Transcript Microarray technology and analysis of gene expression data

Microarray technology and
analysis of gene expression data
Hillevi Lindroos
Introduction to microarray technology
• Technique for studying gene expression for
thousands of genes simultaneously.
• Study gene regulation, effects of treatments,
differences between healthy and diseased cells...
• Comparative Genome Hybridization:
- gene content in related strains/species
- gene dosage in cancer cells
• Microarray: glass slide with spots, each containing
DNA from one gene
Two-colour spotted microarrays
Spot = PCR-product (~500 bp) from one gene or
long oligonucleotide (~50 bp)
Differential expression (two samples compared)
Experimental procedure:
1. Isolate RNA from 2 samples (experiment and control).
2. Reverse transcribe to cDNA with fluorescently labelled
nucleotides, e.g. Cy3-dCTP (control) or Cy5-dCTP
(experiment).
3. Mix and hybridize to microarray.
4. Laser scan: measure fluorescent intensities
Red and green images superimposed:
In principle...
Red spot: up-regulated gene, ratio >1
Green spot: down-regulated gene, ratio <1
Yellow spot: no differential expression, ratio =1
Control
RT
+ green dye
mixing equal amounts
of cDNA
gene A
RT
+ red dye
competitive
hybridization
Sample (e.g. heat shock)
Up-regulation
Microarray
Red dot in image
Why differential expression?
Fluorescent intensities do not directly correspond to
mRNA concentrations, due to:
• different shapes and densities of spots
• different hybridization properties between genes
• different amounts of dye incorporation between genes
 Compare intensities (expression) from two samples.
Data processing and analysis
1. Image analysis
Locate spots in image
Quantify fluorescence intensity (spot + background)
Mean / median of pixel intensities
2. Background correction
– local background for each spot, or global for
whole array
– assuming additive background:
Spot intensity = True intensity + Background
Output
Cy5 (R) and Cy3 (G) intensities
Ratio = R/G
~ [mRNA_experiment] / [mRNA_control]
Up-regulated genes: ratio >1
Down-regulated genes: ratio= 0-1
Assymetry!
 Use logarithm!
M = log2(ratio) is symmetrically distributed around 0
Upregulated 2 times: ratio= 2, M= 1
Downregulated 2 times: ratio= 0.5, M= -1
3. Normalization: correction of systematic errors
(dye bias)
• different amounts of control and experiment
samples
• different fluorescent intensities of Cy3 and Cy5
• different labelling and detection efficiencies
Plot of Cy5 intensity (R) vs Cy3 intensity (G):
Dye bias: Most genes seem to be upregulated
(higher Cy5 than Cy3 intensity).
Corrected for by scaling Cy5 values with
total_Cy3/total_Cy5.
Assumes most genes unaffected by treatment.
Intensity dependent dye bias
Dye bias may depend on total spot intensity A
(A =½(log2R+log2G)), position on array, print-tip…
Correction:
Mnormalized = M – Mtrend(A)
Identify differentially expressed genes
•Simple: cutoff (e.g. |M| > 1)
•Better: statistical test, e.g. t-test (replicate spots or
repeated experiments) => Significance
–Unstable mRNAs may have high ratios – and
high variation!
–Weak spots: small difference in signal may be
big relative difference (high ratio).
Affymetrix genchips
Spots = 25 bp oligonucleotides
Pairs of perfectly matching probe + probe with 1
mismatch for each gene
One sample per array
Radioactive labelling
Expression level computed from difference in intensity
between matching and mis-matching probe
Expression profiles
Plot expression over a series of experiment (e.g.
time series)
Expression profiles
3
M = log2(R/G)
2
1
0
0
1
2
3
-1
-2
-3
-4
Time
4
5
6
Gene_A
Gene_B
Clustering expression profiles
Analyze multiple experiments to identify common
patterns of gene expression
Similar function – similar expression (co-regulation)
Goals:
•Identify regulatory motifs
•Infer function of unknown genes
•Distinguish cell types, e.g. tumors (cluster arrays)
Hierarchical clustering
Expression profile -> vector
Compute similarity between expression profiles (e.g.
correlation coefficient)
Successively join the most similar genes to clusters, and
clusters to superclusters
Serum stimulation of
human fibroblasts,
time series.
A: cholesterol biosynthesis
B: cell cycle
C: immediate-early
response
D: signaling and
angiogenesis
E: wound healing
Distance: correlation coefficient
Agglomeration: average linkage
from: Eisen et al., 1998, PNAS
95(25): 14863-14868
Clustering of arrays:
classification of
cancer cells.
From Chen et al. (2002). Mol
Biol Cell 13(6):1929-39
Exercise:
Normalization (Excel):
R-G plot
M-A plot
most up- and downregulated genes