Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Photomorphogenesis talk
1. A transcriptomic analysis of photomorphogenesis in
Arabidopsis Thaliana
Hugh Shanahan,
Department of Computer Science,
Royal Holloway,
University of London
UCC,
Department of Microbiology
19 September 2007
2. Outline
• Introduction :- A. Thaliana as a model organism
• Photomorphogenesis
• A conservative estimate of differentially expressed genes
• A strategy for examining functional classification
• The picture of photomorphogenesis
• Where to go from here...
3. Why is Arabidopsis thaliana a good system ?
• Short lifetime - around 40 days.
• Large number of genes - ~25,500 (Drosophila around
11,000 genes)
• Compact Genome - 125 Mbases (non-coding regions
~30-50%)
• Large number of array experiments (65 Affymetrix data
sets, with 10’s of raw image files per data set at TAIR for
example)
• A huge number of different strains - over half a million
genotypes.
4. And why perhaps not so good....
• Not much Protein-Protein interaction data (Tandem Affinity Purification experiments
on the way).
• The genomes of near neighbours have not been sequenced.
• A. lytra, C. rubella, B. rapa and T. halophilia are planned or underway
• Not clear how representative genes are for other agriculturally relevant species
• A. thaliana has a huge repertoire of Ubiquitination proteins.
5. Photomorphogenesis in
Arabidopsis thaliana
• Before exposure to light, seeding
grows via skotomorphogenesis
after germination - slow root
growth, no growth in shoot apical
meristem or cotyledon.
• When exposed to light, cotyledon
grows through simple
reproduction (control).
• Meristem (stem cells) grow by
differentiation.
• Meristem source of true leaves.
• Little understood about process.
6. Photomorphogenesis in
Arabidopsis thaliana
• Before exposure to light, seeding
grows via skotomorphogenesis
after germination - slow root
growth, no growth in shoot apical
meristem or cotyledon.
• When exposed to light, cotyledon
grows through simple
reproduction (control).
• Meristem (stem cells) grow by
differentiation.
• Meristem source of true leaves.
• Little understood about process.
7. Photomorphogenesis in
Arabidopsis thaliana
• Before exposure to light, seeding
grows via skotomorphogenesis
after germination - slow root
growth, no growth in shoot apical
meristem or cotyledon.
• When exposed to light, cotyledon
grows through simple
reproduction (control).
• Meristem (stem cells) grow by
differentiation.
• Meristem source of true leaves.
• Little understood about process.
8. The data
• RNA material was gathered from the shoot apical meristem and cotyledon of
Arabidopsis seedlings at
• 0 hour (in darkness)
• 1 and 6 hours (Cot and Mer with replicates)
• 2, 24, 48 and 72 hours (Mer only)
• Samples hybridised with Affymetrix ATH1 GeneChip array.
• No amplification of RNA material !
9. Strategy
• Construct stringent test to determine genes which are clearly differentially
expressed.
• Identify kinetic behaviour of different classes of differentially genes (i.e. try and find a
time line of events).
• Identify functional groupings of genes and then examine how all the genes in that
functional grouping behave (i.e. including those that are not differentially expressed
according to our strict criteria).
10. Finding differentially expressed genes
• Look at three different normalisations for Affymetrix data
• GCRMA
• MAS5
• VSN
• Only consider genes that are differentially expressed in all three normalisations as
being significant.
11. Test for significance
• Apply two-way ANOVA test. Look
for significance with respect to
• tissue
• time
• time and tissue
• Compute F-Ratio
• Only use data with two replicates
(i.e. Cot and Mer at 0, 1 and 6
hours)
12. Finite sample size :-
bootstrapping
• 2 replicates for the ANOVA data
set.
• Cannot trust a p-value from such
data !
• Solution :- create a large set of
artificial data by randomly
selecting expression values from
all of the data.
• Compute histogram of resulting F-
values for ANOVA test to
determine a p-value.
13. Finite sample size :-
bootstrapping
• 2 replicates for the ANOVA data
set.
• Cannot trust a p-value from such
data !
• Solution :- create a large set of
artificial data by randomly
selecting expression values from
all of the data.
• Compute histogram of resulting F-
values for ANOVA test to
determine a p-value.
14. Finite sample size :-
bootstrapping
• 2 replicates for the ANOVA data
set.
• Cannot trust a p-value from such
data !
• Solution :- create a large set of
artificial data by randomly
selecting expression values from
all of the data.
• Compute histogram of resulting F-
values for ANOVA test to
determine a p-value.
15. False Detection Rate
• Bonferroni Correction is very
conservative.
• Estimate FDR by plotting a
histogram of the p-values.
• Fix FDR to 5% and set p-value.
• Important step : employ Present/
Absent filter in MAS5 to filter out
genes. Substantial improvement in
results.
• Final step : remove all genes with
fold change less than 2
16. Initial Results
• Selected 5,620 genes (out of 22,810).
• (Very conservatively) 1/4 of the transcriptome is differentially expressed during
photomorphogenesis.
• Majority selected through time variation (2/3 time, 1/3 tissue).
• Very small number selected using time-tissue variation (10).
17. Functional Classification :- it should be easy....
• Many genes in Arabidopsis have some kind of functional annotation.
• Use Gene Ontology to give a structured functional annotation.
• Enumerate numbers of genes for a given annotation.
• Compute probability of over or under-representation using hyper-geometric
distribution.
18. Functional Classification :- but it isn’t....
• Gene Ontology are useful but far too specific at its lowest nodes - False Discovery
Rate calculation.
• Initially interested in general picture, what is the highest level annotation ?
• GO slim should cover more general cases, however annotations of genes can have
multiple parents, e.g. a gene with kinase function and binds to DNA will sit in both
classes.
• Ultimately, we developed our own general functional annotation.
• Lesson :- GO has a huge amount of information, but when looking at the big picture,
you need to make the decisions !
19. Functional Classification :- but it isn’t....
• Gene Ontology are useful but far too specific at its lowest nodes - False Discovery
Rate calculation.
• Initially interested in general picture, what is the highest level annotation ?
• GO slim should cover more general cases, however annotations of genes can have
multiple parents, e.g. a gene with kinase function and binds to DNA will sit in both
classes.
• Ultimately, we developed our own general functional annotation.
• Lesson :- GO has a huge amount of information, but when looking at the big picture,
you need to make the decisions !
20. Functional Classification :- but it isn’t....
• Gene Ontology are useful but far too specific at its lowest nodes - False Discovery
Rate calculation.
• Initially interested in general picture, what is the highest level annotation ?
• GO slim should cover more general cases, however annotations of genes can have
multiple parents, e.g. a gene with kinase function and binds to DNA will sit in both
classes.
• Ultimately, we developed our own general functional annotation.
• Lesson :- GO has a huge amount of information, but when looking at the big picture,
you need to make the decisions !
25. Clustering :- Understanding the time behaviour
• Important lesson :- cluster early, cluster often.
• While genes have been selected using data at 0,1,6 hours, all data is used for
clustering.
• Results quoted use K-Means (20 clusters) but checked with varying cluster size and
with Hierarchical Clustering (very different clustering algorithm) to check if clusters
are consistent.
• For each cluster, examine functional classes and explore any possible over-
representations.
37. Cell Wall
• Plant cells have rigid cell walls.
• Expansion implies that cell walls
must become less rigid.
• Complicated process between cell
wall modification and internal
turgor pressure.
• Nonetheless, see late expression
in genes controlling this behaviour.
38. Summary of what’s been seen
• After constructing a set of genes which are strictly differentially expressed we find:
• An early burst (0-1 hours after exposure to light) of genes in the meristem that are
regulatory in nature and are in general down rather than up-regulated.
• Around 6 hours after exposure to light evidence for cell division and repression of
growth.
• At later times, parts of meristem are already starting to behave like leaves and we
see growth through expansion rather than division (up-regulation of relevant
hormone-related genes, down-regulation of ribosomal genes).
39. Where to go from here
• So far we’ve put together a picture of genes up and down regulated and looking at
their classification fits in with a picture of particular types of growth.
• Useful at middle and late times (6 hours and beyond)
• We see sets of transcription factors, MAP kinases and Ubiquitination-related genes
which are over-represented.
• Can we identify targets of these regulators ?
• Start off with co-regulated sets and see if we have common upstream promoter
elements.
• How conserved are the above mechanisms elsewhere ?
41. Acknowledgements
School of Biological Sciences, Computer Science,
Royal Holloway Royal Holloway
• Saul Hazeldine
• Enrique López-Juez
• Edyta Dillon
• Safina Khan Department of Molecular Genetics,
• Zoltan Magyar Ghent University
• Laszlo Bögre
• Gerrit Beemster
Institute of Biotechnology,
University of Cambridge
• James A. H. Murray
