Automated creation and SEM imaging of Ultrathin Section Libraries

1
Automated creation and SEM imaging of Ultrathin
Section Libraries
Kenneth J. Hayworth
This work was performed with Narayanan (Bobby)
Kasthuri at Harvard University in the laboratory
of Jeff Lichtman.
2

• To understand the algorithms underlying the
  brains functioning we must address neural
  circuitry at the level of point-to-point
  connectivity - not merely the statistics of
  connectivity
• Neuronal processes, some only 40nm in diameter,
  are tortuously interwoven in the brains neuropil
  and only electron microscopy can trace them
  successfully

3

• In order to really address functional questions,
  these electron microscopic reconstruction
  techniques must be scaled-up to handle much
  larger volumes.
• How large? Large enough to encompass the entire
  circuit of interestI would argue on the order of
  1-10mm3 and greater.
• In an attempt to trace circuits over larger
  volumes, three automated techniques have been
  invented in recent years all based on electron
  backscatter imaging in the Scanning Electron
  Microscope (Denk and Horstmann 2004)
• This technique of electron backscatter imaging
  has really made automation possible since it does
  away with the need for handling thin sections on
  fragile TEM slot grids...

4
TEM Imaging
In transmission electron microscopy sections need
to be held on a gossamer thin substrate to allow
electrons to pass through the sample.
From www.microscopy.ethz.ch
5
Electron Backscatter Imaging

• In electron backscatter imaging, electrons
  bounce off heavy staining atoms (osmium, lead,
  uranium) and so there is no need for the gossamer
  thin substrate.

6
Electron Backscatter Imaging
7
Electron Backscatter Imaging
Electron Backscatter Imaging
Backscatter detector (Solid state diode or
scintillator)
8
Electron Backscatter Imaging
Electron backscatter imaging can also be
performed on thin sections collected on solid
substrates (glass, Kapton tape, etc.)
9
New Techniques for Automatic Imaging of Neural
Tissue
10
Blockface Imaging vs. Collecting Sections for
Later Imaging

• Advantages of blockface imaging
• Intrinsically more reliable since no fragile
  sections need be handled
• Essentially perfect registration for free
• Less sensitivity to embedding parameters
  (especially FIBSEM technique)
• Main disadvantage
• Each section is destroyed after imaging
  requiring an immediate and final decision on what
  is to be imaged in the section and at what
  resolution.
• If too much is imaged, one ends up wasting an
  enormous amount of time imaging processes that
  will never be reconstructed or analyzed later.
• If too little is imaged, tracing of even single
  neurons becomes impossible

11
Blockface Imaging vs. Collecting Sections for
Later Imaging

• Advantages of collecting sections for later
  imaging
• Allows use of random-access directed imaging
• Image only what you need to get the job done.
• Multiscale imaging
• First pass at low resolution to get lay of the
  land.
• Allows post staining of sections with heavy
  metals (better contrast, faster imaging rates)
• Better resolution (higher kV beam without blur
  from deeper parts of block)
• Faster, parallel imaging
• Multimodal imaging possible on same sections
• Fluorescence imaging (GFP etc.)
• Antibody staining (for LM and immunogold electron
  imaging)
• Electron tomographic imaging to obtain better
  depth resolution than section thickness
• SIMS
• Main disadvantage
• Reliability, Reliability, Reliability If
  sections are lost or destroyed during the
  collection process it can render many small
  processes that cross the lost section from being
  successfully traced.

We think that the advantage of directed imaging
warrants research into the development of highly
reliable automated sectioning and collecting
machines, and here is why
12
Comparing Sectioning Time vs. Imaging Time

• The ATLUM can currently collect a 30nm thick
  1.5mm wide and 5mm long section every 2.5 minutes
• At this rate a 1mm3 volume of tissue could be
  reduced to 30nm sections in one week (7.7 days).
• At 5nm pixel resolution (and 30nm thick
  sections), a 1mm3 volume contains 1.3x1015
  voxels.
• Imaging such a 1mm3 volume at 1Mhz (which is 10x
  faster than current blockface imaging rates)
  would require 42 years!
• So what took one week to section takes at least
  42 years to image in total.
• Thus sectioning can be 2000x 20,000x faster
  than imaging

13
Typically conceived volume imaging and analysis
pipeline
14
An alternative pipeline using directed imaging on
Ultrathin Section Libraries
15
Multi-scale directed imaging using an Ultrathin
Section Library
Using an Ultrathin Section Library, the
high-resolution imaging, segmentation, tracing,
and analysis steps can all proceed at a
reasonable pace since they are directed to only a
few neurons out of the millions contained in the
entire volume.
16
Directed imaging using a library of ultrathin
sections
17
Directed imaging using a library of ultrathin
sections
18
Harvard prototypes for the automated collection
of ultrathin sections
19
Overview of ATLUM Process
20
ATLUM process overview
Animal perfused
21
ATLUM process overview
22
ATLUM process overview
Wafer loaded into Scanning Electron Microscope
and imaged using the electron backscatter signal
23
ATLUM process overview
A set of these wafers constitutes an Ultrathin
Section Library. This allows any region within a
volume potentially encompassing many cubic
millimeters to be imaged at any desired
resolution down to 5nm. Allowing directed imaging
to be performed to efficiently trace the neural
circuits of interest.
24
How does the ATLUM work?
25
Automatic Tape-collecting Lathe Ultramicrotome
(ATLUM)
26
(No Transcript)
27
How does the ATLUM work?
28
ATLUM main subsystems
29
ATLUM main subsystems
Lathe-cutting mechanism
30
ATLUM main subsystems
Tissue tape collection mechanism
31
Lathe Cutting Mechanism
32
Lathe Cutting Mechanism
Air bearing Rotary Stage
33
Lathe Cutting Mechanism
Collet chuck holding tissue embedded on steel axle
34
Lathe Cutting Mechanism
Piezo-driven diamond knife stage
35
Piezo-driven Knife Stage
36
Piezo-driven Knife Stage
Rotating steel axle
37
Piezo-driven Knife Stage
Piezo tilt-stage
38
Piezo-driven Knife Stage
Bracket to hold diamond knife and capacitive
sensors
39
Piezo-driven Knife Stage
Diamond knife with attached water boat
40
Piezo-driven Knife Stage
Capacitive sensors (measures distance from knife
stage to surface of steel axle with 5-10nm
resolution)
41
Piezo-driven Knife Stage
Ultrathin tissue sections collected from knifes
water boat on submerged conveyor belt
42
Basic Operation
Air bearing stage rotates axle (slowly when
cutting wedge)
43
Basic Operation
44
(No Transcript)
45
Basic Operation
46
Movie of Lathe Sectioning and Collection Process
Movie of 100 sections collected Fast-forward speed
47
Details of the tape collection drive mechanism
Final tissue-tape collection reel
Note Display actual tissue tape from run
48
Summary of sectioning results

• Our longest ATLUM runs to date have been
• 1000 sections at 39nm thickness (1.5mm x 5.0mm
  blockface sectioned at 0.05mm/s)
• 1100 sections at 29nm thickness (1.5mm x 5.0mm
  blockface sectioned at 0.05mm/s)
• These runs took 40 hours each during which the
  ATLUM ran with no user intervention. Video
  recording of all 40 hours of section collection
  was used to verify that no section was lost
  during these runs.
• Water level in the knife boat was maintained by a
  syringe pump controlled by a video feedback loop
  (keeping the waters meniscus reflection boundary
  at a constant position in the video).

49
Imaging Results
50
Light microscope images Stained with Toluidine
Blue
51
(No Transcript)
52
SEM images Osmium fixed mouse cortex stained
with Uranyl Acetate and Lead Citrate
53
SEM
JEOL JSM-7001F Scanning Electron Microscope
54
How ATLUM-collected sections are imaged
55
50-60nm thick mouse cortex
56
(No Transcript)
57
(No Transcript)
58
(No Transcript)
59
50nm Stack (movie)
60
(No Transcript)
61
(No Transcript)
62
38nm thick mouse cortex
63
(No Transcript)
64
(No Transcript)
65
(No Transcript)
66
(No Transcript)
67
(No Transcript)
68
(No Transcript)
69
(No Transcript)
70
(No Transcript)
71
(No Transcript)
72
(No Transcript)
73
35nm stack
Play ImageJ stack
74
Section thickness

• Section thickness is a crucial parameter for
  ultrathin section libraries.
• If too thick it is impossible to trace the finer
  processes
• If too thin then reliability of ATLUM sectioning
  and collection becomes a concern.

75
Section thickness
20nm
60nm
76
25nm thickness sections
77
(No Transcript)
78
(No Transcript)
79
(No Transcript)
80
(No Transcript)
81
(No Transcript)
82
Tilt images allow tracing of thicker sections

• We have also taken high resolution stereo-pairs
  (SEM backscatter images from two different
  angles). When viewed with stereo glasses these
  provide a sense of depth even in sections 50nm
  thick.

• Within these stereo pairs we can see the relative
  heights of synaptic vesicles, raising the
  possibility that a full tilt series could be used
  to virtually slice each section perhaps down to
  10nm Z-resolution.

83
Good for blue
60nm Section 45o tilt
84
Good for red
60nm Section -45o tilt
85
Summary

• Random access directed imaging on Ultrathin
  Section Libraries offers a time-efficient
  approach to tracing neural circuits over large
  volumes.
• The ATLUM prototype has demonstrated the ability
  to reliably and automatically collect gt1000
  ultrathin sections creating such Ultrathin
  Section Libraries

86
Acknowledgments
This research is funded by a grant from the
McKnight Endowment Fund for Neuroscience and
continuing support from the Center for Brain
Science, Harvard University and the Gatsby
Charitable Foundation.

• USC
• Irving Biederman

• Harvard
• Narayanan (Bobby) Kasthuri
• Richard Schalek
• Juan Carlos Tapia
• Erika Hartwieg
• Jeff Lichtman

87
END
88
Miscellaneous Imaging Results

• The ATLUMs ability to collect sections for later
  imaging raises the possibility of chemical
  analysis overlaid on top of structural EM images.

89

• This is a LR-white embedded muscle (sectioned on
  a regular ultramicrotome) from a transgenic
  mouse.
• The mouse expresses florescent protein (blue) in
  its motor axons.
• The muscle tissue has been stained (red) with
  phalloidin.

90

• This same section was then stained with Uranyl
  Acetate and Lead Citrate and imaged in the SEM.

91
Such overlays of chemical and EM structural
images should be possible on ATLUM collected
tissue as well.
92
Imaging rate

• Imaging speed in the SEM is fundamentally limited
  by the amount of current one can put into the
  electron beam while maintaining a small beam
  diameter (for high-resolution).
• This means low-resolution images can be acquired
  at high data rates, but high-resolution images
  (5-10nm pixel size) are currently limited to data
  rates around 1MHz (for well stained tissue).

Pixel size 18nm
Image size 3k x 2k
Image time 3s
Data rate 2 MHz
93
Directing Imaging to follow a single process