A. Overview |
Vertical Divider
We require large numbers of 3D images of live cells to understand and predict the states and variance of the human cell. We built an automated pipeline based on spinning disk light microscopy. We optimized every step to create high-quality, standardized data sets in high replicates.
|
1. Sample Preparation
|
2. Imaging
|
3. Image Processing
|
4. Data Management
|
BACKGROUND
Human induced pluripotent stem cells (hiPSC) grow as densely packed epithelial-like monolayers on Matrigel coated glass. They are up to 20μm tall.
hiPS Cell Biology Overview >>
We chose spinning disk microscopy and robotics for sample preparation as the central part of the pipeline because it is optimized for:
To add more information to our data, we plan to build additional pipelines based on the Zeiss LSM 880 Airy fast, FCS, and image correlation techniques.
We visualize all structures with diffraction limited optical microscopy and collect four channels:
hiPS Cell Biology Overview >>
We chose spinning disk microscopy and robotics for sample preparation as the central part of the pipeline because it is optimized for:
- Live cell imaging
- 3D microscopy
- Image quality
- Reproducibility
To add more information to our data, we plan to build additional pipelines based on the Zeiss LSM 880 Airy fast, FCS, and image correlation techniques.
We visualize all structures with diffraction limited optical microscopy and collect four channels:
- Structure of interest (eGFP)
- Nucleus as reference (NucBlue)
- Plasma membrane to define shape of cell (CellMask)
- Transmitted light
Standard operating procedure downloads
B. Grow Large Quantities of Cells |
Vertical Divider
For statistical modelling and analysis of variance and to discover rare states, we produce large numbers of images for each cell line. As a first step we manually grow and passage the cell population before seeding it on glass.
|
C. Prepare for Imaging in 96-well Plates |
Vertical Divider
To streamline automation, maximize throughput, standardization and reproducibility, we use robotics to seed cells in 96-well plates with optical glass bottoms for high-resolution optical fluorescence microscopy.
|
Instrumentation
ROBOTICS
Cells are seeded and maintained on a Hamilton Microlab Star (Hamilton Company, Reno, NV, USA) robotic pipetting workstation equipped with:
- Arm with 8 channels for pipetting
- iSWAP plate handler
- Orbit barcode scanner
- 2 MultiFlex tilt modules
- CPAC heating/cooling unit
- Tip and tube carrier and plate stacking modules
- Star HEPA filter with UV light option
- Cytomat 24C (Thermo Fisher) incubator (37℃, 85% humidity and 5% CO2)
- Cytomat 6002 refrigerator (Thermo Fisher)
(4℃) - Venus 2 ver. 4. software
- Venus 1 Dynamic Scheduler 5.1
96-WELL PLATES
For all imaging experiments we use 96-well glass-bottom plates (Cellvis, Mountain View, CA, # P96-1.5H-N):
- Black polystyrene frame
- High performance #1.5 cover glass (0.170±0.005mm)
- Lid
- Individually packed and sterilized
CONTROLS
Each plate has the following controls:
- Dye solutions for illumination profile correction:
- Coumarin 102 (Sigma-Aldrich, #546151)
- Fluorescein (Sigma-Aldrich, #32615)
- Acid Blue 9 (TCI, # CI-42090)
- Rose Bengal (Sigma-Aldrich, #198250)
- Mixed TetraSpeck beads (0.1μm diameter, Thermo Fischer, # T7284) and FocalCheck beads (15μm diameter, Thermo Fischer, # F7239) for resolution and alignment check
- Unlabeled hiPSC (Link to AICS-0)
- hiPSC with eGFP fused to tubulin (Link to AICS-12)
Overview
WORKFLOW
1. Robotic Matrigel Coating
|
2. Robotic Cell Seeding
|
3. Cell Maintenance (once per day)
|
4. Plate LayoutFor standard pipeline experiments, we use a plate layout with 10 wells for experimental lines. We add wells with 6 controls.
|
D. Quality Control at Well Level |
Vertical Divider
The pluri-potent hiPS cells have a natural tendency to differentiate. To ensure a well-defined cell state for pipeline experiments and to avoid cell health issues, we implemented a rigorous quality control regime. One important step is to assess colony health for whole wells before imaging. For this assessment, we use images acquired with a plate scanner and analyze them using CellProfiler.
|
Instrumentation
ROBOTICS
Cells are seeded and maintained on a Hamilton Microlab Star (Hamilton Company, Reno, NV, USA) robotic pipetting workstation equipped with:
- Arm with 8 channels for pipetting
- iSWAP plate handler
- Orbit barcode scanner
- 2 MultiFlex tilt modules
- CPAC heating/cooling unit
- Tip and tube carrier and plate stacking modules
- Star HEPA filter with UV light option
- Cytomat 24C (Thermo Fisher) incubator (37℃, 85% humidity and 5% CO2)
- Cytomat 6002 refrigerator (Thermo Fisher)
(4℃) - Venus 2 ver. 4. software
- Venus 1 Dynamic Scheduler 5.1
Overview
WORKFLOW
- Remove plate from incubator
- Place plate into plate scanner
- Acquire transmitted light images of all wells
- Move plate back into incubator
- Transfer images to cluster
- Process images with CellProfiler
- Segment colonies
- Remove small colonies
- Count colonies
- Rank wells and plates by count
WELLL QUALITY
To make sure that all cell images in a data set come from a visibly homogeneous population, we exclude wells that show:
Link: SOP: WTC culture v1.1
Link: SOP: Cell plating for imaging v1.0
- Balling colonies (>3 balling events/well)
- Differentiation (>2 identifiable events/well)
- High number of dead cells (>50% of cells/well)
Link: SOP: WTC culture v1.1
Link: SOP: Cell plating for imaging v1.0
E. Focus on Reproducible Cell States |
Vertical Divider
hiPSC grow in colonies. We manually select colonies and positions within colonies to ensure all images coming from cells residing in similar environments.
|
Instrumentation
PIPELINE MICROSCOPES
For pipeline operation we use three spinning disk microscopes (Carl Zeiss Cell Observer SD):
- Yokogawa CSU-X1 spinning disk head (micro-lens enhanced Nipkow disk, about 20,000 pinholes (1,000 in imaging area), beam shaping optics to improve illumination profile)
- 2x sCMOS cameras (Hamamatsu ORCA-Flash 4.0 V2+, 6.5µm x 6.5µm pixel size)
- 1.2 magnification in front of cameras
- Zeiss Plan-Apochromat 10x/0.45 air objective
- Zeiss C-Apochromat 100x/1.25 water immersion objective
- Laser lines (nominal laser power at laser):
- 405nm (50mW)
- 488nm (50mW)
- 561nm (50mW)
- 638nm (75mW)
- AxioObserver 7 motorized inverted microscope
- Definite Focus 2
- Pecon environmental chamber and controls (temperature, humidity control, CO2 control)
- Okolab stage incubator for 96-well plates (temperature, humidity, CO2)
- Motorized xy-stage
- Prior Piezo z-stage (100µm z-travel)
- Zeiss ZEN blue software
- Newport air table for vibration isolation
ADDITIONAL MICROSCOPY
We use additional widefield and confocal microscopes for R&D and assay development work. Some of them will be used in future imaging pipelines.
- 3i Yokogawa CSU-W1 spinning disk system
- Additional Carl Zeiss CSU-X1 spinning disk system for assay development
- Carl Zeiss LSM 800 with Airy Detector
Overview
WORKFLOW
- Equilibrate environmental chamber and stage incubator at 37°C, saturated humidity, 5% CO2
- Set-up system for transmitted light imaging
- Select wells based on colony quality and distribution
- Acquire overview images of wells with 10x/0.45 air objective using tile scan
- Manually select well-defined positions within colonies (edge, ridge, or center) and record within Zen software for automated imaging
- Select area without cells for background correction
- Remove 96-well plate and stain nucleus and cell membrane
COLONY QUALITY
We select areas within colonies for imaging based on colony and cell morphology:
Link: SOP WTC culture v1.1
Link: SOP Cell plating for imaging v1.0
- More than 300 cells per colony
- Well packed
- No lifting or signs of differentiation
Link: SOP WTC culture v1.1
Link: SOP Cell plating for imaging v1.0
F. Nucleus & Cell Membrane as References |
Vertical Divider
We want to understand the locations of various subcellular structures. To do this, we use dyes that illuminate the nucleus and cell membrane as positional references markers.
|
Overview
LABELING OF REFERENCE STRUCTURES
We visualize reference structures with organic dyes:
Complete, phenol red-free mTeSR1 culture media: 400 ml basal media with 100ml 5x supplement (Stem Cell Technologies custom order) with added 1% penicillin streptomycin (Gibco #15070-063)
- DNA: NucBlue Live ready probe reagent (Thermo Fisher, # R37605)
- Plasma Membrane: CellMask Deep Red plasma membrane stain (Thermo Fisher, # C10046)
Complete, phenol red-free mTeSR1 culture media: 400 ml basal media with 100ml 5x supplement (Stem Cell Technologies custom order) with added 1% penicillin streptomycin (Gibco #15070-063)
WORKFLOW
- Equilibrate phenol red free mTeSR1
- Add 60μl of NucBlue to 1 ml phenol red-free mTeSR
- Spin for 60min at 20,000g
- Add 100μl of NucBlue solution per well of 96-well plate
- Incubate at 37°C and 5% CO2 for 20min
- Dilute CellMask Deep Red with the NucBlue 1x solution (concentrations are stock dependent, see SOP)
- Add 100μl CellMask/NucBlue in addition to NucBlue solution to well
- Incubated at 37°C and 5% CO2 for 10min
- Wash with phenol red-free mTeSR
- Start high-resolution imaging immediately after washing
Integrated Cell
To determine the locations as well as morphologies of cellular structures, we generate images of cellular structures in reference to the nucleus (labeled with NucBlue) and the cell membrane (labeled with CellMask).
Integrated Cell >>
Integrated Cell >>
G. Imaging at the Resolution Limit |
Vertical Divider
For our applications, high image quality and information content are more important than high throughput. We optimized the microscope setup for optimal sectioning along the optical axes and maximum lateral resolution. We use different degrees of automation to balance reproducibility and manual quality control.
|
Instrumentation
WHAT MICROSCOPE TO USE?
The initial goals for the microscopy pipeline are:
We decided on spinning disk microscopes as the central part of the pipeline. To add more information to our data, we plan to build additional pipelines based on the Zeiss LSM 880 Airy fast, FCS, and image correlation techniques
- Quantify dimensions of structures
- Live cell imaging
- 3D imaging
- Optimized for optical quality
- High reproducibility
- Very robust
- Results are input for statistical modelling
- Wide-field microscopy: fast and reliable, but not 3D
- Deconvolution: fast and 3D, but potential to introduce processing artifacts
- Spinning Disk: robust, 3D, fast, but slower than wide-field microscopy
- Confocal point scanner (incl. NLO): robust and 3D, but slow (except for new implementations like the Carl Zeiss LSM 880 Airy fast)
- TIRF: very good optical sectioning, but limited to structures close to glass surface
- Conventional light sheet: low phototoxicity and high speed, but low spatial resolution and complex sample preparation
- Lattice light sheet: low phototoxicity and high imaging speed, but complex setup and slow sample exchange
- Structured Illumination Microscopy: high resolution but limited acquisition speed
- PALM/STORM: super-resolution but slow acquisition speed and limited 3D
- Scanning Probe Microscopy (STM, AFM, SNOM): super-resolution, but images only surfaces
- Fluorescence Correlation Spectroscopy (FCS) and Image Correlation Methods (e.g. RICS): very information rich but not suitable for structure visualization
We decided on spinning disk microscopes as the central part of the pipeline. To add more information to our data, we plan to build additional pipelines based on the Zeiss LSM 880 Airy fast, FCS, and image correlation techniques
THE ROLE OF PINHOLE
When imaging thick samples like monolayers of hiPSC, with a wide-field microscope, the structures in focus are blurred by out-of-focus light. A confocal microscope overcomes this limitation by rejecting out-of-focus light with a pinhole in the image plane in front of the detector. In the classical point scanning implementation, the system scans a focused laser beam line by line over the sample and records one pixel at a time. This setup creates very high-quality images but is relatively slow
http://www.microscopyu.com/articles/confocal/confocalintrobasics.html
Spinning disk microscopes overcome the speed limitation by scanning multiple beams in parallel. The image quality is lower than for point scanning microscopes and the optical sectioning capability is limited to thinner samples. hiPSC are thin (20μm) enough not to suffer from this limitation.
In a spinning disk microscope, the sample is illuminated through pinhole arrays in a fast spinning disk (Nipkow Disk). This disk creates a pattern of spots sweeping over the sample. Fluorescence or reflected light from the sample is collected through the same pinholes providing optical sectioning.
Typical implementations are:
Spinning disk microscopes overcome the speed limitation by scanning multiple beams in parallel. The image quality is lower than for point scanning microscopes and the optical sectioning capability is limited to thinner samples. hiPSC are thin (20μm) enough not to suffer from this limitation.
In a spinning disk microscope, the sample is illuminated through pinhole arrays in a fast spinning disk (Nipkow Disk). This disk creates a pattern of spots sweeping over the sample. Fluorescence or reflected light from the sample is collected through the same pinholes providing optical sectioning.
Typical implementations are:
- One disk with pinholes: simple and robust setup, but limited illumination efficiency and prone to stray light from back of disk
- Second disk with micro-lenses to focus excitation light at pinholes: better illumination efficiency but more complex and main dichroic close to imaging plane
- Micro-mirrors instead of lenses: issues with imaging quality
https://www.leica-microsystems.com/science-lab/hyvolution-super-resolution-imaging-with-a-confocal-microscope/
Optical sectioning (black, solid line), lateral resolution (black, dashed line), and light collection efficiency (green line), all depend on the diameter of the pinhole. Smaller pinholes decrease the thickness of the optical section, thus increase the resolution of the system along the optical axis. Less pronounced is the increase of lateral resolution for smaller pinholes. Smaller pinholes reject more light and thus the collected image is dimmer.
The diameter of a pinhole is measured in Airy Units (AU) and depends on the wavelength λ of the illumination light and the numerical aperture NA of the objective:
\[1AU = \frac{1.22\cdot \lambda }{NA} \]
For λ = 488nm (excitation for eGFP) and NA = 1.25 (high resolution water immersion objective) 1AU is 476nm. To calculate the physical dimension of the pinhole we multiply this value with the magnification of the optical system in front of the pinhole. E.g. a 1AU pinhole with a 100x Objective corresponds to a pinhole diameter of about 50µm.
For most applications, especially for live cell imaging, a pinhole diameter of 1AU is a good compromise between light efficiency and resolution. At one 1AU the system collects still 90% of the focal light intensity (see green curve) while the z-resolution does not increase significantly for a smaller pinhole.
Yokogawa manufactures two models of spinning disks with micro-lenses:
Literature:
Hideo, H., et al. (2008). New Technologies for CSU-X1 Confocal Scanner Unit. Yokogawa Technical Report English Edition. 45.
Wilhelm, S., et al. Confocal Laser Scanning Microscopy: Principles, Carl Zeiss.
Nyquist
Optical sectioning (black, solid line), lateral resolution (black, dashed line), and light collection efficiency (green line), all depend on the diameter of the pinhole. Smaller pinholes decrease the thickness of the optical section, thus increase the resolution of the system along the optical axis. Less pronounced is the increase of lateral resolution for smaller pinholes. Smaller pinholes reject more light and thus the collected image is dimmer.
The diameter of a pinhole is measured in Airy Units (AU) and depends on the wavelength λ of the illumination light and the numerical aperture NA of the objective:
\[1AU = \frac{1.22\cdot \lambda }{NA} \]
For λ = 488nm (excitation for eGFP) and NA = 1.25 (high resolution water immersion objective) 1AU is 476nm. To calculate the physical dimension of the pinhole we multiply this value with the magnification of the optical system in front of the pinhole. E.g. a 1AU pinhole with a 100x Objective corresponds to a pinhole diameter of about 50µm.
For most applications, especially for live cell imaging, a pinhole diameter of 1AU is a good compromise between light efficiency and resolution. At one 1AU the system collects still 90% of the focal light intensity (see green curve) while the z-resolution does not increase significantly for a smaller pinhole.
Yokogawa manufactures two models of spinning disks with micro-lenses:
- CSU-W1
- larger field of view
- Two disks with pinhole diameters of 50µm and 25µm
- Larger distance between pinholes to reduce lateral light crosstalk on thick samples
- CSU-X1
- One disk with pinhole diameters of 50µm
- Higher pinhole density
- Faster rotation speed of disk
- Robust
Literature:
Hideo, H., et al. (2008). New Technologies for CSU-X1 Confocal Scanner Unit. Yokogawa Technical Report English Edition. 45.
Wilhelm, S., et al. Confocal Laser Scanning Microscopy: Principles, Carl Zeiss.
Nyquist
NYQUIST SAMPLING
The resolution of a camera based light microscope is limited by optical diffraction and the pixel size of the camera chip.
The optical resolution of a microscope is given by the wavelength λ of the light used and the numerical aperture NA of the objective. According to Rayleigh the resolution limit is
\[r_{Airy} = 0.61\frac{\lambda }{NA}\]
For λ = 488nm (excitation for eGFP) and NA = 1.25 (high resolution water immersion objective), rAiry is about 240nm. Thus, the spinning disk microscope we use can distinguish two structures 240nm apart.
To be able to distinguish these structures in a microscope image, the pixel size of the camera needs to be small enough. In our system we use an objective with 100x magnification. Thus, two structures the optics of the microscope can just resolve, will appear 24μm (=240nm*100) apart on the camera chip. Each pixel on the chip we are using is 6.5µm x 6.5μm large and can very well resolve the structures.
A camera with such a small virtual pixel size collects only a few photons per pixel. The accuracy of each photon detector is limited by the fundamental quantum mechanical Poisson or statistical photon noise. The signal-to-noise ratio (SNR) for N photons counted is given by
\[SNR = \frac{N }{\sqrt{N}} = \sqrt{N}\]
Thus, images collected with cameras with larger pixels that receive more photons are less noisy.
Nyquist derived for non-periodic data that the sampling frequency should be at least 2.4x higher than the highest frequency in the data. For microscope images that translates into a pixel size that should be 2-3 times smaller than the optical resolution limit. To achieve this in our setup, we bin 2x2 pixels. This results in an effective pixel size of 13µm x 13µm. Using an additional 1.2x magnifier in front of the camera brings us very close to the Nyquist criterion while collecting sufficient photons to overcome Poisson noise. The camera has a similar resolution as the optical resolution of the system. If we use smaller pixels we will collect less photons per pixel and the signal will be noisier.
Literature:
Pawley, J. B. (2006). Handbook of Biological Confocal Microscopy. New York, Springer Science + Business Media. Chapters 1, 2, 4
The optical resolution of a microscope is given by the wavelength λ of the light used and the numerical aperture NA of the objective. According to Rayleigh the resolution limit is
\[r_{Airy} = 0.61\frac{\lambda }{NA}\]
For λ = 488nm (excitation for eGFP) and NA = 1.25 (high resolution water immersion objective), rAiry is about 240nm. Thus, the spinning disk microscope we use can distinguish two structures 240nm apart.
To be able to distinguish these structures in a microscope image, the pixel size of the camera needs to be small enough. In our system we use an objective with 100x magnification. Thus, two structures the optics of the microscope can just resolve, will appear 24μm (=240nm*100) apart on the camera chip. Each pixel on the chip we are using is 6.5µm x 6.5μm large and can very well resolve the structures.
A camera with such a small virtual pixel size collects only a few photons per pixel. The accuracy of each photon detector is limited by the fundamental quantum mechanical Poisson or statistical photon noise. The signal-to-noise ratio (SNR) for N photons counted is given by
\[SNR = \frac{N }{\sqrt{N}} = \sqrt{N}\]
Thus, images collected with cameras with larger pixels that receive more photons are less noisy.
Nyquist derived for non-periodic data that the sampling frequency should be at least 2.4x higher than the highest frequency in the data. For microscope images that translates into a pixel size that should be 2-3 times smaller than the optical resolution limit. To achieve this in our setup, we bin 2x2 pixels. This results in an effective pixel size of 13µm x 13µm. Using an additional 1.2x magnifier in front of the camera brings us very close to the Nyquist criterion while collecting sufficient photons to overcome Poisson noise. The camera has a similar resolution as the optical resolution of the system. If we use smaller pixels we will collect less photons per pixel and the signal will be noisier.
Literature:
Pawley, J. B. (2006). Handbook of Biological Confocal Microscopy. New York, Springer Science + Business Media. Chapters 1, 2, 4
Overview
WORKFLOW
- Keep environmental chamber and stage incubator at 37°C, saturated humidity, 5% CO2
- Setup system for
- Transmitted light (exposure time 100ms)
- Fluorescence (exposure time 200ms, empty channel with exposure time of 33ms between each channel)
laser power at sample measured with 10x/0.45NA
- i.405nm: 0.28mW
ii.488nm: 2.3mW
iii.638nm: 2.4mW - Switch to C-Apochromat 100x/1.25NA water immersion objectives
- Add immersion water
- Start system to automatically acquire z-stacks at positions selected above with 10x tile scan
- Acquire images of dye controls for flatfield correction
- Acquire image of beads
- Collect black reference image
Limit data acquisition to one hour per well to avoid toxic effects of organic dyes and light.
H. Large Scale Image Processing |
Vertical Divider
All cells are part of cell colonies. To analyze individual cells, we segment the cells based on the cell membrane label, the nucleus based on DNA stain, and structure of interest based on its EGFP signature. To process large numbers of images we implemented our processing pipeline on a cluster in CellProfiler under the control of Airflow.
|
Instrumentation
CELLPROFILER
CellProfiler is an open-source platform for high-content image processing. It was developed by the Broad Institute and is available at http://cellprofiler.org/.
CellProfiler provides:
CellProfiler provides:
- Advanced modules for image processing
- Measurement and feature extraction modules
- True 3D image processing (new in CellProfiler 3.0)
- Graphical user interface to build sequential processing pipelines
- Extensive help features
- Head-less mode for cluster processing
AIRFLOW
Airflow is a workflow management system. We use Airflow to combine processing modules written in Python, MatLab, and CellProfiler and to execute them on our in-house cluster.
Airflow is:
Airflow is:
- Based on Directed Acyclic Graphs (DAG)
- Open source (Apache)