Download this poster

Regulatory Signatures of Cancer Cell Lines
Inferred from Expression Data
1,2
Krishnan ,
2
Ma’ayan
Jayanth (Jay)
Avi
1Mahopac High School, Mahopac, NY 10541
2Systems Biology Center New York and Department of Pharmacology and Systems
Therapeutics, Mount Sinai School of Medicine, New York NY
2
Abstract
While gene expression data is widely available
describing mRNA levels in different cancer cells lines,
the molecular regulatory mechanisms responsible for
these changes are still poorly understood. Here we
developed a rationale approach to infer regulatory
mechanisms governing changes in gene expression by
integrating datasets of protein/DNA interactions, proteinprotein interactions and kinase-substrate interactions
collected from prior biological knowledge. We first
utilize data obtained from genome-wide ChIP-on-chip
and ChIP-Seq experiments to connect mRNA
expression levels of the NCI-60 cancer cell lines to the
transcription factors most likely regulating them. These
identified transcription factors are then “connected”,
using known protein-protein interactions, to form cancer
specific sub-networks. Within these sub-networks we
assess the enrichment for protein kinase substrates to
infer the protein kinases likely regulating these
complexes. Finally, using quantitative comparison of the
up and down regulated genes for each cancer cell line,
and genes affected by FDA approved drugs applied to
cancer cells, we predict the mechanisms of action of
these drugs. Following this path, from changes in gene
expression to transcription factors to protein kinases we
can provide a more thorough understanding of the
regulatory mechanisms behind the observed mRNA
levels in the NCI-60 cancer cell lines and other cancer
cells. This approach proposes mechanisms of action for
drugs. Wet lab experimental validation of this approach
is still necessary, it can be done using single drugs or
combinations of them.
Introduction
• The NCI-60 database provides mRNA profiles
from microarray experiments of 60 commonly
studies cancer cell lines
• Although analyzing these mRNA values is a
reliable method to measure the mRNA level of
many genes within a cell, this method offers
little clues about how cells are regulated
• While mRNA profiles indicates changes
caused by cancer, understanding the
underlying regulatory mechanisms
disregulated in different cancers will bring us
closer to therapeutics
• In this project we aim to identify the
transcription factors, protein complexes and
protein kinases responsible for the aberrant
expression of genes in the various types of
cancer cell lines
Workflow
Top ranked
transcription factors
most likely
responsible for the
observed changes
in expression
Microarray
Analyze mRNA profile from NCI -60
database by using statistical
techniques to compute over/under
expressed genes
Example of Process
Identify protein sub-networks
that “connect” the
transcription factors through
additional proteins
Wet lab experimental validation
Top ranked protein kinases
most likely regulating the
protein sub-networks
PLD1
SART1
WARS
DNAJB1
M6PR
RAE1
GPAA1
SLC29A1
IL13RA1
TRAK2
CHPF
GAS7
CUL4B
SLC6A8
CSNK1E
PRKCD
DHPS
SLC37A4
TCTA
TIMP2
CTSF
CD302
KAT2B
EDNRB
NOV
GJB1
AP1S1
ATP7A
EIF2S3
DCT
NPAS2
SCRG1
TRAIP
UBFD1
TYRP1
HBE1
RXRG
PLXNB3
KCNS3
S100A3
TYR
GLRX
HCG4
GPR143
DGKI
STAU1
MAGEA1
CTAG2
ZNF200
CAST
HLA-DRA
TXN2
TNFRSF14
FAM3A
RAB27A
PLP1
HSP90AA1 MAGEA12
CTAG1B
NRP2
SPIN2B
C1orf144
SLC25A6
ZCCHC24
SUCLG2
C14orf109
CLCN2
C9orf61
CCPG1
MAGEA2B
DPY19L2P2
DUSP10
SLC6A10P
SLC5A4
GK3P
BACE2
MUL1
METT11D1 MRPS18A
KLF11
CRIPT
MTO1
HEY1
FBXL15
CSGALNACT1
CADPS2
PRR7
GAL3ST4
NUDT11
CEP97
MGAT4B
FAM86C
ROPN1B
C20orf30
TRIM48
RPL23AP7
MICALL1
LDLRAP1
C17orf90
LUZP1
LOC348926
PCOTH
FAM86B1
LPCAT2
SURF4
HAGHL
TNFSF13B FAM167B
SPRYD5
DGAT2
ULK3
TOMM40L FAM160B1
SNX30
TMEM55A
HDAC10
LOC400657 AFAP1L1
FAM125A
OLIG1
HSD11B1L SCARNA15
SMYD4
LOC153364
CAMK2N2
CHRM1
AARS2
ANKRD54
KIAA1524
KIAA1586
GPR158
SLITRK4
GNASAS
ELOVL3
ST6GALNAC3
RNF175
C5orf35
LOC147645 LOC730259
TMEM171
GPNMB
SREBF2
SEC11A
AKR7A2
USF2
CNOT8
GTF2H1
PCOLCE
CHMP2B
STX7
SNTA1
PTPN18
HPS5
SMCR7L
ACP5
DYNC1I1
BCHE
GSTT2
TRPM2
DDX18
SLC4A3
ASPA
SLC22A18AS PPP2R4
CGGBP1
CSRP2
SLC25A11
PDIA6
BEST1
MCM7
SLC1A4
KHDRBS3
ART3
CAPN3
GYG2
DLAT
TUBB4
RFNG
MORC3
MAGEA5
AZI1
UAP1L1
SFXN3
HLA-DMA ALDH18A1
ARHGEF3
FAHD2A
RINT1
ITIH5
CA14
C14orf139
LONRF3
TINF2
TP53TG3
TH1L
C5orf54
CDCA3
C3orf64
GPR177
COL9A1
XPO5
PDXP
COPG2
C2orf30
C6orf89
UBL7
GGT7
C12orf34
C3orf38
C11orf82
ENHO
CITED4
PAGE2
LOC730124
GBGT1
ZC3H12C
FSTL5
CLEC2L
LRRC33
NPHP3
HMCN1
DLX1
Top 222 over expressed genes for cancer
cell line MDA_N (melanoma)
Future Research
Identify protein sub-networks
that “connect” the
transcription factors through
additional proteins
Analyzing the mRNA profile from the NCI-60 database
Statistical Methods:
Probes
1
2
…
n
1
M1,1
M2, 1
…
Mn,1
2
M1,2
M2,2
…
Mn,2
Cancer cell lines
3
…
c
M1,3
…
M1,c
M2,3
…
M2,c
…
…
…
Mn,3
…
Mn,c
With gene input, ChEA identified the top
ranked transcription factors
Genes2Networks
output of protein subnetworks when top
10 transcription
factors from ChEA
were given as an
input
…
…
…
…
…
Genes2Networks
KEA
ChEA, Genes2Networks, and KEA are all web-based
tools developed at the Ma’ayan lab to allow users to
predict which transcription factors, protein subnetworks, and protein kinases are most correlated
with their inputted seed list
• By using the identified up and down regulated
genes for each cancer cell line as an input for
ChEA; the top ranked transcription factors (based
on p-value from Fisher’s Exact Test) that most likely
influence the input seed list are given as the output
Genes2Networks
•The transcription factor output for each cancer cell line
from ChEA is used as an input to Genes2Networks
• Genes2Networks connects lists of transcription
factors with other protein intermediates from
mammalian protein interactions databases
KEA
Wet lab experimental validation
• Differentially expressed gene lists from the various NCI-60 Cancer
Cell Lines are used as input.
• Over expressed and under expressed genes are identified for
specific cancer cell lines
• The following algorithm was implemented:
• The NCI-60 database was parsed and 18,133 unique genes
were identified
• The population mean for the expression of each of the genes
across all the 60 cancer cell lines was calculated
• The sample mean and sigma for each (gene, cancer cell line)
pair was calculated
• The two-sided T-test statistic was applied for each (gene,
cancer cell line) pair.
• Whether the gene was over expressed or under expressed
was calculated by checking whether the test statistic
exceeded a critical T score or was a less than a critical T
score determined based on a particular P value.
• A list of genes which are over/under expressed for multiple
cancer cell lines was developed
ChEA
60
M1,60
M2,60
…
Mn,60
Population mean µ = ∑ Mi,j / (n * 60)
i=1,n; j=1,60
Sample mean of gene expressions for cancer cell line “c” = x̄ = ∑ Mi,c / n
i=1,n
•The unique protein sub-networks outputted by
Genes2Networks can then be inputted into KEA which
identifies protein kinases most likely regulating the
proteins from the subnetwork using the Fisher’s Exact
Test.
• At this stage top regulating transcription factors,
protein sub-networks and kinases have been identified
for each of the NCI-60 cancer cell lines
• An integrated matrix can now be created in order to
holistically compare the data by displaying the top
regulating elements and their putative effects on the
different cell lines
Future Research
• Future research involves further analyzing other
cancer datasets
• Cluster analysis will be done to groups
transcription factors or kinases that were identified
• Additionally, by combining such data with data
collected for drug perturbation of these cells, we
may be able to suggest which drugs can reverse
the observed changes
Acknowledgements
Std deviation of gene expressions for cancer cell line “c” = s = sqrt (∑ (Xi – x̄ ) 2 / (n-1))
i=1,n
Test statistic = ( xbar – µ) * sqrt(n) / s
Top ranked kinase proteins identified
from KEA
This research was supported by NIH Grant No.
5P50GM071558