1. Microarrays and Gene Expression
DTC Bioinformatics Course
9th February 2010
Helen Lockstone

2. Overview Background Array design Applications of array technology
Steps in data analysis
Finding differentially expressed genes
Biological interpretation

3. Schedule
Time
Topic
Introduction to microarray technology and applications
Break
Microarray data analysis
Practical 1
Lunch
Biological interpretation
Practical 2

4. Microarrays in the Literature

5. The Central Dogma
Transcriptome measured by microarrays

6. Premise of Microarrays
Compare gene expression between groups
Differentially expressed genes may provide some biological insight
But not magical solutions!

7. Typical Microarray Designs
Disease vs control
Good prognosis vs poor prognosis
Different tumour types
Effect of treatment
Effect of stimulus
Time course
Different tissues/stages of development

8. Criticism of Microarrays
Non-hypothesis driven “fishing expeditions”
Because microarray experiments are expensive and time-consuming to interpret, often published as a stand-alone experiment
Produce large amounts of data, interpretations can be very different (but equally valid)
Further experimental work, following up hypotheses suggested from array data, can produce elegant studies
Perception that data is unreliable – validation

9. Microarray Repositories
GEO –
ArrayExpress -
Excellent resource of microarray data
MIAME guidelines

10. What is a Microarray?
Glass slide consisting of hundreds of thousands of probes arranged in grid layout
Each probe detects a particular RNA species (transcript)
Hybridisation occurs by complementary base-pairing
Make quantitative measurements – signal from each probe is proportional to the amount of hybridised RNA
Interrogate entire genome in single experiment

11. Microarray Technology
Probes
cDNA
Oligonucleotides
PCR products
Design
Targeted to genes
Tiling (chromosomes, promoters)
Fabrication Method
Spotted (robotic printing)
Photolithography (synthesised in-situ)
Type
One-colour (log intensities)
Two-colour (log ratios)
Labelling molecules
Cy3 (green), Cy5 (red), biotin

12. Experimental Protocol

13. Microarray Manufacturers
Company
Established
Main Microarray Technology
Human Whole-Genome Array released
Headquarters
Affymetrix
1992
GeneChip
1994
Santa Clara, CA
Illumina
1998
BeadChip
2005
San Diego, CA
Roche NimbleGen
1999
High-density tiling arrays
Madison, WI
Agilent
aCGH, ChIP-chip, custom
2004

14. Array design

15. Affymetrix Microarrays
Manufacturing microarrays for >15 years
25bp probes – 11 individual probes comprise a probe-set, signal combined to estimate gene expression
Whole human genome array has >50,000 probesets
Size array surface 1.28cm2
3’ expression arrays – probes designed to 3’ end of transcript

16. Recent Developments Limitations of 3’ array design
Assumes representative of entire gene
Assumes well-defined 3’ end of gene
Can’t assess splicing events
Can be difficult to distinguish homologous genes
Whole transcript arrays
4-probe probesets designed to each exon
Gene 1.0 and Exon 1.0 arrays

17. Exon Array Design
Picture from Affymetrix

18. Illumina Beadchip Arrays
Beads randomly occupy wells on surface of array
30-40 replicates of each bead type (probe)
Longer probe length – typically one probe per gene

19. Applications of Microarray Technology

20. Microarray Applications
ChIP-chip
Gene Expression
Alternative Splicing
DNA Methylation
microRNA expression
Comparative Genomic Hybridisation
SNP Genotyping

21. Gene Expression Still most common use for microarrays
Aim to determine differential expression between groups of samples e.g. disease and control
Generate hypotheses about the mechanisms underlying the disease of interest

22. Alternative Splicing
Up to 75% of human genes may produce alternative transcripts
Increases protein diversity from given set of genes
Alternative transcripts from same gene can produce proteins with different, even opposite, functions (e.g. Bcl-x)
Role in disease - mutations can disrupt splice sites or splicing machinery

23. Alternative Splicing
Affymetrix exon array allows investigation of alternative splicing
Custom arrays with junction probes
Additional layer of analysis

24. Alternative Poly-A Sites
Alters length of 3’ UTR - may change which target regions for miRNAs are present

25. Alternative Splicing

26. MicroRNAs Small non-coding RNAs (~22bp)
Sequence-specific binding to 3’ UTRs
Post-transcriptional gene silencing
Picture from He et al. Nature Reviews Cancer 7, (2007)

27. SNP Arrays Illumina and Affymetrix ~6 million SNPs genome-wide
Genotype individuals in high-throughput and cost-effective manner
Genome-wide association studies
eQTL studies

28. Tiling Arrays
Applications so far use arrays with probes designed to genes/miRNAs/SNPs of interest
Tiling arrays consist of high-density probes covering a particular region(s) of the genome
Identify novel transcripts, exons

29. DNA Methylation
Methylation of cytosine bases (CpG islands) in gene promoter regions can silence transcription
Epigenetic mechanism
Two-colour hybridisation

30. ChIP-chip
Method to identify transcription factor binding sites in an unbiased fashion
Cross-link protein (TF) of interest with DNA
Use immuno-precipitation to pull down DNA fragments bound to the protein (enriched sample)
Hybridise with genomic DNA to obtain log-ratio
Again looking for large positive ratios

31. Comparative Genomic Hybridisation
Trisomy 13 in female compared to reference male
Detect regions of amplification/deletion (copy number changes)
Feature of cancer – hybridise sample with reference DNA (copy number=2)
Potential dosage effects on genes in affected regions

32. Analysing Gene Expression Data

33. R and BioConductor
Powerful, open-source software for statistical analysis and graphical visualisation
Greater functionality provided by software packages contributed by researchers
BioConductor packages are specifically for genomic data
affy
limma
vsn

34. Analysis Steps Check quality of the data
Decide if any samples are outliers
Preprocessing and normalisation
Statistical analysis to find differentially expressed genes
Tools for biological interpretation

35. Data Quality
Looking for good signal and similar metrics across all arrays in experiment (after normalisation between arrays)
Poor signal could indicate a hybridisation problem or degraded sample
Control probes for hybridisation, labelling and sample can help identify problems

36. Illumina Array Metrics
Average signal
Number of detected genes
Housekeeping genes signal
Biotin controls
Hybridisation controls
Negative control probe signal

37. Processing Data Background correction
Transform data to log scale (more suitable for statistical analysis)
Normalisation between arrays (adjust for systematic differences such as overall brightness)
Probe-set summarisation (Affymetrix) or across replicate probes (Illumina)

38. Exploring Data – Boxplots Signal Intensity

39. Exploring Data - PCA

40. Outlier Samples
Potential outlier samples will look different to others in the experiment
No definitive rules to decide when to exclude a sample from analysis
Depends on size of experiment
Can be useful to run analysis with and without outlier to assess effect on results
Always re-normalise data excluding any outlier samples before proceeding

41. Outlier Sample

42. PCA indicating outlier sample

43. Filtering
Lose data but signal from low intensity probes is noisy and can give false positives
Detection p-values calculated for each probe based on overlap of signal with negative control probe signal distribution
Criteria
Detected in all samples/at least one sample
Detected in at least one group

44. Detecting Differentially Expressed Genes
Linear Models for Microarray Analysis (limma)
Handles analysis of simple and complex experimental designs
For two-group comparisons, analogous to t-test, otherwise ANOVA
Uses information from all genes to estimate variance
Reduces chance of false positives from very low variance genes
More robust for small sample sizes

45. limma Fits linear model for each gene
Test whether slope = 0 for each gene and assign p-values
Multiple testing correction - FDR
Group 1 Group 2
Log normalised intensity

46. Effect of other variables
Wt and Mut groups
Three different litters
Top gene ~ 5x higher expression in Wt compared to Mut
Similarly expressed across litters in both genotypes

47. Strong litter effect Overlap between groups
Within litters, consistent pattern of higher expression in WT vs Mut
Within genotypes, B>C>A – expression depends on litter
Accounting for this variance increases power

48. Limma Output

49. Limma Output
Small sample size and subtle effects can mean no probes would be considered statistically significant
Ranked in order of evidence for differential expression – can still be explored
Biological interpretation can be most difficult step – tools available