NRF1 and NRF1 target genes and aggressive Breast Cancer

[jramo033]

We processed the fastq files with raw data for MCF10A cells with three different treatments. These experiments were performed at FIU (Dr. Roy's lab). Using GALAXY, we found the NRF1 binding sites and the lists of genes associated with the following results:

Treatment A (NRF1 upregulated + E2): 5,421 candidate target genes

Treatment B (NRF1 upregulated) : 8,280

Treatment C (NRF1 down regulated): 6,219

[jramo033]

We used the NRF1 target genes of HCC1954 (breast cancer cell lines) and HMEC (normal breast cells) to look for enriched KEGG pathways. Some of these pathways are involved in the so called hallmarks of cancer as follows:
1.-Sustaining proliferative signaling

Growth Factors
PI3K-Akt Signaling
MAP-kinase Pathway
mTOR Pathway
Cellular Senescence
Oncogenes & Tumor Suppressor Genes

2.-Evading growth suppressors

p53 Signaling
Apoptosis
TGF-beta Signaling
EMT
Oncogenes & Tumor Suppressor Genes

3.-Resisting cell death

p53 Signaling
Apoptosis
Autophagy
DNA Damage Signaling

4.-Enabling replicative immortality

Telomeres & Telomerase
p53 Signaling
DNA Repair

5.-Inducing Angiogenesis

Angiogenesis
VEGF Signaling
Angiogenic Growth Factors & Angiogenesis Inhibitors

6.-Activating invasion and metastasis

ECM & Adhesion Molecules
EMT
Tumor Metastasis

7.-Reprogramming of energy metabolism

Hypoxia Signaling
Glycosylation
Energy Metabolism

8.-Evading immune destruction

T-cell and B-cell Activation
Innate & Adaptive Immune Responses
Inflammatory Response & Autoimmunity

For next week we should have the list of NRF1 target genes involved in these pathways.

[jramo033]

We identified 78 NRF1 target genes that may be involved in the hallmarks of breast neoplasms. To validate our results, we double checked that these genes had also been identified as NRF1 targets in other studies comparing with results reported by ENCODE. We found 59 genes overlapping our results and ENCODE's list of NRF1 target genes. We are planning in selecting 30 of them to run BENE. Criteria selection would be correlation with breast cancer phenotype of cells HCC1954.

In the mean time, we ran BANJO for 1, 4 and 8hrs with the top 2,000 NRF1 target genes correlated with breast cancer (cell lines HCC1954 and HMEC). We were unable to reproduce the results (Markov blanket genes). BDe scores were -1683.65 / -1629.99 /-1618.99

We are now in the process of repeating the same attempts with BANJO lowering the number to the top 500 genes.

[jramo033]

Since we were unable to reproduce BANJO results with 500 variables due to the small number of observations for the variable "amount of NRF1 binding in the promoter region", we changed the approach to keep the selection of the genes based on the NRF1 binding criteria but the Bayesian structure analysis was changed to use gene expression (RNA seq) from TCGA. We selected as cases, only those with genetic profile similar to the HCC1954 cell line (ER-, PR- and HER2+). This time BANJO was run 4 hrs, 8 hrs and 16hrs with 51 variables and 83 cases. Results will be posted next week.

[jramo033]

Following previous post, Banjo was run 7 times (1hr, 2hrs, 4hrs, 8hrs (three times) and 16 hrs.) Best BDe score reached at 16 hrs. Log norm to compare Bde scores are shown below:
Banjo Running time (hrs)
Log score -1178.6227 is 0.1146% of total score 1
Log score -1182.3458 is 0.0028% of total score 2
Log score -1176.4953 is 0.9620% of total score 4
Log score -1174.8859 is 4.8100% of total score 8
Log score -1172.9652 is 32.8316% of total score 8
Log score -1177.1193 is 0.5155% of total score 8
Log score -1172.3496 is 60.7636% of total score 16

Total score -1171.85142 100.0000%

Planning to do one final run for 32 hrs. Markov blanket genes for the variable breast cancer (BC) will be analyzed as possible involved in HER2+ breast cancer development

[jramo033]

Here I am attaching the information of gene expression (50 genes used in Banjo) for 22 HER2+ BC samples and 61 normal which were downloaded from TCGA , cleaned, sorted and selected :
1.- Normalized gene expression
2.-Categorized
3.-File with the gene expression of Markov blanket which contain both raw and categorized and also shows all the categorization parameters used (+/- 2 std dev of normal samples)

[jramo033]

Files of TCGA microarray gene expression and clinical variables were downloaded using Firehose (Tool developed by Broad Institute of MIT & Harvard). File is attached.

[jramo033]

TCGA RNA-seq breast cancer data set were grouped into eight different clusters for different breast cancer subtype race and ethnicity. Genes were filtered out using NRF1 enrichment analysis, differential expression and Hallmark of cancer signaling pathways to com up with a matrix of genes in the rows and samples on the column (breas cancer and normal samples). Here attached the categorized files based on +/- 1 sd of normal samples

[efrain.gonzalez0]

Good evening,

I have attached the new datasets that I will be using for the MCMC order search. The new datasets are based on the 1st degree Markov Blankets that Jairo provided me with. The new datasets are located on the second sheet in each one of the excel files. Please make sure that I have the right genes/variables in each case.

Respectfully,

Efrain Gonzalez

[efrain.gonzalez0]

Below I have attached a table that shows how long it took to run each one of the new datasets and how many runs were completed within that time. I would not recommend more than 4 parents be used on the larger datasets. If we are going to stick to 3 parents then we should have no problem obtaining the 100 unique iterations that Dr. Yoo and I spoke about.