase selectivity in rational drug design (Fig. 2B). S100 family members are rather acidic proteins (pI average = 5.68 ± 0.92), therefore, basic residues (Arg and Lys) are preferred in their ligands over acidic side chains (Glu and Asp). The enrichments of positively charged residues were found signi cant in our assay for S100A1, S100A2, S100B and S100P; as these S100 family members possess the lowest theoretical pIs (4.39, 4.68, 4.52 and 4.75, respectively). It is notable that neutral polar side chains were also found preferable for some of these family members (e.g. S100A2, S100P).
The binding pattern mostly displayed a diagonal symmetry indicating a closely neutral nature of the template backbone. For some cases, the lack of the symmetric characteristics (e.g. S100A9, S100P) was observed suggesting that the two β 3 -amino acids are not interchangeable with each other, since not only the relative position of the side chains is important, but also the position of the preferred proteinogenic side chains related to the terminals.
Investigating the interactions between the S100ome and the selected foldamers by uorescence polarization HTP (and also low-throughput) measurements generally need to be validated by an orthogonal approach to eliminate experimental artifacts 19 . We selected 11 foldamers (WL, IF, WW, YF, IL, VL, TW, RF, RR, TI, TM) based on the HTP HU assays, and after resynthesizing with a uorescent label at the C-terminus (Fig S14-16, Table S1), the S100ome was tested against the labeled foldamers by direct uorescence polarization (FP) (Fig. 3A). In this assay, the association of the uorescently labeled foldamer and the S100 protein of interest is monitored, through the change in polarization of the emitted light by the uorophore upon the binding event. In direct FP, the presence of the uorophore might change the binding a nity of some foldamers. While it would have been preferable to address the binding capacity of non-labelled foldamers by competitive FP 7 , the limited solubility of the compounds and their low a nity did not allow us to set up a competitive assay. Nevertheless, we assumed that the uorophore would affect all foldamers that target the same binding site to the same degree, because the foldamer scaffold is rigid. De ning the threshold of detection at the dissociation constant of 1 mM, we identi ed 87 interactions between the selected foldamer fragments and the S100ome out of 220 possible interactions (Table 1, Fig S2-12). Table 1 The dissociation constants given as mean ± SEM of the selected foldamers and the S100ome measured by direct uorescence polarization The a nity pro le of the S100ome was depicted as a heat map (Fig. 3B), using the K d values determined in direct FP measurements tted with a quadratic binding equation by the ProFit program 7 . Based on the a nities, the S100ome can be divided into two groups. The upper group shown in Fig. 3B contains S100 proteins (S100A5, S100A2, S100A6, S100A4, S100A10, S100B, S100A1, S100P) with multiple detected interactions, which can be characterized by micromolar binding a nities. Meanwhile, the lower part of the heat map consists of members (S100A3, S100A14, S100A8, S100A11, S100Z, S100A13, S100A12, S100A16, S100A7, S100A15, S100G, S100A9) without a clear binding preference implying only a few or no partners amongst the selected foldamers. The arrangement of the S100ome shows a similar pattern compared to our recent study 7 with the S100ome using natural binding partners.
To ensure that foldamer fragments bind to the hydrophobic binding groove of S100 proteins, which opens upon binding of calcium ions, we performed FP experiments on a selected S100 protein, S100A5, in the presence of EDTA and TRTK12, an S100-binding peptide of 12 amino acids. Our results showed that S100A5 is unable to interact with either fIF or fWL in the absence of Ca 2+ . Moreover, TRTK12 peptide competed with both foldamers with a K d value like our previous results (Fig. 3C) 7 . Based on the similarity and redundancy among the S100 family, it can be assumed that all S100 proteins bind the members of the H14 foldamer library through their hydrophobic binding pocket in a Ca 2+ -dependent manner. Mapping promiscuity in the S100ome Assuming that the LSM library contains all the relevant binary combinations of amino acid side chains covering all the side chain preferences of S100 proteins, our data provide information about the binding promiscuity of each S100 member. Herein, we will refer to "promiscuity" as a parameter capturing both the broadness of exploration of the potential ligand space and the strength of binding to the recognized ligands. Based on the F B values of the HU assays, we de ne a quantitative promiscuity term that is calculated for each S100 protein by dividing its average bound fraction value against the library by the maximal F B value (Eq. 1). As the theoretical maximum of F B is unity (F B is a value between 0 and 1), promiscuity is equal to the average bound fraction. This way, interacting with only one foldamer results in a low average F B value, however, by binding to multiple members of the LSM library in a similar manner leads to a higher value of promiscuity parameter.
The determined promiscuity parameter represents the binding properties of each S100 family member against the applied foldamer library (Fig. 4). Higher values (e.g. in the case of S100A2 or S100A6) implicate a promiscuous behavior with numerous fragments to interact with, while lower values belong to S100 members (e.g. S100A7 or S100A13) with only a few, weak interactions or without a clear binding preference.

Discussion
High-throughput holdup screening with foldamer libraries is a potent tool for speci city pro ling of protein families Using the H14 LSM library, the chemical-binding preferences of the S100ome were screened effectively by the application of a HTP-HU assay, in which numerous strongly interacting foldamers were identi ed.
When considering our overall results, the quantity and quality of the selected foldamer residues were utilized to create the speci city map of the overall S100ome. The detected enrichment of the highly hydrophobic and/or aromatic residues on the interacting surface is not a unique feature of the foldamers; moreover, their side chain binding propensities are biomimetic. Certain aromatic and aliphatic amino acids (i.e. Trp/Phe/Tyr and Leu/Ile/Val) were especially favored on the binding interface and these ndings are in line with literature data from protein-protein interaction interface databases 20 . In general, selective recognition of ligands can be explained with the unique binding patterns of the protein interfaces; therefore, the side chain frequency levels can be different even for proteins having considerably similar structures. Importantly, as other foldamer libraries with different constitution (i.e. the constitutional and/or spatial conformation of the β 3 -amino acid side chains in the foldamer fragments) are available (e.g. the H12 foldamer library), the a nity of the individual S100 members towards the foldamer libraries can vary.
Therefore, it would be interesting to screen the S100ome against other foldamer libraries, which could reveal additional relationships between the S100 members through their binding properties, providing a more re ned speci city map of the family.
To validate the detected interactions, an orthogonal biophysical method, direct FP technique was used.
Importantly, the a nity pro le of the S100ome against the selected foldamers shows good correlation with the speci city map of the S100ome from our previous work using natural S100 partners 7 . While S100 proteins with multiple natural interaction partners (e.g. S100B, S100A6) are keen on binding foldamers, S100 proteins without a clear binding preference (S100G, S100A13) can barely interact with LSMs, either because they only bind to proteins or peptides presenting a different conformation, or because they do not naturally bind to proteins.
Importantly, as S100 proteins are potential therapeutic targets, therefore the concatenation of the smaller foldamer fragments screened here by the HTP HU assay might lead to highly speci c and strong ligands, paving the way to rational drug design.
The promiscuity of the full S100ome is explained using the foldamer library Promiscuity (or its complementary notion, speci city) within a protein family can hardly be de ned against natural partners, owing to still potentially unknown interactions. However, using the LSM foldamer library against the S100ome to screen the binding properties within the protein family, promiscuity can be de ned for each member against the actual library, which eventually may approximate the real, yet unde ned promiscuity pro le. The promiscuity parameter values de ned in this study for each S100 member against the foldamer library are in good correlation with previous works 7,15,21−27 . Promiscuous S100 proteins with several known cellular partners (e.g. S100A6 or S100A4) show more interactions towards the members of the foldamer library, thus displaying a higher value of their promiscuity parameter. Orphan S100 proteins without a clear intra-or extracellular binding preference (e.g. S100A16 or S100Z) exhibit less interactions with lower binding a nity, which is represented by a lower value of the promiscuity parameter.
Orphan members do not interact considerably with the members of the H14 foldamer library, thus suggesting that these S100 proteins might lack the ability to interact with proteins in the real cellular environment and rather play a role in the Ca 2+ -homeostasis 28 . It is still possible that the less promiscuous S100 proteins have highly speci c, yet undiscovered natural interaction partners that may adopt a drastically different, non-helical conformation, explaining their lack of preference for the H14 helical foldamer fragments.
While, in principle, functional redundancy within the S100ome can only be interpreted with natural partners, screening the S100ome against 'non-natural' libraries constitutes a powerful approach to draw a more detailed and re ned picture about binding properties within the family. The promiscuity of the S100 proteins observed herein against the foldamer library, may have high relevance for their actual interactome in the real cellular environment.
Synthesis and puri cation of the foldamer libraries. The foldamer libraries were synthetized and puri ed as described previously 30 . Brie y, the 256-memberd library was divided to four sublibraries (aromatic, charged, apolar, non-charged polar) containing 64 members (Fig S13). Synthesis and puri cation of labeled foldamer sequences. Individual foldamers were synthetized manually using solid-phase peptide synthesis with Fmoc strategy applying HATU as coupling agent 13 . Coupling of the 5(6)-carboxy uorescein to the ε-amino group of a Lys attached to the C-terminus of the foldamers was carried out as the last step of the synthesis. The crude foldamers were cleaved from the resin and then, the samples were precipitated in diethyl ether and puri ed by RP-HPLC (Phenomenex Jupiter C18, 250 x 10 mm). Purity was con rmed by HPLC-MS. The concentration of the foldamers was determined by UVspectrophotometry using the absorbance of 5(6)-carboxy uorescein.
Synthesis and puri cation of the peptide TRTK12. The TRTK12 peptide was synthetized as described previously 7 . Brie y, the peptide was chemically synthetized by solid phase peptide synthesis with a PS3 peptide synthesizer (Protein technologies, Tucson, AZ, USA) using Fmoc/tBu strategy, and puri ed by RP-HPLC using a Jupiter 300 Å C 18 column.
Holdup assay. Screening the interaction between the foldamer libraries and the S100ome was performed by holdup assays as described previously 30 . Brie y, S100 proteins were immobilized in a buffer containing 20 mM HEPES pH 7.5, 150 mM NaCl, 2 mM CaCl 2 , 1 mM TCEP on Co 2+ -a nity resin (~ 2 mg protein / ml resin concentration) via the N-terminal His 6 -tag followed by the addition of the foldamer libraries. After incubation, the resin was centrifuged (Pierce™ Spin Cups -paper lter, Thermo Fisher Scienti c) to separate the unbound fraction of the library. Negative controls were prepared using the procedure described above in the absence of the His 6 -tagged protein. Fluorescence polarization assay. In direct uorescence polarization assays, S100 proteins were diluted in a buffer containing 50 nM labeled foldamer, 20 mM HEPES pH 7.5, 150 mM NaCl, 1 mM CaCl 2 , 0.5 mM TCEP and 0.01% Tween20. The dilution series (50 µl) were divided into three technical repeats and transferred (15 µl) to a 384-well microplate. In competitive uorescence polarization assays, the buffer applied in direct measurements was supplemented with the S100 protein of interest to reach a saturation of 60-80%. This mixture was titrated with the competitor (i.e. the unlabeled peptide). Fluorescence polarization was measured in 8 different S100 concentrations (one of which contained no S100 protein) on a Synergy H4 plate reader using 485 ± 20 nm and 528 ± 20 nm band-pass lters for excitation and emission, respectively. The K d values were obtained by tting the data from the FP measurements with the python-based ProFit software using quadratic and competitive binding equation for direct and competitive FP, respectively 7 . The detection threshold was based on two parameters. First, we rejected all tted dissociation constants above 1 mM. Second, we also rejected all tted data where the experimental window was signi cantly lower (< 80 mP) or higher (> 350 mP), compared to other, stronger interactions of the same labeled foldamer.
Correlation between holdup and FP. The correlation between the holdup assay (F B values) and the uorescence polarization (K d values) was quantitatively described by the Pearson correlation coe cient (PCC) using the standard formula.
Calculation of promiscuity. Promiscuity, de ned as a number between 0 (no interaction with any member) and 1 (the strongest interaction with all the members in the library) was calculated for each S100 protein Author contribution MAS carried out protein expression, FP experiments, FP data analysis, calculations, and wrote the paper.
ÉB prepared the foldamers, executed the HU assays, analyzed the HU data and wrote the paper. BM, LR and EB contributed by carrying out foldamer synthesis and protein expression, HU experiments and data analysis. GG analyzed the data and wrote the manuscript. GT, TAM and LN supervised the research, analyzed the data and wrote the paper.

Competing Interest
The Authors declare no competing interest.

Supplementary information
Supplementary information on holdup data Figures Figure 1 The methodology of the high-throughput (HTP) holdup (HU) assay Panel A: S100 proteins generally prefer helical partners or disordered partners prone to helix formation upon binding (right: S100A4 -NMIIA complex 14, PDB code: 3ZWH). Foldamer fragments (left: H14 foldamer building blocks, each with two proteinogenic side chain) cannot act as perfect α-helical mimetics, as the introduction of β-amino acids results in a bulkier helix with an altered amino acid pattern compared to the α-helix. They rather act as local surface patches, which present the proteogenic amino acids to the complementary binding surface; while the hydrophobic frame exclude the solvent from the binding pocket. The S100 binding surface can accommodate multiple foldamer fragments. Panel B: His-tagged S100 proteins immobilized on Co2+-resin (left panel) are incubated with the H14 foldamer library (256 members). The unbound fraction ( owthrough) is recovered (middle panel) and the ow-through fractions are analyzed by LC-MS.

Figure 2
The interaction between the H14 foldamer library and the S100ome measured by holdup (HU) assay Panel A: The interactions between S100 proteins and foldamers were measured by a high-throughput (HTP) holdup (HU) assay, as visualized in Fig 1B. Bound fraction values were calculated based on the loss of intensity of the foldamer of interest in the ow-through fraction using eq. 1 (see materials and methods); and were depicted as a heat map in linear scale for each S100 protein. FB ranges are color coded as shown on the right. The vertical axis and horizontal axis represents the β-amino acid in the second and fth positions, respectively 12. Some S100 members favored multiple fragments (e.g. S100A6 on the left), while multiple S100 proteins did not show clear binding preference towards the foldamer fragments (e.g. S100A13 on the right). Panel B: S100 proteins exert different amino acid sidechain preference based on the HTP HU measurements. The amino acid preferences were calculated for all S100 proteins using eq. 4 and eq. 5 (see materials and methods). Propagation of uncertainty was calculated using the standard formula. FBroot values (mean ± SEM) were depicted as a bar chart. The residues with high frequency in the bound foldamers have hydrophobic properties as aromatic and aliphatic side chains are the most preferred ones. Importantly, due to the rather acidic nature of S100 proteins, acidic side chains are the least preferred among S100 proteins. It is noteworthy that in some instances polar residues are also favored (e.g. S100A2, S100A5).
Page 17/20 The interactions between the selected foldamers and S100 proteins measured by direct uorescence polarization (FP) Panel A: The interactions between the S100ome and the labeled foldamer molecules were monitored by direct uorescence polarization assay, in which the increase of the polarization (i.e. decrease of the rotation) caused by adding S100 proteins is indicative of the binding event, i.e. the association of the labeled foldamer -S100 complex is monitored. Dissociation constants were calculated by tting the anisotropy values (mP) using quadratic equation with the ProFit program 7. The dissociation constants were given as mean ± SEM. Left panel: S100A6 was added in various concentrations to the uorescently labeled foldamer (fYF) and a signi cant binding event is observed. Right panel: S100A13 was added to the same foldamer and only a minor linear increase of the polarization was noticed con rming the results obtained by the HTP HU assays. Panel B: The -lg(Kd) values of the interactions between the selected foldamers and the S100ome were depicted as a heat map. -lg(Kd) ranges are color coded as shown on the right. The speci city-map of the S100ome towards the H14 library correlate well qualitatively with the results of the HTP HU measurements; i.e. S100 proteins (e.g. S100A2, S100A5, S100A6) interacting with numerous foldamer fragments in the HU assays exhibit the same behavior in direct FP measurements, meanwhile S100 members (e.g. S100G, S100A9, S100A13) imposing fewer interaction with the H14 library in the HU assays form weak, or no bound with the selected foldamers. It is noteworthy that based on the speci city map, the S100 proteins can be divided into two groups; one with numerous detected partners (upper part) and one with few or no detected partners (lower part). Panel C: Foldamer fragments bind to the hydrophobic binding pocket of S100 proteins in a calcium-dependent manner. Left panels: examples of direct titration of fIF and fWL in the presence of Ca2+ with S100A5, respectively, showing signi cant binding. Middle panels: the same titrations in the presence of EDTA resulted in the loss of binding event for both foldamer fragments, providing evidence that the S100foldamer interactions are calcium-dependent. Right panels: Titrating the preformed S100 -labeled foldamer complex with an S100 binding peptide, TRTK12, competition between the labeled foldamer fragment and the unlabeled S100-binding peptide is observed in both cases, providing evidence that the foldamer fragments bind to the binding pocket of S100 proteins. Kd values were calculates as in Panel A using quadratic (left panels) and competitive binding equation (right panels), respectively. All the dissociation constants were given as mean ± SEM.