Topical instillation of cell-penetrating peptide-conjugated melphalan blocks metastases of retinoblastoma

Retinoblastoma is the most common primary intraocular malignancy in infancy with a metastases-related death risk. However, a safe and convenient treatment without enucleation is still an unmet clinical need. In this work, a cell-penetrating peptide, 89WP, was conjugated with melphalan (89WP-Mel), which achieved comparable tumor inhibition effects to intravitreally injected melphalan via topical instillation for the rst time. Notably, the “outside-in” diffusion of instilled 89WP-Mel created a protective shield surrounding the eye, eciently preventing tumor metastases. In contrast, the mice treated with intravitreally injected melphalan suffered more brain metastases related death, probably due to the “inside-out” diffusion of injected melphalan expelling the tumor outside the eye. The ocular absorption of 89WP-conjugated melphalan and other small molecules, both hydrophobic and hydrophilic, occurred via non-corneal pathway with high safety and a prolonged residence duration in retina up to 24 h. The present work paves a new avenue for simultaneous intraocular tumor inhibition and extraocular metastases prevention in a safe and convenient way via topical instillation.


Introduction
Retinoblastoma (RB) is the most frequent primary intraocular malignancy in infancy and childhood, about 9000 new cases every year 1 . Potential extraocular extension, including brain and bone marrow metastases, causes RB a life-threatening disease 2 . Brain metastases is usually with nearly 100% mortality rate, therefore, preventing metastases is the rst priority during treatment of RB 3 . Few years ago, RB was treated only by eye enucleation, leading to a permanent vision loss, but today can be treated by conservative chemotherapy with preservation of functional vision in developed regions 4 . E cient control of RB requires early detection, following timely and appropriate treatment 5 . Besides, blocking the extraocular spread especially the brain metastases is of vital importance 6 .
Interventions such as intra-arterial 7 or intravitreal 8 chemotherapy improve the outcomes in advanced intraocular RB by offering higher doses of drugs in retina and vitreous body. Lack of vasculature structure in the vitreous body makes intravitreal chemotherapy the preferred method for treating sub-retinal and vitreous tumor seeds, and melphalan (20-30 µg/0.1 mL, once a week or on alternate weeks) is the most common used drug 3,9 . Notably, intravitreal chemotherapy is probably with extraocular tumor extension and metastasis risk via the needle tracks or other routes 6 . Besides, frequent intravitreal injections, mainly due to quick elimination of melphalan in vitreous body 10 , request a high-quality medical environment and a practiced clinician, which is scarce in developing regions with limited health care systems 4 . Indeed, a safe and convenient treatment, for example topical instillation, could substantially improve the conditions. Unfortunately, it is di cult for small molecules like melphalan to be delivered into the posterior ocular segment via topical instillation.
Penetratin was previously reported as a more safe and powerful ocular permeation enhancer than several other cell-penetrating peptides 11 and could promote ocular delivery of both small molecules and nucleic acids [12][13][14][15][16] . By hydrophobic amino acids mutation, various penetratin derivatives were obtained with even better ocular permeability, and tryptophan mutation at the eighth (glutamine) and ninth (asparagine) amino acid residues resulted in an excellent ocular permeable peptide (Q8W, N9W-Penetratin, 89WP) 17 . In this work, covalent conjugate of 89WP and melphalan (89WP-Mel) was constructed to achieve e cient inhibition on intraocular tumor proliferation by topical instillation. It was also postulated that the instilled 89WP-Mel would form a protective shield surrounding the eye through the non-corneal absorption route and siege the intraocular tumor. Besides exploring the mechanisms for 89WP-Mel to hinder brain metastases of RB, the potential of turning melphalan from intravitreal injection to eye drops for RB therapy by conjugation with 89WP was also investigated.

Results
Synthesis and characterization of 89WP conjugates. Harnessing previously designed penetratin derivatives with high permeability to the eye 17 , we combined one of powerful derivatives (89WP) with melphalan to explore if topical instillation of the covalent conjugate 89WP-Mel could inhibit intraocular tumor proliferation. Molecular structure, synthetic routes and characterization of conjugates are shown in Fig. 1, S1 and Table 1. ). After 1-month treatment, the sizes of vitreous seeds in different groups were compared based on the whole eye sections stained with hematoxylin-eosin. As shown in Fig. 2A, the vitreous seeds after the treatment of 89WP-Mel (both low-and high-dose) were much smaller than that in the normal saline treated group. Topical instillation of high-dose 89WP-Mel (89WP-Mel-High) achieved comparable antiretinoblastoma effects to the positive control (IVT. Mel). However, mice in IVT. Mel treated group suffered unexpected death. Three out of ve mice died during the treatment, while only one death occurred in the high-dose 89WP-Mel treated group. According to the immunohistochemical section of the whole brain from a surviving mouse in the group treated by topically instilled melphalan (Fig. 2B), an area with distinct GFP signal from the metastasizing retinoblastoma cells emerged in the region of spinal trigeminal nucleus, which connected with the eye via the optic nerve 3 .
To verify the anti-metastasis effect of topical instillation of 89WP-Mel, pharmacodynamical evaluation was further implemented with a larger size of samples (n = 8), accompanying with the optimized dose of instilled 89WP-Mel (containing 3.0 mg/mL melphalan, 10 µL/eye, once daily) and adapted dose regimen of intravitreally injected melphalan (0.5 mg/mL, 2 µL/eye, once on alternate weeks). During the 1-month treatment, 89WP-Mel exhibited e cient inhibitory effects against intraocular tumor, which was comparable to intravitreally injected melphalan (Fig. 3B, Fig. S2). In contrast, topical instillation of melphalan at the same dose was invalid. The mice treated with 89WP-Mel and intravitreally injected melphalan suffered a much lower intraocular tumor proliferation, registering one-fth and one-third of the increasing folds of bioluminescence in comparison to the normal saline treated group on the Day 30 (p < 0.0001, Fig. 3C). However, there was no statistically signi cant difference between the groups treated by normal saline and topically instilled melphalan (p = 0.9087). More importantly, the mice treated with 89WP-Mel resulted in an improved survival rate, which was even better than those receiving intravitreally injected melphalan (Fig. 3D). Brain metastases were evaluated based on immunohistochemical sections of the brains from moribund mice with a sudden decrease of body weight lower than 21 g ( was approximate 100%, which was much higher than the small molecules alone or their mixtures with 89WP (89WP/FAM and 89WP/Cy5). In HCEC cells, the mean uorescence intensity of the 89WP-FAM treated cells was respective 37 (p < 0.001) and 28 (p < 0.001) times higher than the FAM or 89WP/FAM treated cells. In ARPE-19 cells, the increasing folds were respective 342 (p < 0.001) and 272 (p < 0.001). For covalent conjugate Cy5-89WP, the increased folds were 125 (p < 0.001) and 28 (p < 0.001) in HCEC cells, 105 (p < 0.001) and 78 (p < 0.001) in ARPE-19 cells. Notably, there was no statistically signi cant difference (p > 0.05) in cellular uptake between small molecules and their mixtures with 89WP in two cells, suggesting that covalent conjugation was indispensable to improve cellular uptake of small molecules.
According to intracellular distribution of small molecules and the covalent conjugates in HCEC and ARPE-19 cells (Fig. S7, S8), there was no obvious uorescent signal in FAM or Cy5 treated cells. However, much stronger uorescence could be observed in the cells treated with the covalent conjugates. More importantly, most of the uorescence from the covalent conjugates was not co-localized with lysosomes, indicating a potential endosomal escape ability.
To discern cell entry pathways of the covalent conjugates, cellular uptake of 89WP-FAM in ARPE-19 cells was explored in the presence of various inhibitors or at different temperatures 20 . Compared to the uninhibited control group (Fig. 4B), cellular uptake decreased substantially either at low temperature or in the presence of chlorpromazine and hypertonic sucrose, pointing to an energy-dependent and clathrinmediated endocytosis pathway. The endosomal escape capability of 89WP was evaluated by hemolysis assay. At the concentrations ranging from 40 to 640 µM, hemolysis rate induced by 89WP at pH 5.5 was 2-3 times higher (p < 0.05) than that at pH 7.4 ( Fig. 4C), further suggesting enhanced endosomal escape of the covalent conjugates.
Sclera was more permeable than cornea for 89WP-conjugated small molecules. The ARPE-19 cell monolayer model was established to evaluate 89WP-FAM permeability across the posterior ocular barriers (Fig. S9). As shown in Fig. S9B, there was nearly no green uorescent signal in FAM or 89WP/FAM mixture treated monolayer, but conspicuous uorescence was observed in the monolayer after the treatment of 89WP-FAM. From the side view ( Fig. 5A (a)), 89WP-FAM could permeate the whole monolayer. In addition, 89WP-FAM crossed the monolayer in a linearly time-dependent manner within 4 h ( Fig. 5A (b)). After 4 h incubation, up to 16.5% of the 89WP-FAM was found in the basolateral side, which was 5.5 times more than the free FAM and the 89WP/FAM mixture. Apparent permeability coe cient (P app ) of 89WP-FAM was 3.46 × 10 − 5 cm/s, which was 6 times higher than the others (p < 0.0001, Fig. 5A (c)). After penetrating the monolayer, 89WP-FAM remained intact in the extracted sample from the basolateral side, testi ed by fragment ions in ESI-MS spectrum (Fig. S9C). Besides, signi cant cellular uptake of the permeated 89WP-FAM was observed by WERI-Rb-1 cells seeded in the basolateral side ( Fig.  S9E, S9F).
Freshly excised rabbit cornea and sclera were exploited to investigate permeation of Cy5-89WP across the ocular absorption barriers. For Cy5 treated group, nearly no uorescent signal was observed up to 4 h in cornea or sclera (Fig. S10). In contrast, Cy5-89WP was found in corneal epithelium after 1 h treatment, and remained up to 4 h ( Fig. 5B (a)). Cy5-89WP was also clearly observed in sclera in 1 h, and the permeation depth of Cy5-89WP in sclera demonstrated a ve-fold increase from 1 h to 4 h (Fig. S10B).
According to the quantitative results ( Fig. 5B (b)), Cy5-89WP crossed the excised cornea and sclera in a linear manner within 4 h. After diffusion for 4 h, the cumulative amount of Cy5 in basolateral side was respective 0.27% for cornea and 0.85% for sclera. However, they increased to 1.04% and 6.13% for Cy5-89WP. The P app of Cy5 across cornea and sclera was respective 7.3 × 10 − 7 cm/s and 2.3 × 10 − 6 cm/s, while Cy5-89WP demonstrated a 3.8-and 7.4-time increase across cornea and sclera (p < 0.0001, Fig. 5B (c)). It was also worth noting that the P app of Cy5-89WP across sclera was 7.4 time higher than across cornea.
89WP enabled penetration of the conjugates into the back of eye via a non-corneal pathway. After topical instillation, free FAM or Cy5 solution could hardly be absorbed into the mice eyes, evidenced by no perceptible uorescence signal in the whole eye balls (Fig. 6A). In contrast, topical instillation of 89WP-FAM and Cy5-89WP into the conjunctival sac led to obvious uorescent signal, mainly in the posterior ocular segment. According to the semi-quantitative results (Fig. 6B, 6C), the uorescence intensity of the covalent conjugates in retina peaked at 4 h after instillation with a retention time longer than 24 h. However, there was only little FAM distributed in retina from 2 to 4 h after instillation, and nearly no Cy5 could reach retina. In the retinal sections ( Fig. S11), the covalent conjugates abundantly accumulated in the periocular tissue during 4 to 18 h, which served as a depot to achieve extended release of conjugates to the retina. These results suggested that 89WP could alternatively facilitate penetration of covalent conjugates into the back of the eye via a non-corneal pathway.
89WP-conjugation strategy was safe for noninvasive intraocular delivery of small molecules. As the eye is a particularly susceptible organ, safety evaluation of 89WP-Mel both in vitro and in vivo was incredibly important. As shown in Table 2  cells) were about 500 folds higher than that on WERI-Rb-1 cells (0.28 µM). According to the results of the ex vivo tissue sections, covalent conjugates did not cause impairment on corneal epithelium and sclera after 4 h incubation (Fig. S13). After pharmacodynamical evaluation, the treated cornea and visceral organs were sectioned. No any damage occurred during the long-term treatment (Fig. S14, S15). Besides, melphalan is insoluble in normal saline (< 1µg/mL) and therefore organic solvents such as propylene glycol and ethanol were needed in the formulations of intraocular injection or eye drops 18 . However, solubility of melphalan in normal saline increased to over 5 mg/mL after conjugated with 89WP, which could allow abolishing irritant organic solvents in the formulation. Recently, cell-penetrating peptides are emerging as promising tools for improved ocular drug delivery 23,24 .
In this work, we postulate that penetratin derivative-conjugation strategy may improve absorption of eye drops to achieve effective intraocular concentration of drugs, which are only intravitreally applied before. On a retinoblastoma-bearing mice model, 89WP-Mel demonstrated signi cantly better intraocular tumor inhibition effects via topical instillation than free melphalan. Besides, the high-dose 89WP-Mel treatment e ciently restricted the proliferation of vitreous seeds and displayed similar tumor inhibition effects compared with intravitreally injected melphalan. According to the results of the immunohistochemical stained sections of mouse brain, the positive signal in spinal trigeminal nucleus revealed the intraocular tumor would metastasize to brain to induce almost 100% mortality. Based on our previous work 16, 17 , we considered that "outside-in" absorption of 89WP-Mel would con ne the tumor cells inside the eye, while the diffusion of intravitreally injected melphalan compelled the metastases. This hypothesis was further veri ed by the results of pharmacodynamical evaluation. 89WP-Mel treated mice possessed the highest survival rate and the lowest brain metastases rate (Fig. 3).
The ocular adsorption route of 89WP-Mel was explored to verify the probable anti-metastases mechanism. Conjugation with 89WP ensured substantial increase in cellular uptake of small molecules in HCEC and ARPE-19 cells. However, cellular uptake of the mixture with 89WP displayed no signi cant difference compared to the free small molecules, suggesting the indispensable role of covalent conjugation in facilitating the uptake of these molecules. Besides, the covalent conjugates entered cells through an energy-dependent and clathrin-mediated endocytosis pathway, accompanying with an endosomal escape capability enabled by 89WP to further achieve the posterior ocular permeation.
In the in vitro models of ARPE-19 cell monolayer, 89WP-FAM exhibited high permeability across the whole monolayer, and could penetrate the monolayer in an intact structure, followed with entry into tumor cells in the bottom chamber. As a hydrophilic molecule, it was di cult for Cy5 to enter into cornea or sclera. In contrast, the permeability of Cy5-89WP dramatically improved, especially in sclera. Cy5-89WP was also endowed with additional adhesive potential on the corneal surface. In other words, topical instillation of Cy5-89WP or other 89WP-engaged conjugates would adhere to the ocular surface, probably due to electrostatic interaction between cationic 89WP and anionic mucin on the ocular surface, and subsequently permeated into the eye mainly through the sclera 25 .
By topical instillation, neither hydrophobic FAM nor hydrophilic Cy5 alone could permeate into the eye mainly due to tear turnover, corneal epithelial barrier and other ocular protecting barriers, which was a common absorption challenge for most of eye drops in the market 23,26 . Both 89WP-FAM and Cy5-89WP exhibited obvious distribution in the posterior ocular segment after topical instillation, but much less distribution in anterior segment, which further con rmed ocular absorption of the covalent conjugates mainly through the non-corneal pathway. In addition, both conjugates could reside in the retina up to 24 h, which was critical to reduce the administration frequency.
Taken together, conjugation with 89WP rmly improved ocular absorption of small molecules, especially in the posterior segment mainly through a non-corneal pathway. As shown in Fig. 7, we deduced that such "outside-in" absorption manner of topically instilled 89WP-Mel created a protective shield surrounding the whole eye, which killed tumor cells in a "siege" way and therefore would e ciently prevent extraocular tumor metastases, either via optic nerve or the circulating subarachnoid uid tumor cells 27 . In contrast, although with almost 100% drug bioavailability, the "inside-out" diffusion made intravitreally injected melphalan kill tumor cells in a "pursuit" way, which might weaken the e ciency on distant tumor cells and even "facilitate" tumor metastases.
Considering melphalan is a cytotoxic agent, biosafety for ocular application is also important for the druggability 28 . In cytotoxicity evaluation, the IC 50 values of 89WP-Mel on HCEC and ARPE-19 cells were ≥ 500 times more than that on WERI-Rb-1 cells. When the concentration of melphalan was 1-100 µM, 89WP-Mel exhibited nearly no toxicity on HCEC and ARPE-19 cells, but the survival rate of WERI-Rb-1 cells was lower than 50%. Compared to tumor cells WERI-Rb-1, normal ocular cells HCEC and ARPE-19 were relatively low proliferative and would be much less sensitive to melphalan 29,30 . In addition, there was also no obvious tissue injury occurred caused by topical instillation of 89WP-Mel. Due to low solubility of melphalan, the marketed product contains propylene glycol and ethanol in solvent, which might induce the ocular irritation during long-term use 18, 31 . Conjugation with hydrophilic 89WP could increase the solubility of melphalan more than 5000 times in physiological saline, implying that 89WP-Mel could be dissolved in aqueous media as eye drops for intraocular tumor inhibition at a concentration over 5.0 mg/mL (melphalan) without the need of any other co-solvent.

Conclusions
In the present work, melphalan was successfully conjugated with a specially designed ocular permeation enhancer 89WP without impairing the antineoplastic activity. Topical instillation of 89WP-Mel was proven as a safe and e cient anti-retinoblastoma treatment, which exhibited not only the signi cant improvement in infant compliance but also additional advantages over intravitreally injection, especially in preventing brain metastases. The probable mechanism was related to the enhanced ocular absorption and the "outside-in" absorption manner of 89WP-Mel. Besides, conjugation with hydrophilic 89WP increased the solubility of insoluble molecules such as melphalan, which make it possible that the chemotherapeutics could be administered as an ophthalmic solution. Taken together, 89WP-conjugation strategy could change intraocular injections of various small molecules to much safer eye drops, also providing an alternative approach on the treatment of metastatic intraocular diseases.  Solid phase synthesis was used to accomplish conjugation of 89WP and melphalan (89WP-Mel) using Wang resin according to Fmoc methodology. Fmoc-protected melphalan (Fmoc-Mel) was synthesized via mixing melphalan (1.0 eq.) with Fmoc-OSu (1.2 eq.) in dioxane at 50°C with agitation overnight. The mixture was directly used for next procedure after the solvent dried under vacuum. Fmoc-Mel was added to Wang resin (0.5 eq.) after 5 min of pre-activation in dichloromethane (DCM) using DMAP (1.2 eq.) and DIC (1.0 eq.) as catalysts. After washing with dimethylformamide (DMF), acetic anhydride (5.0 eq.) and pyridine (5.0 eq.) were introduced to block unreacted active sites of the resin. The resin was immersed in piperidine/DMF (2:8) solution for 15 min twice to remove Fmoc-group and then rinsed with DMF. Further peptide condensation was facilitated by using excess Fmoc-protected amino acid (8.8 eq. to Wang resin, similarly hereinafter) activated with HBTU (8.8 eq.), HOBT (8.8 eq.) and DIEA (17.6 eq.) for 1 h in DMF.

Materials
The To identify probable internalization mechanisms of 89WP-engaged conjugates, ARPE-19 cells were preincubated at 4°C or treated with inhibitors at 37°C for 0.5 h, then incubated with the 89WP-FAM at 4°C or in the presence of inhibitors at 37°C for another 1.5 h. The cellular uptake of 89WP-FAM at 37°C was used as the positive control (100% uptake e ciency). The results are shown as relative uptake rate of different groups normalized according to the positive control.
Endosomal escape ability of 89WP-engaged conjugates was evaluated via an erythrocyte lysis assay 32 . Brie y, freshly prepared human erythrocytes were resuspended in sodium citrate buffer (300 mM NaCl, 30 mM sodium citrate) at a concentration of 7 × 10 7 cells/mL, following incubation with equal volume of 89WP solution at a series of concentrations for 1 h at 37°C. After centrifugation (1000 g × 5 min), 100 μL of the supernatant were transferred to a 96-wells plate and OD 450 nm was determined via microplate reader (Bio-Tek, United States). The distilled water was used as the negative control (no hemolysis) while 10% Triton X-100 solution was used as the positive control (100% hemolysis).
HCEC, ARPE-19 or WERI-Rb-1 cells at the logarithmic growth phase were cultivated on 96 well plates at a primary cell density of 2,000 cells/well for 24 h. Cells were then incubated with 89WP, melphalan or 89WP-Mel at concentration gradients for 4 h, following a further cultivation in 200 μL complete medium for 20 h except WERI-Rb-1 cells. After that, cells were incubated with 0.5 mg/mL MTT reagent in each well for 4 h at 37°C, and produced formazan was dissolved in 150 μL DMSO, except that WERI-Rb-1 cells were treated with CCK-8 reagent for 2 h after incubation with 89WP-Mel. The OD 490 was measured via microplate reader.
Transport e ciency across in vitro ARPE-19 cell monolayer The in vitro ARPE-19 cell monolayer model was established as previously reported with slight changes 21 . Brie y, ARPE-19 cells were seeded on the front of rat tail collagen type coated 6.5 mm polyester membrane lters (pore size 0.4 μM, Corning) at a density of 5,000 cells/well, respectively. Transepithelial electrical resistance (TEER) was measured by an epithelial volt-Ωm (Millipore, USA) to con rm the barrier integrity. Monolayers with TEER over 100 Ω·cm 2

Ex vivo corneal and scleral permeability
Permeability of the Cy5-89WP through excised rabbit cornea and sclera was evaluated using horizontal diffusion cells as previously reported 11 . Animals were euthanized by an overdose of sodium pentobarbital (150 mg/kg) administered through marginal ear vein. Then the cornea and sclera were excised and immediately transferred into normal saline. Each cornea or sclera was placed vertically between the diffusion cells with epithelium orienting to the donor cells. Both donor and acceptor cells contained 3.5 mL normal saline as the diffusion medium, which was maintained at 34 ± 0.5°C by circulating water bath, and 5 μM Cy5 or Cy5-89WP was added in the donor side, respectively. An aliquot of 500 μL of sample was extracted from the acceptor cell every 0.5 h for 4 h, and followed with supplement of 500 μL fresh normal saline. Fluorescence intensity of the samples was measured via microplate reader. To compare ex vivo tissue permeability of Cy5 and Cy5-89WP, the apparent permeability coe cients (P app , cm·s -1 ) were calculated as mentioned above.
After the diffusion, the cornea and sclera were xed in 4% paraformaldehyde, dehydrated in 30% sucrose solution and dyed with hematoxylin-eosin or DAPI for further observation by microscope.

Intraocular distribution and pharmacokinetics
Male mice (18 ~ 20 g) derived from Institute of Cancer Research, United States were used in this experiment. Both 30 μM 89WP-FAM and Cy5-89WP dissolved in 10 μL arti cial tear uid 33 were instilled into the conjunctival sac of mice, respectively. After instillation, the mice were sacri ced (injection with the fatal dose of pentobarbital sodium, 150 mg/kg) 0.

Inhibition of intraocular retinoblastoma via topical instillation
Intraocular tumor-bearing mice model were established as previously reported 16 . Brie y, the male Balb/c nude mice were anesthetized by intraperitoneal injection with pentobarbital sodium (30 mg/kg) in combination with topical application of 0.4% oxybuprocaine hydrochloride. Topical instillation of 0.5% tropicamide was used to dilate the pupil. Fluc/GFP-Rb-1 cells (2 × 10 4 cells suspended in 2 μL 10 mM PBS) were injected slowly into the vitreous body near retina of the right eye using a microsyringe (33 G, Hamilton), following instillation of 0.25% chloramphenicol. Ocular luminescence of the mice was detected in the following days and the tumor-bearing mice were randomly divided into ve groups (n = 5) in the rst evaluation and four groups in the second evaluation (n = 8).
In the rst evaluation (safety and preliminary pharmacodynamics evaluation), mice were treated with 10 μL of normal saline (N.S.), melphalan (3.0 mg/mL, dissolved in a cocktail of 1, 2-propanediol, ethyl alcohol, sodium citrate and distilled water), 89WP-Mel (containing melphalan 0.3 or 3.0 mg/mL, dissolved in normal saline) via topical instillation once daily from Day 0 to Day 30. Intravitreal injection of melphalan (8.0 mg/mL in 2 μL cocktail mentioned above, once a month) was used as the positive control. After 1-month treatment, the mice were sacri ced (injection with the fatal dose of pentobarbital sodium, 150 mg/kg) and the treated eyes and visceral organs were xed in 4% paraformaldehyde, dehydrated in 30% sucrose solution and dyed with hematoxylin-eosin (HE) for further observation under microscope. Brain from a survival mouse in topically instilled melphalan group was also treated as above and implemented immunohistochemical experiments using HRP-labeled anti-GFP primary antibody (ab183734, Abcam) for metastasizing tumor cells detection.
In the second evaluation (optimized pharmacodynamics evaluation), mice were treated with 10 μL of normal saline, melphalan (3.0 mg/mL), 89WP-Mel (containing melphalan 3.0 mg/mL) via topical instillation once daily from Day 0 to Day 60. Intravitreal injection of melphalan (0.5 mg/mL in 2 μL medium composed of 10% cocktail and 90% N.S.) once on alternate weeks was used as the positive control. The mice were intraperitoneal injected with 150 mg/kg D-luciferin and anesthesia by iso urane 15 min before luminescence signal was detected via the IVIS Spectrum system (PerkinElmer, USA). In vivo imaging was carried out on Day 0, 2, 4, 6, 9, 12, 15, 17, 20, 25, and 30. When the mice was moribund (body weights fast decreased to lower than 21 g) or at the end of experiment (Day 60), the whole brain of sacri ced mice was xed in 4% paraformaldehyde, dehydrated in 30% sucrose solution and was implemented immunohistochemical experiments using HRP-labeled anti-GFP primary antibody (ab183734, Abcam) for metastasizing tumor cells detection.

Statistical analysis
Statistical signi cances of the quantitative data were analyzed by the One-way ANOVA multiple comparison or t-test. Considered that p > 0.05 was no signi cant, p < 0.05 was signi cant, and p < 0.01 even p < 0.001 was highly signi cant.

Declarations
Date availability The data that support the ndings of this study are available within the paper and the Supplementary Information. All other data are available from the authors upon reasonable request.

Figure 2
Intraocular tumor proliferation and brain metastases of retinoblastoma-bearing mice at treatment ending.