Project Topics
Get Complete Project Material File(s) Now! »
Characterization of the subtracted libraries
SSH was used to screen for differentially expressed genes at the very beginning of melanoma regression in tumour cells and the surrounding stroma. Consequently, two SSH experiments were performed which resulted in four subtracted libraries such as SCAW and SCBE representing respectively up and downregulated genes in regressive melanoma tissues and SCBF and SCAX representing up and downregulated genes in cells isolated from regressive tumours. To characterize each tissue derived transcriptomic profile, 1152 (3 x 384 well plates) clones from each tissue library were sequenced, 1853 of which resulted in exploitable gene sequences. These sequences were aligned systematically against databases as described in materials and methods. The remaining sequences were either empty clones or unreadable sequences. 486 different genes could be identified and were classified according to their redundancy. This demonstrated that a majority of genes (74.27%) were present only once in the complete set of sequenced clones, while 17.07% were present at least twice. These proportions reflect the normalizing effect of the SSH procedure. 8.66% of transcripts were found in both tissue libraries and were discarded since they reflect false positives (Table Ia). Regarding the cell derived libraries (SCAX, SCBF), 2016 clones were sequenced, 923 of which resulted in exploitable gene sequences. About 70.97% of the genes were present once in each library, whereas 15.97% were found a least twice and 13.06% of false positives were discarded (Table Ib). To minimize the possibility of false positives, we focused our work on transcripts found at least twice in each library, knowing that this consideration does not replace false positive control for this high throughput technique. Thus, our SSH-gene lists underly no FDR control and are limited in their exploitation. The construction of multiple SSH libraries with biological replicates for each population was technically impossible, therefore the level of biological variation could not be addressed on the SSH level. As a consequence we tested differentially expressed genes on n = 8 progressive and n = 8 regressive melanomas by qRT-PCR.
Analysis of redundant transcripts in each library
The list of redundant genes is shown in Table I a,b. They were classified by Gene Ontology means, taking into account their predicted protein function. The functional category «biological process» was subdivided into 15 subclasses, whereas molecular function and cellular component were split into 3 subclasses (Figure 2). Upregulated genes in regression were overrepresented in classes such as cell -adhesion, -communication, signal transduction, cell -differentiation, -death, -motility, negative regulation of cell proliferation, transport and immune response. Genes involved in cellular and melanin biosynthesis were found to be downregulated. Furthermore, genes with unknown biological process were also overrepresented in the phase of early tumour regression. In the cell libraries, genes upregulated genes at early regression (SCBF) were enriched in the classes of cell –adhesion, – differentiation, -cycle, cellular metabolism and biosynthesis and negative regulation of cell proliferation. Nearly no difference between the cell libraries was detectable regarding classes such as response to stimulus, cell motility and signal transduction. The GO category «molecular function», showed an enrichment of genes upregulated in regressive tissue regarding the subclass catalytic activity. Genes identified from the library of downregulated genes in regressive tissue showed a similar «molecular function» profile. In the «cellular component» category, we observed a stronger expression of genes involved in extracellular space in the library of upregulated tissue genes but a downregulation of those involved in intracellular region.
Confirmation of differential expression of selected genes by real time quantitative RT- PCR
We chose 14 genes for qRT-PCR confirmation (Table II) to conduct multiple comparisons of regressing (n=8) vs. progressing (n=8) melanomas to evaluate the confidence of our SSH results. 6 genes (CAV1, CLU, COL1A2, EMP3, RARRES1, CD9) upregulated in regressive melanoma tissue and 8 (BCHE, CDH1, MAGED2, NUSAP1, TYR, SLC24A5, TYRP1, MITF) downregulated in regressive melanoma tissue were chosen. qRT-PCR analysis and subsequent unpaired t test analysis including Welch’s correction for different populations variances revealed 10 differentially expressed (p<0,05) genes between progressive and regressive melanoma samples (Figure 5). 4 genes did not meet the criteria of significance, even though two of them (TYRP1, COL1A2) showed the same trend in SSH and qRT-PCR. CAV1 and EMP3 expression were almost equal between progressive and regressive melanomas. However, we confirmed the results for the original tumours used for SSH regarding EMP3 expression which seems to be in this case a particularity for these 2 tumours.
RARRES1, CD9, and CLU were all significantly upregulated during regression. CLU showed the most dramatic change with a 11.5 times overexpression during regression. RARRES1 was about 6.6 times higher in regressive versus progressive melanoma tissue and about 4 times higher when comparing regressive melanoma tissue and normal skin (data not shown). The pigmentation related genes such as MITF, TYR, SLC24A5 were all downregulated during early regression by a fold change (FC) of 2. Furthermore CDH1, NUSAP1 and BCHE which were reported to play a role in human tumour progression were downregulated during MeLiM regression. After applying the Benjamini Hochberg correction to control the false discovery rate (FDR=0.05) for multiple testing, 9 genes were significantly differentially expressed between progressive and regressive melanoma (BCHE, CDH1, CLU, MAGED2, NUSAP1, TYR, RARRES1, MITF, SLC24A5).
Gene List comparison
We compared our SSH retrieved gene lists to the one of Hoek et al 2006 (supplementary data) in order to identify common genes. Overlapping genes serve in this case as an indicator of biological relationship. In order to evaluate an eventual gene-overlap with a certain significance, we applied the hypergeometric distribution test (HDT) (Hoek et al., 2006). Briefly, the test calculates the probability of obtaining by chance a number k of annotated genes for a given term among a dataset of size n, knowing that the reference dataset contains m such annotated genes out of N genes. In our case, m was the number of genes in the target gene list, k = the number of genes in our gene list, n = the number of overlapping genes between m and k and N = the number of all possible genes. Since the HDT calculates the probability of having exactly n overlaps, we repeated the calculation up to 4 times by increasing each time n + 1. Adding up these probabilities and using the cumulative results permitted to state the probability of finding at least n overlaps. The cumulative results were rounded up to the next order of magnitude to be more conservative.
Gene Ontology annotation and Ingenuity Pathway Analysis
Functional classifications from Gene Ontology (GO) Consortium were assigned to each identified gene within the different SSH-libraries using GoTree Machine (GOTM) (http://bioinfo.vanderbilt.edu/gotm/) (Ashburner et al., 2000;Zhang et al., 2004). Since gene classification in GOTM is comprehensive and complex, we adopted genes at the default level 4 and higher as the final export results.
SSH identified genes were also analysed by Ingenuity™ Pathway Analysis (www.ingenuity.com), leading to the creation of gene networks and functional clustering. The 135 genes were mapped to genetic networks available in the Ingenuity database and ranked by score. The score is the probability that a group of genes equal or greater than the focus gene number in a network could be achieved by chance alone. (Raponi et al., 2004).
Immunofluorescence analysis
Paraffin sections were dewaxed and rehydrated. Epitopes were retrieved with 10mM citrate buffer (pH = 6) at 95°C for 20min. After blocking non-specific binding sites by incubation with goat serum, melanoma sections were incubated overnight with rabbit anti-CD9 polyclonal antibody (gift from J. Garrido Pavon, Spain), mouse monoclonal anti human MITF (clone C5+D5, Zymed), and goat polyclonal anti human RARRES1 antibody (TIG1 (N-18), Santa Cruz Biotechnology). CD9, MITF, and RARRES1 protein stainings were revealed respectively by: biotin goat anti rabbit IgG (Dako), Alexafluor® 555 labelled, goat anti mouse IgG1 (Molecular Probes) and biotin donkey anti goat IgG (Santa Cruz Biotechnology). For biotin conjugated secondary antibodies subsequent streptavidin, Alexafluor® 555 conjugate (Molecular Probes) treatment was performed. Negative controls were assessed by replacing the primary antibodies with the non immune goat serum at the same concentration. Staining patterns were assessed independently by two different investigators.
Normalization and Statistical Analysis
Normalization and statistical anaylsis of microarray-data have been realized using R resources, Bioconductor statistical packages (http://www.bioconductor.org/), and the ArrayAssist software (Stratagene) for the analysis of variance (ANOVA) and k-means analysis. Raw intensity values were subjected to a pre-processing step using the GCRMA algorithm that summarizes and normalizes data into gene expression values. The log2 scale transformation is integrated into this process, so output values are then log2 transformed and ready to be used for t-test and one-way ANOVA analysis. The time post-birth was considered as central parameter for one-way ANOVA analysis. Multiple hypotheses testing was controlled by applying Benjamini Hochberg FDR correction. P-values of the ANOVA analysis were adjusted using the Benjamini Hochberg algorithm (FDR or adjusted p-value <0.01). For the t-tests, p-value adjustments were performed individually for each comparison. Probe sets were defined as differentially expressed for tn vs. t0 time points if the fold-change (FC) was bigger than 2 and p-value lower than 0.05 after unpaired t-test. Furthermore, probesets also found significant after ANOVA were used for k-means clustering (k=6). We used k = 6 clusters since most of the time the number of clusters is close to the number of time points. Also, we grouped our data by k = 9 clusters. Subsequent functional analysis however showed an “overclustering” of the data, as many genes of the same biological function were arranged in different k-means clusters, a. So, using k = 6 clusters was an experience-based choice but justified by following functional analysis. Microarray data were submitted to ArrayExpress (http://www.ebi.ac.uk/microarray-as/aer/entry). Experiment (E-MEXP-1152).
Table of contents :
CHAPTER 1 Introduction
Melanocytes
Melanoma
Melanoma susceptibility and altered signalling
Conventional anti-melanoma therapy
Anti-melanoma immunotherapy
Gene expression profiling in human melanoma
Spontaneous melanoma regression
Immune escape mechanisms
Animal models of melanoma
Pigs in biodmedical research
Melanoblastoma bearing Libechov Minipig (MeLiM)
Spontaneous melanoma regression in MeLiM
CHAPTER 2 Introduction of: “Identification of differentially expressed genes in spontaneously regressing melanoma using the MeLiM Swine Model”
Comment on Suppression Subtractive Hybridization (SSH) method
Paper I: Identification of differentially expressed genes in spontaneously regressing melanoma using the MeLiM Swine Model
CHAPTER 3 Introduction of: “Gene Expression Signature for Spontaneous Cancer Regression in Melanoma Pigs”
Comment on microarray processing method
Target Production and Hybridization
Reannotation of porcine Affymetrix Probesets
Paper II: Gene Expression Signature for Spontaneous Cancer Regression in Melanoma Pigs 79 Phenotypic characterization of tumor immune infiltrate
1. Characterization of highly pigmented cells
2. Phenotypic characterization of Tumor Infiltrating Lymphocytes (TIL)
CHAPTER 4 General Discussion
Differentially expressed genes between regressive and progressive melanoma
Time dependent gene expression profiling during melanoma progression and regression
Comparison of obtained transcriptomic results
Hypothesis of regression mechanism in MeLiM
Expression profiles and other regressing malignancies
Perspectives
REFERENCES
