The Allen Brain Atlas and Beyond

1
High Throughput Imaging at the Allen Institute
for Brain Science
Pathology Visions 2008 October 28, 2008
Cliff Slaughterbeck, Ph.D. Engineering Manager
2
The Allen Institute for Brain Science

• Who we are
• Launched in 2003 by Microsoft co-founder and
  philanthropist Paul G. Allen with 100 million
  seed funding.
• An independent, 501 (c)(3) non-profit medical
  research organization working to advance brain
  research.
• The Institute takes on leading-edge projects,
  which result in free and publicly available
  resources that exponentially fuel innovation and
  discovery for countless other researchers
  worldwide.
• Apply a business model to scientific research,
  producing results within an agreed upon timeframe
  and within an established budget.
• Our teams are multidisciplinary in our approach
  and reach delivering tools and data free to
  researchers throughout the world.
• Located in 30,000 SF of mixed lab/office space in
  Seattle, Washington.

3
A 19th Century Viewpoint
4
The Most Recent Viewpoint
5
A 21st Century Viewpoint
6
What is a Gene Expression Map?

• A gene is a basic building block of the cell.
• A gene that is turned on typically provides
  instructions to a cell to make a specific protein
  (which in turn performs specialized functions
  within that cell).
• Each cell contains all genes (the genome) on
  chromosomes however, in any given cell, only
  select genes are turned on (expressed).
• The collection of genes that are turned on, and
  the level at which they are turned on, provide a
  unique character to the function of that cell.
• There are approximately 21,000 genes in humans

http//web.jjay.cuny.edu/acarpi/NSC/images/cell.g
if





7
Allen Institute Technology Platform
High throughput histology (up to 4,000
slides/week)
Signal (RNA-single or double, protein,
fluorescent reporters)
High throughput microscopy (brightfield,
fluorescent)
Qdot labeled
4x
Colorimetric
Tissue (brain, spinal cord, tumor, mouse, human)
Automated image processing, data storage
Web-based data viewers
Downloadable 3-D viewers
8
Image Capture System
9
Image Capture System

• The ICS system works with a Ludl 50-slide loader
• A Microscan barcode reader is integrated into the
  unit
• Imaging quality is checked before each slide is
  imaged
• Low resolution images are used to locate tissue

10
ScopeController Imaging Software
11
ScopeController Imaging Software

• ScopeController images each section independently
• Sections are bounded to reduce excessive white
  space
• Focus is performed on left and right edges of the
  tissue

12
Automation and High Throughput

• System is highly parallelized
• We had 10 ICS units working during the Mouse
  Brain Atlas
• System runs 24x7 unattended
• Operators job is to load and unload slides
• Image QC handled via LIMS
• System is designed to be quickly fixed for most
  common problems
• Email notification of any problem that causes the
  scanning to stop
• Email notification that slide processing has been
  completed

13
Informatics Data Pipeline
14
Atlas fun facts

• Imaging lab generates 1Tb of uncompressed images
  per day at full production rates.
• Compression into JP2 files reduces sizes of
  images by a factor of 16x.
• The ABA Project generated over 600 Tb of
  uncompressed data (enough to fill 20,000 iPods,
  or 300 miles of data storage magnetic tape)
• 250,000 microscope slides (10 miles if laid end
  to end)
• 85 million photo tiles (if printed as 4 x 6,
  would cover 325 acres)

11
15
Tools, Resources, Data Free and publicly
available to the worldwide scientific community
Brain-map.org

• Brain-map.org
• Indicates where approximately 20,000 genes are
  expressed in the brain
• Provides researchers with a comprehensive
  resource that reveals where each gene is
  expressed, or turned on, in the brain
• Use of the Atlas is free and open to the public
  to encourage widespread use and scientific
  collaboration
• Most widely used, publicly available databases
  ever generated for neuroscience community
• Brain Explorer
• Downloadable 3-D reference atlas and gene
  expression viewer
• Pre-loaded with high-resolution 3-D color atlas
• Pull summary gene expression information into the
  3-D environment for viewing and analysis

Qdot labeled
4x
Brain Explorer
16
Global findings

• The Atlas reveals that 80 percent of genes are
  turned on in the brain, much higher than the 60
  to 70 percent scientists previously believed
• Several thousand genes show highly regionalized
  expression, but almost none are unique to a
  single region
• By a gene expression definition, there are likely
  to be thousands of unique cell types within the
  brain, more than all the other cell types in the
  body combined

Genome-wide atlas of gene expression in the adult
mouse brain. Lein, et al., Nature 445, 168-176.
11 January 2007
17
Categorizing cell types dFISH
18
Fluorescent ICS System Changes
Invitrogen Alexa555 and Alexa488 spectra
Same basic platform as brightfield ICS unit, but
needed to integrate several other pieces of
hardware
Leica EL6000
Semrock Brightline Zero-pixel shift filter set
19
Changes to ScopeController and IDP

• Needed multi-shot image acquisition system
• Different image acquisition time for each color
• Investigated use of a different camerafound that
  image improvements werent sufficient to justify
  the cost of buying one for each system.
• Changed prescan mode to use 5x image of DAPI
• 1.25x prescan image not bright enough to find
  tissue
• DAPI stain has become our standard counterstain
  use this for finding tissue and finding focus
  height of sample
• Generate 24-bit color images
• Higher bit depth generates massively large file
  (harder to store or process in RAM)
• Standard bit depth makes for easier processing
  using 3rd party tools
• Acquire 8-bit monochrome image of each
  fluorophore
• Merge multiple fluorophore channels into separate
  RGB color channels

20
Issues for Fluorescent ICS systems

• Slow 60 minutes per slide is typical
• Difficult to find focus without over-saturating
  DAPI image channel
• Bleed-through of one fluorophore image into an
  adjacent image channel
• Alexa488 and Alexa555 excitation and emission
  bands not completely separable
• Strong signal in Alexa555 can appear as weak
  signal when imaging probes tagged with Alexa488
  and vice versa
• Can be solved using different fluorophores (e.g.
  Alexa594 or quantum dot-based probes).

• System does work, but not as well as wed like
  for a high throughput environment

21
Aperio FL
Is this the next generation solution to the
problems?
22
Ongoing Major Atlas Projects
23
Allen Brain Atlas The Human CTX Project

• Phase I Gene expression in the primary visual
  and temporal cortex (BA 17/18 and BA 21/22).
  1003 genes studied in male and female
  non-diseased subjects (n2-6 per gene).

• Phase II Variation of gene expression patterns
  in dorsolateral prefrontal cortex (DLPFC) across
  a much larger cohort including both non-diseased
  and schizophrenic subjects. 60 genes studied
  (n50)
• One and a half year project (both phases), some
  data already released on the web

24
Allen Brain AtlasMouse Spinal Cord

• Spinal cord research community and
  disease-related advocacy groups approached the
  Allen Institute in 2006 following the completion
  of the groundbreaking Allen Brain AtlasMouse
  Brain
• Comprehensive gene expression atlas that
  pinpoints where each gene is expressed in both
  juvenile and adult mouse
• Will be the worlds first mouse spinal cord atlas
  that overlays information about gene activity
  onto an anatomical map
• Provides essential baseline of normal spinal cord
  for comparison with models of injury and disease
• Holds great promise for furthering understanding
  of diseases affecting the spinal cord such as
  ALS, MS, Spinal Muscular Atrophy and spinal cord
  injuries
• One-year project

25
Allen Brain AtlasDeveloping Mouse Brain

• First atlas to provide cellular resolution gene
  expression of multiple stages of brain
  development in mouse, ranging from early brain
  formation to adulthood (several thousand genes)
• Will provide four-dimensional atlas describing
  where, what and when a gene is expressed.
• Will show gene expression in various organs
  within the embryo
• Will help further understanding of human
  development disorders such as autism,
  schizophrenia and Alzheimers disease
• Two-year project

26
New Technology Aperio ScanScopes

• Better and faster technology than the ICS systems
• More reliable
• Better image quality (more focus points)
• Capable of higher resolution (although we stick
  with 10x resolution because of file size
  issues)
• ScanScope XT 120 slide loader system.
• Estimate 5-7 minutes per slide to scan at 10x
  resolution.
• Used for high throughput imaging.
• ScanScope CS 5 slide loader system.
• Also includes 2x3 slide image tray
• Used for smaller projects.
• Biggest issue is finding our very lightly stained
  tissue
• Low background
• Requires us to use manual focus points, adding
  about 30 to effective scan time (2 hours per XT
  load)

27
LIMS2 rolled out to replace LIMS 1.0 and IDP

• Informatics data integrated into the LIMS
  database
• New database structure to support new types of
  tissue and associated metadata
• Arbitrary numbers of samples per slide
• Support for new image acquisition platform
• New GUI for entering/retrieving data
• New GUI for image QC

28
Interfacing the ICS and ScanScopes into LIMS2

• ICS Systems
• ScopeController workflow
• Image tiles to incoming tile drop location on the
  SAN
• Writes done.txt metadata files for each section
• Trigger file entries for each section in the
  trigger directory on the SAN.
• LIMS2 does the following
• Starts the Stitch module on each section
• White balance, rotation, and cropping for bright
  field images
• Stretch color contrast and merge color images for
  fluorescent images
• Sends to LIMS QC for approval

29
Web ApplicationsAll data available online at
http//www.brain-map.org
Mouse Spinal Cord Web Application
30
Next Project -- Human Gene Expression Brain Atlas

• Phase I Create Human Brain Atlas containing
  gene expression for all structures, all genes, at
  anatomic resolution.
• Microarray-based atlas anatomic resolution
• 1000 structures in the brain (n10 brains)
• 2.5 years Starts Q1/2008, finishes Q3/2010.
• Phase II Create Human Brain Atlas containing
  gene expression for most structures at cellular
  resolution.
• ISH-based atlas cellular resolution
• 50-500 genes per structure (n4-6 brains per
  structure)
• 3 years Starts Q1/2009, finishes Q4/2011.

31
Human Brain Project Workflow
(Fresh brain images courtesy Mark Vawter and
Preston Cartagena)
Subdivide slab into 2x3 blocks
Cryosection through 3 mm of 2x3 block
Final 1 mm of block
cortex
1 cm sampling
gross dissection
subcortex
0.5 cm sampling
(subset of structures)
32
Imaging Systems for the Human Atlas Project

• New image types
• MRI/DTI images
• Block-face images from serial sectioning
• 6x8 full-size histology slide images
• 2x3 chunk histology slide images
• Images from LCM systems to capture documentation
  of where laser capture sections are taken from
• None of these imaging modalities (other than LCM)
  have been used in our lab before!
• Working with vendors to supply out-of-the-box
  solutions where possible
• Most are uncommon standards at best
• Require customization for high throughput
  automated lab

33
Large Format Cryostat with Block Face Imaging

• Block face imaging system for 6x8 slides is
  complete
• Design concept based around a standard high end
  SLR camera integrated with off-the-shelf control
  system components.
• Each block face image sent directly to the
  computer. Work on LIMS2 interface is just
  starting.

34
6x8 Whole Coronal Brain Slide Imaging

• 6x8 slides will be imaged using Kodak IQ3 flatbed
  scanner
• 4.65 microns per pixel
• 2 hours to image a single 6x8 slide
• Investigating possibility of higher resolution
  imaging system from Aperio

35
2x3 large slide imaging

• 2x3 slides will be imaged using custom 2x3 high
  throughput Aperio ScanScope XT
• Similar to current XT system, but with 2x3 slide
  autoloader
• 1.00 micron per pixel (same as existing Allen
  Brain Atlas data)
• 12 minutes to image a single 2x3 slide

36
Laser Capture Microscopy and Imaging Systems

• Laser capture use restricted to capturing only
  smaller structures
• Primarily off-the-shelf solution.
• LCM software improvements
• Import image overlay from our LIMS2
• Image overlay rotation
• Data export back to LIMS2
• Can upgrade to the Leica LMD7000 when it is
  released.

Leica Microsystems LMD6000 Laser Capture
Microscope
37
Summary

• High Throughput Imaging of Histology slides
• Multiple types of imaging system
• Brightfield and fluorescent image data
• Different size slide formats (1x3, 2x3,
  6x8)
• Block-face image system
• Laser capture microscopy
• Multi-project Data Processing Pipeline
• Extensive use of pre-barcoded slides and samples
  to identify tissue
• Track metadata in LIMS system
• Use LIMS to customize image processing pipeline
  for each project

38
Acknowledgements

• Microscopy Production Team
• Melissa Reding
• Cathy Copeland
• Shiella Caldejon
• Sherry Palubinskas
• Kaylynn Aiona
• Engineering Team
• Suvro Datta
• Viki Phillips
• Derric Williams
• Atlas Operations Management Team
• Paul Wohnoutka
• Amanda Ebbert
• Lon Luong
• Kim Smith
• Andreas Jeromin
• Geri Orta
• Dave Muzia
• Nick Stewart