Database mining for CNF drug targets and mutation-histology correlations in NF1
Isaac Dimitrovsky and Lars Warren Ericson, 15 September 2019
We have two projects
- Mining RNASeq for new CNF drug targets
- Mining COSMIC for histologies primarily associated with NF1 and associating mutation sites with those histologies
In this project we tried to mine the RNASeq gene expression dataset for possible new drug targets. RNASeq includes a variety of tumor types; we concentrated on CNF. Our strategy was to compare gene expression values between CNF samples and as close as we could get to a control group of normal tissue. We then took the set of genes that appeared "most disrupted" vs. normal, and cross-checked it against Synapse's Drug Target Explorer. We found that a high proportion of this "most disrupted" gene set appeared in the Drug Target Explorer as having known associations with copper. This seems intriguing - while copper is apparently novel in the NF1/CNF context it has a well-established role in many disease processes [HPK] - notably, it appears heavily involved in angiogenesis and tumor formation, likely through a variety of pathways. A number of drugs are used to reduce copper levels (ex. in Wilson's disease), including penicillamine and TM (tetrathiomolybdate); TM has also been investigated for cancer treatment. Perhaps this suggests a direction for drug development that could work on a number of pathways simultaneously, and possibly be synergistic with other drug approaches.
CNF tumors are a common complication of NF1, often in large numbers. They have no treatment other than surgery, which may be impractical. Therefore, finding plausible drug targets would be of interest. It's thought that CNF tumors are heterogenous. Therefore, drugs that target multiple pathways may work more effectively. In this project we tried to combine Synapse's Drug Target Explorer (a database of known drug-target gene interactions) with the RNASeq dataset (a set of gene expression levels in a variety of tumors and "normal" cell lines) to search for new CNF drug targets.
For full details, see the shirinRNASeq notebook.
We began with the idea of comparing gene expression in CNF vs. "normal" values. We spoke with Sarah Gosline about our idea to compare expression values of "normal" and CNF cells, and she warned that some of the differences seen could be due to cell line effects, since the "normal" cells in RNASeq came from cell lines rather than biopsies. Based on a number of conversation with Ray Mattingly and Sarah, we used the "normal" values for the two cell lines with specimenID 'ipn02.3' and 'ipn97.4' - these were derived from immortalized normal Schwann cells which are thought to be the main drivers for CNF formation.
We took the RNASeq data from CNF samples (33 specimens, 11 individuals) and the two "normal" (Schwann-derived) cell lines and compared the average absolute difference of zScore for each gene (19,098) between the CNF specimens and the average of the two "normal" samples. We sorted the results and took the 100 "most disturbed" genes (highest average absolute difference values). We then cross-checked these top 100 genes against the Drug Target Explorer database.
A finding that immediately jumped out was the high proportion of "most disturbed" genes with known interactions with copper. Of the genes in the suspect set that appeared in the DTE drug target associations database (v2), 23 of 58 (40%) had a known association with copper; of the total of target genes appearing in the database, only 144 of 3735 (4%) had a known association with copper.
The full heatmap of the "most disturbed" genes (in the copper? column, X indicates a gene not in the Drug Target Explorer, + indicates a gene in the DTE with a known copper association, and - indicates a gene in the DTE with no known copper association):
Perhaps this suggests a direction for drug development that could work on a number of pathways simultaneously, and possibly be synergistic with other drug approaches. Copper is an essential micrometal, and is involved in many fundamental biological processes [HPL]. It's also implicated in a variety of diseases. Of note, there's strong evidence for its role in tumor growth, particularly in the complex process of angiogenesis, likely through a variety of pathways: (the following quotes are taken from the review above)
- "high serum and tissue levels of copper were found in various human tumors"
- "it has been demonstrated that copper is required for angiogenic process"
- "lowering copper level moderately with a drug such as TM produces strong antiangiogenesis and potent anticancer effects in several rodent models"
- "limiting the biological availability of copper by penicillamine or TM administration slows tumor growth"
Copper chelators such as penicillamine and TM (tetrathiomolybdate) are used in treatment of conditions including Wilson's disease, and have been investigated for treatment of cancer.
Sharad Verma mentioned the possible importance of transdermal delivery for CNF drugs. Interestingly, it appears possible to administer penicillamine transdermally using ultrasound-mediated delivery (sonophoresis) [PHLB].
[HPK] Hordyjewska A, Popiolek L, Kocot J. The many "faces" of copper in medicine and treatment. Biometals. 2014;27(4):611–621. doi:10.1007/s10534-014-9736-5
[PHLB] Polat BE, Hart D, Langer R, Blankschtein D. Ultrasound-mediated transdermal drug delivery: mechanisms, scope, and emerging trends. J Control Release. 2011;152(3):330–348. doi:10.1016/j.jconrel.2011.01.006
- try to replicate the results with another (independent) set of CNF samples
- apply the same approach to other tumor types
- analyze the set of genes found for common pathways and clustering
- test CNF tumors for high copper levels
- in vitro or animal model tests of copper chelators such as penicillamine and TM (tetrathiomolybdate)
Run the shirinRNASeq notebook.
We find histologies associated with tissues containing NF1 somatic mutations in the COSMIC database. We find all other gene somatic mutations for those histologies. We assess the main predisposing gene to be the gene with the most mutation-sample pairs associated with the histology. For the NF1 mutation sites associated with each histology, we give FATHM score and PUBMED ID.
NOTE: All of the samples are taken from people with somatic mutations leading to cancer, so there is selection bias in the samples to that extent.
We are looking for how different final protein products of NF1, arising from different mutations, contribute to development of specific histologies noted in the Sanger Institute COSMIC cancer somatic mutation database.
We use Python to search records in the downloaded TSV file.
There are 21 histologies associated with tissues containing NF1 somatic mutations in the COSMIC database. There are 21. We find that 4 histologies are primarily associated with NF1. We find that the mutation sites in NF1 observed to the 4 histologies each are associated with only a single histology. For MPNST, 3 genes are equally scored and Neurofibroma has a very high score for NF1. We find that each histology is associated with multiple mutation sites. We find that each mutation is associated with only one histology. For the NF1 mutation sites associated with each histology, we give FATHM score and PUBMED ID.
The metric we propose seems good at assessing which genes are significant for a histology. It is remarkable that each mutation is associated with only one histology. It is interesting that multiple dispersed mutation sites can support development of a single histology. We need to know more about similarity of proteins realized from replication based on different mutation sites which are associated with same histology, to find common binding sites in the signalling chain for those proteins, and to understand why proteins produced from mutations associated with other histologies differ on those binding sites.
- Understand how different mutation sites associate with the same histology produce proteins which have similar effects, by comparing the protein products of each mutation
- Automating the process of finding, from available online databases and APIs, the variation classification, affected pathways and ClinVar clinical significance score of each mutation
Download a copy of the COSMIC gene expression TSV file. Run the COSMIC NF1 Histologies notebook.