Monosomy 3 Influences Epithelial-Mesenchymal Transition Gene Expression in Uveal Melanoma Patients; Consequences for Liquid Biopsy

Despite outstanding advances in diagnosis and the treatment of primary uveal melanoma (UM), nearly 50% of UM patients develop metastases via hematogenous dissemination, driven by the epithelial-mesenchymal transition (EMT). Despite the failure in UM to date, a liquid biopsy may offer a feasible non-invasive approach for monitoring metastatic disease progression and addressing protracted dormancy. To detect circulating tumor cells (CTCs) in UM patients, we evaluated the mRNA expression of EMT-associated transcription factors in CD45-depleted blood fraction, using qRT-PCR. ddPCR was employed to assess UM-specific GNA11, GNAQ, PLCβ4, and CYSLTR2 mutations in plasma DNA. Moreover, microarray analysis was performed on total RNA isolated from tumor tissues to estimate the prognostic value of EMT-associated gene expression. In total, 42 primary UM and 11 metastatic patients were enrolled. All CD45-depleted samples were negative for CTC when compared to the peripheral blood fraction of 60 healthy controls. Tumor-specific mutations were detected in the plasma of 21.4% patients, merely, in 9.4% of primary UM, while 54.5% in metastatic patients. Unsupervised hierarchical clustering of differentially expressed EMT genes showed significant differences between monosomy 3 and disomy 3 tumors. Newly identified genes can serve as non-invasive prognostic biomarkers that can support therapeutic decisions.

As very little is known about EMT's molecular nature in mesenchymal tumors, including UM, the whole-genome gene expression approach was applied with a focus on EMT-associated genes to clarify the role of EMT in hematogenous dissemination and to identify new, prognostically relevant, differentially expressed genes. Moreover, we assessed the presence of CTCs and ctDNA in the peripheral blood of primary and metastatic UM patients, focusing on the detection of traditional EMT-associated transcription factors (TFs) and driver mutations, previously associated with UM development and progression. These results can help to a better understanding of the factors contributing to hematogenous dissemination in UM.

Clinico-Pathological Characteristics of Patients
In the present study, 53 UM patients were enrolled between August 2018 and September 2020; among them, 10 diagnosed in stage IV. One of the primary UM patients developed metastases 8 months after treatment of the primary tumor (Table 1). Patient sex ratio was similar, with 47.2% (n = 25) being males and 52.8% being (n = 28) females. The median age at the time of diagnosis was 67 years (range 33-87 years). The right eye was affected in 49.1% (n = 26), the left eye in 50.9% (n = 27) of the patients. The median tumor volume was 1.1 cm 3 (ranging between 0.2-2.6 cm 3 ); 71.1% (n = 38) of patients had a tumor volume less than 1.55 cm 3 , while in 28.3% (n = 15) the tumor volume exceeded 1.55 cm 3 . The majority of the tumors, 48.7% (n = 19), were spindle-cell, while 25.6% (n = 10) were classified as epithelioid and the same number as mixed. Altogether 60.4% (n = 32) of patients underwent enucleation without prior treatment; 15.1% (n = 8) underwent enucleation after radiation therapy in the past and 24.5% (n = 13) were treated by stereotactic radiosurgery. Locally advanced disease characterized by vascular cell invasion was diagnosed in 12.8% (n = 5) cases, lymphogenic invasion in 25.6% (n = 10), and perineural spread was detected in 23.1% (n = 9) of patients. Secondary malignancy occurred in six cases, namely auricular tumor, gynecologic cancer, lung carcinoma, prostate cancer, cancer of unknown primary (suspect lipoma confirmed by histopathology as metastasis), and colon cancer in patients UM2, UM20, UM31, UM36, UM44, and UM58, respectively. The presence of secondary malignancies and metastases in individual patient samples are shown in Figure 1. . Schematic presentation of the major conducted analyses. Blood was collected for all patients enrolled except for one (n = 52). Tumor tissues were available for 31 patients who underwent enucleation. M3 and D3 status were assessed by MLPA; M3 is highlighted by brown, D3 by grey color. The quality of tumor material in the samples collected after radiotherapy in the past was poor, hampering subsequent MLPA analysis. Sanger sequencing was done in tumor tissues focusing on GNA11 p.Q209L, p.Q209P, and p.R183C; GNAQ p.Q209P, p.Q209L, p.Q209R, and R183Q; PLCβ4 D630Y, and CYSLTR2 L129Q mutations. Identical mutations were assessed by digital droplet PCR (ddPCR) in tumor tissues and subsequently in DNA extracted from plasma samples.

Multiplex Ligation-Dependent Probe Amplification
MLPA was used to evaluate deletion/duplication status of specific loci located on chromosomes 1, 3, 6, and 8, considered recurrent genetic alterations in UM. As chromosome 3 monosomy strongly correlates with metastatic death, while chromosome 8 gains occur later in UM tumorigenesis, we focus herein on monosomy 3 (M3) only. M3 was detected in 51.6% (n = 16), disomy 3 (D3) in 48.4% (n = 15) of analyzed tumor tissues. Class 2 expression profile was simultaneously detected in all M3 tumors in which gene expression profiling was performed ( Figure 2). The only exception was UM56, where the Class 2 gene expression profile was associated with 1p loss and 8q gain, without M3 presence, therefore classified as D3 based on MLPA results. Figure 2. Unsupervised clustering of 12 Class 1 vs. Class 2 discriminating genes. Multiple gene probes were present twice on the chip. Chromosome 3 status is shown on the left side, M3 tumors are highlighted by brown, D3 tumors are highlighted by gray. The UM56 tumor with 1p loss and 8q gain is highlighted by orange. The red color represents up-regulated gene expression, the blue color represents down-regulated gene expression while the yellow represents no change.

EMT-Associated TF Expression in Peripheral Blood of UM Patients
We assessed the gene expression of four EMT-associated TFs, namely SNAI1, SNAI2, TWIST1 and ZEB1, and epithelial marker keratin 19 (KRT19) in a CD45-depleted fraction of 34 primary and five metastatic UM patients' peripheral blood. CTC enrichment was performed using RosetteSep™ Human CD45 Depletion Cocktail. TaqMan gene expression assays (described in more detail in the Materials and methods paragraph) were used for gene expression analysis. KRT19 and all studied EMT TFs gene transcripts, except for ZEB1, were mostly undetectable in peripheral blood of UM patients (39/39 for KRT19 and SNAI2, 33/39 for SNAI1, and 22/39 for TWIST1). Although gene expression of ZEB1 was higher than the other studied genes, all samples were evaluated as negative compared to cut-off values, set based on the expression of ZEB1 in CD45-depleted peripheral blood fraction of 60 healthy controls [19]. We also compared the ZEB1 gene expression of M3 and metastatic patients to those of D3; however, no significant differences were found ( Figure 3). Relative ZEB1 expression in a CD45-depleted peripheral blood fraction of M3 and five metastatic UM patients (highlighted by red) compared to D3. The difference between D3 and M3 ZEB1 expression was not significant.

Circulating Tumor DNA
Given the negative results of CTCs detection using CD45-depleted peripheral blood, droplet digital polymerase chain reaction (ddPCR) was used to assess if the analytical sensitivity of RT-PCR or other factors were responsible for negative findings. ddPCR has been proven as a highly sensitive detection method allowing for the identification of rare circulating tumor-specific DNA fragments (ctDNA) in peripheral blood. Eight ddPCR assays were used (their numbers are listed in the material and methods paragraph) to detect nine mutations in four genes, namely, GNA11 p.Q209L, p.Q209P, and p.R183C; GNAQ p.Q209L, p.Q209P, p.Q209R, and p.R183Q; PLCβ4 p.D630Y, and CYSLTR2 p.L129Q. These mutations were selected as they have been reported to be present in more than 90% of UM tumors. Firstly, all mutations were interrogated in tumor tissues (available for 31 patients) by ddPCR and then identical mutations were detected in the same tissues using Sanger sequencing to validate assays performance. When 100% agreement was confirmed between the two methods, ddPCR was used to detect the same mutations in peripheral blood of 52 UM patients. GNA11 p.Q209L and GNAQ p.Q209P were the two most frequent mutations. GNA11 p.Q209L was present in 54.8% (n = 17), while GNAQ p.Q209P in 32.3% (n = 10) tumors. GNA11 p.Q209P, GNA11 p.R183C, GNAQ p.Q209L, and GNAQ p.Q209R mutations were identified each in one tumor sample ( Table 2). In plasma samples, tumor-specific mutations were identified in 21.4% (n = 9) of patients, only in 9.4% (n = 3) primary UMs, while in 54.5% (n = 6) of metastatic patients. In primary UM patients, the median value was 0 cfDNA copies per µL (minimum = 0, maximum = 0.13), while in metastatic patients the median was 0.3 copies per µL (minimum = 0, maximum = 108). The number of copies per µL was significantly higher in metastatic patients (p < 0.001, Figure 4). The patient presenting with an extremely high number of copies (n = 108/µL, UM56) was diagnosed with a locally advanced primary disease after the onset of metastatic disease. The p-value remained equally significant even after excluding this extreme value from the analysis. We did not find a correlation between tumor volume and the number of copies. The patient UM56 with the highest number of copies had extremely locally advanced disease at the time of diagnosis. The difference in the number of copies between primary and metastatic UM was significant. The p-value also remained the same after the outlier (UM56) was excluded from the analysis.

Gene Expression Profiling of EMT-Associated Genes
Finally, the whole-genome mRNA expression analysis was performed in 23 tumor tissues based on chromosome 3 status. The data on EMT-associated genes mRNA expression were acquired from the SurePrint G3 Human Gene Expression 8×60K v2 Microarray (Agilent). Altogether 277 genes (322 probes) among 1184 genes from the EMT gene database (http://dbemt.bioinfo-minzhao.org/) significantly differed, but only 127 genes (143 probes) differed 2-fold or more between M3 and D3 tumors ( Figure 5).
Unsupervised hierarchical clustering of differentially expressed EMT genes showed significant differences between M3 and D3 tumors. All M3 and D3 samples clustered together, forming two major distinct clusters (Figure 6), illustrating chromosome 3 loss as the main driving event. A list of all up-regulated and down-regulated genes is provided in Supplementary Table S1. UM56, which is only one sample with 1p loss and 8q gain without M3 in our cohort, showed greater similarity with the M3 group profile.
1 Figure 5. A flow-chart depicting the workflow of the procedure for selecting genes from the EMT database and evaluating the differences in mRNA expression in M3 and D3 tumors. * One gene with opposite regulation of its two transcripts was excluded. Sample UM56 (Class 2 expression profile, monosomy1, D3) was excluded from the statistical analysis. Figure 6. A heatmap of EMT-associated genes mRNA expression in UM tumors with at least 2-fold change. Chromosome 3 status is shown on the left side, M3 tumors are highlighted in brown, D3 tumors are highlighted in gray. UM56 highlighted in orange, a metastatic patient with 1p loss and 8q gain, whose disease status was discussed earlier, showed greater similarity with the M3 group. As the mRNA expression of three other metastatic patients (red numbers) showed a similar pattern as those with primary UM, they were included in the heatmap. The red color represents up-regulated gene expression, the blue color represents down-regulated gene expression, while the yellow represents no change.

Discussion
UM, the most frequent primary intraocular tumor in adults, is a rare disease. Although several treatment options are available for primary disease, most extremes are enucleation and stereotactic radiosurgery. Half of the patients develop metastases, in spite of the efficient treatment of the primary disease, and no treatment is accessible to prevent metastatic disease so far [48]. Even though various tumor tissue-based prognostic markers, including gene expression changes and chromosomal rearrangements, have been discovered, they are not available for patients treated with stereotactic surgery or other eye-preserving techniques.
A liquid biopsy is a non-invasive, promising approach in plenty of malignancies, allowing early diagnostics and disease progression monitoring [49,50]. Metastasis in UM arises from hematogenous spread unless tumor cells infiltrate the conjunctival lymphatics [51]. The presence of tumor cells in intra-tumoral blood vessels also found in our study is associated with poor prognosis [52,53]. Using blood-based markers remains challenging in UM due to several impediments. One main reason that made it unfeasible up to now is the extremely low number of CTCs or insufficient concentration of ctDNA in the blood of primary UM patients. It was reported by several authors using different methods that the number of CTC varies from 1 to 5 in 10 mL of venous blood [54,55]. Based on current findings showing a significantly higher amount of CTCs in metastatic patients, we can hypothesize that metastasis seeds CTCs into circulation. Moreover, there is no significant association between the number of detectable CTCs in primary UM and their propensity to metastasize [56,57]. Interestingly, disseminated melanoma cells detected in the bone marrow of 328 UM patients had a positive prognostic value [58]. Therefore, implementing new liquid biopsy methods for UM patients is needed, which would aid therapeutic decisions [51].
Based on this and our previous success with the quantitative PCR (qPCR)-based detection of EMT-associated transcription factors in CD45-depleted peripheral blood fraction of breast cancer patients [19,59,60], we tested the suitability of this approach in UM patients. We compared the gene expression of epithelial marker KRT19 and four EMT-associated TFs, namely SNAI1, SNAI2, TWIST1, and ZEB1 of the 34 primary and 5 metastatic patients to the expression level of identical genes in the peripheral blood of 60 healthy controls. Except for ZEB1, gene expression was mostly undetectable in the majority of patients' blood, independently of their M stage. ZEB1 expression of any patient did not exceed the cut-off value. Moreover, it did not differ between M3 and D3 patients. Given our previous findings in breast cancer, these surprising results can be explained by low analytical sensitivity and reduced CTCs seeding in primary UM tumors.
An additional important blood-based marker for monitoring disease progression in UM is ctDNA. A high number of tumor-specific recurrent hot spot mutations allows for tumor-specific ctDNA detection [57]. Therefore, we validated our negative findings in patient plasma samples using the highly sensitive ddPCR method. GNA11 p.Q209L and GNAQ p.Q209P were the most frequent tumor-specific mutations, with four more in the same genes identified each in one patient. Consistent with previous reports, ctDNA was identified more frequently in the metastatic than in the primary disease [61,62]. The presence of ctDNA was reported to be 23% in primary UM and 100% of metastatic patients by Beasley et al., while in our case, it was 9.4% and 54.5%, respectively. This discrepancy can be caused by the different clinicopathological status of enrolled patients in two datasets. Moreover, this approach is hampered by the relatively high number of the mutations to be screened, all requiring plasma DNA. Therefore, the total plasma volume for screening individual mutations remains relatively low, decreasing analytical sensitivity. Our findings show that in contrast to patients with localized UM, the majority of metastatic patients had detectable ctDNA in their plasma, confirming the results reported previously [54,57]. Based on this, the authors concluded that given the low proportion of detectable ctDNA in localized disease, ctDNA could not be an adequate marker for the screening of primary UM patients with a high risk of metastasis. However, it may offer an achievable minimally invasive approach for monitoring metastatic disease progression [57].
By ddPCR, we confirmed the low analytical sensitivity of the qPCR method for the detection of selected epithelial and EMT-associated TFs transcripts in the peripheral blood of UM patients, independently of their M stage status. Therefore, we decided to examine the expression of EMT-associated genes in the tumors of UM patients, given that EMT has been studied rarely in UM and other mesenchymal tumors. While E-cadherin (coded by CDH1) is considered a tumor repressor, mesenchymal marker N-cadherin (coded by CDH2) is regarded as a tumor facilitator in carcinomas. In the few published reports, traditional EMT-associated TFs, among them SNAI1/2, TWIST1, and ZEB1, were involved in UM pathogenesis. However, unlike in carcinomas, an epithelial-like phenotype, rather than mesenchymal-like (spindle-shaped), has been associated with the risk of poor prognosis. The reverse switch between EMT and mesenchymal-epithelial transition (MET) was negatively correlated with the inhibitor of DNA binding 2 (ID2) gene expression in the aggressive UMs [22]. Surprisingly in this regard, overexpression of TWIST1 and ZEB1 accompanied by CDH1 down-regulation and CDH2 up-regulation were reported in prognostically poor UMs [14]. Therefore, it was hypothesized that EMT-TFs are necessary for UM tumorigenesis, but not for EMT morphology switch [63]. The authors demonstrated in vitro and in vivo, that spindle UM cells were able to convert to epithelioid cells and that higher ZEB1 expression drives UM progression by inducing cell dedifferentiation, proliferation, invasion, and dissemination without a change in the cell morphology [63]. Moreover, ZEB1 promoted down-regulation of crucial genes involved in melanocyte differentiation, including BAP1, melanocyte inducing transcription factor (MITF9), tyrosinase (TYR), and tyrosinase-related protein 1 (TYRP). Interestingly, ZEB1 expression was only detected in epithelioid cells showing less pigmentation. ZEB1 was therefore shown to be the major oncogenic factor in UM progression. To disseminate, UM cells have to undergo a loss of adherence. CDH1 down-regulation is responsible for cell detachment from each other, while extracellular matrix remodeling is important to allow tumor cell dissemination. It was speculated that epithelioid cells, preferentially expressing epithelial differentiation markers such as cytokeratins, are terminally defined phenotypes with rapid proliferation, quick mobility, high invasiveness, and disseminating abilities. The question remains open if this morphology change is related to the formation of cancer stem cells [63].
Gene expression profiling uncovered an interesting association between poor prognosis, substantiated by the M3/Class 2 expression profile, and deregulated expression of 127 EMT-associated genes. As expected, genes associated with the epithelial-like phenotype (CDH1), cell lineage determination and differentiation (TWIST2), regulation of cell survival and proliferation, stem cell maintenance (KIT), cellular communication, and the normal development and function of the nervous system (ALK), cell proliferation, motility, and invasive activity promoting cancer metastasis (PTP4A3), and cell cycle regulation (CCND2) were up-regulated. Besides these that were previously associated with UM pathogenesis, several new genes were also discovered. Interestingly PTP4A3 (HR 2.54, 95% CI 2.01-3.20) was associated with the risk of liver metastasis development in UM [64]. Moreover, PTP4A3 up-regulation increased UM cells' invasivity and migratory potential [23]. This phenotype was associated with actin microfilament network modifications. Similarly, S100A4, a member of S100 family proteins, may function in motility, invasion, and tubulin polymerization regulation. Chromosomal rearrangements and altered expression of this gene have been implicated in tumor metastasis formation [21]. GDF15 is a divergent member of the TGF-ß superfamily ligands. In normal conditions, it regulates cell survival, proliferation, differentiation, migration, and apoptosis. It can be up-regulated by inflammatory stimuli. Its expression in the plasma of UM patients was associated with metastatic risk [29].
The highest up-regulation was found for the CCL18 gene, highly expressed also in cutaneous melanoma, breast, ovarian, and other cancers [65].
CCL18 is a T-lymphocyte-attractant, having chemotactic activity for naive T-cells, CD4+, and CD8+ T-cells, and thus may play a role in both humoral and cell-mediated immunity responses [66]. In contrast to other malignancies, the presence of tumor-infiltrating lymphocytes in UM has been associated with poor prognosis [39].
LGALS3 is another gene involved in immune-suppressive pathways up-regulated following BAP1 loss [31]. As shown previously, CSPG4 expression was detected in various human cancers, including the majority of human UMs. Given its low expression in non-cancerous tissues, it was studied as a target for antibody therapies [67][68][69]. Hypoxia induces the expression of JAG1, a Notch pathway member, whose pharmacologic inhibition largely blocked the hypoxic induction of invasion in UM. The down-regulation of Jag-1 expression was facilitated by GNAQ knockdown [37,38].
Several newly discovered genes have not been previously implicated in UM pathogenesis, including HTN1, MRC2, VWCE, GJB2, LGALS3, and PRRX1. For example, histatin-1, a product of the HTN1 gene, counteracted the effects of EMT inducers on the outgrowth of oral cancer cell spheroids, suggesting that it affects processes that are implicated in cancer progression [25]. Endo180 protein, coded by the MRC2 gene, plays a role in extracellular matrix remodeling and plasticity in tumor cell movement, cooperating with the matrix metalloproteinases. It has been suggested that stabilization of the Endo180-CD147 EMT suppressor complex and targeting of the non-complexed form of Endo180 in invasive cells could have therapeutic benefit in the prevention of cancer progression and metastasis [70]. Up-regulation of the VWCE gene (synonym URG11) was demonstrated to promote proliferation, migration, and invasion in the prostate, non-small cell lung, and hepatocellular carcinomas [71][72][73]. Moreover, it predicted the poor prognosis of pancreatic cancer by enhancing EMT-driven invasion [34]. Connexins control migration in neural crest and cancer cells, interact with the cytoskeleton, and regulate cell polarity and directional movement. GJB2 gene product connexin 26 was suggested to promote cancer development by facilitating cell migration and invasion [74]. Among the most interesting findings is the up-regulation of PRRX1 TF in the UM tissues. Prrx1 has been previously implicated in developmental processes associated with fibroblast behavior. Recently, it was shown that it is an EMT inducer, conferring migratory and invasive properties. Unlike other EMT-associated TFs, the loss of Prrx1 is required for cancer cells to metastasize in vivo in carcinomas [75].
We found eight significantly more than 4-fold down-regulated genes in M3. RORC is a T helper 17 (Th17)-associated TF. Th17 are CD4+ cells that produce interleukin 17 (IL-17) and are potent inducers of tissue inflammation and autoimmunity [76]. The SPP1 gene is one of the developmental genes down-regulated in Class 2 tumors [77]. It is involved in different aspects of tumor biology, including invasion and metastasis [40]. Angiogenin (ANG), a member of the ribonuclease A superfamily, together with the ID2 gene, were negatively associated with liver metastasis in UM (HR 0.42, 95% CI 0.33-0.55; HR 0.48, 95% CI 0.39-0.6, respectively) [64]. SATB1, another Class 2 discriminating gene, is the global chromatin organizer and TF. It emerged as a key factor in integrating higher-order chromatin architecture [78]. The role of several other down-regulated genes such as cellular communication network factor 5 (WISP2), hook microtubule tethering protein 1 (HOOK1), Cathepsin Z (CTSZ), or ETS variant transcription factor 1 (ETV1) in UM has to be investigated. They were deregulated in the breast, hepatocellular, colorectal, and gastric cancers [42,44,46,47]. The majority of these genes have been involved in cell angiogenesis, growth, proliferation, and migration, while CTSZ is histone methyltransferase involved in epigenetic regulation. Many other EMT-associated genes were significantly up-regulated or down-regulated in M3 tumors. They are all listed in Table S1. Their differential expression could serve for the identification of new blood-born prognostic markers for UM patients. Since for further analysis it is necessary to obtain a sufficient amount of RNA from the CD45-depleted fraction of peripheral blood from both controls and patients, we plan to verify this hypothesis in the following study.
Several metabolic pathways were deregulated in M3 tumors, not surprisingly PI3K-AKT, focal adhesion, or the Hippo tumor suppressor pathways. Angiogenesis, a hallmark of cancer, has been shown to play an essential role in UM development. However, antiangiogenic drugs, similarly to targeted therapies or immunotherapies, have shown disappointing results in clinical trials so far. Therefore, combination therapies and novel experimental approaches such as gene therapy or targeting epigenetic modifications may offer possible clues for more effective patient management options in UM.

Patients and Sample Processing
The 53 UM patients, consecutively examined at the Department of Ophthalmology, Faculty of Medicine, Comenius University in Bratislava (University Hospital Bratislava in Slovakia), were enrolled in the study. Among them, 10 were diagnosed in stage IV, when metastases were already present. One of the patients developed metastases 8 months after primary UM treatment (Table 1, Figure 1). Metastases were located in liver (n = 4), lungs (n = 3), spine (n = 1), skin (n = 2) and pelvis (n = 1). Samples were collected between August 2018 and September 2020. The study was approved on December 12th, 2018, by the Ethics Committee of the Ruzinov Hospital Bratislava, number EK/250/2018, and written informed consent was obtained from all patients. The clinico-histopathological data of enrolled patients are summarized in Table 1. All patients were treated for the primary ocular lesion with radiotherapy or/and surgery (enucleation). Peripheral blood samples were obtained before treatment from all patients. Fresh tumor tissues were collected from 37 patients who underwent enucleation, either without (n = 31) or after stereotactic surgery in the past (n = 6). Immuno-histologic features were assessed in formalin-fixed aliquots of the tissues.

CTCs Enrichment Using CD45 Depletion
The RosetteSep™ Human CD45 Depletion Cocktail (StemCell Technologies, Vancouver, BC, Canada) was used to enrich CTCs from whole blood (10 mL) by depleting CD45+ cells. The antibody cocktail crosslinks unwanted cells to red blood cells (RBCs), forming rosettes. The unwanted cells were then pelleted with the free RBCs when centrifuged over a density centrifugation medium. The enriched cell layer was harvested, and after washing and centrifugation steps, the pellets of CD45-depleted cells were mixed with TRIzolVR LS Reagent (Invitrogen Corporation, Carlsbad, CA, USA) and stored at −80 • C until RNA extraction.

DNA Extraction and Quality Assessment
Approximately 1−1.5 cm 2 large pieces of tumor tissue were snap-frozen in liquid nitrogen immediately after enucleation. Tissue samples were stored at ™80 • C. Before analysis, tissue was mechanically disrupted using mortar and pestle. DNA was subsequently extracted using Gentra Puregene Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. DNA concentration and absorbance at 280 and 260 nm were measured by spectrophotometry (NanoDrop System; NanoDrop, Minneapolis, MN, USA).
For the liquid biopsy, 10 mL of peripheral blood was collected into an EDTA tube from each patient on the day of treatment. Blood samples were processed up to 4 h after sampling. Two centrifugation steps were applied (1500× g for 10 min and 3000× g for 10 min) to remove any residual intact blood cells carried over from the first centrifugation step. Plasma samples were then flash-frozen in liquid nitrogen and stored at −80 • C. QIAamp Circulating Nucleic Acid Kit (Qiagen, Hilden, Germany) was used for circulating DNA extraction from 3 mL of plasma with an elution volume 70 µL following manufacturer instructions. DNA concentration was measured by the Qubit™ 2.0 Fluorometer (Qubit™ dsDNA HS Assay Kit, Thermo Fisher Scientific, Waltham, MA, USA).

Mutation Detection in Tumor Samples by ddPCR
To introduce and optimize the ddPCR method, mutation detection was done firstly on DNA extracted from 31 UM tumor tissues using QX100™ Droplet Digital™ PCR system (Bio-Rad Laboratories, Inc., Hercules, CA, USA). We used 8 assays, namely dHsaMVD2010049 for GNA11 p.Q209L, dHsaMDS961917975 for GNA11 p.Q209P, dHsaMDS314447910 for GNA11 p.R183C, dHsaP2010051 for GNAQ p.Q209L, as well as for the GNAQ Q209R mutation detection, dHsaMDV2516794 for GNAQ p.Q209P, dHsaMDS533896396 for GNAQ p.R183Q" dHsaMDS848188535 for PLCB4 p.D630Y and dHsaMDS821441396 for CYSLTR2 p.L129Q mutations. Based on the premise that mutations are mutually exclusive, samples with one identified type of mutation have not been further tested for the remaining mutations of interest. The 20 µL PCR mix contained 10 µL ddPCR™ Supermix for probes (no dUTP, Bio-Rad Laboratories, Inc., Hercules, CA, USA), 1 µL predesigned assay with primers and probes (450 nM primers/250 nM probes; Bio-Rad Laboratories, Inc., Hercules, CA, USA), 2 µL (2U) of Tru1I enzyme, nuclease-free water, and template DNA up to 7 µL. The droplets were generated in QX200 Droplet Generator cartridges (Bio-Rad Laboratories, Inc., Hercules, CA, USA) and transferred to a 96-well plate (Bio-Rad Laboratories, Inc., Hercules, CA, USA), sealed in PX1™ PCR Plate Sealer (Bio-Rad Laboratories, Inc., Hercules, CA, USA). The PCR program performed in C1000 Touch™ Thermal Cycler (Bio-Rad Laboratories, Hercules, CA, Inc., USA) was used according to manufacturer's recommendations, as follows: initial denaturation at 95 • C for 10 min (ramp rate~2 • C/s), followed by 40 cycles of amplification: 94 • C for 30 s, 55 • C for 1 min (ramp rate~2 • C/s) Final deactivation of the enzyme was performed at 98 • C for 10 min (ramp rate~2 • C/s). For mutation interrogation, the QX200 Droplet Reader (Bio-Rad Laboratories, Inc., Hercules, CA, USA) and the QuantaSoft 1.7. software (Bio-Rad Laboratories, Inc., Hercules, CA, USA) using a two-color detection system FAM and HEX were used. Results are expressed in copies/microliter or events/well, respectively. A mutation-positive control DNA, a negative (wild-type) control, and a no template control were included in each run, also used to set the threshold. Droplets are assigned as positive or negative by thresholding based on their fluorescence amplitude. All positive droplets were evaluated above the threshold. To exclude false positivity, we determined a false positive rate (FPR) on wild-type tissue samples. The tests providing less than 3000 droplets were excluded from the analysis.

Sanger Sequencing
To validate assays' performance, we sequenced all 31 UM tissue samples (Figure 1) to determine the genotypes of the above-mentioned nine primary driver mutations. DNA sequencing was performed on PCR products obtained by amplification with the proprietary designed primers (Table 5) in optimized conditions. The 20 µL PCR reaction mixture contained components from FastStart ™ Taq DNA Polymerase, dNTPack (Roche Diagnostics, Basel, Switzerland): 0.4 µL of dNTPs, 2.0 µL of the buffer with MgCl 2 , 0.1 µL of FastStart Taq DNA Polymerase, and 10 pmol/L of each primer (Generi Biotech, s.r.o., Hradec Kralove, Czech Republic). PCR steps were: initial denaturation at 95 • C for 10 min, followed by 40 cycles of amplification steps at 95 • C for 30 s, 54-64 • C (depending on the primer set) for 45 s, 72 • C for 1 min, and a final extension at 72 • C for 10 min. Amplification products were separated by electrophoresis on 1.75% agarose gel stained with GelRed nucleic acid gel stain (Biotium, Fremont, CA, USA) and visualized on a UV transilluminator. PCR products were purified by NucleoSpin ® Gel and PCR Clean-up (Machery-Nagel, Duren, Germany). Sequencing primers were the same as those used in the PCR reaction. Sequence reaction was performed using the Big Dye Terminator v1.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA, USA) and run on the ABI Prism 3500 Genetic Analyzer (ThermoFisher Scientific, Waltham, MA, USA). In order to exclude analytical errors, all sequencing analyses were carried out on both strands. Obtained sequences were analyzed and compared with reference sequences by the BLAST program (http://blast.ncbi.nlm.nih.gov/Blast.cgi).

Circulating Tumor DNA Detection
After optimizing the method and validating the assays, we analyzed 52 plasma samples by the ddPCR (Figure 1). Identical reaction conditions, procedures, and evaluations were used for tissue analyses. A maximum volume 7 µL of plasma DNA was added to each run. DNA extracted from the plasma of two healthy controls was used as the wild-type control.

MLPA and Data Analysis
The quality of DNA extracted from radiosurgery treated patients in the past was low; therefore, these samples were not used for MLPA analysis. A total of 100 ng of DNA was used to identify chromosomal rearrangements in tumor tissues by SALSA MLPA Probemix P027 Uveal melanoma (MRC Holland, Amsterdam, Netherlands). Besides probes located on chromosomes 1, 3, 6, and 8 (seven probes for 1p, 19 probes for chromosome 3, six probes chromosome 6, and six probes for chromosome 8), the kit contains 12 control probes. The reaction was performed according to manufacturer instructions. After amplification, MLPA products were separated by capillary electrophoresis (Genetic Analyzer 3130XL, Applied Biosystems, Foster City, CA, USA) and analyzed by Coffalyser software (MRC Holland, Amsterdam, The Netherlands).

RNA Extraction and Quality Control
RNA extraction was performed from approximately 34 mg of fresh frozen tumor tissue using the RNeasy Mini Kit (Qiagen, Venlo, Netherlands) following manufacturer instructions. The quality of isolated RNA was analyzed using the Agilent RNA 6000 Nano Kit (Agilent Technologies, Santa Clara, CA, USA), and only RNA samples, where RIN numbers were above 7.5, were selected for subsequent gene expression analysis.

Microarray Assay
Microarray analysis was performed on total RNA isolated from tumor tissues of 23 samples ( Figure 1). Also, 100 ng of total RNA was primed with T7-promoter primer, and cDNAs were synthesized using MoMULV reverse transcriptase. cDNA labeling was done using the Quick Amp Labeling kit (Agilent Technologies, Santa Clara, CA, USA). During the amplification process, Cy3 labeled CTP nucleotides were incorporated into cDNA, generating labeled cRNA. Labeled targets were purified in order to remove nonincorporated nucleotides and reaction components using the GeneJET TM RNA Purification Kit (ThermoScientific, Waltham, MA, USA). Subsequently, samples with specific activity above 8 were fragmented by incubation for 30 min at 60 • C using Gene Expression Hybridization Kit (Agilent Technologies, USA) and proceeded to the hybridization step (17 h, 65 • C, 10 rpm), where 600 ng of the sample were applied onto SurePrint G3 Human Gene Expression 8×60K v2 Microarray Slide (Agilent Technologies, Santa Clara, CA, USA). Parts of labeled cRNA, which have bound nonspecifically or not bind at all, were washed away in two washing steps using Gene Expression Wash Buffer Kit (Agilent Technologies, Santa Clara, CA, USA) according to the manufacturer's instructions. Finally, the microarray slide was scanned at resolution 2 µm using the SureScan Microarray Scanner (Agilent Technologies, Santa Clara, CA, USA).

Image and Data Analysis
TIFF images were converted and processed using Feature Extraction Software 12.0.3.2 (Agilent Technologies, Santa Clara, CA, USA). Acquired data of spot intensities corresponding to each sample were imported into GeneSpring 14.9 GX software (Agilent Technologies, Santa Clara, CA, USA), where gene expression differences were analyzed. Statistical analysis using moderate t-test was performed to detect changes in gene expression between two groups (M3 vs. D6).
The EMT gene list included in our analysis was from the EMT gene database (http://dbemt. bioinfo-minzhao.org/). The array contains all protein-coding genes; only miRNAs were excluded. We considered all different probes for selected genes, e.g., different transcripts (ID 072363). Several genes were selected repeatedly, as they are encoded by different probes in the Agilent platform.

Statistical Analysis
Statistical analyses were performed using the IBM SPSS statistics version 23.0 software for Windows (IBM Corp. Released 2015. IBM SPSS Statistics for Windows, Version 23.0. Armonk, NY: IBM Corp.) The Sapiro-Wilk test was used to assess the normality of data. Depending on the data distribution, the Mann-Whitney U test or t-test was used to assess differences between primary UM and metastatic patients. The categorical variables were tested using χ2 or Fisher's exact test. To calculate tumor volume "TV = π/6 × (largest basal diameter × width × prominence)" we used the following formula by Gass (Gass 1985).

Conclusions
Hematogenous dissemination is believed to be an early event and the major cause of metastatic spread in UM. CTCs have been detected with different success in primary and metastatic patients. Intriguingly, disseminated CTC did not correlate with the prognosis. Therefore, revealing the factors involved in the hematogenous spread of UM would significantly contribute to the development of new therapeutic approaches aimed at prolonging dormancy and delaying the onset of metastatic disease in high-risk patients. Snail family transcription repressors 2 S100A4 S100 calcium-binding protein A4 TF Transcription factor TYR Tyrosinase TYRP Tyrosinase-related protein 1 TWIST1

Abbreviations
Twist family bHLH transcription factor 1 TWIST2 Twist family bHLH transcription factor 2 UM Uveal melanoma VWCE von Willebrand factor C and EGF domains WISP2 Cellular communication network factor 5 ZEB1 Zinc finger E-box-binding homeobox 1 ZEB2 Zinc finger E-box-binding homeobox 2