Transcript Where are Our Computational Bottlenecks?

HIPCAT
Meeting
January 27, 2006
Stephen E. Harris
Where are Our Computational
Bottlenecks?
• Large collections of images such that layers
of resolution are maintained ie. like a
satellite image that can see a grass-blade in
someones backyard.
• 3D imaging of biological processes with high
resolution and animation
• Connecting, utilizing, displaying large gene
expression datasets with all known
information .
• Use of Natural Language Processing (NLP)
technology.
Natural Language Processing
• Our biological data is organized and
abstracted in Medline and Pubmed
• NLP technology can be used to aid in the
search of these large databases for contextdependent hits and links that have meaning
in terms of a biological pathway
• Example: x binds y resulting in…c binding,
z stimulates v.., a inhibits c, only when d is
present, y is in the cytoplasm and moves to
the plasma membrane after…., resulting in…
Introduction to Our Biological Problem
and Where we can Use HIPCAT
• Mechanical loading of bone and Finite Element
Analysis models—associate with select gene
expression
• Osteocytes biology-mechanosenors in bone
• Imaging osteocytes at work in health and disease.
• Pathways and gene networks unique to osteocytes
and the mechanical loading.
• Connect “List of genes” to large databases, such as
Medline/Pubmed
• Derive Virtual Pathways that can lead to a deeper
more systems biology approach to understanding a
given biological system
Bone is Formed Where the
Biomechanical Demands are Greatest
Robling, 2002
•Osteocytes make
up over 90% of all
bone cells
•Osteocytes express
long dendritic
processes
•These cells are
viable for decades in
the bone matrix.
Mechanosensory Cell for Bone
from the “Primer on Metabolic Bone Disease
and Disorders of Metabolism” editor Murray J.
Favus
Fluid Flow Through the Osteocyte Lacunae-Canalicular System-Procian
Red Injection Into the Tail Vein of a Mouse
Mouse Ulnae Loading Model
Courtesy of Alex Robling
(Adapted from Torrance et al., 1994)
Pathway Assist
http://www.ariadnegenomics.com
• Organize complex list of gene expression
patterns and link to Medline/PubMed
Databases
• NLP technology in MedScan, a ResNet
database --includes comprehensive
database of molecular networks—ie 500
pathways and over 1 million biological
interactions
• Construct candidate interaction pathways,
the data is directly linked to Medline and
PubMed.
• Needs improvements and new ideas.
DMP1-MEPE-SPP1-CDC42 Mechanical Loading
Responsive Gene Network
PathwayAssist
http://129.111.78.243/HarrisLab/HarrisLab_home.htm
Mouse Ulna Regions Analyzed for Gene Expression of
DMP1 and MEPE mRNA
MEPE Expression at 3mm Distal to Mid-shaft
24hr after Loading at 2.4N at 60 cycles 2 Hz.
Lateral
Medial
U
Control-Left
Top – In situ, darkfield Bottom-lightfield, HE,
U
Loaded-Right
U =ulnae
Fold Change Relative to
Control
Quantitation of MEPE mRNA in Osteocytes after a 30 sec Load of 2.4N at 2Hz
In the Mouse Ulnae(N=3)
*
2.5
Loaded
Control
2.0
*
*
1.5
*
1.0
0.5
*
*
P < 0.05
R2 = 0.63
d1
d2
0.0
0
p2
p1
MS
d3
Position along Ulnae
d4
8
Strain Gradient
Estimates Along the
Diaphysis of the
Axially Loaded Mouse
Ulna
MEPE mRNA
Control
Loaded
P = 0.038 slope
1.0
GEt = 1350 +/- 350 uE
0.5
40
39
33
59
26
76
20
00
-0.5
13
08
0.0
44
4
Gene Expression
Change
GEtx-GEctr
MEPE Gene Expression Threshold (GEt) and Relative Gene Expression
Change (rGE = GEtx-GEctr) At 24 hr After 30 sec 2.4 N, 2Hz Load of Mouse
Ulnae 1.5
Microstrain uE
The Gene Expression Threshold(GEt) is similar to the Estimated Bone
Formation Threshold in the Mouse and Rat Models
Preliminary Finite Element Model of the Ulnae of Mouse- 4month C57BL/6
A mCT image consisting of 1105
sections at 13 micrometer spacing of
the C57BL/6 female ulnae (4months).
A coarse 2832 element model was
then constructed and analyzed using
LS-DYNA. Proximal and distal
structures of the ulnae have been
removed in the model and idealized
boundary conditions imposed. (a)
the course finite element mesh
superimposed on the CT image, (b)
the shaded finite element model with
idealized boundary conditions, and (c)
representative equivalent strain
contours for a 2.4N idealized static
loading.
Need more work. 1st Reiteration.
How can we study the osteocyte
home and gene expression patterns?
Pathways and Gene Networks in
Osteocytes
WT
Canaliculi
Osteocyte
Lacunae
8KB DMP1 Cis-Regulatory Region Plus Intron 1-GFPtopaz and
Conserved Non-Coding Sequences/ Mouse Human Comparison
A.
E1
8kb Region
E2 GFPtopaz
Intron 1
-7892bp
+4439bp
A
Exon1
B.
-10kb
-7.3kb
-4.6kb
-2.0kb
Exon2 & 3
+3.4kb
+6.1kb
+1.0kb
Exon6
+8.7kb
+10.4kb
Use of the DMP1 cis-regulatory region to target GFP to osteocytes. A. Contruct with the 8kb plus
Intron 1 region of DMP1 ligated to GFPtopaz. Used to make stable osteoblast cells that differentiate into
B
osteocytes and transgenic mice models. B. Conserved nucleotide sequences(CNS) between mouse and
DMP1 cis-REGULATORY REGION - GFPtopaz CONSTRUCTS
human DMP1 genes. 8kb plus Intron 1 contains a large portion of the CNS
10kb REGION
E1 INTRON1
GFPtopaz
8KB Flanking Plus Intron 1 DMP1 Direct Expression to Ostecytes
A
B
C
D
8kb Region
-7892bp
E1
Intron 1
DMP1 Gene
E2
+4439bp
GFPtopaz
Fuorescent activated cell sorting was used to purify Primary osteocytes from Calvariae
A. DMP1 mRNA expression in Unsorted, -GFP and +GFP Cell Fractions.
B. GFP expression in osteocytes of calvarial bone, driven by the 8kb Plus Intron 1 DMP1-GFP
construct.
Gene Expression Studies
•
•
•
•
500 ng of total RNA was 2x amplified.
Affymetrix 430A mouse GeneChips
GC-RMA was used for Normalization
With N=3, LIMMA in Bioconductor was
used to determine significant genes at
a max False Discovery Rate = 5%
• 723 Genes between –GFP cells and
+GFP cells were analyzed, setting –GFP
= 1.0
Cluster 10
Gene Expressed 2-10 times higher in +gfp
Primary Osteocytes
Muscle
Secreted Differentiation
Lipid
Transcription
Summary
• Need new HPC and Imaging tools for
analysis of biological functions in vivo.
• Better tools for connecting complex
dataset from microarray analysis to
other databases, such as Medline,
Pubmed, Protein interaction databases,
and Pathway networks.
Acknowledgements
UTHSCSA
UCONN
Marie Harris
Ivo Kalajzic
David Rowe
Wuchen Yang
Jelica Gluhak-Heinrich
UMKC
Jian Q. Feng
Indiana University Medical School
Charles H. Turner
Alex Robling