Fig 1.
Searching google scholar for “retinoblastoma”, “copy number”, “(a)CGH”, and “SNP-array”, 11 studies were identified that performed genome-wide profiling of retinoblastoma adding up to 290 samples. No duplicates samples were identified. For the Ganguly study [14], copy number results could not be linked to individual tumors and therefore this study, including 25 samples was discarded. The remaining 265 samples were all included in quantitative analysis and were complemented with 45 SNP-arrays from our current study, adding up to 310 samples.
Fig 2.
Genome-wide SCNA-frequencies (DNA copy number gains in red, losses in blue) detected in primary retinoblastoma samples in our current study (Kooi, N = 45) and in ten published studies. Studies are ordered by sample size where the largest study is placed at the top.
Fig 3.
Identification of SCNA peak regions.
For chromosomes that contained commonly altered genes, the gain-loss difference (number of patients with gains minus the number of patients with losses) is plotted for each official HGNC along chromosomal coordinates. For each of these chromosomes, peak regions were defined (also see S3 Table) indicated by the dashed rectangles and were used for retinoblastoma driver discovery.
Fig 4.
Clustering of genome-wide SCNA.
Heat map of genome-wide SCNAs for 310 primary retinoblastoma samples (columns). Gains are indicated in red and losses in blue. (A) Samples are ordered by unsupervised hierarchical clustering (UHC) visualized by the dendrogram on top. The dendrogram was pruned in 4 UHC-groups indicated by UHC-group numbers and colored branches. Color-coded sample information: invasion: red = not-invasive, blue = invasive; rosettes: red = no rosettes, blue = rosettes; differentiation: red = poor, blue = moderate, green = well; focality: red = multifocal, blue = unifocal; laterality: red = bilateral, blue = unilateral; heredity: red = hereditary, blue = non-hereditary, tumor volume by MRI and age at diagnosis: continuous green-to-red scale where green means small tumors diagnosed at early age and vice versa. (B) Instead of ordering by UHC, samples are ordered based on total genomic disruption by SCNAs. P-values indicate the significance of association between total genomic disruption and the clinical variable.
Fig 5.
Relation of total genomic disruption with clinical data.
Total genomic disruption by SCNAs with significantly correlated (p-values in Table 1) clinical and histopathological variables. Total genomic disruption is defined by the number of genes with either increased or decreased copy number. For numerical variables (B) age at diagnosis and (C) SCNA-amplitude, samples were categorized in quartiles based (A) on total genomic disruption. For example, Q1 denotes the 25% of tumors with the least genomic disruption and Q4 vice versa. SCNA-amplitudes are expressed as Log2-ratios. The horizontal red line illustrates the SCNA-threshold used for calling gains and losses (also see paragraph 2.4 in materials and methods and S6 Fig). For factorial variables (D) laterality, (E) heredity, and (F) differentiation grade the number of genes with SCNAs is stratified per factor level.
Table 1.
Total genomic disruption in relation to clinical and histopathological variables.
Fig 6.
Copy numbers in retinoblastoma cell lines.
Genome-wide heat map of SCNAs in 8 retinoblastoma cell lines. Sex chromosomes are not displayed since no matching reference was available. SCNA-amplitudes are expressed as Log2-ratio and mapped to a blue-to-red continuous color-scale.
Fig 7.
Overview of putative candidate genes proposed by previous studies for commonly gained chromosomal arms 1q and 6p and commonly lost 16q.