Network of interactions between the tumor necrosis factor superfamily members and small S100 proteins

Abstract

Tumor Necrosis Factor Superfamily (TNFSF) comprises 20 members of membrane/soluble signaling proteins regulating cell survival, cell proliferation/differentiation, and innate/adaptive immunity. Targeting signaling of TNFSF members is used clinically to treat several autoimmune and oncological diseases, and bone loss. They and their cognate receptors are in clinical trials as targets for treatment of autoimmune, inflammatory, oncological and other diseases. Recently, some representatives of S100 family of pleiotropic calcium-binding proteins were shown to interact with TNFSF members TNF and TRAIL, thereby suppressing their activity. In this work, we explored selectivity of interactions between soluble forms of 13 TNFSF members and 21 non-fused S100 proteins using surface plasmon resonance spectroscopy. A total of 27 interactions were found between CD70, CD30L, 4−1BBL, TWEAK, APRIL, LIGHT, VEGI and AITRL and Ca²⁺-loaded forms of S100A1/A2/A4/A5/A6/A12/A16/B/P proteins, with equilibrium dissociation constants from 2 nM to 24 μM. Removal of calcium leads to disruption of the interactions. Molecular docking indicates presence of well-conserved binding sites of the both interaction partners. Mutagenesis of S100P evidences involvement of its ‘hinge’ region in binding of CD30L, VEGI and AITRL, as well as F89 residue in VEGI recognition. The revealed network of interactions is potentially important for regulation of the cellular communication mediated by TNFSF/S100 proteins, which could be exploited for targeted therapy of socially significant diseases.

1. Introduction

Tumor Necrosis Factor Superfamily (TNFSF) of cytokine-like proteins comprises at least 20 members (including TWE-PRIL [1]; 18 genes) that signal through specific subsets of 29 receptors in humans, thereby affecting fundamental aspects of cell functioning, regulating innate and adaptive immunity [2–4]. The founding member of TNFSF, Tumor Necrosis Factor (TNF) [5], is a potent pro-inflammatory protein that has proven to be one of the most successful targets for treatment of autoimmune and inflammatory diseases (rheumatoid arthritis, Crohn’s disease, psoriasis, etc. [2]), with global market for its inhibitors exceeding US$42 billion (The Business Research Company). Monoclonal antibodies to another therapeutically important TNFSF member, RANKL, are clinically used in treatment of bone loss, bone metastases, and multiple myeloma [6]. Similarly, anti-BAFF antibody is approved for treatment of active systemic lupus erythematosus and lupus nephritis [7]. A fusion protein specific to BAFF/APRIL is also registered for treatment of systemic lupus erythematosus [8]. The antibody-drug conjugates specific to receptors of CD30L or BAFF/APRIL (BCMA) are approved for therapy of lymphomas [9] and multiple myeloma [10], respectively. Other therapeutic modalities using receptors of TNFSF proteins include T cell engagers (BCMA) and chimeric antigen receptor-T cell constructs (BCMA and receptor of 4–1BBL) [11]. Recombinant TNF and circularly permuted TRAIL are approved outside of the US for treatment of soft tissue sarcoma and multiple myeloma, respectively [11]. The success of the drugs that interfere with signaling of the TNFSF members stimulates research into therapeutic potential of this protein family and their receptors, many of which are currently in clinical trials as therapeutic targets [11–13].

TNFSF members are type II transmembrane proteins (except for soluble VEGI) with an extracellular C-terminal domain, termed ‘TNF homology domain’ (THD), which has a ‘TNF-like’ fold (SCOP ID 2000041: β-sandwich, 10 strands in 2 sheets, jelly-roll [14]) [2,3]. Their proteolytic cleavage gives rise to soluble forms of TNFSF members. Both membrane-bound and soluble forms of the cytokines are able to bind to their cognate cell surface receptor(s) via the latter's cysteine-rich domain(s), forming a hexameric complex between 3 ligand molecules and 3 receptor molecules that activates downstream signaling pathways [2]. The reverse signaling is also possible [15]. Soluble forms of the receptors able to modulate activity of their cognate ligands can be produced by proteolysis or alternative splicing [2,3]. Moreover, cytoplasmic tails of the receptors may or may not contain a death domain, resulting in differences in the downstream signaling pathways. In humans, some of the 29 receptors can interact with more than one of the 20 TNFSF members, and some of the latter can recognize more than one receptor, thereby providing at least 40 individual interactions (may be as many as 194) [2,11,16]. Given the broad expression profile of TNFSF members and their receptors (natural killer cells, T, B and mast cells, eosinophils, monocytes, macrophages, dendritic, epithelial and endothelial cells, etc. [2]), they establish a vast network of communications between various cell types, which is important for homeostasis, innate and adaptive immunity. Therefore, they are involved in the processes accompanying a wide range of autoimmune, inflammatory, and oncological diseases [11–13].

Considering medical significance of the TNFSF, knowledge of the factors affecting functional activity of its members is a prerequisite for effective use and development of the therapeutic approaches based on modulation of their signaling pathways. Recently, specific members of the S100 protein family have been shown to alter activity of soluble forms of two TNFSF proteins, TNF and TRAIL, against some cancer cell lines via binding of the S100 proteins to TNF/TRAIL [17,18]. S100 proteins are an evolutionary young family of regulatory calcium-binding proteins (except S100A10, which lacks specificity to Ca²⁺) of vertebrates with broad tissue- and cell-specific expression that are widely involved in a variety of vital physiological processes [19–23]. They contain a pair of EF-hand motifs consisting of two α-helices linked by a Ca²⁺-binding loop: an N-terminal S100-specific EF-hand and a C-terminal canonical EF-hand, connected by a flexible ‘hinge’ region [24,25]. Some S100 proteins also contain distinct Zn²⁺/Cu²⁺-binding sites [26,27]. Human S100 family contains 21 small members (78–113 residues) and 4 long members (146–2849 residues) in which the paired EF-hand domain is fused to a predominantly disordered chain [28]. With the exception of monomeric S100G, S100 proteins tend to form non-covalent homo- or heterodimers (or disulfide-linked dimers, in several cases), as well as higher-order multimeric forms [29–31]. Calcium binding to S100 proteins generally leads to exposure of their hydrophobic residues for target recognition [24]. The localization of S100 proteins both inside and outside cells along with wide range of their targets, including various proteins (enzymes, transcription factors, ion channels, receptors and their ligands, etc.), lipids, nucleic acids and N-glycans, allows S100 proteins to affect numerous (patho)physiological processes [20,23,32–34]. Since most S100 proteins, including S100A subfamily, are located in the epidermal differentiation complex (human chromosome 1q21), which encodes many genes expressed in epidermal cells and is rearranged in some cancers [35,36], dysregulation of S100 proteins is associated with skin diseases and cancer [37–41]. Specific S100 proteins are also implicated in many cardiovascular, pulmonary, neurodegenerative, metabolic, inflammatory, autoimmune and other diseases [21,22,42–47]. Some of them are used in diagnostics and/or being tested as therapeutic targets [23,44,48–51].

Since structural modelling and mutational analysis (in the case of S100P-sTRAIL interaction) argue in the favor of conservatism of the S100-binding sites of TNF/TRAIL [17,18], other TNFSF members are expected to be specific to some S100 proteins. To verify this hypothesis, in the present work we examined specificity of soluble forms of 13 TNFSF members to 21 non-fused S100 proteins by surface plasmon resonance spectroscopy. The identified network of interactions between representatives of TNFSF and S100 family points out complexity of regulation of their activities, which should be taken into account in the therapeutic interventions aimed at their signaling.

2 Materials and methods

2.1 Materials

The samples of recombinant soluble forms of the human TNFSF members used in SPR experiments are listed in S1 Table in S1 File (excluding TNF, TNF-β and TRAIL). With the exception of CD70/APRIL/LIGHT/OX40L and FasL, which were produced in insect and human cells, respectively, they were produced in E. coli. Recombinant tag-free human S100A1/A2/A3/A4/A5/A6/A7/A7L2/A8/A9/A10/A11/A12/A13/A14/A15/A16 /B/P/G/Z proteins and S100P mutants F89A and Δ42–47 (lacks PGFLQS sequence in the ‘hinge’ region) were produced in E. coli as described in [52–57]. Recombinant tag-free human CaM and human PV were produced in E. coli and purified according to [58,59]. HSA was from Merck KGaA (Darmstadt, Germany). Protein concentrations were determined spectrophotometrically as described in [60] (see also S1 Table in S1 File and ref [54]).

HEPES, sodium hydroxide and hydrochloric acid were purchased from PanReac AppliChem (Darmstadt, Germany). Sodium chloride was from Helicon (Moscow, Russia). Calcium chloride, EDTA and TWEEN 20 were from Sigma–Aldrich, Inc. (St. Louis, MO, USA). Sodium acetate, EDAC, sulfo-NHS and ethanolamine were bought from Bio-Rad Laboratories, Inc. (Hercules, CA, USA). All solutions were prepared using ultrapure water (Simplicity 185 system from Merck KGaA, Darmstadt, Germany).

2.2 Affinity of TNFSF members to S100 proteins

The interactions between soluble forms of TNFSF members and S100 proteins at 25°C were studied by SPR spectroscopy using ProteOn™ XPR36 system and ProteOn GLH sensor chip (Bio-Rad Laboratories, Inc.), essentially as previously described for soluble forms of TNF and TRAIL [17,18]. TNFSF members (30–50 μg/mL in 10 mM sodium acetate pH 4.5 buffer) were immobilized on the sensor chip surface (up to 5,000–14,500 RUs) by amine coupling (EDAC/sulfo-NHS), followed by blocking of the remaining activated amine groups with 1 M ethanolamine solution. Analyte (62.5 nM to 16 μM) in a running buffer (10 mM HEPES-NaOH, 150 mM NaCl, 0.05% TWEEN 20, 1 mM CaCl₂ or 5 mM EDTA, pH 7.4) was passed through the SPR sensor for 300 s (flow rate of 30 μL/min), after which the sensor was washed with the buffer for 200–2,400 s. The ligand was regenerated with 20 mM EDTA pH 8.0 solution for 100–200 s.

The double-referenced SPR sensograms were approximated mostly using a heterogeneous ligand model:

$$\begin{array}{c}{\text{L}}_{\text{1}}+\text{A}\stackrel{k_{a\mathit{1}}}{\underset{\begin{array}{c}\hfill k_{d\mathit{1}}\hfill \\ \hfill K_{d\mathit{1}}\hfill \end{array}}{\rightleftarrows }}{\text{L}}_{\text{1}}\text{A}{\text{L}}_{\text{2}}+\text{A}\stackrel{k_{a\mathit{2}}}{\underset{\begin{array}{c}\hfill k_{d\mathit{2}}\hfill \\ \hfill K_{d\mathit{2}}\hfill \end{array}}{\rightleftarrows }}{\text{L}}_{\text{2}}\text{A}\end{array}$$ (1)

where L₁ and L₂ denote two populations of the ligand binding one analyte molecule (A); k_d and K_d correspond to kinetic and equilibrium dissociation constants, respectively. Alternatively, the SPR sensograms were described using either a one-site binding model:

$$\begin{array}{c}\text{L}+\text{A}\stackrel{k_a}{\underset{\begin{array}{c}\hfill k_d\hfill \\ \hfill K_d\hfill \end{array}}{\rightleftarrows }}\text{LA}\end{array}$$ (2)

or a two-state model:

$$\begin{array}{c}\text{L}+\text{A}\stackrel{k_{a\mathit{1}}}{\underset{\begin{array}{c}\hfill k_{d\mathit{1}}\hfill \\ \hfill K_{d\mathit{1}}\hfill \end{array}}{\rightleftarrows }}{\text{LA}}_{\text{1}}\stackrel{k_{a\mathit{2}}}{\underset{\begin{array}{c}\hfill k_{d\mathit{2}}\hfill \\ \hfill K_{d\mathit{2}}\hfill \end{array}}{\rightleftarrows }}{\text{LA}}_{\text{2}}\end{array}$$ (3)

where LA₁ and LA₂ correspond to two distinct states of the ligand-analyte complex. The simplest fitting model ensuring the lowest χ² value was used. The kinetic/equilibrium dissociation constants were calculated using ProteOn Manager™ v.3.1 software (Bio-Rad Laboratories, Inc.) for each analyte concentration and the estimates were averaged (standard deviations are shown, n = 2–9).

2.3 Structural modeling of the complexes between TNFSF members and S100 proteins

Structural modelling of the complexes of soluble forms of TNFSF members with S100 proteins was performed within rigid body approximation using ClusPro docking server [61], as described previously [54]. The tertiary structures of the TNFSF members and Ca²⁺-loaded S100 dimers were either taken from PDB [62] (including complexes with their binding partners) or predicted using AlphaFold 3 [63] (https://alphafoldserver.com, accessed 1 June 2025) in the cases where direct structural information was missing or unresolved residues were present in PDB structures (S2 Table in S1 File). For each TNFSF-S100 pair, 10 docking models were built, and the contact residues included in at least 5 models were considered as residues of the probable binding site. The contact residues were mapped onto the amino acid sequences of the binding partners aligned using Clustal Omega [64] (https://www.ebi.ac.uk/jdispatcher/msa/clustalo, accessed 1 June 2025). The predicted tertiary structures of the complexes were drawn with PyMOL v.2.5.0 molecular graphics system [65] (https://pymol.org, accessed 1 June 2025).

3 Results and discussion

3.1 Selectivty of the interactions between TNFSF members and S100 proteins

Screening of 21 recombinant human small S100 proteins (S100A1/A2/A3/A4/A5/A6/A7/A7L2/A8/A9/A10/ A11/A12/A13/A14/A15/A16/B/P/G/Z) for affinity to 13 recombinant soluble forms of human TNFSF members (S1 Table in S1 File) at a physiological Ca²⁺ level of 1 mM by SPR spectroscopy (TNFSF members are immobilized on the SPR sensor chip via amino groups) revealed 27 specific ligand-analyte interactions, represented as analyte concentration-dependent SPR sensograms with characteristic association-dissociation patterns (Figs 1-3). Since in most cases the dissociation phase is a combination of fast and slow processes, the SPR sensograms were mainly described by the heterogeneous ligand model (1) (Figs 1-3, Table 1), which has been previously successfully used to describe SPR data on S100 binding to soluble TNF/TRAIL [17,18]. In the several cases accompanied by lower SPR signals, the sensograms were approximated with the simplest one-site binding model (2) or the two-state model (3) (Figs 2-3, Table 1). The estimates of equilibrium dissociation constants, K_d, range from 2 nM (TWEAK-S100A12 interaction) to 24 μM (LIGHT-S100A2) (Table 1, Figs 4-5). Similar K_d values were previously reported for soluble TNF (2 nM – 0.7 μM [18]) and TRAIL (0.16-1.8 μM [17]). The range of K_d values for the TNFSF-S100 complexes overlaps with the K_d values reported for complexes of cell-bound members of the TNF receptor superfamily with their ligands (0.01 nM - 19 nM [66]) and for complexes of the S100 proteins with extracellular domains of their receptor, RAGE (2 nM - 23 μM [67]).

Fig 1. Kinetics of CD30L interaction with Ca²⁺-bound S100A1/A2/A4/A5/A6/B/P proteins or S100P F89A at 25°C, monitored by SPR spectroscopy (gray), and theoretical curves (black; see Table 1).

CD30L is immobilized on the SPR sensor by amine coupling; concentrations of the S100 proteins are indicated for the SPR sensograms. The association phase is indicated by the vertical dashed lines.

Fig 3. Kinetics of interaction of CD70 (panel A), 4−1BBL (B), TWEAK (C, D), APRIL (E-H), LIGHT (I-K), VEGI (L) with Ca²⁺-bound S100A1/A2/A5/A6/A12/B/P proteins at 25°C, monitored by SPR spectroscopy (gray), and theoretical curves (black; see Table 1).

The TNFSF members are immobilized on the SPR sensor by amine coupling; concentrations of the S100 proteins are indicated for the SPR sensograms. The association phase is indicated by the vertical dashed lines.

Table 1. The parameters of interaction between soluble forms of TNFSF members (used as a ligand) and Ca²⁺-loaded S100 proteins (analyte) at 25°C, as determined from the SPR data (Figs 1-3), within the heterogeneous ligand model (1), one-site binding model (2) (marked by *) or two-state model (3) (marked by **). See also Figs 4, 5; n/a, not applicable.

TNFSF member	S100 protein	k d1 , s -1	K d1 , M	k d2 , s -1	K d2 , M
CD70	S100A2	(7.97 ± 3.70) × 10−4	(7.63 ± 4.85) × 10 −7	(1.24 ± 0.02) × 10−2	(1.29 ± 0.24) × 10 −6
CD30L	S100A1	(1.25 ± 0.62) × 10−4	(3.05 ± 2.10) × 10 −8	(2.47 ± 1.60) × 10−2	(2.25 ± 1.54) × 10 −7
	S100A2	(3.88 ± 3.52) × 10−4	(1.29 ± 0.32) × 10 −7	(1.69 ± 1.64) × 10−2	(4.81 ± 2.94) × 10 −7
	S100A4	(6.71 ± 1.66) × 10−2	(4.43 ± 1.08) × 10 −6	(2.38 ± 1.31) × 10−3	(4.51 ± 3.34) × 10 −6
	S100A5	(1.75 ± 0.11) × 10−4	(3.53 ± 0.94) × 10 −8	(1.31 ± 0.39) × 10−2	(3.34 ± 1.82) × 10 −7
	S100A6	(8.04 ± 1.03) × 10−4	(4.64 ± 2.58) × 10 −7	(4.39 ± 0.53) × 10−2	(1.59 ± 0.48) × 10 −6
	S100B	(3.59 ± 1.09) × 10−2	(1.08 ± 0.29) × 10 −7	(6.64 ± 2.65) × 10−4	(1.95 ± 0.96) × 10 −8
	S100P	(1.27 ± 0.23) × 10−3	(3.40 ± 1.31) × 10 −7	(3.50 ± 0.48) × 10−2	(8.58 ± 1.03) × 10 −7
	S100P (F89A)	(4.60 ± 0.66) × 10−4	(1.03 ± 0.44) × 10 −7	(2.86 ± 1.18) × 10−2	(3.33 ± 0.57) × 10 −7
4−1BBL	S100A6*	(2.73 ± 0.06) × 10−2	(3.03 ± 0.16) × 10 −6	n/a	n/a
TWEAK	S100A5*	(3.29 ± 0.49) × 10−5	(7.57 ± 1.14) × 10 −8	n/a	n/a
	S100A12	(3.52 ± 2.33) × 10−4	(1.86 ± 0.17) × 10 −9	(1.71 ± 1.20) × 10−2	(1.40 ± 1.08) × 10 −7
APRIL	S100A1*	(2.48 ± 0.03) × 10−3	(3.79 ± 0.16) × 10 −7	n/a	n/a
	S100A2*	(1.22 ± 0.02) × 10−3	(6.84 ± 0.15) × 10 −7	n/a	n/a
	S100A5*	(4.85 ± 1.96) × 10−6	(1.10 ± 0.45) × 10 −8	n/a	n/a
	S100B**	(6.69 ± 2.78) × 10−2	(1.40 ± 0.70) × 10 −6	(2.77 ± 0.81) × 10−4	n/a
LIGHT	S100A2	(2.71 ± 0.13) × 10−1	(2.38 ± 0.76) × 10 −5	(1.81 ± 0.05) × 10−3	(1.41 ± 0.06) × 10 −6
	S100A5	(5.79 ± 0.76) × 10−5	(6.44 ± 0.97) × 10 −8	(1.81 ± 0.67) × 10−2	(1.07 ± 0.47) × 10 −6
	S100B	(3.19 ± 1.20) × 10−4	(1.21 ± 0.80) × 10 −7	(1.07 ± 0.34) × 10−2	(1.11 ± 0.76) × 10 −7
VEGI	S100P*	(1.23 ± 0.06) × 10−2	(1.11 ± 0.14) × 10 −5	n/a	n/a
AITRL	S100A1*	(7.12 ± 3.34) × 10−4	(3.98 ± 1.50) × 10 −7	n/a	n/a
	S100A2	(5.76 ± 1.53) × 10−4	(6.29 ± 3.22) × 10 −7	(4.08 ± 2.80) × 10−2	(4.60 ± 2.79) × 10 −6
	S100A4	(1.49 ± 1.17) × 10−3	(1.22 ± 0.71) × 10 −5	(4.95 ± 3.20) × 10−2	(1.07 ± 0.43) × 10 −5
	S100A5*	(2.75 ± 0.86) × 10−4	(4.09 ± 3.22) × 10 −7	n/a	n/a
	S100A6	(1.66 ± 0.52) × 10−3	(1.59 ± 1.21) × 10 −6	(6.16 ± 1.18) × 10−2	(3.77 ± 2.75) × 10 −6
	S100A16	(4.15 ± 0.10) × 10−5	(4.43 ± 2.31) × 10 −8	(5.09 ± 0.05) × 10−3	(2.98 ± 2.48) × 10 −7
	S100B*	(6.26 ± 0.34) × 10−3	(1.20 ± 0.12) × 10 −5	n/a	n/a
	S100P	(4.75 ± 1.47) × 10−4	(6.75 ± 0.87) × 10 −7	(4.98 ± 1.63) × 10−2	(3.50 ± 1.66) × 10 −6
	S100P (F89A)	(9.54 ± 1.36) × 10−4	(6.47 ± 2.51) × 10 −7	(1.18 ± 0.23) × 10−1	(7.46 ± 1.92) × 10 −6

Fig 2. Kinetics of AITRL interaction with Ca²⁺-bound S100A1/A2/A4/A5/A6/A16/B/P proteins or S100P F89A at 25°C, monitored by SPR spectroscopy (gray), and theoretical curves (black; see Table 1).

AITRL is immobilized on the SPR sensor by amine coupling; concentrations of the S100 proteins are indicated for the SPR sensograms. The association phase is indicated by the vertical dashed lines.

Fig 4. The free energy changes accompanying the binding of Ca²⁺-loaded S100 proteins to TNFSF members at 25°C, calculated from the SPR estimates shown in Table 1: ΔG_i = −RT ln(55.3/K_di), i = 1,2.

The K_d values are indicated by the horizontal dotted lines. See also Fig 5.

Fig 5. (A) Heat map of the highest affinity of soluble forms of TNFSF members for S100 proteins calculated from the SPR spectroscopy data (Table 1, Fig 4) and reported for soluble TNF/TRAIL [17,18]. The lowest ΔG value (kJ/mol) is given for each S100-cytokine interaction. (B) Correlation between the promiscuity parameter of S100 proteins [72] (columns) and number of TNFSF members specifically recognized by a S100 protein according to SPR data (filled circles).

The columns for “promiscuous” and “orphan” S100 proteins [73] are colored gray and white, respectively.

Note that conditions of the SPR experiments favor monomerization of the TNFSF members, whereas formation of active signaling complexes with their cognate receptors involves trimerization of the ligand [2]. However, monomeric forms of TNFSF members are presumably of physiological significance, given that the prototypical TNFSF member, TNF, is known to exist in a monomeric form at physiologically relevant picomolar levels [18,68]. The previous study of stoichiometry of the interaction between soluble TRAIL and S100P using intrinsic fluorescence and chemical crosslinking revealed a complicated picture, namely, presence of multiple equilibria between several complexes differing in the molar ratio of S100P to sTRAIL (1:3, 1:1, and 2:1, respectively [17]). Nevertheless, the resulting estimates of the apparent equilibrium dissociation constants of the complexes were in qualitative agreement with the SPR estimates.

Comparison of the K_d values for the complexes between TNFSF and S100 proteins with their maximum serum/plasma concentrations reported in literature and found in the PeptideAtlas project (Fig 6) indicates a potential interaction in vivo in the cases of CD30L binding to S100A1/A2/A5/B, TWEAK-S100A12 and APRIL-S100A5. It is noteworthy that comparison of the blood concentrations of the both reagents in these cases (excluding the interactions with S100A5, for which this information is missing) shows that both situations are possible: an excess of S100 protein over a cytokine (TWEAK-S100A12 interaction) and vice versa (CD30L-S100A1) (Fig 6). Therefore, it is expected that the extracellular S100 proteins may affect activity of the TNFSF members, as exemplified by S100-induced suppression of cytotoxicity of soluble TNF/TRAIL toward human cancer cell lines [17,18], and vice versa. Importantly, local extracellular concentrations of the TNFSF/S100 proteins in tumors or inflamed/damaged tissues are expected to exceed those in the blood, thereby favoring their interactions.

Fig 6. The concentration ranges of the interacting TNFSF (panel A) and S100 (B) proteins in serum/plasma under normal (blue bars) and pathological (red bars) conditions, according to the immunoassay data found in literature and the PeptideAtlas project (Human Plasma 2023−04 build [76]) (see Supplementary materials SM1).

‘ND’ denotes lack of the data.

The fact that about 10% of the tested cytokine-S100 interactions were positive could be related to notable affinity of EF-hand motifs of S100 proteins for cytokines. This possibility is not supported by the SPR data showing no specific binding of the TNFSF members to recombinant human CaM (S1 Fig in S1 File), a prototypical member of EF-hand superfamily with four EF-hand motifs that recognizes dozens of protein targets [69]. Meanwhile, excess Ca²⁺ is a prerequisite for efficient interaction of the TNFSF members with S100 proteins, since replacement of Ca²⁺ with EDTA [70] leads to disappearance of the SPR effects (S1 Fig in S1 File), which allowed regeneration of the SPR sensor by passage of 20 mM EDTA pH 8.0 solution. The strong calcium dependence of these interactions is in line with previous reports on S100 binding to soluble forms of TNFSF members TNF/TRAIL [17,18] and four-helical cytokines [33,52–54,57,71].

The network of conformation-dependent interactions of TNFSF members with S100 proteins revealed in this work and previously [17,18] is shown in Fig 5A. It includes 31 interactions between ten TNFSF members and eleven S100 proteins with equilibrium dissociation constant values from 2 nM to 24 μM. These interactions are unique due to the considerable evolutionary distance within each group of the binding partners: pairwise sequence identities for the TNFSF members and the S100 proteins are 13%−29% and 25%−61%, respectively. The cross-reactivity of S100 proteins has previously been quantified by introduction of the so-called “promiscuity parameter” [72]. The number of TNFSF members bound per a S100 protein correlates well with the promiscuity parameter (Fig 5B), thus confirming the intrinsic ability of S100 proteins to share protein targets. Notably, some of the S100 proteins that are considered “orphans” [73] (prone to selective target recognition), namely S100A12 and S100A13, exhibit highest affinity for a very limited set of the TNFSF members (TNF/TWEAK; Fig 5), indicating high selectivity and potential physiological significance of these interactions. This observation in the case of TWEAK-S100A12 interaction is consistent with the fact that serum/plasma concentrations of the both reagents (Fig 6) are close to or intersect with the dissociation constant of their complex of 2 nM (Table 1).

Examination of IntAct [74] and BioGRID [75] databases (accessed on May 20, 2025) shows different propensities of the TNFSF members specific to S100 proteins to bind extracellular soluble proteins (the data mainly from large-scale interactome studies): TRAIL and VEGI have no such interaction partners, TWEAK, APRIL, LIGHT and AITRL have 2−4 binding partners, whereas CD70/4–1BBL/CD30L/TNF have 6/11/16/21 interaction partners, respectively. Of these, AITRL and CD30L interact most frequently with S100 proteins (8 and 7, respectively – see Fig 5A), which correlates with a higher propensity to recognize multiple proteins in the case of CD30L. Thus, the ability to CD30L to bind numerous S100 proteins appears to be due to its promiscuous nature, while AITRL is equally prone to recognition of S100 proteins despite its much higher selectivity of protein binding.

Analysis of the binders of the S100-specific TNFSF representatives found in IntAct/BioGRID databases, excluding S100 proteins, shows that only 3 binders contain EF-hands (UniProt IDs Q9H4F8, Q14515 and Q08629). Furthermore, CD30L and LIGHT bind three cytokines, sharing two of them, Bone morphogenetic protein 10 and C-X-C motif chemokine 9, suggesting a possible link between S100 proteins and signaling of these cytokines.

3.2 Structural modelling of the complexes between TNFSF members and S100 proteins

Considering low level of amino acid sequence identity within the interacting representatives of TNFSF (13%−29%) and S100 family (25%−61%), their cross-reactivity (Fig 5A) needs a structural explanation. Since calculated pI values of the interaction partners are in the range of 4.5-7.1 (S100) and 6.1-9.8 (TNFSF members, S1 Table in S1 File), their interaction at pH 7.4 cannot be due solely to electrostatic factor. To elucidate structural determinants of this phenomenon, models of the complexes between TNFSF members and Ca²⁺-loaded S100 dimers were constructed in the rigid body approximation using ClusPro docking server [61]. The predicted contact residues of S100 proteins (S3 Table in S1 File) are predominantly located at the N- and C-termini, in helices α1 and α3, C-terminal half of helix α4, and in the ‘hinge’ region between helices α2 and α3 (Fig 7A). Notably, these regions of S100 proteins are often involved in recognition of their targets [77], including RAGE [78] and many four-helical cytokines [55,57]. For the TNFSF members, the predicted contact residues (S3 Table in S1 File) are mainly located at the N-terminus, in β-strands A, A’, С, F and H (Fig 7B). Note that some residues of the β-strands A’, С and F are involved in receptor binding [3], suggesting that S100 binding to the TNFSF representatives may affect signaling of the latter’s. Importantly, despite noticeable variation in positions of the contact residues in the aligned sequences of the both interaction partners, locations of their binding sites are well conserved, as illustrated by the models of the least stable (AITRL-S100B) and most stable (TNF-S100A12) complexes (Table 1, Fig 4) presented in Fig 8, and the models of AITRL complexes with eight S100 proteins (S2 Fig in S1 File). Meanwhile, the modeling of S100A2 complexes with five TNFSF members shows that the latter can bind to S100A2 from the opposite sides of its dimer (S3 Fig in S1 File). In fact, the rigid body modelling may underestimate number of the contact residues, since it does not account for possible conformational changes during the interactions. Nevertheless, this approach is valuable for assessment of spatial arrangement of the interacting molecules.

Fig 7. The predicted contact residues (orange) of the TNFSF-S100 complexes mapped onto the amino acid sequences of the binding partners aligned with Clustal Omega [64].

Panels A and B correspond to the aligned amino acid sequences of the S100 proteins (Ca²⁺-binding loops are shown in yellow) and TNFSF members (β-strands are labelled according to ref [3]), respectively. The α-helices and β-strands are marked in cyan and green, respectively.

Fig 8. The structural models of the least stable (AITRL-S100B, panel A) and most stable (TNF-S100A12, B) complexes of TNFSF members (green) with Ca²⁺-loaded S100 dimers (grey and cyan) built using ClusPro docking server [61].

The Ca²⁺-binding loops and the contact residues are shown in yellow and orange, respectively.

To experimentally validate the docking results, S100P mutants Δ42–47 (lacking PGFLQS in the ‘hinge’) and F89A (C-terminal half of helix α4), which were previously shown not to bind TRAIL [17] despite retention of their structural properties [57], were examined for their affinity to CD30L, VEGI and AITRL using SPR spectroscopy. Although residue F89 of S100P protein is predicted to bind all these cytokines (Fig 7A, S3 Table in S1 File), its replacement with alanine caused suppression of S100P affinity to them only for VEGI (S1G Fig in S1 File), whereas the affinity for CD30L and AITRL remained virtually unchanged (Figs 1-2, 4A, 4B, Table 1). Meanwhile, absence of ‘hinge’ region in S100P Δ42–47 leads to reduced affinity of S100P affinity for CD30L, VEGI, AITRL (S1B, S1G, S1H Fig in S1 File), although this effect is predicted by molecular docking only for AITRL (Fig 7A, S3 Table in S1 File). Hence, the theoretical contact residues are incompletely correct, which may be due to noticeable structural rearrangements during some S100-TNFSF interactions that are not taken into account in the rigid body approximation used for the predictions. Overall, the experimental data evidence involvement of S100P’s ‘hinge’ in binding to CD30L, VEGI and AITRL, as well as the residue F89 in VEGI recognition, similarly to the previous data on S100P binding to TRAIL [17], many four-helical cytokines [55,57], and RAGE [78]. Importantly, namely these regions of S100P (‘hinge’ and C-terminal fragment) have previously been shown to be most affected by Ca²⁺ binding [79,80], which may explain the strong Ca²⁺ dependence of these S100P interactions.

3.3 Intrinsic disorder predspositions of TNFSF members and S100 proteins

To get a potential clue of sequence-encoded features defining differential interactability of TNFSF members and S100 proteins we analyzed the intrinsic disorder predispositions of these proteins. Results of these analyses are summarized in Figs 9 and 10, respectively.

Fig 9. Comparative analysis of structural disorder in the TNFSF members interacting (A and C) and non-interacting with S100 proteins (B and D).

Aligned intrinsic disorder profiles of the interacting (A) and non-interacting cytokines (B). Disorder propensity was evaluated by PONDR^® FIT [81], and amino acid sequences were aligned using Clustal Omega multiple sequence alignment tool [82] available at https://www.ebi.ac.uk/jdispatcher/msa/clustalo?stype=protein [64]. Aligned 3D structures of the interacting (C) and non-interacting cytokines (D). 3D structural models for all proteins were generated by AlphaFold [83]. Multiple structural alignment was conducted by SALIGN web server freely accessible to the academic community at http://salilab.org/salign [84].

Fig 10. Aligned intrinsic disorder profiles of the S100 proteins capable of interaction with the TNFSF members (A) and of the non-interacting S100 proteins (B).

Based on the global disorder analysis, the cytokines interacting with S100 proteins were predicted to be a bit less disordered than the non-interacting cytokines (median average disorder scores (ADS): 0.254 vs. 0.299; median predicted percent of intrinsically disordered residues (PPIDR): 16.4% vs. 17.04%). However, the observed differences in the median values between the two groups were not significant, as evidenced by the results of the Mann-Whitney rank sum test (p = 0.704 for median ADS and p = 0.957 for median PPIDR, respectively). Figs 9A and 9B represent aligned disorder profiles of the cytokines. Since amino acid sequences of the TNFSF members are very different, reliability of their sequence alignment is questionable. However, some similarity is observed for the peculiarities of disorder distribution within the C-terminal parts of these proteins (i.e., within the TNF domain region), whereas the N-termini are completely different in this sense. Comparison of Figs 9A and 9B shows that the cytokines interacting with S100 proteins show higher variability in disorder of TNF domains than the non-interacting cytokines. In fact, the aligned intrinsic disorder profiles in the TNF domain containing C-terminal parts in the cytokines interacting with S100 proteins differ from the corresponding regions of the non-interacting cytokines. It appears that in general TNF domains of the S100-interacting cytokines are somewhat more disordered than the C-termini of the non-interacting cytokines (median ADS: 0.189 vs. 0.148; p = 0.092, based on the Mann-Whitney rank sum test) (see also Figs 9A and 9B). Some additional information supporting this notion can be retrieved from analysis of the structurally aligned sets of the cytokines interacting with S100 and the non-interacting cytokines (Figs 9C, 9D). The non-interacting cytokines align along their C-terminal TNF domains reflecting stronger similarities of these domains, whereas the cytokines interacting with S100 proteins show greater variability of their TNF domains, as they align along the long N-terminal helices and show no structural similarity in their C-terminal TNF domains.

Next, we conducted disorder analysis of the S100 proteins interacting and non-interacting with the TNFSF members. Global disorder analysis indicated that the S100 proteins interacting with the cytokines are somewhat less disordered than the non-interacting cytokines based on their median PPIDR values 27.55% vs. 30.85%, whereas median ADS values are almost indistinguishable (0.407 vs. 0.398). In the both cases, the observed difference between the two groups in the median values were not statistically significant (p = 0.193 for median PPIDR and p = 0.788 for median ADS). Fig 10 represents aligned disorder profiles of the S100 proteins and shows that the disorder profiles of the S100 proteins capable of interaction with the cytokines seems to be more similar to each other than the disorder profiles of the S100 proteins that cannot interact with the cytokines (see Fig 10A and 10B, respectively).

Overall, results of these analyses provided some interesting hints on the potential involvement of intrinsic disorder in the S100-cytokine interactions. Our data suggest that the TNF domains of the TNFSF members capable of interaction with S100 proteins are somewhat more disordered than the TNF domains of the non-interacting cytokines, but the TNF domains of the non-interacting cytokines are more similar to each other. On the other hand, the S100 proteins capable of interaction with the cytokines are somewhat more ordered than the non-interacting S100 proteins and show more similarity in their disorder profiles, suggesting that some specific disorder-related sequence peculiarities may play a role in selectivity of the S100-cytokine interactions.

4 Conclusions

Among soluble forms of the 13 representatives of TNFSF family studied in this work and 3 TNFSF members previously studied [17,18] for their specificity to non-fused S100 proteins (S1 Table in S1 File), 10 cytokines demonstrated the ability to bind up to 8 S100 proteins, with a total number of positive interactions reaching 31, and the corresponding dissociation constants estimated by SPR spectroscopy ranged from 2 nM to 24 μM (Table 1, Figs 4−5 and ref [17,18]). The idea of existence of the highly intertwined network of the cytokine-S100 interactions is illustrated by Fig 11 that assembles in one plot all the interaction data known so far. It is obvious that most of the TNFSF members (with the exception to VEGI, CD70, and 4−1BBL) are able to bind more than one S100 protein each. Similarly, most S100 proteins (with the exception to S100A11, S100A13, and S100A16) interact with more than one cytokine.

Fig 11. Protein-protein interaction network between soluble forms of the tumor necrosis factor superfamily members and small S100 proteins.

This network is assembled based on the data reported in this study and previous studies [17,18].

Overall, about 9% of the TNFSF-S100 interactions examined are positive, and ca 63% of the TNFSF members specifically bind at least one S100 protein, which is similar to the 71–73% of the four-helical cytokines shown to be specific to S100A6/S100P [33,57]. Molecular docking and partial experimental verification of the theoretical predictions evidence that S100 proteins bind TNFSF and four-helical cytokines in a similar manner, characteristic for target recognition by S100 proteins. Although we found some weak correlations between the disorder predispositions of the members of the TNFSF superfamily and S100 proteins and their capability to be involved in mutual interactions, structural reasons for the selectivity of the TNFSF-S100 interactions remain unclear, highlighting the need for direct structural studies of this aspect.

The ability of certain S100 proteins to recognize multiple TNFSF members reflects the inherent potential of S100 family to share binding targets, as demonstrated by correlation of the number of TNFSF members bound per an S100 protein with the “promiscuity parameter” [72] (Fig 5B). Meanwhile, some of the S100 proteins with lowest promiscuity parameter (namely, S100A11/A12/A13) exhibit highest affinity to TNFSF members, suggesting a possible (patho)physiological significance of these selective interactions. This assumption is supported by results of cellular studies in the case of S100A12/A13 binding to TNF [18]. Similarly, TRAIL-S100P interaction shows signs of physiological importance in the cellular assays [17] despite much higher promiscuity parameter of S100P and moderate values of the dissociation constant. This fact suggests that other TNFSF-S100 interactions may have physiological significance under certain conditions, particularly in the extracellular milieu of damaged cells.

Since the interactions of S100 proteins with TNFSF representatives characterized to date inhibit activity of the latter [17,18], it is likely that the identified TNFSF-S100 interactions serve as a mechanism for suppression of extracellular signaling, especially in the case of cell damage. Considering that the disease spectra associated with the both protein families overlap significantly (cancer, autoimmune and inflammatory diseases; see also [17,18]), the TNFSF-S100 interactions are attractive for targeted pharmacological interventions modulating these disorders. The elevated levels of S100 proteins in most cancers [38–41] may suppress activity of individual TNFSF members, which may interfere with the therapies aimed at activation of their signaling (for instance, treatment of soft tissue sarcoma and multiple myeloma by recombinant TNF and circularly permuted TRAIL, respectively [11]). Thus, targeted inhibition of such negative interactions could be a promising therapeutic option. Therefore, S100 protein levels may need to be considered upon use and development of the drugs targeting TNFSF member signaling to ensure their maximum efficacy.

Supporting information

S1 Data. The following supporting information can be downloaded at: www.mdpi.com/xxx/s1: Supplementary Materials SM1: Data for Fig 6.

(XLSX)

S1 File. Figure S1: SPR spectroscopy data on kinetics of interaction between the TNFSF members immobilized on SPR sensor by amine coupling and S100 proteins or human CaM/PV/HSA (1 μM) at 25°C, in the absence/presence of Ca²⁺; Figure S2: The structural models of complexes between AITRL and Ca²⁺-loaded S100 dimers built using ClusPro docking server; Figure S3: The structural models of complexes between Ca²⁺-loaded S100A2 dimers and TNFSF members built using ClusPro docking server; Table S1: The samples of soluble forms of TNFSF representatives studied in this work and refs [17,18] for affinity to non-fused S100 proteins and their characteristics; Table S2: The structures of the Ca²⁺-loaded S100 dimers and TNFSF members used for structural modelling of their complexes using ClusPro docking server; Table S3: The contact residues of the complexes between Ca²⁺-loaded S100 dimers and TNFSF members predicted using ClusPro docking server.

(DOCX)

Abbreviations

4-1BBL

4-1BB ligand (TNF ligand superfamily member 9)

ADS

average disorder score

AITRL

activation-inducible TNF-related ligand (TNF ligand superfamily member 18)

APRIL

A proliferation-inducing ligand (TNF ligand superfamily member 13)

BAFF

B-cell-activating factor (TNF ligand superfamily member 13B)

BCMA

B-cell maturation protein

CaM

calmodulin

CD30L

CD30 ligand (TNF ligand superfamily member 8)

CD70

CD70 antigen (TNF ligand superfamily member 7)

DR

death receptor

EDAC

N-(3-dimethylaminopropyl)-N′-ethylcarbodiimide hydrochloride

EDTA

ethylenediaminetetraacetic acid

HEPES

4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid

HSA

human serum albumin

IFN

interferon

LIGHT

TNF ligand superfamily member 14

NMR

nuclear magnetic resonance

PDB

protein data bank [62]

PPIDR

percent of intrinsically disordered residues

PV

α-parvalbumin

RAGE

receptor for advanced glycation end products

RANKL

receptor activator of nuclear factor κ-B

RU

resonance unit

SPR

surface plasmon resonance

sulfo-NHS

N-hydroxysulfosuccinimide sodium salt

THD

TNF homology domain

TNF

tumor necrosis factor (TNF ligand superfamily member 2)

TNFSF

TNF superfamily

TRAIL

TNF-related apoptosis-inducing ligand (TNF ligand superfamily member 10)

TWEAK

TNF-related weak inducer of apoptosis (TNF ligand superfamily member 12)

VEGI

vascular endothelial cell growth inhibitor (TNF ligand superfamily member 15)

Δ42–47

human S100P mutant lacking PGFLQS sequence in the ‘hinge’ region [57]

Data Availability

All relevant data are within the manuscript and its Supporting Information files.

Funding Statement

This research was funded by Russian Science Foundation, grant № 25-14-00223 to S.E.P. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.