detection technologies for cryo-electron microscopy...16 sensitivity • minimum detectable signal...
TRANSCRIPT
1
Detection Technologies for Cryo-Electron
Microscopy
S2C2 Workshop – Cryo-EM Training
Christopher Booth
Director of Product Management
Gatan Inc.
May 29, 2020
2
Detection Technologies in Cryo-EM
• Detection of Electrons
• Measuring Detector
Performance
• Energy Filtering and cryo-EM
• Framerate: Motion Correction,
CDS and Co-incidence loss
• Throughput in Cryo-EM
• Practical Considerations
3
Detection Technologies in Cryo-EM
• Detection of Electrons
• Measuring Detector
Performance
• Energy Filtering and cryo-EM
• Framerate: Motion Correction,
CDS and Co-incidence loss
• Throughput in Cryo-EM
• Practical Considerations
4
Detection Technologies in Cryo-EM
• Detection of Electrons
• Measuring Detector
Performance
• Energy Filtering and cryo-EM
• Framerate: Motion Correction,
CDS and Co-incidence loss
• Throughput in Cryo-EM
• Practical Considerations
5
Detection Technologies in Cryo-EM
• Detection of Electrons
• Measuring Detector
Performance
• Energy Filtering and cryo-EM
• Framerate: Motion Correction,
CDS and Co-incidence loss
• Throughput in Cryo-EM
• Practical Considerations
6
Detection Technologies in Cryo-EM
• Detection of Electrons
• Measuring Detector
Performance
• Energy Filtering and cryo-EM
• Framerate: Motion Correction,
CDS and Co-incidence loss
• Throughput in Cryo-EM
• Practical Considerations
7
Detection Technologies in Cryo-EM
• Detection of Electrons
• Measuring Detector
Performance
• Energy Filtering and cryo-EM
• Framerate: Motion Correction,
CDS and Co-incidence loss
• Throughput in Cryo-EM
• Practical Considerations
8
Detection Of Electrons
9
Key Concepts in Detecting Electrons
• CMOS and CCD
• “indirect” vs direct detection cameras
• Sensitivity
• Linearity and dynamic range
• Dynamic range
• Pixel size and field of view
• Electron counting
• Co-incidence loss
10
Recording Images In Electron MicroscopyA little bit of history
• Oldest recording medium: photographic film
• 1970: Charge coupled device (CCD) was invented
• 1976: CCD camera was used for astronomy
• 1982: 100 x 100 CCD was directly exposed to 100 kV electrons…radiation damage
• 1988: 576 x 382 CCD used with scintillator and optical coupler
• 1990: Gatan made the world’s first commercial CCD camera
• 2002: 128 x 128 direct detection camera developed
• 2008 – 2009: commercial complementary metal-oxide semiconductor (CMOS) cameras and radiation hard CMOS cameras were introduced
11
CCD vs. CMOS
Both CCD and CMOS use photo diodes to convert photons to electrons,
the difference is how they store charge and transfer it.
• CCD: Charge is transferred between neighboring cells, and read-out
• CMOS : Charge immediately converted to voltage (read out with digital output)
http://meroli.web.cern.ch/meroli/lecture_cmos_vs_ccd_pixel_sensor.html
12
Detectors in Electron Microscopy
A. Optically coupled
B. Fiber-optic coupling
C. Direct detection
D. Transmission direct detection
13
Traditional Indirect Detection
e-
Scintillator
Light transfer
Light sensitive CCD or CMOS
1. Convert
electrons to
light
2. Transfer light to
detector
3. Detect light and
convert to
signal
14
Direct Detectione-
Radiation hard CMOS
1. Convert
electrons to
light
2. Transfer light to
detector
3. Detect electron
and convert to
signal
15
Transmission Direct Detectione-
Radiation hard, thinned CMOS
1. Convert
electrons to
light
2. Transfer light to
detector
3. Detect electron
and convert to
signalMinimize back scattered
electrons that add noise
16
Sensitivity
• Minimum detectable signal in terms of the number of incident electrons.
• Single-electron sensitivity
• if the gain of the system is such that the output of a single incident electron is above the
noise floor
Deposited Energy
# p
ixel
s
Noise floorSignal from incident electrons
17
Linearity
• Relationship between output (image
intensity in digital units) and the input
(number of incident electrons).
• CMOS and CCDS are much more linear
than film
• Counted cameras have a special kind of
non-linearity called co-incidence loss
Journal of Microscopy, Vol. 200, Pt 1, 2000, pp. 1±13.
18
Dynamic Range
Dynamic range: The range of values that can be distinguished between a maximum level
(saturation) and zero (noise)
• Driven by combination of max allowable charge in each and noise floor
• One pixel can have 16 bit dynamic range (values between 0 – 16000)
• Used to be a very important factor for cameras, now frame rate is much more important
• A camera with only 12 bit dynamic range (0 – 4095) might accumulate 40 frames in a
second.
• 4095 x 20 = 163,800 counts of dynamic range
Journal of Microscopy, Vol. 200, Pt 1, 2000, pp. 1±13.
19
3,71
0 pi
xels
3,838 pixelsK2
,4.0
92 p
ixel
s
5,760 pixelsK3
23.6 Mpixels(94 Mpixels super-resolution)
14.4 Mpixels
How Many Pixels are Enough?
20
Rio™ 16: 4k x 4k, 9 µm
Rio 9: 3k x 3k, 9 µm
How Important is FOV?
OneView®: 4k x 4k, 15 µm
Rio 16: 4k x 4k, 9 µm
K3™ : 6k x 4k, 5 µm
K2®: 4k x 4k, 5 µm
Side-mount camera
Bottom-mount camera
CCD
21
Electron Counting Makes All the Difference
Single high speed frame using conventional
CCD-style charge read-out
Same frame after counting
Counting removes the variability from scattering,
rejects the electronic read-noise, and restores the DQE.
22
Traditional Integration
Similar to indirect detection cameras, direct detectors can integrate
the total charge produced when an electron strikes a pixel.
Electron enters
Detector.
Electron signal is
scattered.
Charge collects in
each pixel.
23
Counting
In counting mode, individual electron events are identified at the time
that they reach the detector.
Electron enters
Detector.
Electron signal is
scattered.
Charge collects in
each pixel.
Event reduced to
highest charge pixels.
To do this efficiently the camera must run fast enough so that individual electron
events can be identified separately.
24
Super-Resolution
The theoretical information limit defined by the physical pixel size is
surpassed when you use super-resolution mode.
Electron enters
Detector.
Electron signal is
scattered.
Charge collects in
each pixel.
Event localized to
sub-pixel accuracy.
High-speed electronics are able to recognize each electron event
(at 400 fps) and find the center of event with sub-pixel precision.
25
Measuring Detector Performance
26
PSF, MTF, NTF and DQE?
• PSF: Point spread function
• Blurring of a single point in the camera
• MTF: Modulation transfer function
• PSF as a function of spatial frequency• Most often estimated using a “knife
edge”
• NTF: Noise transfer function
• Noise power spectrum• Noise as a function of spatial
frequency
DQE s
=SNRout s
SNRin s
=MTF(s)2
ΤNPSout(s) Dosein(s)
27
DQE Challenges
• Signal challenges: Edge image non-ideality
• Charging and edge cleanliness• Scale• Edge dose• Motion• Fields• scatter
• Noise challenges:
• Fixed pattern noise• Calibration of noise power• Measurement of incoming beam level
• Counting-related challenges:
• spatial effects of coincidence loss: high-pass filtering• Non-linear counting due to coincidence loss – calibration.• background
𝐷𝑄𝐸 𝑠
=𝑆𝑁𝑅𝑜𝑢𝑡 𝑠
𝑆𝑁𝑅𝑖𝑛 𝑠
=Τ𝑆𝑃𝑆𝑜𝑢𝑡(𝑠) 𝑆𝑃𝑆𝑖𝑛(𝑠)
Τ𝑁𝑃𝑆𝑜𝑢𝑡(𝑠) 𝑁𝑃𝑆𝑖𝑛(𝑠)
=𝑀𝑇𝐹(𝑠)2
Τ𝑁𝑃𝑆𝑜𝑢𝑡(𝑠) 𝐷𝑜𝑠𝑒𝑖𝑛(𝑠)
28
Measuring MTF with a Physical Edge (1)
AuPd-coated and plasma cleaned edge with canted face mounted in the entrance aperture of a GIF Quantum®
imaged on the K2 at the end of the GIF in super-resolution mode at 200 kV. (particularly bad example – not
always this bad)
Effect of charging
Short traversal close to
surface charge
Long traversal
close to surface
charge
Canted
edge Wire
29
Measuring MTF with a Physical Edge (2)
Motion – edge creepNoise
A noise-tolerant method for measuring MTF from
found-object edges in a TEM, Paul Mooney,
Microscopy and Microanalysis 15:1322-1323 CD
Cambridge University Press (2009). Figure 4:
Simulated MTF with various amounts of shot noise
added.
Good: difference
between two 20 s
edge images
showing no “motion
fringe”.
At low dose rates, need long exposures to get
enough dose → have to be careful about edge
creep.
30
Other Things to Avoid with DQE Measurements
• Missing dose in a Faraday cup holder: overestimates DCE and
therefore DQE (in same proportion)
• Using the TEM screen calibration
• Drifting beam current
• Over- or under- values MTF(0) or DCE
• Leaving specimen holder inserted during MTF measurement
Charging of specimen and/or holder can move image of edge shadow.
31
Measuring Image Performance Using Thon Rings
Thon rings are a indicator of
performance.
However, they are a system
test and really hard to
compare quantitatively.
32
Energy Filtering and cryo-EM
33
Inelastic scattering (EELS)
TEM Beam-Specimen Interactions and Signals
Elastic scattering
(Diffraction)
X rays
(EDS)Auger
electrons
BSE
Secondary
electrons
TEM specimen
EBIC
Light
(CL)
34
Atomic-Scale View of Electron Energy Loss in TEM
• Primary electron transfers
both energy and
momentum (angle)
• Primary electron is then
measured in the EELS
system
• Both final energy and
scattering angle are
important
Incident beam electron
E0 (60 to 300 keV)
Excited specimen electron
E = ΔE
Scattered beam electron
E0 - ΔE
36
Features in an Energy-Loss Spectrum
1000
2000
3000
4000
0 50 100 150
Energy Loss (eV)
x10
Cou
nts
Zero-loss peak
“Fine structure”
37
K edges L2,3 edges M2,3 edges M4,5 edges N4,5 edges O2,3 edges
Which Edges Can be Used?
x 1
0^3
eV
0
10
20
30
40
50
x 1
0^3
1800 1900 2000 2100 2200 2300 2400 2500 2600
eV
Si
Bkgd
Edge
x 1
0^3
eV
0
20
40
60
80
100
120
140
160
x 1
0^3
700 720 740 760 780 800
eV
Fe
Bkgd
Edge
eV
0
1000
2000
3000
4000
5000
6000
1200 1300 1400 1500 1600 1700
eV
Ge
Bkgd
Edge
eV
0
5000
10000
15000
20000
25000
30000
900 950 1000 1050 1100
eV
Ce
Bkgd
Edge
Slide thanks
to G. Kothleitner
38
Relative Intensities of EELS Signals
Ni M2,3
O KNi L2,3
Feature
Zero loss peak
Plasmon peak
Ni M edge
O K edge
Ni L edge
Integrated
counts
8.9 x 1010
6.5 x 1010
4.3 x 109
1.1 x 108
9.9 x 107
Inte
nsity (
CC
D c
ou
nts
)
C K (contamination)
39
EFTEM analysis of the Visual Cortex of the Occipital Brain – contrast
enhancement
ZLP
Low loss Core loss
ZLP C K-edge
Unfiltered Zero Loss
filteredPre-Carbon
• Filtering enhances the contrast in the TEM image.
• All Bio Quantum or Bio Continuum are sold for zero loss filtered imaging
40
EFTEM analysis of the Visual Cortex of the Occipital Brain
Ca elemental map P elemental map N elemental map
• Elemental maps taken using EFTEM SI
41
The Zero-Loss Signal
• Specimen thickness
• Energy reference
• Intensity reference
• Beam current
1000
2000
3000
4000
Be EELS
0 50 100 150
Energy Loss (eV)
x10
Co
un
ts
42
In-column filterPost-column filter
Basic Principles of TEM Energy Filters
43
Details of a Post-Column Spectrometer
Electron
Detector
Electrically
isolated
drift tube
44
Electron Detection
Direct Detection
scintillator:
~2000 photons
one 100 keV
electron
fiber optic plate:
~200 photons reach CCD
CMOS or CCD detector:
~60 electron-hole
pairs produced
electronics:
measure and digitize
charge signal
one 300 keV
electron
Support:
Back thinned
CMOS detector:
~electron-hole
pairs produced by primary
Incident electron
electronics:
measure and digitize
charge signal
Indirect Detection
45
Basic Principles of TEM Energy Filters
Final EELS
readout
Energy loss spectrum at camera
Spectrum mode
46
Basic Principles of TEM Energy Filters
Unfiltered EFTEM
47
Basic Principles of TEM Energy Filters
Zero-loss image projected
onto CCD detector
Zero-loss EFTEM
Energy-selecting
slit inserted
48
Energy Filtering helps improve 2D crystallography contrast
50
Energy filtering improves image contrast even for (small) particles
• Filtered Image (left)
• Unfiltered Image (right)
Unfiltere
d
Filtered
51
Energy filtering improves image contrast even for (small) particles
• Red = filtered Images
• Blue = unfiltered Images
• Few % improvement in
contrast even with TMV
52
Framerate: Motion Correction, CDS and
Coincidence Loss
53
How Frame Alignment Works
Raw counted frame Final aligned image
54
+ + ... +
* ... **
=
=
Motion correction with raw frames is hard
Motion correction with sub-frames is very good
1 sub-frame
1 final image
55
MotionCor2 -InMrc Stack.mrc -OutMrc CorrectedSum.mrc -Patch 5 5 -FtBin 1.2 -Iter 10 -FmDose 1.2 -bft 1.1 -Tol 0.5
MotionCor2 on the K3
56
Better Data Through Motion Correction
• Sample damage increases
during the exposure
• The first frames have the
least damage but the most
drift
57
Better Data Through Motion Correction
• Sample damage increases during the
exposure
• The first frames have the least
damage but the most drift
• Today the first 2 –3 frames are
excluded
• The K3 should be fast enough to let us
use frames 1 – 3 !
58
K3 lowers Read Noise with Correlated Double Sampling (CDS)
59
Standard mode (K2 and K3)
Reset…expose… Signal read
CDS mode (K3 only)
Reset…Reference read…expose…Signal read… subtract
K3 lowers Read Noise with Correlated Double Sampling (CDS)
60
Standard mode (K2 and K3)
Reset…expose… Signal read
CDS mode (K3 only)
Reset…Reference read…expose…Signal read… subtract
K3 lowers Read Noise with Correlated Double Sampling (CDS)
61
Readout noise Complete absorption of electron by detector (only for low E electrons or very large pixels)
Traversing electrons
False negativesFalse positives
Improved Noise Also Allows us to Improve Electron Countability
Deposited Energy
# p
ixel
s
62
Co-incidence loss is when it is impossible to tell if a signal comes from one
electron or more
This is one electron event
What about this?
or this?
63
Coincidence Loss, Framerate and Throughput
0
30
60
90
120
150
180
0 30 60 90 120 150 180
Measure
d d
ose r
ate
(e
lectr
ons/p
ixel/second)
Input dose rate (electrons/pixel/second)
K3
250 fps camera
Perfect Detection
DQE = 1 with no coincidence loss
• Co-incidence loss is the inability of counting direct electron detectors to detect electrons that
strike the detector near the same location
64
Electron Counting Requires that Electrons Don’t Overlap on the Sensor
Faster frame
rate
Lower beam
intensity
Short Exposure
Time
Long Exposure
Time
65
Throughput In cryo-EM
66
Cryo-EM and Structural Biology
• Cryo-EM is still growing at an
exponential pace
• Efficient use of each
microscope is that much
more important
2019*201620132010200720042001199819951992
2019*201620132010200720042001199819951992
NMR X-ray EM
67
Improving Throughput: Larger Sensors
• One chance to expose a specimen area
• If the pixel quality is high, larger sensors
reduce the number of images needed
K3
K2
68
Image Shift To Improve Throughput
K2: 86 Mpix/hole K3: 122 Mpix/hole
69
Automation with Coma-correction
Mode Low Mag High Mag Focus
Mode 1 Stage Stage Stage
Mode 2 Stage Stage Optical
Mode 3 Stage Optical Optical
Mode 4 Optical Optical Optical
70
How much area is imaged per hour with a camera
Imaging
efficiency=
Area per
image
Images per
hourX Area imaged per hour=
0
50
100
150
200
250
0 2 4 6 8 10 12 14
Ima
gin
g A
rea
(u
m2/h
r)
Coincidence Loss (%)
56 e-Å-2
As electron dose increases,
imaging efficiency increases
and so does the
coincidence loss
30 e-Å-2
K3
Increasing Quality
Imaging Efficiency
71
How much area is imaged per hour with a camera
Imaging
efficiency=
Area per
image
Images per
hourX
10.6 x
Area imaged per hour=
250 fps camera
at coincidence loss of 3.9%
Imaging
efficiency= =
images per
hourXµm20.167 86 µm2/hr14.4
Imaging
efficiency ==images per
hourXµm20.167 150*at coincidence loss of 7.1% µm2/hr25.1
Imaging
efficiency = =images per
hourXµm20.235 650ϕK3
at coincidence loss of 3.9%µm2/hr152.7
Φ Experimental data from Ametek Gatan, Inc.
* Acta Crystallogr D Struct Biol. 2020 Apr 1;76:313-325
Imaging Efficiency
72
TMEM16F
Resolution (Å) 3.15
Camera K2 summit
Total dose (e-/Å2) 56
Dose rate (e-/pix/sec) 7.8
Exposure time (sec) 8
Imaging area (µm2) 714
Coincidence loss (%) 9.3
Looking at a real example
Cell Reports 2018;28(2):567-579
73
4.09
26.13
-
10
20
30
Imagin
g tim
e (
hrs
)
Ion channels ~3 Å | Symm=C4
K3 (2701 images)
250 fps camera (3794 images)
0
50
100
150
200
250
0 5 10 15
Ima
gin
g A
rea
(u
m2/h
r)
Coincidence Loss (%)
6.4x boostK3
250 fps camera
Looking at a real example
74
Standard Magnification
• 49kX Mag
• 1.6 A/pixel
• 0.8 A/super-pixel
• 233 particles per frame
• 240 collected
• 2.7 A resolution
75
Low Magnification
• 39kX Mag
• 2.0 A/pixel
• 1.0 A/super-pixel
• 427 particles per frame
• 260 collected
• 3.0 A resolution
76
Solving Structures of COVID-19
At least 6 different structures of various COVID-19 components
"The K3 camera not only gave us great images…, it also prevented data collection from becoming a bottleneck, ultimately letting our team concentrate on getting the structure as quickly as possible”
Sample to publications in 3 weeks!
77
Practical Considerations in Data Collection
78
Dark Subtraction
• Removes the noise
baseline from the image
• New dark references are
often taken once a day
Dark image
79
Gain Correction
• Gain correction normalizes the response of each pixel to an electron
• This is why images are often floating point values
• In K3 we are allowing integer gain normalization
• Each electron is 32 counts
• Usually collected once per week
Gain map
80
Defect Correction
• Removes poorly performing pixels
• Hot• Dark• Unstable
• Defect pixels contribute to fixed pattern noise
• Usually updated with Gain Reference
Defect map(enlarged detail)
81
Defect map
(enlarged detail)
Uncorrected image
Corrected image
Dark image
Gain map
-
Typical Gain Correction Scheme
82
Counted Gain Correction Scheme
-
Linear Image Correction Counted Image Correction
Electron
CountingRaw
Linear
Dark
Linear
Gain
Linear
Defect
LinearUn-proc
Counted
Counted
Gain Ref
Counted
Defect
Final
Counted
Electron
Counting
83
Image: Uniform intensity FFT: White noise
Checking the Quality of Image Correction
84
Measurement of Fixed Pattern Noise (FPN)
Uniform A
Uniform B
Center of Xcorr
FPN = 0.0106
Uniform A Uniform B⊗
Uniform illumination
Common defects, dark image and gain image
Frame rate = 75 fr/s, (0.0133156 s/fr), all images.
Total dose = 14 e/pix, all images
FPN = peak pixel valueCross-correlation map
85
Dose rate on the detector
Mean number of electrons hitting
a detector pixel per unit time.
Total dose at the sample
Number of electrons that traverse a
unit area of the sample during the
exposure of this image frame.
86
At 8-bit/pixel, gain-corrected data saturates with a value of 255.
The saturation monitor reports the percentage of pixels that have reached saturation in a single frame.
Keeping track of Pixel Saturation in K3
87
Annealing Prevents Contamination Buildup
• A cold sensor is essentially a vacuum pump. Contamination builds up on its cold surface and, if left unchecked over prolonged
times, will accumulate to the point of degrade data quality
• Severe contamination may even become evident on the gain reference images, as in the example below of a K2 sensor
Without
annealing
After
annealing
88
• Regular annealing of the sensor reduces
background levels and surface
contamination
• It can also help to repair some radiation
damage
• If the camera is used for a prolonged time
without warming, the electron beam can
harden any contaminants, making the
CMOS detector difficult to clean
• Annealing should be done at the end of
every session one day long or longer
• Waiting a week is no problem, waiting 3
months is too long
• Every time you do a cryo-cycle is a good
rule of thumb
Camera Heating/Annealing
89
Thank you!