Transcript Characterization of Gene Expression Profiles Associated

Characterization of Gene Expression Profiles
Associated with Glioma Progression using
Oligonucleotide-Based Microarray Analysis and
Real-Time RT-PCR
Van Den Boom et al
American Journal of Pathology, V 163, No 3, Sep 2003
Presented by: Puneet Kumar
CMDB 281
Introduction & Background Information
• Glioma accounts for ~ 80% of all brain tumors in adults
• Annually ~ 36000 cases of glioma reported
• No complete cure possible. Best chance in surgery or
surgery+Radiotherapy
• Even after surgery chance of recurrence very high
• Avg. age of survival (with treatment) after detection is 1.5-5 yrs
Glioma  Astrocytoma
Purpose of this study
• Molecular basis of astrocytoma progression not fully understood
• Most prev. studies have only shown changes at the chromosomal
level(genetic aberrations like homozygous deletion, amplification,
LOH)
• This study was undertaken to determine the differential transcriptional
profile of ~ 6800 genes in primary WHO grade II gliomas &
corresponding recurrent high grade (WHO grade III or IV) gliomas
from 8 patients using oligonucleotide array analysis.
Point to note : PTEN mutations and EGFR/MDM2/CDK2
amplifications absent in spite of the wide occurrence of these
genetic aberrations in the gliomas.
Experimental design of the array
• Authors chose a made-to-order HuGene FL oligo microarray
(Affymetrix) consisting of 7129 probe sets corresponding to 6800
genes (the web site lists this array as having 5600 genes).
Sample extraction & sample application
• RNA extracted from frozen tumor samples
Tumor samples chosen in such a way that most of them (all except
one) had not received chemotherapy/radiotherapy.
• Nice point- Since chemotherapy/radiotherapy could have altered
the gene expression profile in the tumor samples.
Sample application was according to Affymetrix described protocol.
Image scanning & Image extraction
• Image scanning: Gene array scanner (provided by Affymetrix)
• Image Extraction: Microarray Suite 4.0 (provided by Affymetrix)
Normalization
• Good Point- Instead of using proprietary software for normalization &
further data analysis, they used their own developed & made publicly
available software.
PM-MM = Signal
PM
Antilog[Mean of log PM
Measured
S tan dard
] x PMstandard = Imputed PM
Same way calculation for MM.
Probe pairs with PM-MM  -200 are removed from analysis
25 % of the highest & lowest differences also removed. The remaining
values are averaged.
This becomes the “trimmed mean” analogous to the “ avg. difference”
used by Affy.
Normalization (contd.)
• Intensities for the remaining arrays are normalized using quantile
normalization whereby the distribution of trimmed means is adjusted to
match the distribution of a standard chip by making 99 individual
quantiles having the same values using a piecewise linear function.
Pretty Good way of normalization: all the negative signal values are
not directly dumped but some of them are taken into acct.
All the self developed software & documentation available at:
http://dot.ped.med.umich.edu:2000/pub/glioma/index.html
Pattern recognition/Clustering
• To judge whether there was any difference in clustering primary &
recurrent tumors, performed Principal Component Analysis &
hierarchical cluster analysis on the overall expression data.
Overall expression data do not completely separate the primary low
grade from the recurrent high grade tumors.
• Performed paired t-tests on the log transformed data for the 8 tumor
pairs (to be able to distinguish between primary & recurrent tumors).
Yielded 266 probe sets with P  0.01 (pretty stringent condition).
Included another level of stringency as taking only those probe sets
with  2 fold change between the low grade & the high grade tumors.
Yielded 70 probe sets( corresponding to 66 genes).
• Out of these 23 genes were upregulated & 43 were downregulated.
A heat map of these genes & their identities are shown :
Upregulated genes: Most encode either proliferation related proteins
(e.g. FOXM1, TOP2A, CDC20, MCM6, KNSL6, TFRC, and MCM4)
or proteins associated with ECM formation and/or angiogenesis (e.g.
COL4A2, COL4A1, ACTA2, MGP, COL5A2, JAG1 and BGN).
Downregulated genes: More heterogeneous with respect to functions
of gene products. e.g. genes involved in ECM formation, cell adhesion
and/or cell motility (e.g. SELL, SELPLG, CD37, TJP2, ITGAM, and
GSN), development of the nervous system (e.g. MAP1A, ID1,
SEMA3B, and NTRK2) or signal transduction (e.g. GJA1, CAMK2G,
MAP3K5, ARL3, S100A13, and ACCN2).
Confirmation of differential expression of selected
genes by RT-PCR analysis
Selected 8/66 genes initially.
Added ADD3 & ABLIM which had P < 0.01 but just less than 2 fold
expression level change (saying that they represented interesting
candidates from 10q).
Also selected 2 more genes CENPF & VEGFA because they showed
large increases in expression without being statistically significant.
Finally RT-PCR validation of data for 12 selected genes was done.
Points to note:
• Took a larger % age of upregulated genes(7/23) than downregulated
genes(5/43). Going after the upregulated genes more than the
downregulated genes.
• Inclusion of additional genes not supposed to have been there by
microarray analysis alone reveals that they did not have too much
faith on their own microarray data.
Point to note: The stringency for significance of data for RT-PCR
analysis was different than that for microarray analysis (P <0.05 for
RT-PCR as compared to P <0.01 for microarray analysis).
Could be due to the fact that RT-PCR analysis is more sensitive
than a microarray analysis.
RT-PCR analysis of those 12 selected genes in an
independent set of 43 gliomas(9A’s, 7sGB’s,
17pGB’s, 10 AA’s)
Individual comparisons between “A” grp. Vs. the “AA” grp. Or “GB” grp. was
done.
Observations from Table 3
• Comparison of “A” grp. Vs. “AA” grp.& “GB” grp. reveals lower
expression of FOXM1, TOP2A, COL4A2, CENPF & MGP as well as
higher levels of ADD3 in “A’s”.
• VEGFA & IGFBP4 mRNA levels higher in “GB’s” than in “A’s” &
“AA’s”.
• Variation in CAMK2G transcripts only between “GB’s” & “A’s”.
Thus found the differential expression of mRNA between “A” grp. vs.
the “AA” grp. or “GB” grp.in 9 out of 12 genes selected.
Correlation of expression data with genetic
aberrations
• Authors cite the amplification of CDK4 in recurrent tumor sample of
patient 8. They say that array reveals the overexpression of several
genes in the close proximity of CDK4 (like SAS, TSFM, MARS,
METTL1, PIKE, GALGT, and PRIM1).
However, none of these genes are there in the list of the 23 upregulated
genes.
Analysis for amplification of candidate genes
overexpressed in high grade gliomas
Amplification of FOXM1 & MGP genes : Duplex PCR and Southern Blot in 8
glioma pairs & 24 independent “GB’s”.

No amplification of these genes was found.
Only one “sGB” patient (GB15D) & one “AA” patient sample (AA106D) show
 signal intensities for both FOXM1 & MGP (GB15D) or just FOXM1
(AA106D) suggesting copy no. gains of about 3 fold in these tumors.
Point to note- No patient GB15D exists in the list in Table 1.
Discussion
• Identified 66 genes whose expression varies with the progression of
the disease. Out of these, some had prev. been identified like VEGFA,
CENPF, and TOP2A.
In addition, many novel genes were also identified (like COL4A2,
FOXM1, MGP, CAMK2G).
For many genes like FOXM1 & MGP the differential expression was
huge but no amplification of the gene was found implying there were
other mechs. of mRNA overexpression operating(as stated by the
authors)
• Alternatively, there is even the possibility of their techniques used for
checking gene amplification could be erroneous since they did not find
gene amplification in any tumor sample for any gene (except one case)
in spite of the fact that they keep mentioning about the overexpression
& amplification of these genes again and again.
Final Critique :
• The authors do a good job at trying to analyze the data using nonproprietary software & making their techniques & software freely
available.
• The authors did not use any sample replicates in their study(may be
due to prohibitive costs). This raises a question mark on the reliability
of their statistical data.
• The authors show a slight bias while selecting genes for RT-PCR
validation (are more concerned about the targets they have in mind
than are being observed according to the microarray analysis.
• The authors fail to confirm some of the data that has previously been
shown ,like PTEN mutations & CDK4/MDM2/EGFR amplification in
gliomas. (may be due to inconsistent benchwork practices).
• The authors found some novel genes (in addition to some previously
identified genes) whose expression changed during the course of
progression of the disease & said that these gene targets could be used
for future therapeutic purposes.
Thank You