UM remains a devastating disease with a high propensity for metastasis and limited therapeutic options. To discover novel therapeutics, we optimized our xenograft zebrafish model, developed a robust algorithm to consistently determine model output measures and carried out a detailed analysis of zebrafish xenografts using multiple cell lines, which, to our knowledge, has not been previously investigated in this model. Previous UM-zebrafish xenografts typically utilized cell lines without known secondary mutations (OMM1[11], OMM2.3[11, 18, 22, 23], Mel270[11], and Mel285[22]) or patient-derived spheroids (spUM-LB008[26] and spXmm66[25]). Here, we describe the novel cellular behavior of established cell lines with known secondary mutations in all three clinically relevant molecular subclasses (EIF1AX, SF3B1 and BAP1) in both primary and metastasis-derived cell lines. Notably, all established cell lines were derived from patients who eventually developed metastatic disease, including the 92.1 cell line, which harbors an EIF1AX mutation.
Due to the variety of inoculation sites in zebrafish xenografts, we first evaluated which inoculation site would provide robust data that would allow investigation of the cellular behavior and effectiveness of therapeutic compounds. By generating 691 xenografts with primary and metastatic UM cell lines via retro-orbital, perivitelline space or yolk-sac injections, this study revealed that the perivitelline space is the most robust site for UM zebrafish xenografts. The quantification of zebrafish xenografts has varied across laboratories, for which we have developed Z-Tada. This provides a method for standardized analysis of the tumor volume and cell dissemination of zebrafish xenograft larvae, which accurately detects tumor cells and provides migration distances. Although yolk sac inoculation typically yielded the highest tumor burden (Supplementary Fig. 3), this site lacked robust cell dissemination (Fig. 1). Retro-orbital inoculation provided robust cell dissemination, yet the number of disseminated cells and overall tumor burden were significantly lower than perivitelline space inoculation (Supplementary Fig. 3). Interestingly, cell lines behave differently depending on the molecular subclass. The cell line 92.1 (primary EIF1AXmut) typically yielded a high tumor burden, but the size of the disseminated cell clusters was relatively small, whereas MP38 (primary BAP1mut) and MM28 (metastatic BAP1mut) developed more disseminated cell clusters with a large size (Fig. 2). Although this model provides robust data on tumor volume and cell dissemination, healthy and hTERT-immortalized melanocytes were also able to proliferate and disseminate, suggesting that this model is highly prone to cell dissemination regardless of cell type. Nonetheless, molecular subclasses with a greater probability of developing metastatic disease (SF3B1mut and BAP1mut) behave more aggressively in this model.
To evaluate whether this model system can reliably be used for novel drug screening for UM, we investigated previously used compounds that were shown to effectively inhibit tumor growth or cell dissemination in UM cells. A potential therapeutic option for high-risk UM is histone deacetylase (HDAC) inhibitors, which are able to differentiate UM into a more melanocyte-like state[34]. In mice, the histone deacetylase inhibitors inhibitor quisinostat (targeting HDAC3, 5, 8 and 9) was shown to selectively inhibit BAP1mut xenografts (MP46 and MM28)[35]. However, in zebrafish xenografts, quisinostat was also found to effectively inhibit the dissemination of metastatic cell lines (OMM2.3) or spheroids (spXmm66) that do not harbor BAP1 mutations[11, 25]. In this study, we confirmed inhibition of overall tumor burden in quisinostat-treated BAP1mut zebrafish xenografts (MP46). However, quisinostat was also able to inhibit the cell dissemination of 92.1 cells (EIF1AXmut) but failed to reduce the overall tumor burden (Fig. 4); suggesting quisinostat effectivity is dependent on molecular subclass. Another HDAC inhibitor, ricolinostat (targeting HDAC6), was previously shown to inhibit tumor burden in zebrafish xenografts using the metastatic UM cell line OMM2.3[23]. Here, we showed that ricolinostat was also able to inhibit other types of UM cells in vitro (Fig. 3) but was effective only in 92.1-based xenografts. Ricolinostat was not effective against BAP1mut-cells (MP46) in vivo. Interestingly, combining HDAC inhibitors was shown to be either synergistic (92.1) or antagonistic (MP46) based on the molecular subclass (Fig. 3, Table 1). Our zebrafish xenografts demonstrated that the synergistic effects of quisinostat and ricolinostat on 92.1 cells strongly inhibited cell dissemination but did not affect overall tumor burden. However, its antagonistic effect seen in vitro was also reproduced in MP46-xenografts, as the inhibitory effect of quisinostat was lost in the combination-treated xenografts (Fig. 4). Targeting HDACs is a promising therapy for many cancers and has been combined with several compounds that target other pathways[36]; however, combining multiple HDACs is not a strategy that has been studied in detail before. Our study suggests combining multiple HDAC inhibitors can be mutation-dependent; and should be investigated pre-clinically in detail to validate efficacy. In addition to HDAC inhibitors, we evaluated withaferin A, a compound tested only in mice. Withaferin A is able to inhibit the MET and MEK1/2 pathways, making this an interesting compound for UM because it acts on the primary driver pathways[33]. In vitro, withaferin A inhibited the growth of all the tested cell lines. However, withaferin A in combination with the spliceosome inhibitor E7107 had synergistic effects on Mel202 cells, yet failed to illustrate synergistic effects on other cell lines tested (Fig. 3, Table 1). The synergistic effect was reproduced in Mel202-xenografts, suggesting that this combination could provide novel therapeutic options for SF3B1mut UM patients (Fig. 4).
Zebrafish xenografts hold potential as a drug screening platform for UM, where synergism and antagonism can be studied in detail. However, this study only treated xenografts for a total of 3 days, which could be too short to see a more defined inhibitory effect. Other important elements in this platform that could affect drug efficiency are the number of fish treated per well and the efficiency of compound uptake by zebrafish larvae. However, an important advantage of this model is functional liver metabolism in zebrafish larvae[37], which can metabolize compounds and therefore alter their effects. Nonetheless, this platform illustrates that molecular subclasses of UM are important parameters for discovering novel therapeutic compounds.
In summary, this study illustrates that optimal inoculation of uveal melanoma cells in the perivitelline space allows for robust tumor burden and cell dissemination analysis in a short time span of 3 days. Using Z-Tada, this model is able to identify differences in cellular behavior depending on genetic background in a standardized fashion. Furthermore, this methodology allows standardized read-out parameters for high-throughput screening of novel compounds and can identify synergistic or antagonistic effects. Using this model, we provide evidence that clinically prognostic subclasses of UM are key to developing effective therapies. Synergistic compound screenings for UM should take this into account, as the same combination can be synergistic in low-risk UM but antagonistic in high-risk UM.