Whole Brain Mapping with X-rays
Yeukuang Hwu 胡宇光
Institute of Physics, Academia Sinica
台大物理 27/10/2020
Collaborators: A true international effort!
• Taiwan
• Keng S. Liang, Ting-Kuo Lee, Maw-Kun Wu, Shih-Chang Lee, En-Te Hwu, Ming-Li Chu, Chih-Hsiun Lin (Academia Sinica): phase retrieval, reconstruction, magnetic nanoparticles, microfluidity
• Ann-Shyn Chiang, Chia-Wei Li (Life Science, Tsing Hua U.): cell biology, shell fish, fossil, fire fly
• Hong-Ming Lin (Tatung U): electron chemistry, battery
• Yu-Tai Ching (NCTU), David Chien (UCS-St. Marcos): Image Processing, reconstruction
• Chung-Shi Yang, Kelvin K. C. Tsai (NHRI): nanomedicine, microangiogenesis
• Ann Chen, Maria Ka, Dueng-Yuan Hong (TSGH): CKD vessel imaging
• Y. F. Hu, S. K. Tsai (TTY Pharmaceutics): (drug delivery, pharmacokinetics)
• Y. C. Yang, Hong-I Yeh, Yu-Jen Chen (Mackay Memorial Hospital): (artery disease, tumor development)
• Japan: T. Ishikawa, Y. Kohmura, K. Sawada, Y. Joti (RIKEN/SPring-8 Center)
• Korea: Jung-Ho Je (POSTECH), Doyoung Noh (GIST), Jun Lim, Jae-Hong Lim (PAL)
• China: Jun Hu, Chunhai Fan, Lihua Wang, Xiaoqing Cai, Chichao Zhang, Ying Zhu
• Singapore: Eng Soon Tok, Alvin Teo, Chian-Ming Low, Gan-Moog Chow, H. O. Moser (NUS)—drug delivery, polymer blend, electrochemistry, neurobiology.
• France: Cyril Petibois, Sophie Javerzat, Michele Moenner (U. Bordeaux, brain tumor angiogenesis), Patrick Soukiassian
• USA: Yong S. Chu, Wah-Keat Lee, Qun Shen, B. Lai, (APS), Wen-An Chiou (U. Maryland), John Boeckl (US Air Force Lab)–TXM
• Switzerland:
• G. Margaritondo (EPFL)
• Rolf Gruetter (EPFL)
NanoX Team (Current Members)
• X-ray microscopy:
• Hsiang-Hsin Chen (陳翔欣), Shun-Min Yang (楊舜閔), Tsung-Tse Lee (李宗 澤), Cheng-Huan Hsu (徐晟桓), Ching-Yu Chiou(邱鏡宇)
• X-ray Nanosynthesis:
• Ming-Tsang Lee (李旻倉)
• Nanofabrication of X-ray optics:
• Mei-Chun Chen (陳玫君), Yu-Ting Jian (簡郁庭)
• Bioimaging:
• Shun-Min Yang (楊舜閔), Hsiang-Hsin Chen (陳翔欣), Yi-Yun Chen (陳怡云), Chia-Ru Chang (張家如), Ya-Sian Wang (王雅嫻), Shiou-Jin Chiou (邱繡謹), Cheng Jyun Yang (楊程鈞)
•
Visitors: Cyril Petibois, Pei-Feng Chen, Keng S. Liang, Benoit RecurInnovation in physics and instrumentation has opened new eras of biology and medicine
• Microscopy
• Spectroscopy
• X-rays scattering
• Mass spectrometry
• Nuclear Medicine
• NMR, SPECT, PET, CT, ultrasound, OCT
• …
And, More Recently..
•
Laser confocal scanning microscopy + fluorescent protein•
Protein crystallography•
Cryo-electron microscopyThe breakthroughs in x-ray imaging
•
Phase contrast•
Nano-resolution•
Elemental contrast•
High speed for 3D imagingHistorical increase of the x-ray source brightness (in standard units)
Taiwan Photon Source 3rd Gen. SR
sources
It took decades to obtain the first x-ray free
electron lasers, but the results are fantastic:
The second breakthrough: nanoscale
resolution
Transmission x-ray microscopy with
Fresnel zone plate optics
20/450 nm & 40/950 nm Au zone
plates
Star Pattern Image
Finest line width: 20nm Au thickness: 300nm Pixel size: 2.51nm FOV: 5.14um No binning 8 sec exposure 60 frames averaged 4 pieces of diffuser
Zone Plate applied (ZP m2-6):
8.000keV
20nm outermost zone width 50um diameter
400nm thickness
SEM
Enlarge
Finest line width = 25nm 20 nm 15 nm
Si/WSi2
Si/W multilayer test patterns
H. R. Wu, et al. Nanoresolution Radiology of Neurons, J. Phys. D 45, 242001 (2012).
HeLa cells with AuNPs on culture dish
Element mapping with nanoresolution
J Electrochem Soc 157, B783 (2010) Figure 2
• Ni–YSZ (yittria-
stabilized zirconia) anode
• Ni K-edge: 8.333 keV
• Spatial resolution: 38.5 nm
• Box size: 5 µm
16 eV below Ni K-edge 24 eV above Ni K-edge
Historical increase of the x-ray source
brightness (in standard units)
Taiwan Photon Source 3rd Gen. SR
sources
Phase contrast: the first breakthrough
The XFEL: Toward the ultimate x-ray
source
“Detect and Destroy”
In collaboration with Prof. Yoshinori Nishino of Hokkaido University and SACLA
Blank liposome ((NH
4)
2SO
4inside)
4
Doxorubicin loaded liposome
Coherent diffraction imaging (XCDI)
A single 40 nm AuNP Two 10 nm AuNP
A single 200 nm liposome
Tumor Induced Angiogenesis
Cyril Petibois et al. (U. Bordeaux), DY Hong et al. (TSGH),
Large tissue, high resolution and speed
Whole-brain blood vessels in mouse
Res. < 0.5 mm Size > 400 mm3 Time < 10 min
X-ray imaging of angiogenic
microvasculature of glioblastoma
Solid glioma tumors
Diffused glioma tumors
TXM Images of a microvessel
What do we expect to learn?
• Small differences in microvasculature between different models
• Effect of treatment
• Inhomogeneity in microvasculature with respect to tumor
• Microvasculature with respect to the growth and metastasis of tumor
• Drug development and usage
There are more problems in brains with neurons (dementia) than vessels (cancer)
• Can X-rays image neurons?
• Can X-rays image blood vessels at the same time?
• Can X-rays study neuro-vascular interaction?
• On dead specimens, of course.
WHY BRAIN?
Mapping the Brain:
an Historical Mission for Science and Technology
• A fundamental research objective
• An effective way to understand and cure brain diseases
• Potential for a broad social impact
• Important technological byproducts:
- Advanced imaging technologies - New computational strategies
- Artificial intelligence
THE CHALLENGES:
…we must do better!
An effort to image the neural network
of a whole human brain with proven
performance of synchrotron x-ray
tomography and a collective effort
SYNAPSE: an international
partnership with a milestone research mission
First objective:
mapping the neuron
connections of an
entire human brain
by 2023
Mapping the Brain:
an Historical Mission for Science and Technology
• A fundamental research objective
• An effective way to understand and cure brain diseases
• Potential for a broad social impact
• Important technological byproducts:
- Advanced imaging technologies
- New computational strategies
- Artificial intelligence
IT TAKES DECADES FOR
MAPPING ONE HUMAN BRAIN WITH THE PRESENT X-RAY
TOMOGRAPHY PERFORMANCES!
•
Parallel image taking at multiple facilities•
Automated specimen sectioning•
Automated specimen mounting and alignment•
High-speed data transfer and sharing for all partners•
New, efficient strategies for reconstructionSYNAPSE is the solution
SYNAPSE:
core partners with top-level
synchrotron facilities
NUS/SSLS
AS
RIKEN/Spring8 SARI/SSRF POSTECH/PAL
In addition to x-ray microtomography facilities, the core partners bring with them a very powerful
extended coalition
CONCENTRATE ON 3D IMAGING:
New radiology techniques combined with other advanced microscopies
• For overall connection mapping: phase-contrast microtomography (0.3 µm resolution in all 3D directions)
• To explore in detail synapses and neuron connections: nanotomography with <10 nm resolution
• To obtain detailed 3D maps of special regions: electron microscopy/continuous sectioning, optical super-resolution microscopy, with nm resolution
• To analyze the whole drosophila brain and large parts of the mouse brain
obtaining functional information: Confocal (or Light Sheet) + FocusClear imaging
• For single-molecule mapping of drosophila and mouse brains: Super-resolution microscopy with FocusClear
• high resolution functional imaging: Raman, IR spectromicroscopy
• Fusing-in low resolution functional imaging, i.e. fMRI
• …plus other tasks using other frontline techniques
Pohang Light Source II 7C
SPring-8 BL32B2
Taiwan Photon Source BL2A
Singapore Synchrotron Light Source PCXT
Shanghai Synchrotron Radiation Facility BL 09
SYNAPSE:
core partners with top-level
synchrotron facilities
NUS/SSLS
AS
RIKEN/Spring8 SARI/SSRF POSTECH/PAL
In addition to x-ray microtomography facilities, the core partners bring with them a very powerful
extended coalition
Concentrate on 3D imaging: new radiology techniques combined
with other advanced microscopies
• For overall connection mapping: phase-contrast
microtomography (0.3 µm resolution in all 3D directions)
• To explore in detail synapses and neuron connections:
nanotomography with <10 nm resolution
• To obtain detailed 3D maps of special regions: cryo-electron microscopy with nm resolution
• To analyze the whole drosophila brain and large parts of the mouse brain obtaining functional information: Confocal
FocusClear imaging
• For single-molecule mapping of drosophila and mouse brains:
Super-resolution microscopy with FocusClear
• …plus other tasks using other frontline techniques
Lightsheet Localization super-resolution Microscopy
Communications Biology, 2, 177 (2019) 2.7 × 104 µm3/s
250 image averaging, 15 min 75 nm resolution
Lattice lightsheet microscopy w/ tissue clearing
Brain size
Volume (mm3)
Number of Neurons
brain/body Ratio
Number of Synapse
Density of neurons
(/mm3 ) drosophila 2x10-2 2.5x105
(1.3x105) 1.25x107
mouse 450 7.1x107 1:40 1011 1.57x105
marmoset 6.4x108
human 1.2x106 8.6x1010 1:50 1014-15 7.2x104 Source of information: Wikipedia
Big Brain, Big Data
Nature
http://www.nature.com/nature/journal/v541/n7638/full/541559a.html#references
Nature blogs http://blogs.nature.com/naturejobs/2017/01/26/new-neuroscience-tools- for-team-science-in-big-data-era/
Scientific American https://www.scientificamerican.com/article/neuroscience-big- brain-big-data/
Small Brain, Big Data
Fluorescence confocal microscopy on connectome mapping
• Drosophila Brain: ~135,000 neurons
• Resolution:500 nm
• Data amount: ~10 TB
• Time to complete: ~20 year/10 microscopes
• Human brain: 85 billion neurons
• Time to complete: ~17,000,000 年
500 mm × 300 mm × 200 mm
3D X-ray data
• Single fly brain: ~100GB
• Single mouse brain: ~360TB
• Complete the map with 100 brains: ~36 PetaByte (PB)!
• Human/mouse brain volume: 2500 !
1 m 100 10 1 mm 100 10 1 µ m 100
SuperRes
*+
Pr ob e D ep th
CT
SuperRes
µCT
TEM TEM+
SuperRes*
10 1 100 10 1 100 10 1
mm µ m nn
- - - - - - - -
Phase Contrast
CT
PC – phase contrast
* – Tissue clearing
+ – Continuous sectioning
Resolution
10
010
-110
-210
-310
-410
-510
-310
-210
-110
010
110
210
3Resolution (mm)
Specimen size (mm)
400 nm Mouse brain 20 nm
Mouse brain
Mouse brain Mouse brain
X-ray imaging — From Animals, Organs, Neurons to Synapses
< 0.5 mm resolution (mCT) 1μm 0.5μm
ccelerated -ray
bservation for
eurons
AXON Implementation: TLS, APS, PLS-II, TPS..
Whole brain imaging without sectioning
Merging Different Brains
DAL neurons
enzmet/gold toning/gold enhance
Large tissue, high resolution and speed
Whole-brain blood vessels in mouse
Res. < 0.5 mm Size > 400 mm3 Time < 10 min
AXON on Mouse Brains
X-ray imaging of a mouse Purkinje neuron with 20 nm resolution
X-ray imaging of a whole mouse with 0.5 µ m resolution.
Tomography image shows blood vessels (golden) and neurons (green)
2 mm
200 µm 100 µm
20 µm
Recent progress – whole mouse brain imaging
Mouse brain slice, 400 µm thick
Mouse Perkinji cell in a thick tissue slice
Same specimen, different resolution
10 µm
Same specimen, different resolution
20x, NA = 0.75, resolution ~0.5 µm 40x, NA = 0.95, resolution ~0.3 µm
And speed..
Mouse brain, (0.8
µm)
3Technology impacts — Beyond connectome
•
X-ray imaging: phase contrast radiology and tomography, nanotomography•
Super-resolution imaging•
3D continuous sectioning•
3D spectromicroscopy•
3D pathology•
Big Data analysis•
Neuromorphic computing & AIBrain inspired computation, neurosynaptic
computing, neuromorphism and artificial brains
•
Power consumption• 86 billion neurons, ~1000 synapses per neuron
• >trillion transistor, >10 GW
• Human brain 20 W!
•
Simulate Brain from neuron (HBP)•
Neurosynaptic computer using neuromorphic chips•
Artificial IntelligenceFlybrain Simulators
•
130,000 neurons•
Build with real connectome map•
With input and output terminals for simulations•
Functional information included in the simulations•
Understand how fruit fly brain function (with a super computer)•
Better algorithm and computer architecture for AI•
Simulate and understand brain diseases, and the cureNext steps:
• Sub-10 nm imaging
• Multi-modality imaging
• X-ray molecular/functional imaging
• X-FEL applications?
• Laser wavefront accelerators?
Toward Sub-10 nm
•
Must be an international effort•
Mobilization of resources and user base•
Focus on novel configurations and applications•
A consortium of beamlines at NSRRC, SPring-8, PLS-II,BSRF (Beijing), SSRF (Shanghai), SSLS, NSLS-II (Singapore)
…
Objectives – to map in 3 years:
• One whole human brain
• 200 whole mouse brains
• Develop the microtomography technique for >100x improvement in speed to enable complete human connectome mapping
Whole brain w/ 1 m m resolution
SYNAPSE
2023: groundbreaking progress in human brain knowledge brought by the Asia-Pacific countries
Thank you!!
Go fast and go far with friends!
(from China, France, Germany, Japan, Korea, Singapore, Switzerland, US, ..)
Thanks for the funding support from:
• Ministry of Science & Technology
• National Program for Nanoscience and Nanotechnology
• Academia Sinica Thematic Projects
• ANR-MOST, INSERM-MOST
• RIKEN-MOST
• USAF-MOST
d
The SYNAPSE strategy:
data acquisition
and management
Data size:
• Our experience from the tomography of a 0.53 mm3 volume containing one drosophila brain:
Size of each raw projection image: 32 MB
Tomography set (1000 projection images): 32 GB
Reconstructed images for volume rendering: ≈128 GB
• Scaling to one mouse brain: [450/(0.5)3] x 128 GigaByte ≈ 460 TB
• Limited staining rate (~5%) 100 mouse brains required for complete map: ~46 PB
• Scaling to one human brain (2,700 times the mouse volume): 460 TB x 2,700 ≈ 1,240 PB
Total image-taking times at current speed:
• ~5 min for a (0.5)3 mm3 volume:
For one mouse brain: (450/0.53) x 5 min
≈ 1.8 x 104 minutes ≈ 12.5 days
For 100 mouse brains: 1250 days ≈ 3.4 years
• ~15 min for a 1 mm3 volume (with a 4K x 4K detector):
For one human brain: 1.2 x 106 x 15 min
≈ 1.8 x 107 minutes ≈ 34 years
For 100 human brains: 3,400 years
We must increase the throughput.
Without new technologies, by:
• Increasing the number of synchrotron beamlines used in parallel, coordinated within SYNAPSE
• Reducing the number of projection images for each tomography, from 1000 to 100.
With new algorithms
With artificial intelligence
It becomes thus realistic to map one human brain in 4 years – the first
SYNAPSE objective
Further throughput increases
possible with new technologies:
• Staining rate increase from 5% to 30% (demonstrated but not optimized). The number of brains for a
complete map could decrease to <20.
• Faster imaging: Higher x-ray flux (10x), higher detector sensitivity (5x) and better algorithms (to further
reduce the number of projections) – overall, the image acquisition speed could be increased by two orders of magnitude.
The image taking time for full human
brain mapping could be reduced to a
few years
Image taking is not all: data managing is a big challenge!
• Tomography reconstruction for drosophila: currently, 10 times slower than image acquisition
• To perform one automatic segmentation and tracing within a comparable time, it takes now the largest computing facility in Taiwan, NCHC
• Commercial graphic workstations cannot handle the
rendering and visualization of the reconstructed data set
• Another big challenge: creating and managing the database
• Further complication: adding functional information to the map
Crucial data handling tasks:
• Image morphing, warping and fusion
• Correlation identification
• Tracing, segmentation
• Database architecture and optimization
• Database access
• Visualization
Adding functional information:
• Using the connectome as the skeleton/grid.
• Adding local high-resolution 3D information on synapse connections with:
X-ray tomography with 10 nm resolution.
Super-resolution 3D fluorescence microscopy (with tissue clearing)
Super-resolution vibrational spectromicroscopy (Nano-IR and Nano-Raman)
Continuous sectioning with electron and optical microscopy
• Chemical information in large region:
Functional MRI, PET & SPECT imaging (with 100 µm resolution)
3D IR (with <3 µm resolution)
• Simulations based on the structural and functional information
Other technological developments targeted by SYNAPSE:
• Algorithms and related software
• Computer hardware: machine-learning graphic station
• Automation in specimen handling
• Automation in image acquisition
• Artificial Intelligence for automation of image processing
• Image taking hardware: specimen manipulation, detectors and high-brightness X-ray sources
Why us?
We are uniquely qualified!
• We pioneered the special techniques required for the SYNAPSE initiative, including:
X-ray phase-contrast tomography World-record x-ray microscopy
Deep tissue super-resolution microscopy
• We have a solid record of relevant previous
accomplishments and all the required know-how for the core and support techniques
• The new partnership is built on long-standing and very successful collaborations
Sizes of Brains:
Volume (mm3)
Number of neurons
Brain/
body ratio
Number of synapses
Density of neurons
(mm-3 ) drosophil
a 2x10-2 2.5x105
(1.3x105) 1.25x107
mouse 450 7.1x107 1:40 1011 1.57x105
marmoset 6.4x108
human 1.2x106 8.6x1010 1:50 1014-1015 7.2x104
Physics opens the new era of
biology and medicine, again and again.
• Microscopy
• Spectroscopy
• X-rays
• Protein crystallography
• NMR, SPECT, PET, CT, ultrasound, ..
• Mass spectrometry
• Nuclear Medicine
• Digital Camera
• …
540 million year old embryo fossil
Concentrate on imaging
Structure and Network:
• X-ray phase contrast microtomography, 0.3 µm resolution in all 3D directions: to provide the wiring map
• X-ray nanotomography: 10 nm to provide the structure and
location information of targeted synapse and neuron connections
• Confocal + FocusClear® : Full drosophila brain and regions of mouse brain with functional information
• Cryo-EM: small region 3D map down to nm resolution.
• Super-Res + FocusClear® : Full drosophila brain and small region of mouse brain but concentrate on single molecule detection
Stereographic 3D
Volumetric view
ONE STEP FURTHER…
Chian Ming Low, Eng Soon Tok, Alvin Teo, Tin-wee Tan (National University of Singapore)
Zhong Zhong & Hua Hau (中中與华华)
CA 1 CA
3 DG Sub
EC CA
2 Hippocampal
Formation
DG – Dentate Gyrus CA – Cornu Ammonis Sub – Subiculum EC – Entorhinal Cortex
Note: hippocampal formation consists of these regions – DG, CA3, CA2, CA1, Subiculum &
Entorhinal
cortex…you can also call it hippocampus (a more ‘general’ term)
Monkey Hippocampus
(Macaca Fascicularis)
CA 1
CA1 Rat Monkey
# Divisions 1-2 7
# Cell Thick 5 10-15
Boundary between PCL -SR
Clear Less clear
Boundary between CA1- Subiculum
Clear Less clear
Entorhinal Cortex – laminar organization
Less clear (e.g.
Layers V- VI) 2 divisions
Clear (e.g. Layers
V-VI) 7 divisions
hf
hf – hippocampal fissure (a ‘line’ that DG is separated from CA1)
Arrowhead – neuron soma
CA 3
Dentate Gyrus
Mouse brain 2x2x20 mm
3Besides Image Taking:
• Full biological characterization
• Specimen preparation: novel labeling procedures
• Coordinated use of different imaging facilities at partner countries, to reach the required
throughput
• Standardization and mutual validation for all participating facilities
• Advanced image processing with new computer techniques
• Full data access for all partners and eventually to public
The core partners bring with them a very powerful extended coalition:
• Through NUS/SSLS: research teams from university (NUS, NTU, NYP), public institutions (NCSS)
• Through SARI/SSRF: research teams from universities (Shanghai Jiao Tong U., Shanghaitech U.), public institutions (SINAP, IoN)
• Through POSTECH/PAL: research teams from universities (POSTECH, KAIST), medical institutions (ASAN, Samsung, SNU, Yonsei)
• Through RIKEN/Spring8: public institutions (RIKEN QBiC-BDR, BSI)
• Through ANSTO: research teams from universities (U. Sydney, U.
Wollongong), public institutions (Australian Synchrotron, Brain & Mind Center)
• Through AS: research teams from universities (NTU, NTHU, NCTU, NCKU), public institutions (ITRI, NHRI, NHPC), medical institutions (NTUH, Chang Gun H, Mackay H, TSGH, VGH, Chinese Medical UH, Taipei, Medical UH, Kaohsung MU, Cheng Kung UH) and from the private sector (TTY Pharma, Delta Electronics)