Identification of compounds to affect radiosensitivity of cells

Pellegrini Lab—UCLASoCalBSI 2007

Joshua SmithBazyl Nettles

2

Outline

Biological Significance

Overall Objectives

Basic Methodology

Tools

Background

Experimental Approach

3

Biological SignificanceResults from our project could be used in

development of drugs to affect cells’ radiosensitivity

– Decreased radiosensitivity possibly beneficial to people that have been exposed to radiation

– Increased radiosensitivity beneficial to potentially increase effectiveness of radiotherapy (cancer treatment)

4

Project ObjectivesFrom gene expression information from

cells exposed to 167 bioactive compounds:

– Identify transcription factors that are activated in response to drugs

– Identify which compounds activate the same factors as those activated by exposure to radiation

5

Basic Methodology

• Changes in gene expression are regulated by the binding of transcription factors to promoters

• The activity of a transcription factor often depends on co-factors and post translational modification and cannot therefore be reliably estimated from mRNA levels of the factor

• Transcriptional regulation is inherently combinatorial

6

Basic Methodology

Thus, we use multivariate regression to estimate transcription factor activities

7

Major Tools

Matlab 2007– Bioinformatics Toolkit

MS ExcelPerl

www.mathworks.com

www.perl.orgwww.microsoft.com

8

Background

Data taken from “The Connectivity Map: Using Gene-Expression Signatures to Connect Small Molecules, Genes, and Disease” by Lamb, et al

“…we have created the first installment of a reference collection of gene-expression profiles from cultured human cells treated with bioactive small molecules…”

“Connectivity Map” can be used to find connections among small molecules, expression, genes, etc.

9

Experimental Approach

Five basic steps:– Ordering and Gathering Data

– Probe, Gene and Promoter Identification

– Transcription Factor Data

– Generate Models

– Compare Model

10

Ordering and Gathering Data

“Connectivity Map” data retrieved from NCBI’s Gene Expression Omnibus (167 compounds)

Using Matlab’s Bioinformatics Toolkit, imported 564 expression profiles

Using Matlab, MS Excel, and Perl divided data into 453 “experiments”

Using SQL, detected and averaged duplicate experiments, leaving us with 314 experiments

http://www.ncbi.nlm.nih.gov/geo/

TheConnectivity

Map

TheConnectivity

Map(GSE5258)

^SAMPLE = GSM119282!Sample_title = 5202764005789148112904.A10!Sample_geo_accession = GSM119282!Sample_status = Public on Sep 27 2006…ID_REF VALUE ABS_CALL1007_s_at 495.3 P1053_at 278.2 P117_at 3713.4 P121_at 44.7 P1255_g_at 2.6 A1294_at 16 A1316_at 5.2 A1320_at 4.4 A1405_i_at 16 A1431_at 21.2 A1438_at 7.6 A…

GSM118720

GSM118721

GSM119282

GSM119282

…

22280probes

564 Samples (microarrays)

22280probes

453 Drug/Contro Ratios

11

Probe, Gene and Promoter Identification

Retrieved human promoters from UCSC Genome Browser

Retrieved microarray and probe information from GEO for our data

Found variance for each probe across 314 unique experiments

Using top 2000 by variance, revealed 1704 probe/gene/promoter sets

12

Probe, Gene and Promoter Identification

22280Probes

314 Unique Experiments

Normalizedexpression

ratio

Variance

Keep Top 2000

Probe Variance------------------------3 5423.53512799 3647.58217745 550.77435991 253.09153192 250.4694

Withpromoter

data1704 genes& promoters

13

Transcription Factor Data

TRANSFAC® and JASPAR® are the databases of transcription factors, their genomic binding sites and DNA-binding profiles.

For each TF PWM, we move along the promoter sequence calculating the probability of binding

The maximum binding probability calculated along a sliding window is kept for each promoter

Promoter ATGCCCTTGCTATCTGCATGCTATCTGCACTGGACGT…

14

Transcription Factor Data

Then the maximum score for each promoter is compiled into a matrix

1704gene

promoters

~940 TFs from TRANSFAC & JASPAR

probability aTF will bind toa particularpromoters

15

Model Generation

Generate models using Multivariate Adaptive Regression Splines (MARS) to correlate– occurrences of TF binding motifs in the promoter DNA– their interactions to the gene expression levels

“Model” refers to set of Transcription factors that can explain a high percentage of the current variance in expression activity

16

Model Generation

An overabundance of data led to a predicted modeling time of 20 hrs for each of our 314 experiments

This led to a decision to reduce the number of TFs used for computation from all 940 to ~40 “relevant” TFs

This could be used as to identify likely experiments that could be run with all TFs

17

Relevant Factors

Ataxia telangiectasia mutated– Protein kinase that plays a critical role in response to

certain types of DNA damage– Produced in all cells, it is activated once DNA damage has

occurred. (Hawley and Friend, 1996; Banin et al., 1998; Canman et al., 1998)

Used list of ATM dependent factors that are activated in response to radiation damage (prior work)

– Compare with models– Attempt to find experiments (compounds) activate the same

factors as those activated by exposure to radiation

18

Results

We generated models for these relevant factors and found several experiments with a high reduction in variance (RIV)

RIV– The percentage of variance in expression

accounted for by the factors in a model

19

ResultsRIV for our 314 experiments

20

Results The top 10 drugs by RIV were:

1.HC toxin

2.Pirinixic acid

3. Ionomycin

4.Phenanthridinone

5.Tioguanine

6. Fasudil

8. Prochlorperazine

9. Amitriptyline

7. Valproic acid1

1. Valproic acid enhances brain tumor cell radiosensitivity. Immunotherapy Weekly (2005-06-01)

2. Modification of Radiation Response of Tissue by Colchine. International Congress of Radiology (1965-09-27)

210.Colchicine

21

References

Debopriya Das, Nilanjana Banerjee, and Michael Q. Zhang. Interacting models of cooperative gene regulation. PNAS, 2004.

Justin Lamb, et al. The Connectivity Maps: Using Gene-Expression Signatures to Connect Small Molecules, Genes, and Diseases. Science, 2006.

Debopriya Das, Zaher Nahle, and Michael Q. Zhang. Adaptively inferring human transcriptional subnetworks. Molecular Systems Biology, 2006.

Shawn Cokus, et al. Modelling the network of cell cycle transcription factors in the yeast Saccharomyces cervisiae. BMC Bioinformatics, 2006.

22

Acknowledgements

UCLA and the Pellegrini Lab– Dr. Matteo Pellegrini– Dr. David Casero Díaz-Cano

SoCalBSI Instructors and Fellow Students– National Institutes of Health– National Science Foundation– LA / Orange County Biotechnology Center

www.ucla.edu