Selective inhibitors of the TrkC.T1 receptor reduce retinal inflammation and delay neuronal death in a model of retinitis pigmentosa

Abstract

The heterogeneity of receptor isoforms can cause an apparent paradox where each isoform can promote different or even opposite biological pathways. One example is the neurotrophin receptor TrkC. The trkC mRNA translates a full-length receptor tyrosine kinase (TrkC-FL) whose activation by the growth factor NT3 promotes neuronal survival. In some diseases, the trkC mRNA is spliced to a kinase-truncated isoform (TrkC.T1) whose activation by NT3 up-regulates tumor necrosis factor alpha (TNF-α) causing neurotoxicity. Since TrkC.T1 expression is significantly increased at the onset of neurodegeneration, we hypothesized that in disease TrkC.T1-mediated toxicity prevails over TrkC-FL-mediated survival. To study this, we developed small molecules that selectively antagonize NT3-driven TrkC.T1 neurotoxicity without compromising TrkC-FL survival. In a genetic mouse model of retinitis pigmentosa, therapeutic administration of TrkC.T1 antagonists prevents elevation of TNF-α and reduces photoreceptor neuronal death. This work demonstrates the importance of accounting for functional and structural heterogeneity in receptor–ligand interactions, illustrates chemical biology strategies to develop isoform-selective agents, validates TrkC.T1 as a druggable target, and expands the therapeutic concept of reducing neurotoxicity as a strategy to achieve neuroprotection.

Introduction

TrkC receptors are expressed in neurons (1–3), in vascular endothelium (4, 5), and in glia and microglia (6, 7). The trkC mRNA can translate a full-length protein (TrkC-FL) which has an intracellular tyrosine kinase catalytic domain and is the major TrkC isoform in normal healthy adult tissues. The trkC mRNA is alternatively spliced in exons, resulting in the deletion of the kinase domain and the gain of a short unique intracellular sequence. The resulting truncated isoform is called TrkC.T1 (8, 9).

Both TrkC-FL and TrkC.T1 share the same ligand-binding ectodomain primary sequence (8–11). Both isoforms bind the same activating ligands: the soluble growth factor neurotrophin-3 (NT3), and the membrane-bound protein tyrosine phosphatase-sigma (PTPs). The affinity of NT3 for each isoform and the affinity of PTPs for each isoform are indistinguishable (12, 13).

Activation of neuronal TrkC-FL by NT3 or by PTPs mediates “positive” signals through the tyrosine kinase catalytic activity that phosphorylates downstream signals in neurons (pPI3K, pErk1,2, pAkt), resulting in neuronal survival. In neurons, in addition to survival, NT3•TrkC-FL also promotes neuronal growth/differentiation (11, 14), and PTPs•TrkC-FL also promotes synaptic stabilization (13). In contrast, NT3 activation of TrkC.T1 expressed in glia stimulates glial tumor necrosis factor alpha (TNF-α) production to toxic levels, which causes neuronal death (10, 15). However, PTPs activation of TrkC.T1 expressed in glia does not stimulate TNF-α production (11). Together, these data indicate a signaling bias by NT3 and PTPs as agonists of each isoform, and the ability of each receptor isoform to induce different and sometimes opposite biological outcomes. This is in part due to the intrinsic signals of each isoform, to the biased activation of signaling pathways by each ligand, and to the cell types in which each isoform is expressed (14).

Early at the onset of certain neurodegenerative diseases (e.g. experimental models of glaucoma, amyotropic lateral sclerosis (ALS), and retinitis pigmentosa [RP]; and human sporadic ALS) TrkC.T1 mRNA and protein, as well as NT3, are significantly up-regulated in glia and microglia (6, 10, 11, 15, 16). This results in production of high levels of TNF-α which causes neurotoxicity. Reducing TrkC.T1 expression that up-regulates in vivo during disease prevents TNF-α up-regulation and slows down disease progression (6, 10, 11, 15, 16), demonstrating an etiological role for TrkC.T1. In this context, we proposed a strategy to differentially target TrkC-FL and TrkC.T1 therapeutically (14).

We previously reported selective agonists of TrkC-FL (that do not activate TrkC.T1). These agents promote survival signals in TrkC-FL-expressing neurons, without promoting TNF-α toxicity by TrkC.T1-expressing glia (10). Selective TrkC-FL agonists are highly effective therapeutically in vivo in a mouse model of ALS, protecting TrkC-FL-expressing motor neurons even though TrkC.T1 is up-regulated in spinal cord glia (11).

The focus of the present work is a second and complementary strategy, to develop antagonists of TrkC.T1 (that do not antagonize TrkC-FL) to address the remaining problem of TrkC.T1-mediated toxicity induced by NT3. The optimal antagonists should inhibit NT3•TrkC.T1, but not inhibit PTPs•TrkC.T1 because PTPs does not induce toxicity.

Using chemical biology strategies, we developed biased antagonists of NT3•TrkC.T1-driven neurotoxicity, that do not impact on PTPs ligand binding or activity. Antagonism is TrkC.T1 isoform specific, with no effect on TrkC-FL activation by any ligand. Computational modeling and docking suggest that the small molecules bind at the interface of the immunoglobulin-like D4–D5 subdomains of TrkC.T1, adjacent to the D5 domain, where NT3 binds but distant from the leucine-rich motifs-like D1–D2 subdomains where PTPs binds. Two structurally related antagonists are therapeutic in a mouse model of RP where TrkC.T1 and NT3 are up-regulated in Müller glia (15, 16).

This work supports the concept that accounting for the functional and structural heterogeneity of receptor–ligand interactions can improve targeted therapies, including novel neuroprotection strategies to “enhance survival” or to “reduce neurotoxicity,” and validates TrkC.T1 as a druggable therapeutic target.

Results

TrkC.T1-based screening of previously reported TrkC-FL ligands

We have reported agents displaying amino acid side chains R¹R²-R³R⁴ (17) in a β-turn conformation mimicking the i + 1 and i + 2 residues of the β-turns of neurotrophins (17–20). Agents 2c’, 2d’, and 2e’ (respectively, GT-EK, RI-TG, and SM-GK amino acid side chains) mimic a β-turn conformation and pharmacophore of neurotrophins (17–20). 2c’, 2d’, and 2e’ bind to TrkC-FL and functionally modulate TrkC-FL•NT3 activity (20–23). The structures are shown in Table 1.

However, our original work evaluated the agents only with regards to the TrkC-FL isoform and was not evaluated on the TrkC.T1 isoform. This omission arose because at that time, a differential binding of agents to the isoforms was unexpected. As, we demonstrated, it is possible to generate TrkC-FL isoform-selective agents (11), we searched for that TrkC.T1 isoform-selective agents.

2c’, 2d’, and 2e’ were re-screened on TrkC.T1 isoforms. For direct receptor-binding assays, the agents were tagged with biotin (Fig. 1A). In quantitative flow cytometry–binding assays using isogeneic transfected cells, biotinylated 2c’, 2d’, and 2e’ bind to cells expressing stably transfected TrkC-FL (confirming earlier reports; Fig. 1B), and they also bind to cells expressing stably transfected TrkC.T1 (Fig. 1C). Quantification of three independent assays standardizing vs. control parental nontransfected cells (not expressing any TrkC) showed that binding is significantly higher to TrkC-FL or to TrkC.T1-expressing cells (Fig. 1D).

Next, we studied the functional impact of these agents on the signals promoted by NT3. Agent 2d’ significantly inhibits TrkC.T1•NT3-promoted up-regulation of TNF-α mRNA, but 2c’ and 2e’ do not inhibit significantly (Fig. 2A, 2e’ is P = 0.09, Table 2). The effective antagonistic concentrations of 2d’ are nontoxic (Fig. S1A); hence, inhibition is not due to cell death. All the agents are antagonists of TrkC-FL•NT3, and lower the cell survival promoted by NT3 in serum-deprived cultures (Fig. 2B). Hence, all agents bind to both isoforms and antagonize NT3•TrkC-FL, but only 2d’ is a functional antagonist of NT3•TrkC.T1. Note that these agents are TrkC selective, because in counter assays, they did not have any effect on the survival promoted by growth factor and receptor family members NGF•TrkA or BDNF•TrkB (17).

Next, we evaluated whether 2c’, 2d’, and 2e’ antagonize PTPs. PTPs binds to TrkC at the LRR-like D1–D2 subdomains, an epitope different than NT3, which binds to the IgG-like D5 subdomain. PTPs•TrkC binding was studied by quantitative flow cytometry using the ectodomain of PTPs tagged with the Fc domain of immunoglobulin (PTPs-Fc). The agents 2c’, 2d’, and 2e’ do not block PTPs-Fc binding to either TrkC.T1 or TrkC-FL (Fig. 3A) and do not inhibit PTPs•TrkC-FL promotion of cell survival in serum-free cultures (Fig. 3B). Note that the biological action of PTPs•TrkC.T1 cannot be evaluated, because PTPs does not promote up-regulation of TNF-α (11).

In sum, 2d’ binds to TrkC-FL (reported data) and to TrkC.T1 (new data), and functionally 2d’ is an antagonist of NT3•TrkC-FL (reported data) and of NT3•TrkC.T1 (new data). Interestingly, 2d’ does not antagonize PTPs, in terms of PTPs•TrkC-FL binding or function and in terms of PTPs•TrkC.T1 binding (new data). Therefore, functionally 2d’ is a biased antagonist of NT3.

Analogs of 2d’ are selective TrkC.T1 antagonists

We screened analogs of the 2d’ pharmacophore (RI-TG) for NT3•TrkC.T1 selectivity. Analogs were coupled on a backbone in the “forward” and “reverse” orientation (e.g. R¹R²-R³R⁴ or R²R¹-R⁴R³) (17). The structures of the compounds are shown in Table 1. Class A analogs related to 2c’ (GT-EK) are 16 (KE-GT) and 38 (TG-EK). Class B analogs related to 2d’ (RI-TG) are 24 (GT-RI) and 42 (TG-IR). Class C analogs related to 2e’ (SM-GK) are 13 (KG-SM) and 45 (MS-GK). These agents were screened for antagonism of TrkC-FL and/or TrkC.T1 (Fig. 4), at concentrations that are nontoxic to cells (Fig. S1B).

Using functional assays, 16 and 42 significantly inhibit NT3•TrkC.T1 function, blocking up-regulation of TNF-α mRNA, but 24, 38, 13, and 45 do not (Fig. 4A). In the absence of NT3, none of the agents impacts the baseline TNF-α expression, further supporting the notion that the agents are not partial agonists and are not toxic. Using quantitative flow cytometry–based competition assays, 16 and 42 do not inhibit PTPs-Fc binding to TrkC.T1 or TrkC-FL (Fig. 4B), indicating that these agents are biased to inhibiting NT3 and not PTPs.

Regarding isoform selectivity, functional counter assays demonstrated no impact on NT3•TrkC-FL-induced biochemical signals (Fig. 4C, quantified in Fig. 4D) or PTPs•TrkC-FL–induced biochemical signals (Fig. 4E, quantified in Fig. 4F). Moreover, in cell survival bioassays using cells induced to die by culture in serum-free conditions, the agents do not antagonize NT3•TrkC-FL-induced survival (Fig. S2A) or PTPs•TrkC-FL-induced survival (Fig. S2B). These biological assays are consistent with the biochemical data, and further confirm that the effective concentrations that inhibit NT3•TrkC.T1 are nontoxic to cells (Fig. S1A and B).

In sum, these data show that 16 (KE-GT) and 42 (TG-IR) selectively antagonize NT3•TrkC.T1, but do not impact on NT3•TrkC-FL. Antagonism is biased to NT3 because the agents do not affect PTPs binding to either isoform or PTPs•TrkC-FL function. Moreover, the agents do not have intrinsic partial agonistic activity at TrkC.T1 (e.g. in the absence of NT3), and TrkC.T1 functional inhibition is not due to toxic effects. These are desirable pharmacological properties for a selective NT3•TrkC.T1 antagonist.

Converting a nonselective TrkC ligand into a selective TrkC.T1 antagonist

As a second chemical biology strategy for developing antagonists, we investigated analogs of a reported small molecule 1Aa (20) which binds both TrkC-FL and TrkC.T1 and is an agonist of both isoforms (11). We aimed to develop an analog of 1Aa that is selective for TrkC.T1 and acts as an antagonist.

1Aa displays IK amino acid side chains in a β-turn conformation (20). We designed cyclic analogs displaying IK on a constrained ring of different sizes. The linear counterparts (not in a β-turn conformation) were synthesized as controls (Fig. 5A). Two analogs, cyclic-14-KI (CP14) and cyclic-15-KI (CP15) (respectively, on a 14-member ring or a 15-member ring) are functional antagonists of NT3•TrkC.T1 and significantly block up-regulation of TNF-α (Fig. 5B). In NT3•TrkC-FL, counter assays for selectivity CP14 and CP15 do not affect survival-promoting activity (Fig. 5C). These agents are likely biased to NT3 antagonism because in quantitative flow cytometry competition assays, they do not inhibit PTPs-Fc binding to TrkC-FL or to TrkC.T1 (Fig. 5D) and they do not impact on PTPs•TrkC-FL survival-promoting activity (Fig. 5E). Moreover, in receptor selectivity counter assays, the cyclic analogs do not have any effect on receptor family members TrkA or TrkB that are activated by their respective ligands NGF or BDNF (Fig. S3A and B). The relevance of the β-turn structure in the agents is highlighted by the fact that their linear counterparts (not cyclized) have no impact on any assay (NT3•TrkC-FL function; Fig. 5C), PTPs•TrkC-FL binding (Fig. 5D), or NT3•TrkC.T1 function (Fig. S3C).

Together, the data indicate that CP14 and CP15 are selective NT3•TrkC.T1 antagonists, do not affect NT3•TrkC-FL, and do not affect PTPs binding to either of the TrkC receptor isoforms. These results support the concept that it is possible to develop an isoform-selective small molecule antagonist starting with a nonselective TrkC ligand 1Aa.

CP14 selectively reduces the affinity of NT3 for TrkC.T1

We tested the binding of biotin-labeled NT3 (NT3-bio) to TrkC-FL or to TrkC.T1 in flow cytometry–based assays. NT3-bio binds to the receptors in a dose-dependent manner reaching maximal binding (B_max) at ∼5–10 nM (0–80 nM concentration range was tested). Representative data are shown for NT3-bio binding (0, 0.625, 1.25, and 5 nM) on TrkC-FL and TrkC.T. Binding is specific and competed by nonbiotinylated NT3 (“cold”), where preexposure of cells to cold NT3 (10 nM) blocks >80% of the binding of NT3-bio (5 nM) (Fig. S4A and B).

Using this binding assay, we evaluated the impact of compound 16 or CP14 on the binding of NT3-bio on the TrkC isoforms (Table 3). A range of different agent:ligand stoichiometries was tested (0, 1, 5, and 15 µM agents each vs. 0–80 nM NT3-bio). Representative binding data are shown for 5 µM CP14 vs. 5 nM NT3-bio (Fig. S4C and D). From the binding constants at different stoichiometries, we determined the K_d of NT3 ± antagonist (Table 3).

NT3 binds to TrkC.T1 with a K_d 1.8 nM, and CP14 reduces the affinity in a dose-dependent manner to K_d 11.5 nM (a 6.4-fold loss in affinity). Compound 16 causes a less significant 1.7-fold reduction in affinity for TrkC.T1. In contrast, and as internal specificity control, the agents do not affect the affinity of NT3 for TrkC-FL (Table 3). The data suggest that the NT3 inhibitory mechanism of CP14 is noncompetitive and allosteric, and specific for TrkC.T1. Moreover, it is noteworthy to restate that CP14 is also biased to inhibiting NT3 and does not impact the binding of PTPs-Fc to either receptor isoform (see above Figs. 4B and E, 5C, D, and E).

Molecular docking of CP14 suggests noncompetitive and allosteric mechanism

To gain insight into the mode of action of CP14, we performed molecular docking using the 3D structure of TrkC as a target. While there are crystal structures of isolated subdomains of human TrkC, including the NT3-binding IgG-C2 (D5) subdomain (PDB 1WWC) (24), there are no experimental 3D structures of the entire human TrkC extracellular ligand-binding domains (ECD), nor for its T1 splice isoform, nor structures bound to the ligands NT3 or PTPRS. To model these structures of TrkC, we thus used AlphaFold3, a machine-learning and diffusion-based algorithm for high-accuracy prediction of biomolecular structures (25). The structure of the isolated TrkC shows that the ECD is folded, with the LRR-like and D4–D5 subdomains displaying restricted motions, as indicated by the low position error between the subdomains (Fig. S4A and B). Next, we modeled the complex of TrkC with the ligands NT3 and PTPRS. NT3 forms a dimer nearly identical to the crystal structure (e.g. PDB 1B8K) (26, 27) and binds two copies of TrkC via the D5 domain, while PTPRS binds the LRR-like subdomain, as expected (Fig. S4C and D). Next, the predicted TrkC and TrkC–NT3–PTPRS structures were used to model interactions with the small molecule CP14 using molecular docking. CP14 docked preferentially on the D5 subdomain, adjacent to the NT3-docking site in both structures (Fig. S4E and F). The docking prediction is consistent with the NT3 bias of CP14, because PTPs bind at the LRR subdomain near the ectodomain N-termini (see Fig. S4C), and it is also consistent with the experimental finding suggesting that CP14 lowers NT3 affinity in a noncompetitive allosteric manner. Additional work and an experimentally accurate 3D model of TrkC.T1 as monomer or as a homodimer would be required to confirm this.

Therapeutic efficacy of TrkC.T1 antagonists in a model of RP

For testing therapeutic efficacy in vivo in the RHOP347S transgenic mouse model of RP, we selected compound 16 from the first chemical biology approach and CP14 from the second chemical biology approach. The RHOP347S mice recapitulate an aggressive form of human RP by transgenic expression of a rhodopsin gene mutation. In RHOP347S transgenic mice (but not in wild-type [WT] mice), TrkC.T1 and NT3 are significantly up-regulated in Müller glia shortly after the mice open their eyelids at postnatal day 15 (PN 15) and retinas are exposed to light, causing increased TNF-α production, and photoreceptor neuronal death. RHOP347S mice suffer ∼50% photoreceptor death by PN 18, and nearly total photoreceptor loss by PN 28 (28). We have reported that in this mouse model a knockout of one TrkC.T1 allele or the silencing of the TrkC.T1 mRNA normalizes TNF-α production and reduces photoreceptor death (15). Hence, we investigated whether small molecule TrkC.T1 antagonists would yield comparable therapeutic results.

Initially, we used explanted organotypic cultures of retinas obtained from PN 18 RHOP347S transgenic mice or from control WT mice. The photoreceptors of RHOP347S retinal explants recapitulate neurodegeneration with the same kinetics as in vivo, whereas WT retinas remain healthy within the time frame of the experiment (15, 29). Retinal TNF-α protein was quantified by western blots, and the death of photoreceptors was quantified by confocal microscopy of TUNEL⁺ staining of the photoreceptor layer (Fig. 6).

Treatment of RHOP347S organotypic retinal cultures with TrkC.T1 antagonists 16 and CP14 significantly reduced TNF-α protein levels (Fig. 6A, quantified in Fig. 6B) and significantly decreased photoreceptor death across all areas of the retina (quantified in Fig. 6D), when compared with the control vehicle-treated RHOP347S organotypic retinas. There are no statistical differences in efficacy between antagonists 16 and CP14 in either bioassay. Control treatment with exogenous NT3 (20 nM) did not impact positively or negatively on TUNEL⁺ counts. This was expected because NT3 is already significantly up-regulated in RHOP347S retinas (15, 30). Control organotypic retinal cultures from healthy WT mice (without significant TUNEL⁺ counts) treated ± antagonists 16 or CP14 shows no affect in TUNEL⁺ counts. Hence, the antagonists do not impact positively or negatively healthy retinas that do not have detectable TrkC.T1. Representative pictures of TUNEL⁺ images in the photoreceptor layer are shown (Fig. 6E). Data are shown separately for each retinal area concentric to the optic nerve head, to account for the architecture of the retina (thicker in the central area; and thinner in the periphery, see Fig. 6C).

Based on the data resulting from organotypic culture assays, we next tested antagonists in vivo. RHOP347S mice at PN day 18 (a time point where there is already photoreceptor damage) were treated by intravitreal injection of either 16 or CP14 in one eye, and control vehicle in the contralateral eye. In this experimental paradigm, each mouse has its own internal control, which reduces mouse-to-mouse variability.

Two days after intravitreal injections, the retinas were processed for TUNEL⁺ and counts in the photoreceptor layer were quantified by area to account for the natural thinning of the retinal architecture (thicker in the central area; and thinner in the periphery, see Fig. 6C). Compounds 16 or CP14 significantly reduce photoreceptor death in areas 1 and 2 of the RHOP347S retina (Fig. 7A).

Quantification of photoreceptor nuclei represents surviving photoreceptors and is another way to evaluate reduced photoreceptor death. RHOP347S eyes treated with CP14 or compound 16 have significantly more surviving photoreceptors when compared with vehicle (Fig. 7B). Moreover, the thickness of the photoreceptor layer showed a significant preservation of the outer nuclear layer (ONL) structure in the drug-treated RHOP347S eyes when compared with vehicle-treated RHOP347S eyes (Fig. 7C). These data are with the interpretation of increased surviving photoreceptors and preservation of the structure of the ONL retinal layer where photoreceptor cell bodies reside.

Together, the data are consistent in showing reduced photoreceptor death, increased numbers of photoreceptor nuclei, and increased photoreceptor layer thickness and preservation of structures by treatment of RHOP347S mice in vivo with compound 16 or with CP14. We note that compared with WT retinas, the drug-treated RHOP347S retinas had significantly lower photoreceptor counts and ONL thickness. This was expected, because therapy was initiated at PN 18 and retinas were analyzed at PN 20; a stage when there is preexisting ∼40% photoreceptor loss in RHOP347S retinas (15, 16) which treatment does not regenerate.

Discussion

We used two chemical biology approaches to analog reported nonselective TrkC antagonists (20–23) and reported TrkC agonists (17, 31). Screening identified selective antagonists of NT3•TrkC.T1, that do not inhibit PTPs•TrkC.T1 and that do not inhibit prosurvival signals by NT3•TrkC-FL or PTPs•TrkC-FL.

Receptor selectivity and ligand bias

Some of the agents (16 and 42; CP14 and CP15) selectively antagonize TrkC.T1•NT3 by reducing ligand affinity in an allosteric manner, but do not impact on PTPs binding to TrkC.T1, and do not affect TrkC-FL activation by either NT3 or PTPs, or the function of other Trk-receptor family members. Hence, the agents are biased antagonists of NT3•TrkC.T1 (see summary in Table 2).

With regards to mechanisms, analyses of the NT3-binding constants demonstrated that CP14 reduces the affinity for TrkC.T1 but not for TrkC-FL; and the data also suggest an allosteric noncompetitive mechanism. Modeling and docking predictions also suggest an allosteric mode of binding for CP14 at the D4–D5 interface of TrkC.T1. We note, however, that a fundamental limitation of AlphaFold is that it cannot capture alternative or multiple low-energy conformations that could results from different disulfide bonding patterns. As expected, the AlphaFold-model of the receptor structure is identical between TrkC-FL and TrkC.T1, because the sequence of the extracellular region is identical; yet, the structures of the ectodomains differ as we reported that their disulfide bonding pattern differs as TrkC.T1 disulfides are rearranged by Protein Disulfide Isomerase enzymes (11). Additional work and an experimentally accurate 3D model of TrkC.T1 would be required to confirm that CP14 binds at this site, but the modeling at least suggests that it is plausible for our small molecules to bind TrkC and affect the interaction with the NT3 ligand.

Therapeutic efficacy in RP

The etiology of glial-derived TNF-α in many degenerative retinal dystrophies is well-known (32, 33). Intravitreal injection of anti-TNF-α mAb delays retinal degeneration in RP (34) and in other retinal dystrophies where TrkC.T1 is up-regulated in glia [e.g. glaucoma (6, 35, 36)]. For this reason, we posit that modulating a target receptor that induces TNF-α, such as TrkC.T1, may be relevant therapeutically in many pathologies.

In a genetic mouse model of RP where TrkC.T1 is up-regulated at the onset of disease (15, 16), two TrkC.T1 selective antagonists 16 and CP14 were applied therapeutically 1× after disease onset. The agents prevent TNF-α up-regulation and reduce photoreceptor neuronal death in vivo. These data validate TrkC.T1 as a therapeutic target and demonstrate the efficacy of an antineurotoxic strategy that results in neuroprotection.

TrkC.T1 is a naturally occurring receptor isoform that is significantly up-regulated in RP (degeneration of photoreceptors) and other neurodegenerative diseases, such as glaucoma (degeneration or retinal ganglion cells) (30), ALS (degeneration of motor neurons) (10, 15), and noise-induced neurosensory hearing loss (degeneration of inner ear neurons) (2) (7, 37–39), and in some types of cancer (40, 41). The pharmacological efficacy of TrkC.T1 inhibition in a model of RP suggests possible therapeutic options to ameliorate pathologies in these other diseases.

The efficacy of TrkC.T1 antagonists is statistically significant, and comparable with other reported experimental or applied approaches such as vitamin and nutritional supplementation (beneficial but only for a limited group of patients). This is important especially considering the early onset and rapid progression of the transgenic mouse model of RP evaluated in our work, and the fact that TrkC.T1 antagonists were applied only once in challenging therapeutic paradigms, after onset of disease and with ongoing inflammation and neurodegeneration. As comparison, gene therapy (voretigene neparvovec) specific for the RPE65 mutation causing Leber congenital amaurosis received approval from the FDA. However, hundreds of mutations are reported in the Rhodopsin gene and other genes of the visual system that can lead to RP or neuronal degeneration, making gene therapy approaches a one-at-a-time and a mutation specific.

Since TrkC.T1 is expressed in many forms of retinal dystrophies and it is etiological to inflammatory pathways upstream of the well-known target TNF-α, the TrkC.T1 target may offer therapeutic opportunities for larger patient populations. The TrkC.T1 antagonists may be effective in the retina due to a small volume of distribution and slow clearance of the vitreous; and in future work, we will test them in other sites such as intrathecally in the spinal cord in ALS.

The TrkC.T1 antagonists reported here have potential for medical translation, because they are small molecules (range 500–1,200 Da) chemically and structurally related to small molecules that have proven safe, suitable, and effective for therapy of retinal diseases (42–46) or for the experimental therapy of brain diseases requiring CNS penetration (47–50).

Chemical biology strategies for developing TrkC.T1 small molecule antagonists

Small molecule inhibitors were identified using two different chemical biology approaches combined with screening for effects on NT3•TrkC.T1, and lack of effects on NT3•TrkC-FL or PTPs•TrkC-FL, as well as lack of effect on Trk-receptor family members.

In the first approach, known antagonists of TrkC-FL were reevaluated, to yield compound 2d’ (RI-TG) as a TrkC antagonist that is not isoform specific. Two analogs of 2d’, agents 16 (KE-GT) and 42 (TG-IR), are specific antagonists of TrkC.T1, biased to antagonizing NT3•TrkC.T1 without inhibiting PTPs•TrkC.T1 binding; and not impacting NT3•TrkC-FL or PTPs•TrkC-FL activation.

A comparison of nonisoform-selective 2d’ (RI-TG) to the isoform-selective TrkC.T1 antagonist 42 (TG-IR) suggests that sequence inversion enables target specificity, at least with regards to NT3-dependent events. Consistent with this notion, related agents 2c’ (GT-EK) and 2e’ (SM-GK) antagonize NT3•TrkC-FL (also see Ref. (17)). The relevance of bivalency in these agents is shown by the fact that the monomers do not have antagonistic activity against any isoform. These results support the concept that it is possible to convert a nonselective TrkC ligand into a selective TrkC.T1 antagonist. Perhaps this strategy may be used for related scenarios or for making analogs of other small molecule ligands, though the strategy is of course not expected to be universal.

In the second approach, we converted a small molecule agonist of TrkC, 1Aa, into a selective TrkC.T1 antagonist. The analogs CP14 and CP15 significantly blocked the NT3•TrkC.T1-induced up-regulation of TNF-α, with no impact on other parameters. The importance of a β-turn displaying the IK pharmacophore is demonstrated because these agents are antagonists only in their cyclic (β-turn) configuration but are not antagonists in their linear or unstructured forms. The binding and the biological data are summarized in Table 2.

The chemical biology approaches we used for developing selective TrkC.T1 antagonists may be useful also for targeting other receptor families, where isoforms, mutations, or posttranslational modifications create an opportunity to exploit selectivity.

Conclusions

Our results support the concept that accounting for the functional and the structural heterogeneity of receptor–ligand interactions is useful to validate novel therapeutic targets and to enable new pharmacological strategies. In future work, we will explore combining the complementary approaches of direct promotion of neuronal survival by activating TrkC-FL in neurons, together with reduction of the neurotoxicity induced by glial TrkC.T1; aiming to achieve optimal neuroprotection.

Materials and methods

Cell lines

HEK 293 or NIH3T3 cells were transfected with plasmids encoding human full-length TrkC (293-TrkC-FL and NIH-TrkC-FL) or with human or rat TrkC.T1 (293-rTrkC.T1, 293-hTrkC.T1, and NIH-TrkC-T1). The rat Müller glial cell line rMC-1 (51) was transfected with plasmids encoding human TrkC.T1 (rMC-1-TrkC.T1). Stably transfected subclones that express TrkC-FL or TrkC.T1 receptors were drug selected (depending on the vector, 0.5 mg/mL G418, 2 μg/mL puromycin, or 10 μg/mL blasticidin). Expression was verified by qPCR using specific primers of each isoform (10), by flow cytometry, by IHC, and by western blots using our own in-house mAbs 2B7 (52), 1E11 (11), the mAb TrkC 1 (2), and a commercial mAb C44H5 antibody (catalog 3376; Cell Signaling). The mAb 2B7 binds only to TrkC-FL at the IgG-C2 (D5) domain (52) at an epitope conserved in mouse, rat, and human TrkC-FL. The mAb TrkC 1 binds to both TrkC-FL and TrkC.T1 (2). The mAb 1E11 is selective for TrkC.T1 ectodomain (11). For western blots, anti-p-ERK1/2 (catalog 4377S; Cell Signaling), anti-phospho-Akt (catalog 4060S; Cell Signaling), anti-actin (catalog A5316; Sigma), and anti-TNF-α (catalog 500-P64; PeproTech) were purchased.

Protein receptor ligands

NGF, BDNF, and NT3 growth factors were purchased (Alomone Labs and ProSpec-Tany). Biotinylated NT3 was prepared using Sulfo-NHS-LC-Biotin (Thermo Fisher Scientific) by incubation of 2 μM of NT3 with 400 μM of Sulfo-NHS-LC-Biotin for 40 min at room temperature in PBS (20-fold molar excess of biotin reagent). Biotinylated NT3 was concentrated, and unbound biotin was removed, using a desalting Amicon Ultra-0.5 mL centrifugal filter (Ultracell: 10 kDa; Millipore). PTPs soluble ectodomain protein fused to IgG-Fc was prepared, as described previously (13).

Synthesis and characterization of small molecules

The synthesis, purification, and characterization were reported (17, 20, 22, 31). For direct binding assays to receptor isoforms, the TrkC-FL antagonists 2c’, 2d’, and 2e’ were tagged with biotin, and analogs were tagged with polyethylene glycol as described (17). 1Aa is a reported TrkC agonist (20), and analogs were screened to find antagonists. The 1Aa analogs were synthesized as cyclic agents to mimic a β-turn conformation of neurotrophins, and as controls their linear (unstructured) forms were prepared as well. Synthesis started with a linear peptide (K-I-D) assembled on Asp(Wang resin)-OAll (loaded 0.51 mmol/g; Novabiochem) on a reaction using Fmoc-coupling chemistry, and deprotection was performed with 20% piperidine in DMF. To avoid incomplete reactions, the couplings and deprotections were repeated twice at each step. The attachment to AMPA, respectively ABMA, was done with a TBTU, followed by a HBTU as a coupling reagent. The Allyl protecting group of the Asp was removed after treatment with Pd (PPh₃)₄/PhSiH₃. The head-to-tail cyclization of the linear peptide-resin was performed with double coupling with HATU (in the case of a 14-membered ring peptide) or with PyBOP (in the case of a 15-membered cycle). The peptides were cleaved and removed from the resin by incubation in 95% trifluoroacetic acid at RT with agitation for 2.5 h. Purification was done with Varian high-pressure liquid chromatography (HPLC) apparatus and a Chromasil C18 reverse-phase column (10 × 250 mm). Buffer A was 10% ACN in water with 0.1% TFA and buffer B was 70% acetonitrile with 0.1% TFA. Gradient was buffer B from 0 to 40% in 20 min with a 3.5 mL/min flow rate. HPLC analysis of the pure peptides was done with a Phenomenex C18 reverse-phase column (4.6 × 75 mm) with similar buffers, but a gradient from 0 to 100% B in 10 min and a 0.1 mL/min flow rate.

Electrospray mass spectrometry (positive mode) was carried out with a Thermo Q Exactive at a resolution of 35,000, which confirmed the expected masses. Cyclic peptide AMBA-14-KID: MS-direct infusion (ESI): m/z [M + H]⁺ calculated for C₂₄H₃₆N₅O₆, 490.259; found 490.260. Cyclic peptide AMPA-15-KID: FTMS (ESI, +mode): m/z [M + H]⁺ calculated for C₂₅H₃₈N₅O₆, 504.274, found 504.281. Cyclic peptide AVA-15-OID: MS-direct infusion (ESI): m/z [M + H]⁺ calculated for C₂₀H₃₆N₅O₆, 441.266, found 442.265. Linear peptide AMBA-14-KID: MS-direct infusion (ESI): m/z [M + H]⁺ calculated for C₂₄H₃₈N₅O₇, 508.269, found 508.270. Linear peptide AMPA-15-KID: MS-direct infusion (ESI): m/z: [M + H]⁺ calculated for C₂₅H₄₀N₅O₇, 522.285, found 522.293.

Dimers on a triazole core

Bivalent analogs in a triazole core were assembled with different orientation and sequences, as described previously (17, 20–22). Combinations were prepared in three different sets. Set 1 (forward only): R¹R² and R³R⁴ = i + 1−i + 2 and i + 1−i + 2, e.g. TG-TG, set 2 (opposite only): R¹R² and R³R⁴ = i + 2−i + 1 and i + 2−i + 1, e.g. GT-GT, and set 3 (forward-opposite): R¹R² and R³R⁴ = i + 1−i + 2 and i + 2−i + 1, e.g. TG-GT. In silico evaluation of the small molecules using the Pan Assay Interference Compounds program (53) showed no concerns.

Direct TrkC-binding assays using biotin-conjugated small molecules

Agents were tested by quantitative flow cytometry assays (52). Briefly, WT HEK293 or WT NIH, or their transfected counterparts expressing stable TrkC-FL or TrkC.T1 was resuspended in binding buffer (PBS, 0.2% bovine serum albumin [BSA], 0.1% Na Azide) and incubated with 10 µM of the indicated biotinylated agents for 30 min at 4 °C. Cells were washed in cold binding buffer, and exposed to secondary agent fluorothiocyanate-labeled avidin (avidin-FITC, catalog: A 2050; Sigma) for 20 min at 4 °C. After washing two times in binding buffer, cells were analyzed immediately by flow cytometry (Fortessa) and FlowJo software. The mean channel florescence values of WT cells (not expressing any TrkC) were standardized to 1 for each experiment, and data shown are the mean ± SD of at least three independent experiments.

Competition assays to block binding of PTPs or anti-TrkC mAb to TrkC

Agents were assayed by quantitative flow cytometry (52) for their ability to compete the binding of a PTPs-receptor-body containing PTPs-ectodomain linked to a human antibody Fc domain (PTPs-Fc) (11) or to compete the binding of anti-TrkC mAb TrkC 1 (2) that binds to both isoforms.

Cells were incubated ± the indicated agents (10 µM) or vehicle control for 20 min at 4 °C in binding buffer, followed by addition of saturating concentrations of primary reagent: PTPs-Fc (40 nM) or TrkC 1 mAb (67 nM). After incubating for 20 min, cells were washed, and resuspended for 20 min at 4 °C in binding buffer with the proper secondary antibodies with a fluorescent label (F9512; Sigma). After two washing steps, cells were resuspended in binding buffer and data were acquired immediately by flow cytometry and analyzed using FlowJo software. For negative staining controls, the primary reagent was omitted (background fluorescence). For cellular and receptor specificity controls, WT cells or cells expressing TrkC-FL or TrkC.T1 receptors were analyzed. The mean channel fluorescent (MFC) values of vehicle control treatment (no competition) are standardized as 100% binding. Data are the mean ± SD of at least three independent experiments.

Competition assay to block binding of NT3

Competition of NT3-biotin binding to TrkC isoforms was performed, as described previously (11, 52) using compound 16 or Cyclic-14-KI, or “cold NT3” (not biotinylated) as a positive control for self-competition. Cells in binding buffer (PBS + 0.2% BSA + 0.1% Na Azide) were treated with a concentration range (0–1–5–15 µM) of compound 16 or Cyclic-14-KI. Then, a concentration range of NT3-biotin (0–40 nM) was added to each concentration of test compound, with a 15-min incubation at 4 °C. After washing in ice-cold binding buffer, cells were treated with the secondary reagent conjugated avidin-fluorescein (Sigma) for 15 min at 4 °C. After two washing steps, live cells were analyzed immediately by flow cytometry. Analysis in GraphPad was performed to determine the K_d of NT3•TrkC and the mode of NT3•TrkC-binding inhibition (allosteric, noncompetitive, or competitive).

Cell metabolism/survival

The growth/survival profile of the cells was quantified in 96-well plates using the tetrazolium salt reagent 4-[4,5-dimethylthiazol-2-yl]-2,5-diphenyltetrazolium bromide (MTT; Sigma) 48 h after plating; by reading the optical density, as previously described (54). Cells cultured in serum-free medium (SFM) die by apoptosis, but they can be rescued if they express TrkC-FL and are supplemented with NT3 or PTPs. Wells received vehicle control or were supplemented ± NT3 or PTPs (cognate agonists) ± test agents at the indicated concentrations.

Studies in the absence of NT3 or PTPs or NT3 assess the possibility of (partial) agonism (limited survival) and whether agents cause toxicity (accelerated death). Studies in the presence of NT3 or PTPs respectively at their optimal (2 and 40 nM) and suboptimal (0.2 and 8 nM) concentrations (11) assess the pharmacological mechanism of (partial) antagonism, or (partial) agonism. Specificity cellular controls used cells expressing receptor family members TrkA or receptor TrkB, treated with respective positive controls NGF or BDNF. Assays were repeated n ≥ 3 independent times, each independent assay in quadruplicate. MTT data are standardized to optimal dose of neurotrophin = 100% survival, and SFM = 0% survival, using the formula $[(ODtest−ODSFM)×100/(ODoptimalNTF−ODSFM)]$⁠.

Quantification of TNF-α mRNA

rMC-1 cells expressing human TrkC.T1 were treated as indicated for 6 h, with or without NT3 (2 nM) and with or without test agents (10 µM). Quantification of TNF-α mRNA was performed by real-time quantitative PCR and primers used for rat TNF-α and rat RNAs18 reported (10). The values (relative changes in mRNA expression levels) were standardized to the positive control NT3. Data are the mean of at least three independent experiments ± SD, each in triplicate.

Signal transduction assays

The activation of signaling proteins Trk, Akt, and Erk1,2 was studied by western blotting of their phosphorylated states relative to the total Trk, Akt, Erk1,2 proteins, and standardized to actin control. NIH-TrkC-FL cells were cultured in SFM for 120 min (to reduce background and serum signals) and then were treated ± NT3 (optimal 2 nM, or suboptimal 0.2 nM) or PTPs (optimal: 40 nM, suboptimal 8 nM) ± the indicated agents (10 µM) for 20 min. Detergent lysates were analyzed by western blotting with anti-pTyr mAb 4G10, anti-phospho-pErk1,2 or anti-phospho-Akt (Cell Signaling). After stripping, membranes were re-probed with anti-actin (Sigma) or total anti-Erk1.2 or total anti-Akt (Cell Signaling) to standardize loading.

In silico modeling of TrkC bound to NT3 and PTPs, and docking of CP14

We used AlphaFold3 server (25) to model structures of the TrkC.T1 monomer (Uniprot Q16288-2) unliganded or bound to NT3 (Uniprot P20783) and PTPs ligands (PTPs is labeled here as PTPRS). Because TrkC is activated as homodimer, and NT3 is a symmetrical homodimer, we modeled the complex with a 2:2:2 stoichiometry. Five models were calculated for each structure. The result gave good prediction scores for the interaction (ipTM > 0.5) and low position error among the subdomains (≤5 Å). Furthermore, the five models converged, with root mean square deviations between 0.8 and 1.2 Å for the extracellular domain of monomeric TrkC.T1 (a.a. 30–400), and 2.0–3.9 Å for the TrkC.T1–NT3–PTPRS dimer (again using only the extracellular portion of TrkC.T1). The binding of NT3 to the D5 subdomain (also known as IgG-C2), and PTPRS bound to the D1–D2 subdomain (also known as LRR), agree with previous published reports (12, 55).

For docking, we used the SeamDock server and the AutoDock Vina protocols (56) to create unbiased docking to the different receptor structures. Using this computational docking tool, we sought to determine the binding site for Cyclic-14-KI on both the TrkC-T1 monomer and the dimeric TrkC.T1–NT3–PTPRS complex. Initially, we used a large docking box to generate an unbiased hierarchy of docking sites, and then we used a smaller box on the top site centered on the D4–D5 subdomains. Ten binding modes were calculated for each run. The top three models had binding energies between −6.2 and −6.8 kcal/mol for the monomeric TrkC structure, and −6.4 to −6.7 kcal/mol for the dimeric TrkC.T1–NT3–PTPRS complex.

RP animal model

All animal procedures respected the IACUC guidelines for use of animals in research, and protocols approved by McGill University Animal Welfare Committees. All animals were housed 12 h dark–light cycle with food and water ad libitum. We used the “RHOP347S” transgenic mouse (expressing the human rhodopsin mutated at amino acid position 347) in a C57BL/6J (B6) background (28). This model of RP faithfully replicates features of disease progression in humans. Both male and female animals were used in the experiments, with no sex selection of the litters. Similar findings are reported for both sexes. Gender issues do not apply.

Retinal organotypic cultures

For retinal cultures, mice at PN 18 (weighing between 10 and 12 g) were used. Whole explants of retinas were used for organotypic culture experiments. Procedures were as described previously (15, 45, 57). Whole eyes were enucleated, and whole retinas dissected from WT and RHOP347S mice at PN 18 and immediately transferred into 24-well plates containing 500 μL of culture medium (DMEM/F12 supplemented with 10 mM NaHCO₃, 100 μg/mL transferrin, 100 μM putrescine, 20 nM progesterone, 30 nM Na₂SeO₃, 0.05 mg/mL gentamicin, 2 mM L-glutamine, and 1 mM sodium pyruvate). Then, the media were removed and replaced with fresh media containing the treatments or controls and incubated at 37 °C and 5% CO₂ for 24 h. Agent 16 or CP14 was tested at a concentration of 10 μM, NT3 control at 20 nM. Cell-grade DMSO was used for vehicle control and was 0.5% by volume. Retinas were processed for TUNEL staining to evaluate photoreceptor cell death. Retinal flat mounts prepared form WT mice were used as negative controls, as they do not exhibit photoreceptor death in the time frame of the assay.

Quantification of TNF-α protein

RHOP347S retinas were cultured as organotypic explants and treated for 24 h with antagonists 16, or CP14, or vehicle control. Lysates of whole retinas were quantified for TNF-α protein by western blots with anti-TNF-α mAb (PeproTech), standardized to protein loading by western blots using anti-actin mAb (Sigma).

TUNEL staining

Staining was performed as described (45, 57) using the DeadEnd Fluorometric TUNEL system (Promega). RHOP347S retinas in culture (n = 3) were first fixed in 4% PFA in PBS and kept at 4 °C overnight. Three quick washes in PBS-0.2% BSA were done the next day, followed by three 30-min permeabilization steps, using 2% Triton X-100 in PBS. Retinas were then incubated with 20 μg/mL proteinase K in PBS for 15 min, briefly refixed in 4% PFA in PBS for 30 min and washed again with PBS-0.2% BSA before being transferred into Eppendorf tubes. Samples were incubated with 50 μL of equilibration buffer for 20 min and then 25 μL of terminal deoxynucleotidyl transferase reaction mixture for 2.5 h at 37 °C. The reaction was terminated using a 30-min incubation of 2× saline sodium citrate solution. The retinas were mounted with the ganglion cell layer facing up, using Vectashield with 4′,6-diamidino-2-phenylindole.

Fluorescence microscopy, image acquisition, and data analysis

Pictures were taken as Z-stacks of confocal optical sections using a Leica confocal microscope at a 20× objective. Images were equally adjusted using Adobe Photoshop CS 8.0 to remove background signals. The retinas were divided into four quadrants and three concentric areas as depicted in Fig. 6C. For quantification, image acquisition accounted for the different topological density of photoreceptors (higher numbers in the central region and lower numbers in the periphery). In the photoreceptor layer one picture was taken for each of the three areas (central, middle and periphery) in each quadrant for a total of 12 images per each retina, at 20×. Total TUNEL⁺ cells were counted semi-automatically (ImageJ) (45, 57). Analyses compared the drug-treated explants vs. vehicle-treated cultures, in explants from WT control mice and from RHOP347S transgenic mice.

In vivo therapy in RHOP347S mice

Intravitreal injections were performed in RHOP347S mice, at PN18, with ongoing disease and significant preexisting photoreceptor loss. Injection procedures are as previously described (44). Briefly, a total volume of 1 μL of vehicle (PBS) or compound 16 (1 mM) or CP14 (1 mM) was slowly delivered into the RHOP347S mouse vitreous chamber using a Hamilton syringe under a stereotactic microscope. After the injection, the syringe was left in place for 30 s and slowly withdrawn to prevent efflux. Given the severity of the disease model, with rapid degeneration, only 1× drug delivery was done at PN18 and retinas were evaluated at PN20.

Analyses of photoreceptors after in vivo therapy in RHOP347S mice

At PN20 (on day 2 after intravitreal injection), the eyes were enucleated, and serial retinal sections were prepared with distances from the ON recorded to account for quadrants and location (Fig. 6C). The presence of dead photoreceptors was evaluated in TUNEL assays described above. In the same tissues from the vehicle or CP14-treated animals, the presence of surviving photoreceptors was quantified using hematoxylin and eosin staining (H&E, catalog H-3502; Vector) of retinal sections (12 μm). The stain reveals the nucleus and cytoplasm of the live photoreceptors in the photoreceptor layer, morphologically distinct from Müller cells and retinal ganglion cells which in any case are in a different anatomical layer. Bright field Imaging of the stained RHOP347S retinas under 40× magnification was done using a Zeiss apotome three microscope. Sections were prepared as defined for the central (area 1), middle (area 2), and peripheral (area 3) regions (respectively 100–1,000 μm, 1,000–2,000 μm, and >2,000 μm from the center of the optic nerve head). The number of photoreceptor nuclei on retinal sections was evaluated by manually counting H&E-stained nuclei in the ONL. The counts were averaged from 8 to 12 regions of interest from 3 to 4 sections of the same retina, and the mean were calculated for each group.

Statistical analyses

The quantitative data were subjected to statistical analyses using GraphPad Prism 5 software and are presented as mean ± SD of the indicated independent experiments. One-way ANOVA was used for all comparisons between groups with significance α < 0.05 followed by Bonferroni's correction for multicomparisons. For ex vivo and in vivo studies, data were subjected to statistical analyses using GraphPad Prism 5 software and are presented as mean ± SD for a whole retina or per each area of the retina to account for the natural thinning of the photoreceptor layer (thicker in center and thinner in periphery). One-way ANOVA was used for all comparisons between groups with significance α < 0.05 followed by Bonferroni’s correction for multicomparisons.

Acknowledgments

Dr T. Li provided the RHOP347S transgenic mice. Dr Enrique de la Rosa optimized for us the retinal explants and TUNEL assays. Some of the first-generation peptidomimetics were published, and the original references are listed. Selected agents and second-generation analogs and were synthesized in house.

Supplementary Material

Supplementary material is available at PNAS Nexus online.

Funding

This work was supported by grant PJT 162291 from the Canadian Institutes of Health Research (to H.U.S.).

Author Contributions

F.B., H.N., A.G., and C.O. designed, planned, and carried out experiments, curated, analyzed, and cross-validated their data, and prepared most figures and tables after conducting statistical analysis. R.G. statistically analyzed and prepared a figure. F.B. and H.U.S. produced the monoclonal antibodies. H.N. carried out the synthesis and characterization of the small molecules. J.-F.T. carried out the computational modeling, docking, and corresponding figures. H.U.S. generated the concept, helped design experiments, supervised the work, and lead the organization and writing of the manuscript. The iterative editing of versions of the manuscript had contributions primarily from F.B. and H.N. All authors reviewed and approved the final version of the manuscript. Authors agreed to the order in which they are listed. All authors contributed to the article and approved the submitted version.

Data Availability

The data that support the findings of this study are openly available in Figshare.com at https://figshare.com/articles/dataset/My_Data/28024736/1 reference number. The research materials published in this study may be available by signing an institutional Material Transfer Agreement, that can be requested from the corresponding author, H.U.S. at [email protected].

References

Author notes