Direct Detection TEM Cameras - EMBO 2017

Direct Detection TEM Cameras

FOR MORE INFORMATION:BENJAMIN BAMMES, Ph.D.

DIRECTOR OF APPLICATIONS, MARKETING, & SALESEMAIL: [email protected]

PHONE: +1 858-384-0291 x105SKYPE: bbammes

CONFIDENTIAL: Do not share or distribute without written permission.Copyright © 2017 Direct Electron, LP. All rights reserved.

Slide #2 November 201

Factors Affecting Cryo-EM ResultsFernandez-Leiro & Scheres,

Nature 537 (2016).

Microscope

Detector

Sample Preparation

Processing



Photographic Film

• Difficult and time-consuming.• No immediate feedback.• No ability for automation.• No motion correction, dose

filtering, etc.



Charge-Coupled Device (CCD)

• Charge-coupled device invented in 1969 at Bell Labs.

• Scientific-grade (low noise, high efficiency) CCDs developed at CalTech for astronomy in the 1970s (first reported by Janesick, et al., 1981).

Hiraoka, Sedat, & Agard, Science 238 (1987).



Charge-Coupled Device (CCD)



Drawbacks of CCDs for Electrons

• Electrons deposit lots of energy = low dynamic range.

• Polysilicon gates damage (within a couple hours), causing significant loss of efficiency.

De Ruijter, Micron 26 (1995).



Strategies for Using CCDs for Electrons

• Lenses and a mirror to record viewing screen (Mochel & Mochel, 1986).

• Fiber-optic coupled YAG scrintillator (Sedat, Agard, & Krivanik, 1987).

Faruqi & Subramaniam, Quart. Rev. Biophys. 33 (2000).



Unique Applications of TEM CCDs

• Microscope autotuning (Krivanek & Fan, 1992).

• Automated tomography (Dierksen, et al., 1992; Koster & de Ruijter, 1992; Koster, et al., 1992).

• Telemicroscropy (Fan, et al., 1993).• Protein electron crystallography

(Brink & Chiu, 1994).• Electron holography (Daberkow, et

al., 1996; Duan, et al., 1998).Brink & Chiu, J. Struct. Biol. 113 (1994).



Drawbacks of a Scintillator

Booth, et al., J. Struct. Biol. 156 (2006). Bammes, et al., J. Struct. Biol. 175 (2011).

Scintillator imposes a point spread function of ~22.5 μm.



Need for Better TEM Detectors

Glaeser & Hall, Biophys J. 100 (2011).

“Although photographic film has long been the standard to beat (especially for 300 keV electrons),

it leaves much room for improvement in terms of detective quantum efficiency under low—exposure

conditions. New types of area detectors that are currently being developed for EM not only improve

on the readout speed of CCD cameras but also promise to improve the point—spread function (i.e., resolution) relative to the pixel size of the detector.”



Complementary Metal Oxide Semiconductor (CMOS)

• Complementary metal oxide semiconductor (CMOS) invented in 1963 at Fairchild Semiconductor.

• CMOS active pixel sensors (containing in-pixel amplifiers) were first described in 1968.

• High noise and difficulty of manufacturing led CCDs to be much more popular for image sensors.

• In the early 1990s, Jet Propulsion Laboratories (JPL) pushed CMOS image sensors forward, launching Photobit Corp. in 1995 to commercialize CMOS sensors.



CMOS Active Pixel Sensor (APS)

Milazzo, et al., Ultramicrosc. 104 (2005).



CMOS Direct Detection DevelopmentUCSD & UCI

Xuong, Ellisman, Kleinfelder, Jin

MRCTurchetta, Faruqi,

Henderson, McMullan

UCSD, LBNL & UCI

Xuong, Ellisman, Kleinfelder, Matis,

Denes

LBNLDenes

DIRECT ELECTRO

N

GATAN

FEI

20052004 2006 2007 2008 2009 2010



Our Story

Founded in 2007 to provide premium performance CCD cameras for electron microscopy and to commercialize the DDD technology.

DE-Series FC-Series LV-Series SoftwareDirect Detection Device

(DDD®) TEM camerasFiber-coupled CCD cameras

for TEMLow-voltage (e..g, 10-40 kV)

Direct Detection Device (DDD®) cameras

Software to supportour cameras



Direct Electron’s Development of DDD CamerasGeneration 1 2 3 4 5 6 7 8 9 10

Sensor EM1 EM2 EM3 EM4 EM5 12M 12M-2 12M-3 20M Series 6

Camera(s) Proto Proto Proto Proto Proto DE-12 DE-12 DE-12 DE-20 DE-16, DE-64

Year 2002 2002 2003 2005 2006 2008 2010 2011 2012 2014

Pixel Size (μm) 20 5, 10,20, 30 5 5 5 6 6 6 6.4 6.5

Array Size (pixels) 128 × 128 Various 550 × 512 1024 ×

1024 560 × 460 4096 ×3072

4096 ×3072

4096 ×3072

5120 ×3840

8192 ×8192

Pixel Design 3T 3T 3T 3T 3T > 3T > 3T > 3T > 3T > 3T

Feature/Milestone

4 quad 4 sectionsSingle pixel

design; MTF/DQE

Largerformat;

cryo-tomo

ADC per column; electron counting

Large formatFaster; more

radiation hard

Thinned; improved at

200 kV

Larger;reduced noise;

improved MTF

Ultra-largeformat;

improved SNR,

dynamic range, and radiation hardness

Developer/Funding

University of California, San DiegoNIH RR018841

Direct Electron, LPNIH RR024964 & NIH GM103417



CMOS Direct Detection Technology

Scintillator-Coupled Detector

TRADITIONAL TEM DIGITAL IMAGING:Performance is limited by the

scintillator and fiber optic coupling.

Direct Detector

NEXT GENERATION TEM IMAGING:High-performance direct detection

of primary electrons.



Resolution (Edge Spread Function)



Missing Features on a Scintillator-Coupled Camera



DE Cameras : High-Resolution Signal at Nyquist

The Fourier transform (FFT) of a 37,000× magnification image of a line-grating grid. The sampling is 1.01 Å/pixel. The 2.36 Å and 2.04 Ågold lattice spacing are clearly visible in the FFT, even at Nyquist frequency (which is the theoreticalbest resolution of a camera).Data collection courtesy of Scott Stagg, Florida State University (Tallahassee, FL USA).



Modulation Transfer Function (MTF)



Resolution (Modulation Transfer Function)



Signal-to-Noise (Detective Quantum Efficiency)

How well does the detector faithfully record the input imagewithout adding more noise at various resolutions?



Signal-to-Noise (Detective Quantum Efficiency)



Similar Performance at 200 and 300 kV



Counting Boosts DQE for Low-Dose Imaging



Complementary Metal Oxide Semiconductor (CMOS)



Movie-Mode Imaging

NEW IMAGING METHODS:• Exposure setting ex post facto• Motion correction• Radiation damage compensation (dose filtering)• Electron counting• High speed movies (in situ TEM, 4D STEM)

Continuous streamingat a high frame rate with

no dead time between frames



Movie-Mode Motion Correction

Brilot, et al., J Struct Biol 177 (2012), 630-637.



GroEL+GroES at 1.1 um Defocus20 e-/Å2 (conventional exposure) 100 e-/Å2 with damage compensation

Courtesy of Wah Chiu, Baylor College of Medicine, USA.



~200 kDa Protein at 1.25 um Defocus



The Story of High-Resolution Cryo-EM

0

10

20

30

40

50

60

70

80

90

100

1/08 1/09 1/10 1/11 1/12 1/13 1/14 1/15

Num

ber o

f EM

DB D

epos

ition

s

Deposition Date

2.x Å 3.x Å 4.x Å



Challenges for High-Resolution Cryo-EM

- DATA COLLECTION -

POOR DETECTORS: “[There is] much room for improvement in terms of detective quantum efficiency.”

SPECIMEN MOTION: “Beam-induced movement that occurs while the image is recorded is responsible for the steep falloff of signal.”

RADIATION DAMAGE: “The primary limiting factor is radiation damage and, more specifically, the poor signal/noise ratio (SNR) in images recorded with optimal electron exposures.”

Glaeser & Hall, Biophys J 100 (2011), 2331-2337.



The Story of High-Resolution Cryo-EM

First DDD Deposition

0

10

20

30

40

50

60

70

80

90

100

1/08 1/09 1/10 1/11 1/12 1/13 1/14 1/15

Num

ber o

f EM

DB D

epos

ition

s

Deposition Date

2.x Å 3.x Å 4.x Å



“Routine” Near-Atomic Resolution Cryo-EMData collected with a DE-20 Camera System on an FEI Titan Krios fully automated with Leginon software.

Total time required for data collection:3 days.

Courtesy of Scott Stagg, Florida State University, USA.

Published in Spear, et al., J Struct Biol(2015).



Rigorously-Validated Atomic Model of P22

Courtesy of Wah Chiu, Baylor College of MedicineCONFIDENTIAL: Do not share or distribute without written permission.

Copyright © 2017 Direct Electron, LP. All rights reserved.Slide #3

2 November 201

What is Next for Cryo-EM Detectors?



Unrivaled Features

IntegratedFaraday platefor exposuremeasurement

Integrated 2k×2ksurvey camera

Fullyretractable

High-speedDDD sensorprotectionshutter

Custom-designedDirect Detection Devicewith global shutter & CDS

Compatible with nearlyall TEM configurations



The Next Big Thing in Cryo-EMHigh-throughput single particle.

DE-64



DE-64 : Massive Field-of-View

A low magnification image at 120× of a line-grating grid. The entire 8k × 8k field-of-view is shown at the left, and the middle and right images show digital zoom to display the level of detail in this massive image

(note that individual latex beads are visible even at this very low magnification).Data collection courtesy of Scott Stagg, Florida State University (Tallahassee, FL USA).

Digital zoom Digital zoomActual image



The Largest Field-of-View67 MP 67 MP 67 MP

16 MP 16 MP 16 MP

14 MP 14 MP 14 MP



DE-64 Covers an Entire Hole in One Shot1.47 Å/pixel sampling; 1.2 μm diameter holes




Cellular tomography. DE-64



Other Applications : Electron Tomography

Data collected with a DE-20 Camera System on a JEOL JEM-2200FS automated with SerialEM software.

CLEM (correlated light and electron microsopy) tomographic reconstruction shows HIV core pseudotypes with ASLV envelop infected CV-1 cells.

Courtesy of Elizabeth Wright, Emory University, USA.



Motion Correction for Tomography

Data collected with a DE-20 Camera System on a JEOL JEM-2200FS automated with SerialEM software.

Z-slice from a CLEM (correlated light and electron microsopy) tomographic reconstruction shows HIV core pseudotypes with ASLV envelop infected CV-1 cells.

Courtesy of Elizabeth Wright, Emory University, USA.



Damage Compensated Tomography

Courtesy of Elizabeth Wright, Emory University

Conventional

With Damage Compensation




Cellular tomography. DE-64The best DQE at all costs.



DE-64 : Electron Counting

A cropped field of view of a single frame from the DE-64 with no binning (left), showing electron events after thresholding. <4% of the events are larger than 4 unbinned pixels (which corresponds to the size of

one binned-2× pixel). The distribution of counts from incident primary electrons is shown at the right. The most-probable signal-to-noise ratio (SNR) is ~50, and the average SNR is ~98. Based on the fit of the

Landau distribution, only ~1% of incident electrons will be missed due to low signal.CONFIDENTIAL: Do not share or distribute without written permission.

Copyright © 2017 Direct Electron, LP. All rights reserved.Slide #4

2 November 201

Photon Counting



Photon Counting



Electron Counting



Electron Counting



DE-64 : Near-Perfect MTF

A cropped region of a beamstop edge with bin-2× counting (right), with the corresponding MTF (left) shown in orange. For comparison, the unbinned integrating mode MTF is shown in blue.



SNR from Each Incident ElectronCompeting Direct Detector (SNR=19.6) Direct Electron DE-20 (SNR=49.6)

Figure courtesy of Greg McMullan (MRC).

Single incident electrons

2.5× improvement!



DE-64 : Electron Counting

A cropped field of view of a single frame from the DE-64 with no binning (left), showing electron events after thresholding. <4% of the events are larger than 4 unbinned pixels (which corresponds to the size of

one binned-2× pixel). The distribution of counts from incident primary electrons is shown at the right. The most-probable signal-to-noise ratio (SNR) is ~50, and the average SNR is ~98. Based on the fit of the

Landau distribution, only ~1% of incident electrons will be missed due to low signal.



DE-64 : Detective Quantum Efficiency (DQE)

The DQE of bin-2× counting (orange) and unbinned integrating mode (blue).









High-resolution at 200kV. DE-20



Fast, High-Resolution at 200 kV (F20 TEM)A representative cryo-EM micrograph and selected 2D class averages of Thermoplasmaacidophilum 20S Proteasome imaged on a DE-20 camera on a 200 kV FEG Tecnai F20 TEM without an objective aperture. The micrograph shown is at1.5 μm defocus. A total of 350 micrographs were collected using automation in SerialEM, at a rate of approximately 80 acquisitions per hour. Pixel size = 0.91 Å/pixel.

Data courtesy of Radostin Danev, Max Planck Institute of Biochemistry (Martinsried, Germany).



Fast, High-Resolution at 200 kV (F20 TEM)

Each acquisition contained 20 frames, which were motion corrected and dose filtered using MotionCor2 (Zheng et al., 2017) with 5×4 patches. CTF parameters were determined using CTFFind4 (Rohou & Grigorieff, 2015). Boxing was completed using EMAN2 (Tang et al., 2007). 2D class averages and the 3D reconstruction were generated using Relion (Scheres, 2012). All image processing was completed on a desktop computer with a single NVidia GeForce GTX 1060 GPU.

A total of 43,637 particles were boxed, of which 31,324 particles were retained after selecting reasonable-looking 2D class averages (examples shown above).



Fast, High-Resolution at 200 kV (F20 TEM)

The final 3D reconstruction (shown at left) reached 3.76 Å resolution by gold-standard FSC. An alpha-helix and beta-sheet from the reconstruction are shown below, with the X-ray crystal structure (PDB: 1PMA) rigid-body fit into the cryo-EM density map.



3.6 Å Resolution at 200 kV (JEOL 2100F)

Courtesy of Wei-Hau Chang, Academia Sinica





High-resolution at 200kV.

Publish with 120kV LaB6.

DE-20

DE-12



Cryo-EM with 120 kV Lab6



Carbon Film with 120 kV Lab6 (0.8 μm)









Carbon Film (1.5 μm) without Motion Correction



Mm-Cpn with 120kV Lab6 (0.6 μm)









Mm-Cpn with 120kV Lab6 at 10 Å Resolution







DE-20

DE-12

Electron diffraction, 4D STEM, & other techniques.



Cryo-ED of Catalase

~ 5 e-/Å2



Cryo-ED of Catalase Movie (25 fps)



Electron Diffraction “Motion Correction”

Three selected frames from a 1 second movie at 25 fps of electron diffraction of frozen-hydrated catalase crystal, collected on a DE-12 Camera mounted on a JEOL 2010F microscope. For reference, the diffraction pattern from the final frame is overlaid in

light green. Data collected in the lab of Wah Chiu, Baylor College of Medicine (Houston, TX USA).



Electron Diffraction “Motion Correction”

The integrated 1 second diffraction pattern from above, without (left) and with (right) motion correction. Motion correction used the DE_process_frames Python script, which uses cross-correlation to align prominent features in a movie acquired with a DE camera. In this case, instead of correcting specimen motion, we were correcting the motion of the beam during diffraction,

which noticeably improved the signal-to-noise ratio (SNR) of diffraction spots.Data collected in the lab of Wah Chiu, Baylor College of Medicine (Houston, TX USA).



4D STEM : Capturing All Electrons

Courtesy of Miaofang Chi, Oak Ridge National Laboratory, USA.

Bright-Field Disc







DE-20

DE-12

Electron diffraction, 4D STEM, & other techniques.

Documents

Direct Detection TEM Cameras - EMBO 2017