A Potential Off-Target Effect of the Wnt/β-Catenin Inhibitor KYA1797K: PD-L1 Binding and Checkpoint Inhibition

Small molecule inhibitors of the PD-1/PD-L1 immune checkpoint are actively searched. Such compounds would complement the armamentarium against cancer, providing orally active immune regulators to treat advanced cancers [1]. At present, the therapeutic arsenal is made exclusively of monoclonal antibodies (mAbs) targeting PD-1 or PD-L1 [2]. Thirteen mAbs have been registered thus far, targeting PD-1 (ten) or PD-L1 (three), but there is no approved small molecule capable of blocking the checkpoint. Numerous compounds have been identified, principally directed against PD-L1, but none has demonstrated a marked anticancer action in humans. Only a very few PD-1 or PD-L1 small molecule binders have been advanced to clinical trials, such as compounds CA-170 (but it is a dual antagonist of PD-L1 and VISTA) [3, 4] and MAX-10181 (from Maxinovel Pty., Ltd., China) which is a PD-L1 binder currently undergoing phase 1 clinical trial in patients with advanced solid tumor (NCT0412233) [5]. New small molecules targeting PD-L1 are regularly discovered, such as our pyrazolone-type PyrDLones [6] and the pro-drug YPD-30 which was shown recently to bind to PD-L1 and to exhibit significant antitumor activity in vivo [7]. Many other compounds could be cited, all in early preclinical development [8, 9]. These newer molecules remain far from clinical trials at present. Efforts must continue to implement new compounds. Oral medication targeting PD-L1 could be extremely useful to reach tumors not easily accessible to mAbs, such as brain tumors, and to reduce costs of treatment.

Various approaches have been proposed to screen for PD-L1 inhibitors, using experimental and/or computational methods [10]. The knowledge of the tridimensional structure of PD-L1 has greatly helped the identification of compounds capable of binding to the interface of two head-to-head PD-L1 monomers [11‒13]. The structural information has been largely exploited to identify new binders, based on biophysical and computational approaches [14, 15], and for the design of PD-L1 binders containing a biphenyl core, such as the compounds initially discovered by Bristol-Myers Squibb like BMS-202 and analogs [16]. There are now different classes of PD-L1 inhibitors, with varied structures and mechanisms of action [17]. For our part, we have exploited two complementary approaches. On the one hand, we set up a biophysical method based on microscale thermophoresis (MST) to characterize drug-PD-L1 interaction and to guide the design of new binders [6, 18, 19]. On the other hand, we set up a molecular docking approach, based on a deconvolution of the biphenyl scaffold originally proposed by BMS, to identify new potential PD-L1 binders [20‒22]. The two approaches contributed to the identification of novel PD-L1-binding agents. In the frame of our drug screening effort, we have now identified the compound KYA1797K as a weak PD-L1 binder. This synthetic compound, with an atypical 2-thioxo-4-thiazolidinone motif derived from the natural product rhodanine (Fig. 1), is known as a potent and selective Wnt/β-catenin inhibitor (IC₅₀ = 0.75 μM) [23, 24]. We discovered that the compound can also bind to PD-L1, although with a relatively low but non-negligible affinity. The PD-L1-binding characteristics of the compound are reported here and discussed in light of the biological properties of the compound.

The procedure used to identify PD-L1-binding small molecules has been reported previously [20, 21]. In brief, the 3D structure of PD-L1 (PDB code 5J89 [25]) was used in docking experiments performed with the GOLD software (Cambridge Crystallographic Data Centre, Cambridge, UK). The structures of the various ligands have been optimized using a classical Monte Carlo conformational searching procedure as described in the BOSS software [26]. The side chain flexibility of amino acids within the hydrophobic pocket (Y56, Y123, M115, D122 for both chains) is taken into account while computing the free energy of hydration (Monte Carlo search) and the empirical potential energy of interaction with the SPASIBA force field. Ligands are defined as flexible during the docking procedure. For each ligand, up to 30 poses that are energetically reasonable were kept while searching for the correct binding mode of the ligand. The decision to keep a trial pose is based on ranked poses, using the PLP fitness scoring function (which is the default in GOLD version 5.3 used here). In addition, an empirical potential energy of interaction ΔE for the ranked complexes is evaluated using the simple expression ΔE (interaction) = E (complex) − (E [protein] + E [ligand]). The spectroscopic empirical potential energy function SPASIBA and the corresponding parameters were used [27, 28]. Molecular graphics and analysis were performed using the Discovery Studio 2020 Client software, Dassault Systemes Biovia Corp.

The MST procedure used to quantify inhibition of PD-L1 binding to PD-1 has been described previously [18]. Briefly, MST was conducted using an NT.115 Pico MST instrument (Nano Temper Technologies GmbH) equipped with red and blue filter sets. Recombinant Human PD-L1 His-tag protein (R&D Systems, reference #9049-B7-100), diluted to 200 nM in PBS-T buffer (supplied by the vendor), was labeled with Monolith His-Tag Labeling Kit RED-tris-NTA (Nano Temper). The RED-tris-NTA dye was diluted in PBS-T to 100 nM. The mix was incubated at room temperature in the dark for 30 min. Ligands were diluted with a serial 1:1 ratio of 16 gradients. Then the labeled protein and ligands were mixed with 1:1 ratio and incubated at room temperature in the dark for 15 min. Capillaries are then filled individually and loaded into instrument. Data were acquired using high MST power and 100% LED. Data were analyzed using MO Control Software (Nano Temper). For binding affinity analysis, ligand-dependent changes in MST are plotted as Fnorm values versus ligand concentration in a dose-response curve. Fnorm values are plotted as parts per thousand (‰).

A cellular model was developed using CHO-K1 cells overexpressing the protein of interest, PD-1. The model involves the cellular expression of a PD-1-YFP fluorescent protein (fluorescent emission of acceptor yellow fluorescent protein; λ_Ex = 514 nm λ_Em = 527 nm) and a second fluorescent protein SHP-2 tagged with CFP (donor cyan fluorescent protein, λ_Ex = 433 nm, λ_Em = 475 nm). CFP and YFP are variants of the green fluorescent protein (GFP) from Aequorea victoria [29]. Src homology 2 (SHP-2) is a tyrosine phosphatase selectively recruited by PD-1 to initiate T cell inactivation [30]. To measure the fluorescence resonance energy transfer (FRET) signal, 10,000 cells co-transfected with the PD-1-SHP-2 vector were used. A Spectramax i3 (Molecular Devices®) was used for the FRET measurements, configured according to the selected fluorochromes with their excitation and emission spectra. Endpoint reading was performed at the center of the well. The reading time is 2 min for a 96-well plate. The drug was tested at various concentrations (typically from 0.5 to 400 μM), and IC₅₀ values were calculated by curve fitting using the GraphPad Prism v8 software.

Two years ago, we initiated in silico analyses of drug-PD-L1 binding with a series of molecules containing a biphenyl unit, as found in the reference products BMS-202 and BMS-1166 previously characterized as potent PD-L1 binders [14]. The biphenyl core is an essential element for the binding of the molecules at the interface of the PD-L1 dimeric structure [31]. Different biphenyl-containing molecules binding to PD-L1 were identified, such as the anti-inflammatory drugs flurbiprofen, for example [21]. The unfused bicycle can be extended, without losing affinity for PD-L1. The drug-binding pocket can accommodate longer molecules such as extended p-terphenyl compounds [32] and molecules containing pyridine or imidazopyridine rings, for example [33, 34]. These considerations prompted us to investigate a variety of linear poly-heterocyclic molecules as potential PD-L1 binders.

A molecular docking analysis was performed starting from the crystal structure of PD-L1 bound with the biphenyl molecule BMS-202 which is known to fit well into the binding pocket formed by two adjacent PD-L1 monomers [25]. The PD-L1/BMS-202 complex provides a convenient structural model to look for other potential PD-L1 binders. We selected compounds presenting two essential characteristics: (i) an extended heterocyclic scaffold with at least two or three conjugated aromatics and (ii) a known activity to modulate the PD-1/PD-L1 checkpoint. The capacity of the compounds to bind to PD-L1 was analyzed by molecular docking, using the PDB structure 5J89 as a template [25]. The docking analysis led to the identification of the small molecule KYA1797K as a potential PD-L1 binder. A model of KYA1797K bound to the PD-L1 dimer is presented in Figure 2. The compound fits into the cavity formed by the two juxtaposed PD-L1 monomers, closed on one side by the two tyrosine residues Y123 and Y56. These two residues on each monomer delimitate a pocket in which the phenyl-furan moiety of KYA1797K resides, projecting the thiazolidinone ring toward the surface of the binding pocket, in a zone more accessible to the solvent (Fig. 3). The phenyl-furan moiety of the drug occupies exactly the position of the biphenyl core of BMS-202. The 2-thioxo-4-thiazolidinone motif emerges at the entry of the cavity, positioning the propionate chain toward the external extremity of the binding pocket in an area rich in H-bond donor and acceptor groups (Fig. 4).

The complex between PD-L1 and KYA1797K is stabilized by a panel of molecular contacts, including H-bonds, van der Waals, and π-alkyl interactions (Fig. 3). In particular, the acid moiety and the 2-thioxo group are implicated in H-bonds with the amino acid triad Ala18-Phe19-Thr20. An empirical potential energy of interaction ΔE of −91.96 kcal/mol was calculated for KYA1797K binding to the PD-L1 site, comparable to the energies calculated with BMS compounds (−73.39, −94.80, and −91.31 kcal/mol with BMS-202, BMS-1001, and BMS-1166, respectively). Here the empirical potential energy function reflects the contributions brought by the interaction of atoms which are not covalently bonded, from van der Waals interactions, electrostatic interactions, and hydrogen bonds. The interaction of the 2-thioxo group with Ala18 is important to underline because this residue is known to be essential for the interaction between hPD-L1 and peptides. An A18G mutation was shown to abolish binding of a peptide to PD-L1 [35]. This A18 residue seems to also be a major contributor to the interaction of KYA1797K with PD-L1.

The ability of KYA1797K to interact with the PD-L1 protein was evaluated via a specific MST assay using a recombinant human PD-L1 protein commercially available. The results are presented in Figure 5. KYA1797K showed a dose-dependent binding to PD-L1 in solution. A standard monophasic binding curve was observed during the thermophoresis experiments, and a dissociation constant (K_D) of 59 ± 8 μM was calculated. The value is modest compared to that measured with a human anti-PD-L1 antibody (atezolizumab [Tecentriq®], K_D = 92 ± 11 nM) but highly reproducible and significant for such a small molecule. For comparison under strictly identical conditions, the anti-inflammatory compounds nitroflurbiprofen and lonazolac gave K_D values of 114 ± 29 μM and 24.3 ± 2.7 μM, respectively, whereas K_D values in the nanomolar range have been measured with specifically designed pyrazole derivatives using the same experimental procedure [6]. This in vitro assay with purified recombinant hPD-L1 confirms that KYA1797K can bind to PD-L1 but with a modest affinity. Next, we investigated the cellular consequence of the interaction.

Binding of PD-L1 to PD-1 triggers a downstream signaling cascade, initiated by the activation of the tyrosine phosphatase SHP-2 upon its interaction with PD-1. We conducted a FRET assay using a pair of fluorescent fusion proteins, SHP-2-CFP and PD-1-YFP (expressed in CHO-K1 cell), for which the emission spectrum of CFP significantly overlaps the excitation spectrum of YFP. During the CFP-to-YFP FRET, there is a quenching of the emission of the donor CFP protein that is directly related to the efficiency of energy transfer and inversely proportional to the sixth power of the distance between the donor and acceptor proteins. It is a powerful method for elucidating protein interactions in many cellular processes and their drug regulation. We used this approach to determine the effect of KYA1797K on PD-1 signaling. The addition of recombinant exogen PD-L1, which binds to cell surface PD-1, triggers a signal (fluorescence transfer) via interaction of PD-1 with SHP-2. The addition of an interaction inhibitor – in our case, a PD-L1 binder – deactivates the signal. A similar method has been developed for high-throughput screening of PD-1/PD-L1 checkpoint inhibitors [36]. As shown in Figure 6, KYA1797K induces a dose-dependent inhibition of the FRET signal. The fitting of the dose-response curve has yielded an IC₅₀ of 94 ± 4.2 μM. For comparison, IC₅₀ values of 243 μM and 43 ± 4.6 μM were calculated with nitroflurbiprofen and lonazolac, under identical conditions. Here again, this is weak compared to the effects measured with mAbs, such as a human anti-PD-1 antibody (nivolumab, IC₅₀ = 14 ± 4 nM) and a human anti-PD-L1 antibody (atezolizumab, IC₅₀ = 47 ± 11 nM). However, the effect observed with KYA1797K is highly reproducible, novel, and important to consider.

KYA1797K is a synthetic molecule obtained from the Knoevenagel condensation of a functionalized rhodanine derivative with nitrophenyl-furfural [37]. The compound is a ligand of the regulator of G-protein signaling domain of the axin protein which is essential to control the Wnt/β-catenin signaling pathway. The targeting of the axin-regulator of G-protein signaling domain with KYA1797K leads to a destabilization and degradation of β-catenin and Ras proteins, via activation of protein kinase via GSK3β. As a consequence of this drug-induced degradation process, the compound exerts major anticancer effects against colorectal cancers harboring KRAS mutations [24, 38]. This mechanism is useful to reduce or to avoid drug resistance (notably resistance to anti-EGFR therapeutics like cetuximab and gefitinib) in KRAS-mutated colorectal cancer and in non-small cell lung cancer [39, 40]. The process has been evidenced in leukemia also [41, 42], and the same mechanism can operate in gastric cancer, triple-negative breast cancer, and pancreatic cancer, reducing resistance to standard chemotherapy (FOLFOX, gemcitabine) [43‒45]. The compound has been used as a pharmacological tool to block the Wnt/β-catenin pathway in different studies [46‒49]. KYA1797K represents a prototypic inhibitor used to modulate the Wnt/β-catenin signaling pathway in various cancer models [50, 51].

In these different pharmacological studies, the compound is classically used in the 5–50 μM range for in vitro studies. It is qualified as a potent and selective Wnt/β-catenin inhibitor (IC₅₀ = 0.75 μM), but the maximal effects are usually reported using the compound at 25 μM in many cases [23, 52]. For in vivo studies with xenograft cancer models, KYA1797K was injected intraperitoneally at a dose of 25 mg/mL/day daily [43], but the pharmacokinetic properties of the compound have not been reported. Apparently, the drug is safe and no off-target effect has been reported. It is considered a highly selective Ras destabilizer. A possible contribution of PD-L1 binding to the mechanism of action of KYA1797K is not known at present. However, it is remarkable to note that KYA1797K has recently been shown to downregulate PD-L1 in colon cancer stem cells. The compound was found to inhibit glycosylation and stability of PD-L1, via inactivation of the β-catenin/STT3 signaling pathway [52]. STT3 is the subunit of oligosaccharyl transferase (OST) which harbors the catalytic center of the enzyme. The authors demonstrated that inhibition of β-catenin/STT3 by KYA1797K reduced the stability of PD-L1, thereby inhibiting immune evasion and inducing apoptosis of colon cancer stem cells [52]. No direct PD-L1 effect was mentioned, but we can now postulate that there may be a combination of indirect (axin mediated) and direct (PD-L1 dependent) effects, at least in the in vivo experiments when the drug was injected ip at a high daily dose for an extended period (KYA1797K was given at 15 μg/mouse, 6 days/week for 7 weeks). It will be interesting to investigate this option.

The approach deployed here to discover a novel PD-L1 binder can be underlined. On the one hand, the combination of a docking analysis followed with an experimental validation has proved useful to identify KYA1797K as a new PD-L1 binder. This approach can thus be recommended. On the other hand, we must admit that the docking study has largely overestimated the binding PD-L1 capacity of KYA1797K. The calculated empirical potential energy of interaction appeared unreliable, with a ΔE value largely overestimated compared to the affinity measured experimentally. The docking process helped identify the binding site, but it provided only an approximation of the free energy of binding. This is not entirely surprising as conventional structure-based molecular docking methods largely depend on the accuracy of scoring or energy functions for each pose of the protein-ligand docking complex. Other methods and models can be used to improve accuracy of the docking, such as convolutional neural network-based model, for example [53].

Beyond the compound KYA1797K itself, the study also raises novel ideas for the design of PD-L1 inhibitors. The atypical 2-thioxo-4-thiazolidinone motif of KYA1797K is not frequently used in drug design strategies and rarely found in commercially available drugs. This motif certainly warrants further investigation and could represent a novel starting point to build immune suppressors targeted against PD-L1. Recently, antiviral compounds with a similar furanyl-methylidene-rhodanine core have been described [54]. It would be interesting to investigate their capacity to bind to PD-L1 and to modulate the activity of the PD-1/PD-L1 checkpoint.

In conclusion, we have identified the β-catenin inhibitor KYA1797K as a weak PD-L1 binder capable of modulating the functioning of the PD-1/PD-L1 checkpoint. The potency of the compound is modest, but the effect is noticeable and should be kept in mind when using the compound in biological systems. The study also points to the interest of the rhodanine motif as a building element for drug design.

Ethics approval was not required for this in vitro study (no animal study, no human samples used, and no clinical investigation).

The authors have no relevant financial or nonfinancial interests to disclose.

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

All authors contributed to the study conception and design. Material preparation and data collection and analysis were performed by Xavier Thuru, Romain Magnez, and Gérard Vergoten. The first draft of the manuscript was written by Christian Bailly, and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

All data generated or analyzed during this study are included in this article. Further inquiries can be directed to the corresponding author.