Functional Genomics Journal Club presentation (02/19/19)

1. FunGen Journal Club
February 19, 2019
By Katia Lopes

2. Affiliation
1Department of Statistics, Department of Medical Genetics, University of British
Columbia, Vancouver, BC, Canada.
2Canadian Institute for Advanced Research, Toronto, ON, Canada.
3Rush Alzheimer’s Disease Center, Rush University Medical Center, Chicago, IL, USA.
4Department of Neurology, Brigham and Women’s Hospital, Boston, MA, USA.
5Broad Institute, Cambridge, MA, USA.
6Center for Translational & Computational Neuroimmunology, Department of Neurology,
Columbia University Medical Center, New York, NY, USA.
7University of Sydney, Sydney, NSW, Australia.
8Harvard Medical School, Boston, MA, USA.
9Harvard T.H. Chan School of Public Health, Boston, MA, USA.
2

3. • The incidence of late-onset AD is expected to triple in the
US by 2050, yet no therapies are available to treat or
prevent the disease.
• Recent genome-wide association studies (GWAS) have
identified new potential therapeutic targets involved in
endocytosis, metabolism and inflammation.
• However, possible reasons for the continued failure of AD
trials include the biological complexity of the disease and
its phenotypic heterogeneity.
3
Introduction

4. • The authors hypothesized that RNA-seq data from the Dorsal
Lateral Prefrontal Cortex (DLPFC) would enable to identify
coherent intermediate cellular mechanisms associated with
cognitive decline and/or neuropathological changes.
• They created an approach, called gene Module Trait Network
analysis (MTN), constructs gene expression modules and
identifies those that are directly associated with cognitive
decline.
4
Introduction
• Use of the network in prioritizing amyloid and
cognition-associated genes for in vitro
validation in human neurons and astrocytes.

5. 5
Fig. 1 | Schematic of the implementation of
the module–trait network (MTN) method to
prioritize modules and genes directly related
to AD-related traits in our study.
a - Inputs to the MTN method are
(i) AD pathological traits of amyloid and tau
measurements;
(ii) slope of cognitive decline before death;
(iii) average expression of coexpressed gene
sets (modules).
b - These three inputs are combined using
conditional independence relationships (via
Bayesian networks) to identify direct
relationships among coexpression modules,
AD traits and cognitive decline (cog).
c - The disease relevance of top predicted
genes is tested experimentally in an astrocyte
and iPSC-induced neuron in vitro system.
Schema

6. • RNA was sequenced from the gray matter of DLPFC of 542 samples
• Sequencer: Illumina HiSeq
• QC with Parallelized pipeline
• Trimming of the reads
• Reads alignment: Bowtie
• Estimate expression level: RSEM
• Remove outliers samples
• Excluded 30 samples with incomplete assessment (for trait analysis)
• Data normalization:
• Quantile normalization and Combat algorithm
• Keep only genes >= 4 reads in 100 individuals (13,484 g)
• Linear regression to remove batch effects
They chose to only account for know covariates and not any hidden
covariates
6
Methods

7. • Networks:
• SpeakEasy and WGCNA
• 257 modules | 47 of them > 20 genes (98% total)
• Module Enrichment:
• All DLPFC expressed genes
• 29 of the 47 modules were significantly enriched for at least one GO
category
• Replication of gene modules:
• Module preservation (Z-summary)
• 4 datasets as validation data
(I) Microarray Zhang et. al
(II) Microarray mouse Matarin et. al.
(III) Test dataset from ROSMAP
(IV) H3K9ac
7
Methods

8. • Bayesian Network used to estimated the Directed
Acyclic Graph (DAG):
• Included 11 modules associated with at least one of the
3 traits (β-amyloid, tau tangles and cognitive decline)
• Verified cell-type-specific genes
• Chose 4 cell types - neurons (m187), astrocytes (m107),
microglia (m116) and oligodendrocytes (m123)
• Experimental validation of target genes
• iPSCs cell culture
• Perturbation with lentivirus
• qPCR
• Immunocytochemistry and microscopy
• ELISA
8
Methods

9. ROSMAP dataset:
• 478 participants, with a mean age at death of 88.7 years
• At the time of death 32% remained cognitively unimpaired
• 27% had mild cognitive impairment
• 39% had a diagnosis of AD dementia
• 2% had another form of dementia.
Pathological AD (58.6% n = 280) | Clinical AD (38.7% n = 185)
9
Results
5 phenotypic traits related to AD
Clinical measures:
• Clinical diagnosis of AD dementia proximate to death;
• Continuous measure of cognitive decline over time;
Pathology variables:
• Continuous measures of PHFtau tangle density and β-amyloid burden;
• Binary diagnosis of pathologic AD.

10. • 478 individuals;
• Average of 95 million paired-end reads for each subject;
• Normalized to account for the effects of many known biological and
technical confounding factors;
• Genes with low expression were removed, resulting in 13,484 unique genes
10
RNA-seq Analysis
TWAS study:
They identified the proportion of
genes whose expression is associated
with each pair of AD traits
3,025 genes at FDR<0.05 (Cognitive
decline)
π1 statistic, estimated that 55–90% of
the associated genes are shared
among correlated AD-related traits
1
1
1
1
1
0.55
0.95

11. Fig. 2 | Characterization of human cortical RNA-seq data and their relation with AD traits and
a) Genes associated with AD-traits
b) Module enrichment for cell-
specific signature.
c) Association strength: 47 modules & traits.
Bonferroni (p < 0.001)
d) Strength and direction of each module’s
associated with AD diagnostic.
neurons (m187)
microglia (m116)
oligodendrocytes (m123)
astrocytes (m107)

12. MTN consists of 3 steps:
12
RNA-seq Analysis
1 – Identified modules of genes and tried to validated with other
datasets:
• SpeakEasy and WGCNA algorithms.
2 – They determined which modules have direct relationships with
cognitive decline and other AD traits using Bayesian networks
5 perspectives: (i) Functional enrichment analysis, (ii) module preservation
with an independent cohort, (iii) concordance with co-regulation in
epigenomic data, (iv) concordance with brain gene expression data from
multiple AD mouse models, and (v) cell-type-specific expression.
3 – Prioritize genes for validation in vitro model systems
Criteria: Gene network connectivity, sufficient expression levels, gene-level
association with AD phenotypes and gene function. Total = 21 genes.

13. 13Sup Fig 4. Module preservation. X = z-summary | Y = modules
Strongly preserved
Kruskal Wallis test(10)
Moderately preserved
Kruskal Wallis test(2)
𝑍𝑠𝑢𝑚𝑚𝑎𝑟𝑦 =
𝑍𝑑𝑒𝑛𝑠𝑖𝑡𝑦 + 𝑍𝑐𝑜𝑛𝑛𝑒𝑐𝑡𝑖𝑣𝑖𝑡𝑦
2
Preservation in 45 of the 47
modules in the Zhang study
All 47 modules preserved in
the separately-processed
ROSMAP dataset
Results

14. 14
Fig. 3 | The AD network model prioritizes
m109 as being directly associated with
cognitive decline. a) A directed acyclic
graph, obtained using Bayesian network.
Modules
Cell type
AD traits
Bayesian network
Bayesian network consisted of
18 nodes: 11 nodes
representing trait-associated
modules, 3 trait nodes, and 4
‘cell type modules’.
Module 109 (m109) was the
module most strongly
associated with cognitive
decline. It consists of 390
genes with diverse functions.

15. b) Trajectories of cognitive decline for people with low (left) or high (right) levels of m109
expression.
c) Mean expression of m109 for individuals who have no cognitive impairment (NCI; red), mild
cognitive impairment (MCI; green) or an AD diagnosis (AD; blue). d, Expression of m109 for
individuals without (red) and with (turquoise) amyloid deposition at autopsy.
Prioritizing genes in module 109 and testing their effect on extra-cellular -
amyloid levels.

16. 16
Fig. 4 | identifying specific genes within m109 for experimental follow-up. a) The estimated
gene regulatory network (Bayesian network) for 112 selected genes in m109. b) Coexpression
values for the 112 genes shown in Fig. 3a, highlighting the substructure within the
coexpression pattern of m109.
Yellow: tested in astrocytes
and iPSC derived-neurons
Blue: Tested only in astrocytes
Orange: Tested only in iPSC
neurons
Genes that are tested in wet lab

17. 17
Fig. 5 a) In iPSC-derived neurons (black dots) the outcome measure was not
altered. For astrocytes, 2 shRNA constructs targeting different genes: INPPL1 and
PLXNB1.
b) Replication study: results of INPPL1 and PLXNB1 knockdown on Aβ42 secretion
were measured in additional experiments using multiple shRNA constructs targeting
each of these genes. In these experiments, knockdown of both genes led to reduced
Based on those possible targets found using the network they knockdown 12 genes in
neurons, 14 genes in astrocytes and 11 genes on both using short hairpin RNAs. Then, they
measured the effect of each shRNA on the protein Aβ42 levels.
Bonferroni (p < 0.0012)

18. 18
c) They immunostained frontal cortex from subjects with pathologic AD and showed that
both INPPL1 and PLXNB1 were expressed at the protein level in astrocytes confirming that
these two genes were expressed in vivo in the human cell type used in the validation
experiments.
d) The proportion of variance in cognitive decline that is explained by different factors.
As shown, PLXNB1 and INPPL1 capture much but not all of the effect of m109, and
Results
Proportion of variance in cognitive decline
Green = gene marker
Red= astrocyte marker
5.5%
4.4%
5.4%

19. The authors used a network-based approach, to identify biological
processes and specific genes associated with multiple AD traits;
The central finding of this project is the existence of a robust set of
coexpressed genes, supported by other datasets;
Module m109 was associated with β-amyloid pathology;
Genes INPPL1 and PLXNB1 are intriguing candidates connected to
amyloid biology in vitro, however...
… they do not appear to account for the effects of the entire module, suggesting
that further validation work will be needed to identify additional driver genes for
m109.
In summary… they’ve illustrated the use of network in prioritizing
module and a subset of genes.
19
Conclusion