Transcript here

Journal club
Wouter
10 dec 2013
Why
• Interest in autism
• Follow-up of gene-finding
• Interesting: two papers in same issue Cell similar findings
Overview paper
1.
2.
3.
4.
Select hcASD-genes (9) and pASD-genes (122)
Use data Kang & reduce spatial and temporal number of windows
Find enrichment of pASD in coexpression networks in 4 areas
Test enrichment with:
1) hypergeometric test 2) hcASD permutation 3) pASD permutation
4) number of genes selected in network 5) cross validation
6) single period weighted 7) excluding TBR1
5.
6.
7.
8.
9.
10.
Focus on TBR1
TADA confirming pASD genes higher chance in midfetal period
Further improve spatial resolution to layers
Analyze temporal behavior of layer found
Find cell type
Immunostaining in midfetal CPi cortex
Introduction
• No common genetic variation reproducible linked to autism
• However, sequencing has recently led to discovery of de novo loss
of function (LoF) mutation.
• De novo LoF mutations are expected to play role in 15% of patients
• List of associated genes is steadily growing
• Associated loci heterogeneous with respect to biological function
 challenge for translation
Hypothesis
Goal
Gene selection
• Total of 1043 families (987 previously published, 56 additional
exome sequenced)
• LoF = premature stop codon, splice-site disruption, or frameshift
insertion/ deletion
• 144 LoF de novo mutations identified
Chance of true ASD gene
• Subset of 599 quartets: 75 LoF in 72 affected versus 34 in 32
unaffected (OR=2.21, p=5e-5)
FDR of gene ≥ 2 independent cases with LoF
• Permutation: p=0.1975 to find 2 LoF in same gene by chance
• 9 genes with ≥ 2 LoF genes found
• 45.6 more often than expected (9/0.1975)
• FDR = 0.022 (1/45.6)
• Chance of true ASD gene is 0.978
• Analogue chance of true ASD for 1-hit gene (0.55) and 3-hit gene
(0.9998)
hcASD / pASD genes
hc = high confidence (m=9)
• LoF in gene in two unrelated cases (FDR 0.02)
• LoF in three cases (FDR 0.0002)
p = probably (m=122)
• LoF in one case (FDR 0.45)
Use these genes to construct spatiotemporal coexpression networks
Transcriptome data
Transcriptome data
Transcriptome data
• Expression in
– 16 brain regions
– 57 clinically unremarkable postmortem subjects (31M 26F)
– 15 periods from 5.7 PCW to 82 Y
(Thus, 16*14=240 spatiotemporal units)
• Partitioned in subsets
– Temporal partitioning: 13 sliding windows of three consecutive time periods
– Why?
Coexpression network
• Network = hcASD gene + max 20 top correlated genes + edges
• For each gene (M = 16,947 + 9), vector of expression values, by
brain-region and brain-sample
• Per spatiotemporal window, correlation of expression-vectors
between gene-pairs
• Per hcASD, select 20 top correlated genes with abs. cor. ≥ 0.7
• Edges are are correlations between each gene-pair of network with
abs. cor. ≥ 0.7
Spatial partitioning – step 1
• Why?
• Select period, in which networks are most
enriched for pASD genes  period 3-7 (1038 PCW)
Spatial partitioning – step 2
• Select coherent subsets of brain regions based on period 3-7
• Summarize gene-expression per brain region by median expression across all
samples
• Compute pairwise correlation between brain regions
• Subsequent, hierarchical clustering (distance is 1-corr2)
 4 clusters of brain regions
Thus, 4*13 = 52 spatiotemporal windows, with coexpression networks
constructed
Results
Results
Hypergeometric test
• Probability of k successes in n draws without replacement
k = number of successes drawn (nr pASD-genes in network)
K = total number of successes (total nr pASD = 122)
n = number of draws (genes in network, ≤ 20)
N = population (16,947 genes)
Problem: larger genes more chance of de novo LoF mutations
Permutation test 1
• Tests if true hcASD genes are crucial to enrichment with pASD
found
– Select 9 pseudo hcASD genes (based on the likelihood of observing 2-hit de
novo LoF mutations by chance, taking gene size and GC-content into account)
– Build corresponding coexpression networks in concerning spatiotemporal
windows & test enrichment with pASD genes
– 100,000 iterations
Permutation test 1
Permutation test 2
• Identical, but with true hcASD, and permutation of pASD
Permutation test 3
• Permutation of hcASD, with true pASD
• For varying number of genes in coexpression network
Cross-validation
• Remove 1 hcASD and 12 pASD (10%)
• Reconstruct 52 spatiotemporal coexpression networks
• Success = 1 of top three networks most enriched for pASD
– top three PFC-MSC 3-5 & 4-6, MD-CBC 8-10
• Success in 100% of 200 iterations
Single period weighted analyses
• Before, 3 periods equally weighted
• Now, middle period weight 1, periods immediately before and after
weighted 0.5
Questions
• How does “increasing resolution” influence subsequent results?
• Why take expression in subjects older than say 1 year into account?
• Why not report correlation between hcASD gene-expression?
About brainregions
• V1C, ITC, IPC, A1C, STC: non-significant in permutation: dropped
• PFC-MSC: 107 sample (period 3-5) & 140 (period 4-6)
• MD-CBC: only 26 samples (period 8-10): dropped
• Two PFC-MSC networks referred to as midfetal networks
PFC-MSC = Pre-Frontal-Cortex & Primary-Motor-SomatosensoryCortex
TADA
• = transmission and the novo association- test
• Why? To test if pASD in midfetal network are more likely true ASD
genes than estimated with FDR (55%)
• TADA combines family and case-control data
TADA
• Additional, case (935) control (870) data included from ARRA (Liu)
(Liu 2013. Analysis of rare, exonic variation amongst subjects with autism spectrum disorders and
population controls. PLoS Genet.)
Biological interpretation
T-box, brain, 1 (TBR1)
• TBR1=hcASD
• Known transcription factor involved in forebrain development
• In mice
–
–
–
–
–
–
Postnatal day 0 ( = human midfetal)
RNA-seq of cortex
Compare expression in TBR1-/- & TBR1+/+ (n=?)
4 of differentially expressed genes (DEX) in coex- network
TBR1 previously known to regulate these DEX- genes
(not mentioned if DEX- genes are pASD- genes)
Analysis excluding TBR1
Laminar-Specific Expression Data
•
•
•
•
To improve spatial resolution
PFC-MSC (Pre-Frontal-Cortex & Primary-Motor-Somatosensory-Cortex)
NB: cortex is grey matter and contains cell bodies
Test nine cortex-layers from 4 brains from www.brainspan.org
• Apply original coexpression networks and estimate connectivity per
layer ( = sum correlations, weighted for mean correlation in layer)
• Permute rijk over mean(rk) = null distribution of connectivity
Laminar-Specific Expression Data
Subsequent analyses of inner cortical plate (CPi)
• Why? To test if localization to CPi is specific to period 3-5.
(might change over time due to neuronal migration in early brain development)
• How?
–
–
–
–
Two mice brains (m&f)
Expression at six time points
Three zones of layers: select genes upregulated in 1 zone only
Test per zone, the zone-specific genes for enrichment in period 3-5 PFC-MSC
network (hypergeometric test)
Subsequent analyses of inner cortical plate (CPi)
•
•
NB: CPi corresponds to deep mouse layer
Thus, finding of CPi as specific layer is not driven by neurons eventually migrating
to superficial layer
Cell-Type-Specific Markers
• Five cell-types specific marker genes from independent dataset
• Enrichment for cortical glutamergic projections neurons (100,000
permutations of hcASD)
Immunostaining / In situ hybridization
• Staining hcASD genes: TBR1, POGZ, CHD8, DYRK1A, SCN2A (i.s.h.)
• TBR1 restricted to CPi (inner cortical plate)
Discussion Willsey et al
• Results suggest marked locus heterogeneity point to a much
smaller set of pathophysiological mechanisms
• Clear evidence role synaptic proteins. Indeed, the CPi neurons of
midfetal PFC-MSC are among first to form synapsis.
• Findings suggest that ASD genes converge at additional time points
and brain regions
• Small set of hcASD genes: prioritizes specificity over sensitivity
• Results important to subsequent further understanding of
pathophysiology
Parikshak et al.
• Compares ASD to intellectual disability (ID)
• Maps ASD and ID genes on coexpression networks
• ASD genes enriched in superficial cortical layers & glutaminergic
projections neurons
• Distinct patterns of ASD and ID
Journal club
Wouter
10 dec 2013