Evaluation of neuroprotective agents acting via the BDNF–TrkB pathway using AI‐enabled predictions of ligand–receptor interactions

Abstract

Glaucoma is the leading cause of irreversible blindness globally and is associated with retinal ganglion cell (RGC) death. Brain‐derived neurotrophic factor (BDNF) is a potent neurotrophin that promotes neuronal survival via its receptor, tropomyosin receptor kinase B (TrkB) encoded by NTRK2. Our current understanding of the mechanism of action and therapeutic potential of the BDNF pathway is limited by the lack of knowledge of its interaction with TrkB at atomic resolution. We developed an artificial intelligence (AI) model to predict the three‐dimensional protein structures of BDNF and TrkB, as well as their interaction. The AI model was further applied to compare small‐molecule drugs that mimic BDNF–TrkB interaction, leading to the identification of 7,8‐dihydroxyflavone (DHF) as an agonist of TrkB. We verified the neuroprotective effects of DHF in an in vivo acute glaucoma model in which RGC apoptosis caused by acute elevation of intraocular pressure was prevented by the intraocular application of DHF and to a lesser extent by BDNF. Our results provide AI‐enabled prediction of ligand–receptor interactions between BDNF and TrkB at the atomic level and demonstrate the great potential for AI‐enabled drug discovery.

INTRODUCTION

Glaucoma is the leading cause of irreversible blindness in the world and is characterized by loss of retinal ganglion cells (RGCs). In a variety of ocular illnesses, neuroprotective methods that delay or halt neuronal cell death may be able to save RGCs and avoid vision loss. A number of neuroprotective agents have been identified for RGCs, including brain-derived neurotrophic factor (BDNF), N-methyl-D-aspartate (NMDA) receptor antagonists, and calcium channel blockers.

One of the most effective neurotrophic factors for both neuroprotection and functional rescue is BDNF, which belongs to a family of neurotrophins (NTs), including BDNF, nerve growth factor (NGF), neurotrophin 3 (NT3), and neurotrophin 4 (NT4). Their corresponding receptors are tropomyosin-related kinase (Trk) receptors, TrkA, TrkB, and TrkC, encoded by the neurotrophin receptor kinase genes, NTRK1, NTRK2, and NTRK3, respectively. NTs play important roles in the differentiation, survival, axonal transport, and axonal growth of many neurons, including RGCs. Numerous studies have reported that BDNF exerts neuroprotective effects in Alzheimer's diease, Parkinson's disease, and glaucoma. BDNF binds to the extracellular domain of its receptor TrkB and stimulates autophosphorylation of crucial intracellular tyrosine residues, thus initiating intracellular signaling cascades. However, the three-dimensional (3D) protein structures of TrkB receptor alone or in complex with BDNF have not been resolved, hindering the development of small-molecule agents for this important drug target.

The BDNF–TrkB signaling pathway plays a vital role in the survival of RGCs. BDNF has been detected locally in the RGCs and astrocytes. It is retrogradely transported from the lateral geniculate nucleus and the superior colliculus to the cell bodies via RGC axons in higher mammals. In an experimental glaucoma model, retrograde transport of the BDNF–TrkB complex to the retina was obstructed at the optic nerve head, resulting in reduced RGC survival. The exogenous application of BDNF has demonstrated the potential to promote RGC survival in rat models of acute glaucoma. However, the therapeutic potential of BDNF is constrained by its brief half-life (<10 min) and low blood–retina barrier penetration. The effect of BDNF on RGC survival appears transient. Combined ligand–receptor gene supplementation of BDNF and TrkB has demonstrated therapeutic potential in glaucoma and tauopathy models. Small-molecule agents that mimic BDNF neurotrophic signalings, such as 7,8-dihydroxyflavone (DHF) and TrkB receptor antibody, may provide significant therapeutic advantages over BDNF itself.

Rapid protein structure predictions made possible by recent developments in AI-based 3D protein structure predictions have created a completely new field for biology and medicine and have the potential to drastically shorten drug discovery processes. We used this method to predict and understand BDNF–TrkB interaction and screened for small molecules that mimic this ligand–receptor interaction as potential neuroprotective agents acting via the TrkB receptor.

RESULTS

AI-based 3D protein structure prediction for interactions of neurotrophins and their receptors

There is currently no complex structure available for the NTs or Trks in the PDB database. To identify potential structural similarities shared by the NT–TrK interactions and gain broad insights into their functions, we used an AI-based method to predict the 3D protein structures of these proteins and their complexes. Our AI model, KeystoneFold, was able to predict the structures of the various types of natural phenomena, such as the Trks and the NT, similar to those of RoseTTAFold and AlphaFold2 (Figure ). Our predictions also showed that the four neurotrophins and the extracellular domains of their Trks adopted very similar structures (Figure ). There are three obvious compact structures within the extracellular domain of the Trks, two of which are Ig-like C2-type structures. To identify the part of the Trk extracellular domain responsible for NT ligand binding, we generated multiple sequence alignments (MSAs) for each of the three subdomains of TrkB and paired the MSAs with that of BDNF to predict potential protein–protein interactions (Figure ). The AI model predicted a favorable interaction between the Ig-like C2-type 2 domain (283–420) of TrkB and the C-terminus of BDNF.

[IMAGE OMITTED. SEE PDF]

Evaluation of small-molecule agonists for TrkB activation

We then evaluated small-molecule agonists for TrkB activation as potential neuroprotective agents. A range of small-molecule agonists has been described in the literature. We performed molecular dynamics modeling to assess and compare potential docking interactions of these small molecules with TrkB (Figure ). This succeeded in docking several known TrkB agonists onto the extracellular domain of TrkB. Interestingly, these compounds tend to bind similarly to the same interface as that involved in BDNF–TrkB extracellular domain interaction, consistent with competitive agonist activity (Figure ). We then focused on 7,8-dihydroxyflavone (DHF), which showed the highest predicted binding affinity (Figure  and Supporting Information: Table ). Furthermore, our molecular docking modeling predicted for the binding of DHF to the interface between BDNF–TrkB interaction (Figure ). The modeling predicted that DHF could form strong interactions with the binding site of TrkB mainly through hydrophobic and aromatic interactions (Figure ).

[IMAGE OMITTED. SEE PDF]

DHF is a member of the flavonoid family and a selective TrkB agonist which activates the TrkB receptor with high affinity and initiates activation of the BDNF/TrkB pathway. Compared to BDNF, DHF has a longer half-life (134 min in plasma following 50 mg/kg oral administration) and smaller molecular size at 254 Da, which allows for greater permeability. DHF provided neuroprotection and neuroplasticity in animal models of various neurological diseases, including Alzheimer's disease, Parkinson's disease, amyotrophic lateral sclerosis, Huntington's disease, stroke, Rett syndrome and traumatic brain injury (TBI). Gupta et al. reported that DHF provides neuronal protection against oxidative and excitotoxicity stress-induced apoptosis of rat retinal ganglion cells in cell cultures (RGCs). Since DHF provided neuroprotection against the injury-induced apoptosis of RGCs in vitro, we evaluated the in vivo neuroprotective effect of DHF on RGCs against acute intraocular pressure (IOP) elevation, a model for acute glaucoma.

Neuroprotective effect of DHF in an acute glaucoma model

We investigated whether DHF can protect RGC in a rat acute glaucoma model with a predictable and reproducible RGC death time course. We used Brn3a, which is an important transcription factor required for RGC cell fate determination and survival, to label RGCs. The density of Brn3a-positive RGCs in a retinal flat mount was measured under various conditions, including a control group, acute IOP elevation (HIOP group), acute IOP elevation plus intravitreal injection of BDNF (BDNF group), and acute IOP elevation plus intravitreal injection of DHF (DHF group). Figure  shows the Brn3a-positive RGCs in the mid-central regions of the retina flat-mounts at each time point for all four groups.

[IMAGE OMITTED. SEE PDF]

We used the density of RGCs (RGCs/mm²) in the control group (2212.67 ± 36.24 cells/mm²) as the baseline. Compared with the control group, the RGC density of the HIOP group was significantly reduced at all time points (p < 0.001). The proportion of remaining RGCs decreased to 34.9 ± 2.8% at Day 28 after acute IOP elevation, while BDNF and DHF significantly protected the RGCs after acute IOP elevation with 50.52 ± 0.81% and 79.83 ± 2.45% cell survival at Day 28, respectively (Figure ). Moreover, DHF elicited a more potent neuroprotective effect than BDNF on both Day 14 and 28, with 20% more RGC survival on Day 14 and 30% more RGCs survival on Day 28, respectively.

We also assessed the effect of DHF on reducing RGC apoptosis more directly. Quantification of apoptotic RGCs by TUNEL staining showed that acute IOP elevation resulted in a large number of TUNEL-positive cells compared with the control group at all time points (Figure  and Supporting Information: Table ), especially in the first 24 h (p < 0.001, Figure ). The IOP-induced RGC apoptosis was reversed by BDNF treatment as indicated by 11.51-fold reduction of apoptosis at Hour 12, 13.40-fold at Day 1, 10.89-fold at Day 3, 9.10-fold at Day 7, 5.85-fold at Day 14, 7.88-fold at Day 28 when comparing to that of HIOP group (all groups p < 0.001, except p = 0.008 at Day 28). Furthermore, DHF significantly rescued RGCs after acute IOP elevation within the first 24 h (p < 0.001 vs. HIOP, Figure ) and with the effect lasting over 28 days (p = 0.002 vs. HIOP, Figure ), as indicated by 10.84-fold reduction of apoptosis at Hour 12, 12.18-fold at Day 1, 5.76-fold at Day 3, 4.53-fold at Day 7, 3.61-fold at Day 14, 4.62-fold at Day 28 when compared to that of HIOP group (all groups p < 0.001, except p = 0.002 at Day 28).

[IMAGE OMITTED. SEE PDF]

DHF increased NTRK2 mmessenger RNA (mRNA) expression after acute IOP elevation

We next conducted a mechanistic study on DHF's RGC protective effect. Compared with the control group, acute IOP elevation resulted in a 2.21-fold increase of NTRK2 mRNA level at 12 h (p = 0.009), which was not sustained and decreased continuously to 80.01% at Day 3 (p = 0.002), 71.61% at Day 14 (p = 0.012), 51.26% at Day 28 (p = 0.031) comparing to baseline, respectively (Figure ). In contrast, BDNF treatment led to increased NTRK2 mRNA levels by 2.43-fold at 12 h (p = 0.008) and continued to be elevated at 3.15-fold at Day 3 (p = 0.005), 4.37-fold at Day 14 (p < 0.001), 1.39-fold at Day 28 (p = 0.13), respectively. Interestingly, DHF treatment led to higher levels of NTRK2 mRNA than those induced by BDNF treatment, with a 3.24-fold increase at 12 h (p < 0.001), and 4.20-fold at Day 3 (p < 0.001), 5.30-fold at Day 314 (p < 0.001), 5.18-fold at Day 28 (p < 0.001), respectively.

[IMAGE OMITTED. SEE PDF]

DHF suppressed expression of the proapoptotic bax gene after acute IOP elevation

To gain further insight into the molecular mechanism underlying DHF's RGC protective effect, we surveyed changes in expression of the proapoptotic gene Bax following acute IOP rise. After acute IOP elevation, Bax mRNA levels were elevated by 2.13-fold (p = 0.002), 5.80-fold (p < 0.001), 3.31-fold (p = 0.012), and 3.11-fold (p = 0.033) at 12 h, Day 3, 14 and 28, respectively (Figure ). BDNF did not prevent this peak in proapoptotic Bax expression at any timepoint (all p > 0.05). In contrast, the IOP-induced Bax expression was reversed by DHF treatment as indicated by 4.64-fold reduction of Bax mRNA expression at Day 3 (p = 0.004), 1.94-fold at Day 14 (p = 0.01), and 1.91-fold at Day 28 when compared to that of HIOP group (p < 0.001), respectively.

DISCUSSION

In this study, we present the molecular and cellular mechanisms of neuroprotection by the TrkB receptor agonist, DHF, based on AI-enabled prediction of protein–drug interactions and drug screening followed by in vivo validation. We first predicted the interaction between the native neurotrophin BDNF and receptor TrkB using AI-based modeling of protein-protein interactions. We then identified DHF among a group of purported TrkB agonists to bind with the highest affinity to the same interaction domain as BDNF using AI-based prediction of protein–drug interactions. We went on to show that intraocular delivery of DHF provides superior neuroprotection against RGC loss compared with BDNF in a rat model of acute glaucoma.

The acute IOP elevation model used in our study creates a form of retinal ischemia-reperfusion injury, which is common in acute glaucoma and retinal ischemic diseases, leading to a rapid and slower phase of RGC loss. As Brn3a is considered an RGC-specific nuclear marker protein, we labeled the Brn3a-positive RGCs by immunofluorescence. We discovered that the average RGC densities dramatically dropped on Days 1 and 3 following the insult and kept going down for another 4 weeks. These findings are consistent with early studies in the literature. In this study, we found that the DHF partially protected against RGC death. On Day 14, the RGC density in the HIOP group dropped to 42.3% of the control value. Compared to 35% in the HIOP group and 56% in the BDNF group, the application of DHF saved up to 79% of the RGCs present at the end of Day 28. We found that with BDNF injection, the number of RGCs decreased continuously after 14 days. These results indicated a limited effective duration of exogenous BDNF, while DHF rescued more RGCs than exogenous BDNF did.

Consistent with morphologic changes, IOP elevation also induced apoptosis of RGCs, which started 18 to 24 h after IOP elevation and persisted for 4–28 days. In our study, the apoptosis of RGCs induced occurred at 12 h, peaked at 24 h, and sustained for 28 days in the HIOP and BDNF groups. However, DHF protected the RGCs from apoptosis from 12 h to 28 days. DHF itself did not appear to induce significant apoptosis of retinal neurons. The findings indicate a protective effect by DHF against RGC apoptosis induced by acute IOP elevation.

By binding the TrkB receptor and activating mitogen-activated protein kinase, BDNF rescues RGCs. Our study also demonstrates the effects of DHF on the TrkB signaling in rat RGCs in vivo. We found that elevation of IOP at 12 h upregulated NTRK2 mRNA expression in the retina, but the level decreased later. In the DHF group, the expression of TrkB increased at 12 h, peaked at 14 days, and sustained for 28 days. While in the BDNF group, the level of TrkB significantly increased at 4 days and decreased to the basic level at 28 days. This result indicated that DHF had a longer effect on RGC survival compared to that of exogenous BDNF. These findings are consistent with previous reports that DHF, as a TrkB agonist, induced TrkB activation in neuronal cells, including retinal ganglion cells.

We evaluated the retinal expression of the apoptotic cell death protein Bax to clarify the mechanistic effects of IOP increase and DHF on apoptosis. Acute IOP elevation upregulated the expression of Bax in a time-dependent manner, which was almost completely reversed by DHF, not BDNF. These results suggested a superior RGC protective effect of DHF. One of the limitations of our study is that the neuroprotective effect of DHF was tested only in the acute glaucoma model but not the chronic glaucoma model.

The molecular dynamics docking prediction has been an active area of interest for decades to investigate how a ligand binds to a particular protein target. However, its progress is hindered by the availability of known 3D protein structures. The protein structure prediction based on deep learning technology now provides far more 3D protein structures, therefore providing more information on potential drug targets and facilitating structure-based drug design, especially for those important drug targets without experimentally determined structures. This may greatly accelerate novel drug discovery. Future challenges will be using AI for the prediction of agonistic versus antagonistic ligand–receptor interactions.

In summary, AI-based prediction of protein-drug interactions suggests that DHF is a strong competitive agonist to BDNF at the TrkB receptor. This prediction was validated in vivo by DHF-mediated rescue of retinal neurons following acute IOP elevation, superior to exogenous BDNF. Our results support DHF as a neuroprotective agent for glaucoma and other disorders associated with RGC degeneration and open a new paradigm for rapid drug discovery and evaluation.

METHODS AND MATERIALS

MSA and sequence embedding

To generate features for the proteins, we used HHblits in the HH-suite3 package to build MSAs. HHblits is an iterative sequence search tool based on hidden Markov models. It could sensitively and accurately find homologous sequences from protein sequence databases and then build MSAs. We initially generated one MSAs for each protein sequence by running HHblits on the UniRef30 (released in June 2020) and BFD sequence databases with four iterations. The E value cutoff is iteratively set to 1 × 10⁻³⁰, 1 × 10⁻¹⁰, 1 × 10⁻⁶, and 1 × 10⁻³. The resulting MSA has at least 2000 sequences with 75% coverage or 5000 sequences with 50% coverage (both at 90% sequence identity cutoff). From each individual MSA, we derived sequential features that include sequence profile and secondary structure prediction by PSIPRED. Template structures were obtained by searching the generated MSA against the PDB100 database with HHsearch.

The input MSAs were represented as m × n matrixes, where m corresponds to the number of sequences in the MSAs, and n is the residue position length in the aligned sequence. The individual amino acids and gaps of the input MSAs were tokenized as 21 characters for further processing. The tokenized matrixes were mapped to vectors through an embedding layer. Pairwise features including positional similarity and alignment confidence were calculated by extracting residue pairwise distances.

AI-based structure prediction

We adopted a deep learning model with an attention mechanism that was inspired by the RoseTTAFold architecture. In the hidden layers, the input sequence feature matrixes were converted to 3D matrixes. The long-range sequential context of residues from certain MSA traits was captured using transformer-inspired attention architectures. The outer product can capture the correlation information between two residues. The attention mechanism is suitable for protein structure prediction as it could efficiently learn the relationship of residue pairs distant in sequence. The attention layers provide an efficient MSAs representation. The hidden layer outputs were then fed into a transformer-based architecture followed by a linear transformation that was employed to generate initial Cartesian coordinates of protein backbone atoms. Multiple SE(3)-transformer layers were used to refine given 3D coordinates based on original and updated MSA and pairwise features. The average of transposed and untransposed feature maps was used to ensure the symmetry of the network predictions. Predicting protein structure with a long sequence can exceed the memory of a single GPU. A domain segmentation has been applied to Trk receptors. Using pyRosetta, full-atom structure models were produced based on gradient-based folding. Final models were chosen using the residue-wise C_ɑ-lDDT scoring function from all the sampled structures.

Small-molecule docking

The protonation states of TrkB were assigned using the “Calculate Protein Ionization and Residue pK” module of Discovery Studio 3.1 to correspond to pH 7.0, optimizing the rotameric states for histidine, asparagine, and glutamine residues. The protein model of TrkB was prepared for docking, minimizing, and optimizing hydrogen placement and side-chain atoms with CHARMM. Docking was performed using the “Docking Ligands (LigandFit)” module of Discovery Studio 3.1. The docking sites and spheres were generated automatically and then manually inspected and modified where necessary to achieve a more uniform distribution across the binding site. The docking scores LigScore1, LigScore2, PLP1, PLP2, and PMF were used to rank the binding of compounds.

Animal model

The studies were conducted using young adult male Sprague–Dawley (8 weeks old, 180–220 g body weight), which were provided by the Experimental Animal Center of Sichuan University in Chengdu, People's Republic of China. They were housed with standard chow and water ad libitum and sustained on a 12 h:12 h light and dark cycle at a temperature of 21–25°C. The rats were divided into four groups consisting of nine animals per group: control group (without intervention), HIOP group (acute elevation of IOP with phosphate-buffered saline [PBS] injection), DHF group (acute elevated IOP model with DHF injection), and BDNF group (acute elevated IOP model with BDNF injection). All procedures were conducted strictly following the ARVO statement for the Use of Animals in Ophthalmic and Vision Research. The use of laboratory animals was approved by the institutional review board of Guangzhou Women and Children's Medical Center.

Experimental elevation of IOP

Rats were anesthetized by intraperitoneal injection of chloral hydrate (400 mg/kg; Kelong). Topical anesthetic of 0.4% hydrochloric oxybuprocaine (Santen) and pupil dilation drop of 1% tropicamide (Mydrin-p; Santen) were applied. Transient acute IOP elevation was induced according to methods described in previous studies. Briefly, the anterior chamber of the left eye was penetrated with a 30-gauge needle attached to a sterile saline reservoir. During the process, IOP was elevated to 130 ± 20 mmHg, which was monitored by TonoLab (Tiolat Ltd.). Whitening of the iris and blanching of the fundus indicate retinal ischemia. After maintaining the ischemic state for 60min, the needle was withdrawn, and reperfusion was observed by reddening of the fundus. The eyes were inspected daily, and the animals were kept in individual cages for the following experiments.

Intravitreal injection

An intraperitoneal injection of chloral hydrate (400 mg/kg body weight; Kelong) was used to first anesthetize the rats. A topical anesthetic was used first, followed by pupil dilatation and intravitreal injection. A volume of 5 μl of sterile, 0.1 M PBS were injected intravitreally into the treatment groups (pH 7.4). In the DHF and BDNF groups, after acute IOP elevation was induced by PBS, 5 μl of sterile DHF (total injected amount, 100 nanomoles; Sigma-Aldrich), or BDNF (total injected amount, 100 nanomoles) was injected intravitreally. The injection site was 3 mm posterior to the supertemporal limbus. The ophthalmic antimicrobial ointment was applied to the eyes following surgery. Only eyes without cataracts, intravitreal hemorrhages, or any postoperative problems were included. At different times (12 h, 3, 7, 14, and 28 days), the animals were killed by an intraperitoneal injection of an overdose of 10% chloral hydrate.

Retinal flat mount and quantification of Brn3a-positive RGCs

To preserve the orientation of the rats' eyes, a suture was tied around the superior rectus muscle's insertion. The rats' eyes were enucleated and fixed with 4% paraformaldehyde in 0.1 mol/L PBS (pH 7.4) at 4°C for 2 h, then washed two times in 0.1 mol/L PBS. After removal of the anterior segment, the retina was carefully detached from the sclera by making cuts along the ora Serrata and optic nerve. The isolated retina was mounted on a glass slide that had been coated with 1% gelatin before being flattened using four radial cuts with the vitreous side facing up. The retina was permeabilized (0.3% Triton X-100 in 5% BSA), blocked, and stained with primary antibodies (goat anti-Brn3a Antibody, 1:300, Santa Cruz) at 4°C overnight (18 h) and with secondary antibodies (Alexa Fluor-488 donkey anti-goat IgG antibody, Invitrogen) at room temperature for 2 h.

The superior-temporal, inferior-temporal, superior-nasal, and inferior-nasal quadrants were used to partition each retina into four sections. Central, mid-central, and periphery regions were then further separated into each quadrant (at 1, 2, and 3 mm from the optic disc in the central region). The images of the retinas were captured with a fluorescence microscope (Carl Zeiss), each measuring 0.04 mm². According to Biermann, to determine the density of Brn3a-positive RGCs, we multiplied the number of analyzed cells per 0.04 mm² by 25. Three investigators blinded to the study design counted Brn3a-positive RGCs in 12 regions of each retina. By averaging all of the RGC counts in each retina, which are shown as (cells/mm²) ± SEM, the mean RGC density was determined.

Apoptotic cell count by TUNEL assay

Three eyes were used for analysis at each time point. Apoptotic cells were stained by TUNEL in the retinal flat mounts, according to the procedure described by the manufacturer in the In-Situ Cell Death Detection Kit (Roche Applied Science, Cat 11684817910). Proteinase K (Sigma-Aldrich) was used to permeabilize the flat-mounted retinas for 30min at 37°C, followed by 60 min in the TUNEL reaction mixture at 37°C for 2 h. TUNEL-positive cells were observed under a fluorescence microscope at ×200 magnification. The number of apoptotic cells was evaluated the same as that described for RGC count.

Quantitative reverse transcription–polymerase chain reaction (RT-PCR)

Quantitative RT-PCR amplification of Bax and NTRK2 was carried out with Applied Biosystems 7500 Real-time PCR System using the 2X SYBR Green PCR Master Mix (Applied Biosystems). Total RNA was extracted from the rat retina using a purification kit (RNeasy Mini Kit; Qiagen Inc.) and reverse-transcribed using a cDNA synthesis kit (SuperScript III; Invitrogen, Carlsbad, CA, USA). The cDNA was first denatured at 95°C for 15 min, then subjected to a series of PCR cycles at 94°C for 15 s, T_m—60°C for 30 s, and 72°C for 30 s. Data were analyzed according to the comparative C_t method. The C_t was automatically determined for each reaction by the Applied Biosystems 7500 Real-time PCR System set with default parameters. The mRNA levels of certain targets for each sample were normalized using the housekeeping gene glyceraldehyde 3-phosphate dehydrogenase (GAPDH) mRNA level.

Statistics

The statistical analyses were carried out by SPSS 22. software (SPSS 22.0 statistical analysis software for Windows, SPSS Inc), and the data were presented as mean ± standard error of the mean (SEM). Comparison between the groups was analyzed using Student's t tests or one-way analysis of variance (ANOVA). Results with two-sided p values less than 0.05 (p < 0.05) were considered statistically significant.

AUTHOR CONTRIBUTIONS

Jing Zhu, Jun Zou, Fei Li, Yuanxu Gao, Lijun Wang, Yi Sun, Jie Zhu, Xiaomeng Zhang, Kanmin Xue, Gen Li, Nga Man Cheng, Juan Guo, Xiulan Zhang, and Kang Zhang collected and analyzed the data. Kang Zhang conceived and supervised the project. All authors have read and approved the article.

ACKNOWLEDGMENTS

The study was supported by National Key R&D Program of China (2017YFE0103400 and 2017YFC1104600), NSFC/FDCT joint grant, and the Recruitment Program of Leading Talents in Guangdong Province (2016LJ6Y375).

CONFLICTS OF INTEREST

Jun Zou, Yuanxu Gao, Kanmin Xue, and Kang Zhang are editorial board members of MedComm - Future Medicine. None of them is involved in the journal's review or decisions related to this manuscript. The other authors declared no conflict of interest.

DATA AVAILABILITY STATEMENT

The main data supporting the results in this study are available within the paper and its Supplementary Information. The customized codes for predicting 3D protein structure and molecular interaction are available upon reasonable request to the corresponding authors.

ETHICS STATEMENT

The study was approved by the institutional review board of Guangzhou Women and Children's Medical Center.

References

Quigley HA. Neuronal death in glaucoma. Prog Retin Eye Res. 1999;18(1):39‐57.

Buckingham BP, Inman DM, Lambert W, et al. Progressive ganglion cell degeneration precedes neuronal loss in a mouse model of glaucoma. J Neurosci. 2008;28(11):2735‐2744.

Zhang K, Zhang L, Weinreb RN. Ophthalmic drug discovery: novel targets and mechanisms for retinal diseases and glaucoma. Nat Rev Drug Discov. 2012;11(7):541‐559.

Kido N, Tanihara H, Honjo M, et al. Neuroprotective effects of brain‐derived neurotrophic factor in eyes with NMDA‐induced neuronal death. Brain Res. 2000;884(1‐2):59‐67.

Kimura A, Namekata K, Guo X, Harada C, Harada T. Neuroprotection, growth factors and BDNF‐TrkB signalling in retinal degeneration. Int J Mol Sci. 2016;17(9):1584.

Rovere G, Nadal‐Nicolás FM, Wang J, et al. Melanopsin‐Containing or Non‐Melanopsin‐Containing retinal ganglion cells response to acute ocular hypertension with or without brain‐derived neurotrophic factor neuroprotection. Invest Ophthalmol Vis Sci. 2016;57(15):6652‐6661.

Valiente‐Soriano FJ, Nadal‐Nicolás FM, Salinas‐Navarro M, et al. BDNF rescues RGCs but not intrinsically photosensitive RGCs in ocular hypertensive albino rat retinas. Invest Ophthalmol Vis Sci. 2015;56(3):1924‐1936.

Peinado‐Ramon P, Salvador M, Villegas‐Perez MP, Vidal‐Sanz M. Effects of axotomy and intraocular administration of NT‐4, NT‐3, and brain‐derived neurotrophic factor on the survival of adult rat retinal ganglion cells. A quantitative in vivo study. Invest Ophthalmol Vis Sci. 1996;37(4):489‐500.

Mansour‐Robaey S, Clarke DB, Wang YC, Bray GM, Aguayo AJ. Effects of ocular injury and administration of brain‐derived neurotrophic factor on survival and regrowth of axotomized retinal ganglion cells. Proc Natl Acad Sci USA. 1994;91(5):1632‐1636.

Vidal‐Sanz M, Lafuente MP, Mayor‐Torroglosa S, Aguilera ME, Miralles de Imperial J, Villegas‐Perez MP. Brimonidine's neuroprotective effects against transient ischaemia‐induced retinal ganglion cell death. Eur J Ophthalmol. 2001;11(Suppl 2):S36‐S40.

Doozandeh A, Yazdani S. Neuroprotection in glaucoma. J Ophthalmic Vis Res. 2016;11(2):209‐220.

Tejeda GS, Diaz‐Guerra M. Integral characterization of defective BDNF/TrkB signalling in neurological and psychiatric disorders leads the way to new therapies. Int J Mol Sci. 2017;18(2):268.

Vidal‐Sanz M, Galindo‐Romero C, Valiente‐Soriano FJ, et al. Shared and differential retinal responses against optic nerve injury and ocular hypertension. Front Neurosci. 2017;11:235.

Xiao J, Wong AW, Willingham MM, van den Buuse M, Kilpatrick TJ, Murray SS. Brain‐derived neurotrophic factor promotes central nervous system myelination via a direct effect upon oligodendrocytes. Neuro‐Signals. 2010;18(3):186‐202.

Hachisu M, Konishi K, Hosoi M, et al. Beyond the hypothesis of serum anticholinergic activity in Alzheimer's disease: acetylcholine neuronal activity modulates brain‐derived neurotrophic factor production and inflammation in the brain. Neurodegener Dis. 2015;15(3):182‐187.

Zhang F, Kang Z, Li W, Xiao Z, Zhou X. Roles of brain‐derived neurotrophic factor/tropomyosin‐related kinase B (BDNF/TrkB) signalling in Alzheimer's disease. J Clin Neurosci. 2012;19(7):946‐949.

Igarashi T, Miyake K, Kobayashi M, et al. Tyrosine triple mutated AAV2‐BDNF gene therapy in a rat model of transient IOP elevation. Mol Vis. 2016;22:816‐826.

Martin KR, Quigley HA, Zack DJ, et al. Gene therapy with brain‐derived neurotrophic factor as a protection: retinal ganglion cells in a rat glaucoma model. Invest Ophthalmol Vis Sci. 2003;44(10):4357‐4365.

Klocker N, Kermer P, Weishaupt JH, Labes M, Ankerhold R, Bahr M. Brain‐derived neurotrophic factor‐mediated neuroprotection of adult rat retinal ganglion cells in vivo does not exclusively depend on phosphatidyl‐inositol‐3’‐kinase/protein kinase B signaling. J Neurosci. 2000;20(18):6962‐6967.

Chitranshi N, Dheer Y, Mirzaei M, et al. Loss of Shp2 rescues BDNF/TrkB signaling and contributes to improved retinal ganglion cell neuroprotection. Mol Ther. 2019;27(2):424‐441.

Fu QL, Hu B, Li X, et al. LINGO‐1 negatively regulates TrkB phosphorylation after ocular hypertension. Eur J Neurosci. 2010;31(6):1091‐1097.

Dekeyster E, Geeraerts E, Buyens T, et al. Tackling glaucoma from within the brain: an unfortunate interplay of BDNF and TrkB. PLoS One. 2015;10(11): [eLocator: e0142067].

Herzog KH, von Bartheld CS. Contributions of the optic tectum and the retina as sources of brain‐derived neurotrophic factor for retinal ganglion cells in the chick embryo. J Neurosci. 1998;18(8):2891‐2906.

Crish SD, Dapper JD, MacNamee SE, et al. Failure of axonal transport induces a spatially coincident increase in astrocyte BDNF prior to synapse loss in a central target. Neuroscience. 2013;229:55‐70.

Pease ME, McKinnon SJ, Quigley HA, Kerrigan‐Baumrind LA, Zack DJ. Obstructed axonal transport of BDNF and its receptor TrkB in experimental glaucoma. Invest Ophthalmol Vis Sci. 2000;41(3):764‐774.

Quigley HA, McKinnon SJ, Zack DJ, et al. Retrograde axonal transport of BDNF in retinal ganglion cells is blocked by acute IOP elevation in rats. Invest Ophthalmol Vis Sci. 2000;41(11):3460‐3466.

Ko ML, Hu DN, Ritch R, Sharma SC, Chen CF. Patterns of retinal ganglion cell survival after brain‐derived neurotrophic factor administration in hypertensive eyes of rats. Neurosci Lett. 2001;305(2):139‐142.

Leibrock J, Lottspeich F, Hohn A, et al. Molecular cloning and expression of brain‐derived neurotrophic factor. Nature. 1989;341(6238):149‐152.

Jang SW, Liu X, Yepes M, et al. A selective TrkB agonist with potent neurotrophic activities by 7,8‐dihydroxyflavone. Proc Natl Acad Sci USA. 2010;107(6):2687‐2692.

Di Polo A, Aigner LJ, Dunn RJ, Bray GM, Aguayo AJ. Prolonged delivery of brain‐derived neurotrophic factor by adenovirus‐infected müller cells temporarily rescues injured retinal ganglion cells. Proc Natl Acad Sci USA. 1998;95(7):3978‐3983.

Almasieh M, Wilson AM, Morquette B, Cueva Vargas JL, Di Polo A. The molecular basis of retinal ganglion cell death in glaucoma. Prog Retin Eye Res. 2012;31(2):152‐181.

Khatib TZ, Osborne A, Yang S, et al. Receptor‐ligand supplementation via a self‐cleaving 2A peptide‐based gene therapy promotes CNS axonal transport with functional recovery. Sci Adv. 2021;7(14): [eLocator: eabd2590].

Khalin I, Alyautdin R, Kocherga G, Bakar MA. Targeted delivery of brain‐derived neurotrophic factor for the treatment of blindness and deafness. Int J Nanomedicine. 2015;10:3245‐3267.

Jang SW, Liu X, Chan CB, et al. Deoxygedunin, a natural product with potent neurotrophic activity in mice. PLoS One. 2010;5(7): [eLocator: e11528].

Bai Y, Xu J, Brahimi F, Zhuo Y, Sarunic MV, Saragovi HU. An agonistic TrkB mAb causes sustained TrkB activation, delays RGC death, and protects the retinal structure in optic nerve axotomy and in glaucoma. Invest Ophthalmol Vis Sci. 2010;51(9):4722‐4731.

Jumper J, Evans R, Pritzel A, et al. Highly accurate protein structure prediction with AlphaFold. Nature. 2021;595(7873):583‐589.

Baek M, DiMaio F, Anishchenko I, et al. Accurate prediction of protein structures and interactions using a three‐track neural network. Science. 2021;373(6557):871‐876.

Liu C, Chan CB, Ye K. 7,8‐dihydroxyflavone, a small molecular TrkB agonist, is useful for treating various BDNF‐implicated human disorders. Transl Neurodegener. 2016;5:2.

Liu X, Obianyo O, Chan CB, et al. Biochemical and biophysical investigation of the brain‐derived neurotrophic factor mimetic 7,8‐dihydroxyflavone in the binding and activation of the TrkB receptor. J Biol Chem. 2014;289(40):27571‐27584.

Zhang Z, Liu X, Schroeder JP, et al. 7,8‐dihydroxyflavone prevents synaptic loss and memory deficits in a mouse model of Alzheimer's disease. Neuropsychopharmacology. 2014;39(3):638‐650.

Sconce MD, Churchill MJ, Moore C, Meshul CK. Intervention with 7,8‐dihydroxyflavone blocks further striatal terminal loss and restores motor deficits in a progressive mouse model of Parkinson's disease. Neuroscience. 2015;290:454‐471.

Korkmaz OT, Aytan N, Carreras I, et al. 7,8‐Dihydroxyflavone improves motor performance and enhances lower motor neuronal survival in a mouse model of amyotrophic lateral sclerosis. Neurosci Lett. 2014;566:286‐291.

Jiang M, Peng Q, Liu X, et al. Small‐molecule TrkB receptor agonists improve motor function and extend survival in a mouse model of Huntington's disease. Hum Mol Genet. 2013;22(12):2462‐2470.

Wang B, Wu N, Liang F, et al. 7,8‐dihydroxyflavone, a small‐molecule tropomyosin‐related kinase B (TrkB) agonist, attenuates cerebral ischemia and reperfusion injury in rats. J Mol Histol. 2014;45(2):129‐140.

Johnson RA, Lam M, Punzo AM, et al. 7,8‐dihydroxyflavone exhibits therapeutic efficacy in a mouse model of Rett syndrome. J Appl Physiol (1985). 2012;112(5):704‐710.

Wurzelmann M, Romeika J, Sun D. Therapeutic potential of brain‐derived neurotrophic factor (BDNF) and a small molecular mimics of BDNF for traumatic brain injury. Neural Regen Res. 2017;12(1):7‐12.

Gupta VK, You Y, Li JC, Klistorner A, Graham SL. Protective effects of 7,8‐dihydroxyflavone on retinal ganglion and RGC‐5 cells against excitotoxic and oxidative stress. J Mol Neurosci. 2013;49(1):96‐104.

Nadal‐Nicolás FM, Jiménez‐López M, Sobrado‐Calvo P, et al. Brn3a as a marker of retinal ganglion cells: qualitative and quantitative time course studies in naive and optic nerve‐injured retinas. Invest Ophthalmol Vis Sci. 2009;50(8):3860‐3868.

Goldblum D, Mittag T. Prospects for relevant glaucoma models with retinal ganglion cell damage in the rodent eye. Vision Res. 2002;42(4):471‐478.

Johnson TV, Tomarev SI. Rodent models of glaucoma. Brain Res Bull. 2010;81(2‐3):349‐358.

Guo L, Salt TE, Maass A, et al. Assessment of neuroprotective effects of glutamate modulation on glaucoma‐related retinal ganglion cell apoptosis in vivo. Invest Ophthalmol Vis Sci. 2006;47(2):626‐633.

Selles‐Navarro I, Villegas‐Perez MP, Salvador‐Silva M, Ruiz‐Gomez JM, Vidal‐Sanz M. Retinal ganglion cell death after different transient periods of pressure‐induced ischemia and survival intervals. A quantitative in vivo study. Invest Ophthalmol Vis Sci. 1996;37(10):2002‐2014.

Bui BV, Batcha AH, Fletcher E, Wong VH, Fortune B. Relationship between the magnitude of intraocular pressure during an episode of acute elevation and retinal damage four weeks later in rats. PLoS One. 2013;8(7): [eLocator: e70513].

Schmeer C, Gamez A, Tausch S, Witte OW, Isenmann S. Statins modulate heat shock protein expression and enhance retinal ganglion cell survival after transient retinal ischemia/reperfusion in vivo. Invest Ophthalmol Vis Sci. 2008;49(11):4971‐4981.

Sandalon S, Könnecke B, Levkovitch‐Verbin H, et al. Functional and structural evaluation of lamotrigine treatment in rat models of acute and chronic ocular hypertension. Exp Eye Res. 2013;115:47‐56.

Cheng L, Sapieha P, Kittlerova P, Hauswirth WW, Di Polo A. TrkB gene transfer protects retinal ganglion cells from axotomy‐induced death in vivo. J Neurosci. 2002;22(10):3977‐3986.

Iwabe S, Moreno‐Mendoza NA, Trigo‐Tavera F, Crowder C, Garcia‐Sanchez GA. Retrograde axonal transport obstruction of brain‐derived neurotrophic factor (BDNF) and its TrkB receptor in the retina and optic nerve of American Cocker Spaniel dogs with spontaneous glaucoma. Vet Ophthalmol. 2007;10(suppl 1):12‐19.

Levin LA, Schlamp CL, Spieldoch RL, Geszvain KM, Nickells RW. Identification of the bcl‐2 family of genes in the rat retina. Invest Ophthalmol Vis Sci. 1997;38(12):2545‐2553.

Ju WK, Lindsey JD, Angert M, Patel A, Weinreb RN. Glutamate receptor activation triggers OPA1 release and induces apoptotic cell death in ischemic rat retina. Mol Vis. 2008;14:2629‐2638.

Bender A, Cortés‐Ciriano I. Artificial intelligence in drug discovery: what is realistic, what are illusions? part 1: ways to make an impact, and why we are not there yet. Drug Discov Today. 2021;26(2):511‐524.

Steinegger M, Meier M, Mirdita M, Vohringer H, Haunsberger SJ, Soding J. HH‐suite3 for fast remote homology detection and deep protein annotation. BMC Bioinformatics. 2019;20(1):473.

Remmert M, Biegert A, Hauser A, Soding J. HHblits: lightning‐fast iterative protein sequence searching by HMM‐HMM alignment. Nat Methods. 2011;9(2):173‐175.

Buchan DWA, Jones DT. The PSIPRED protein analysis workbench: 20 years on. Nucleic Acids Res. 2019;47(W1):W402‐W407.

Fuchs FB, Worrall DE, Fischer V, Welling M. SE(3)‐Transformers: 3D Roto‐Translation Equivariant Attention Networks; 2020.

Chaudhury S, Lyskov S, Gray JJ. PyRosetta: a script‐based interface for implementing molecular modeling algorithms using Rosetta. Bioinformatics. 2010;26(5):689‐691.

Mariani V, Biasini M, Barbato A, Schwede T. lDDT: a local superposition‐free score for comparing protein structures and models using distance difference tests. Bioinformatics. 2013;29(21):2722‐2728.

Krammer A, Kirchhoff PD, Jiang X, Venkatachalam CM, Waldman M. LigScore: a novel scoring function for predicting binding affinities. J Mol Graph Model. 2005;23(5):395‐407.

Gehlhaar DK, Verkhivker GM, Rejto PA, et al. Molecular recognition of the inhibitor AG‐1343 by HIV‐1 protease: conformationally flexible docking by evolutionary programming. Chem Biol. 1995;2(5):317‐324.

Muegge I, Martin YC. A general and fast scoring function for protein‐ligand interactions: a simplified potential approach. J Med Chem. 1999;42(5):791‐804.

Zhu Y, Ohlemiller KK, McMahan BK, Gidday JM. Mouse models of retinal ischemic tolerance. Invest Ophthalmol Vis Sci. 2002;43(6):1903‐1911.

Ren R, Li Y, Liu Z, Liu K, He S. Long‐term rescue of rat retinal ganglion cells and visual function by AAV‐mediated BDNF expression after acute elevation of intraocular pressure. Invest Ophthalmol Vis Sci. 2012;53(2):1003‐1011.

Salinas‐Navarro M, Mayor‐Torroglosa S, Jiménez‐López M, et al. A computerized analysis of the entire retinal ganglion cell population and its spatial distribution in adult rats. Vision Res. 2009;49(1):115‐126.

Kwong JM, Caprioli J, Piri N. RNA binding protein with multiple splicing: a new marker for retinal ganglion cells. Invest Ophthalmol Vis Sci. 2010;51(2):1052‐1058.

Biermann J, Lagreze WA, Dimitriu C, Stoykow C, Goebel U. Preconditioning with inhalative carbon monoxide protects rat retinal ganglion cells from ischemia/reperfusion injury. Invest Ophthalmol Vis Sci. 2010;51(7):3784‐3791.