Cameras for
Fluorescence
Microscopy How to choose
the best solution
Dr Orla Hanrahan
18
ISSUE 36 DECEMBER 2014
Choosing the correct camera
for fluorescence microscopy
can be difficult when faced
with the variety of sensors
which are available and when
the key parameters which
guarantee a sensitive camera
are not fully understood by
the user. These days, the life
science researcher can choose
between interline CCD’s, single
photon sensitive EMCCD’s and
superfast and sensitive sCMOS
cameras, a choice which can
be daunting if the camera
specifications are not known
or understood. Ultimately,
the
choice
of
camera
depends on the application
under investigation and the
requirements of the end-user,
i.e. fast frame rates, single
photon sensitivity, large field
of view or resolution. One of
the key factors to consider
when choosing the camera is
sensitivity.
19
Sensitivity is a key performance feature of any
market by Andor in 2000 and represents a significant
move sub-pixel distances during the time taken for
detection system. When assessing the sensitivity
leap forward in combining ultra-sensitivity with
rolling readout to traverse them.
of any CCD sensor it is the achievable Signal-to-
speed. EMCCD (Electron Multiplying Charge
Noise Ratio (SNR) which is of key importance. This
Coupled Device) cameras employ an on-chip
Sensitivity Comparisons
encapsulates the capacity to have the signal stand
amplification mechanism called “impact ionization”
out from the surrounding noise. The approach to
that multiplies even single photon events well above
ensure the best possible SNR is to a) develop a
the read noise floor. Importantly, this renders the
sensor with the highest possible quantum efficiency
EMCCD capable of single photon sensitivity at
and b) reduce the various sources of noise to a
fast frame rates. This attribute has rapidly gained
minimum.
recognition for EMCCD technology in demanding
camera has dominated cell microscopy.
This
1.4 MP sensor offers 5-6 electrons read noise at
approximately 11 frames/sec (down to 2.4 e- at 1
frame/ sec). The 6.45 μm pixel size represents a
distinct “sweet spot,” an ideal balance between the
photon collection area per pixel and the ability to
oversample the diffraction limit for better resolution
of fine intracellular detail.
The electron multiplying CCD (Charge Coupled
Device) camera was introduced into the scientific
20
ISSUE 36 DECEMBER 2014
of the number of photons incident per 6.5 µm
pixel. The read noise differences between the two
floor” detector, the down side is that the signal
RMS (Root Mean Square) read noise, without
amplification mechanism carries an additional noise
amplification, while reading out 5.5/4.2 megapixels
source called multiplicative noise which effectively
at 100 frames/ sec. Significantly, under extremely
increases the shot noise (or poisson noise) of the
low light conditions a sCMOS camera can be readily
signal by a factor of x1.41, manifested as an increase
operated with pixel binning, thus creating larger
in the pixel to pixel and frame to frame variability of
“super-pixels” for improved photon collection area
Figure 3a shows a further plot of SNR against
signals. Furthermore, the popular back-illuminated
when required. It is worth noting that, unlike with
number of photons per pixel, this time for sCMOS
EMCCD sensors are limited to 13 μm smallest pixel
interline CCDs, under this 2x2 binning condition
(non-binned and 2x2 binned) versus a 13 μm back-
size, which while offering good photon collection
the read noise of sCMOS will double, i.e. 1 e- rms
area, tends to limit the ability to resolve fine
becomes 2 e- rms. This noise increase is indeed
intracellular detail.
appropriately factored into comparative SNR plots
Scientific CMOS (Scientific Complementary Metal-
shown in this article.
Oxide-Semiconductor) technology is based on
By way of a unique dual amplifier sensor architecture,
a new generation of CMOS design and process
the sCMOS camera offers much higher dynamic
technology. This device type carries an advanced
range than would be expected from a CCD with
set of performance features that renders it
similarly small pixel size.This design circumvents the
entirely suited to high fidelity, quantitative scientific
need to choose between high or low gain amplifiers,
measurement. sCMOS can be considered unique
in that signal can be sampled simultaneously by both
in its ability to simultaneously deliver on many key
high and low gain amplifiers. As such, the lowest
performance parameters, whilst also overcoming
noise of the sensor can be harnessed alongside
the performance drawbacks that have traditionally
the maximum well depth, affording widest possible
been associated with conventional CMOS imagers.
dynamic range.
The 5.5/4.2 megapixel sensor offers a very large field
Figure 1: Plots comparing SNR of sCMOS vs Interline CCD cameras
(neither with pixel binning) as a function of incident photons per
square micron on the sensor. sCMOS provides higher SNR throughout
the intensity range due to a significantly lower read noise floor.The
specifications for sCMOS are based on the Andor Neo camera,
operated in rolling shutter mode with 560MHz readout speed and
fan cooled to -30°C.The CCD interline performance is based on the
specifications of the Andor Clara camera @ 20MHz readout speed,
fan cooled to -55°C. Exposures of 1 sec are assumed, such that the
darksignal contribution from each cooled camera is relatively small.
camera technologies. The x-axis is a representation
a well optimised EMCCD is close to a “zero noise
detection and photon counting experiments. While
For more than a decade the small pixel interline
photons per pixel for sCMOS vs. interline CCD
Figure 2: Comparative low light images taken with Andor Neo sCMOS
(1.2 electrons read noise @ 400 MHz) vs Andor Clara interline CCD
(5.5 electrons read noise @ 20MHz) of a fluorescently labelled
fixed cell using a CSU-X spinning disk confocal microscope (x60 oil
objective), each 100ms exposure, same laser power per channel,
displayed with same relative intensity scaling. Note, the field of view is
limited by the aperture size of the CSU-X spinning disk unit, which is
matched to the 1.4 megapixel interline sensor.
ultra-low light measurements, such as single molecule
Overview of camera
technology
Figure 1 shows a plot of SNR against number of
of view and high resolution, without compromising
read noise or frame rate and a 6.5 μm pixel size
is again ideally suited to cell microscopy. The read
noise is exceptionally low, even when compared
to the highest performance slow CCDs, but not
as low as the effective read noise of EMCCDs. The
sCMOS device can achieve down to 1 electrons
technologies is reflected in the notable separation
between the respective SNR curves across the
intensity range shown.
Figure 2 demonstrates
clear distinction in low light signal contrast, arising
from read noise differences between sCMOS
and interline CCD cameras. The comparative
images were recorded on a spinning disk confocal
fluorescence microscopy set-up (an inherently low
light modality).
The sCMOS can read out in both “rolling” and
“global” (snapshot) shutter modes (5.5 MP sensor;
4.2 MP sensor is a rolling shutter only sensor).
Dark signal generated in each mode is very similar,
but the absolute fastest frame rates and lowest read
noise can be achieved in rolling shutter mode, which
in reality will suit the vast majority of biological
imaging applications since objects will typically
Figure 3: Plots comparing SNR of sCMOS (non-binned and 2x2
binned) and larger pixel EMCCD (non-binned) as a function of incident
photons per square micron on the sensor.Two ranges are given, the
expanded low light range (B) is where the ‘zero read noise’ operation of
EMCCDs offers a distinct sensitivity advantage.
21
area (i.e. up to ~ 85 photons per 13 μm pixel). Here
exception to this may become apparent in the area
noise to approximately 2.5 electrons (b) a cooled
we can see that at signal intensities below 0.36
of super-resolution microscopy by single molecule
Sony interline sensor maintains an extremely
low dark current performance relative to cooled
photons per μm , the back-illuminated EMCCD will
localisation
indeed offer an improved SNR compared to the 2x2
Localization Microscopy (PALM), Stochastic Optical
binned sCMOS. This in effect is the region in which
Reconstruction Microscopy (STORM)), in that
the “zero read noise” properties of an EMCCD
since it is required to reach a threshold SNR in
outweigh the negative effect of multiplication
order to yield sufficient localisation accuracy, then
noise. By way of demonstration, Figure 4 shows
the threshold number of photons per exposure
comparative images recorded of an LED illuminated
required may occur at or beyond the cross-over
resolution chart in a light tight imaging chamber
region between the two technology curves. While
at a series of light intensities, marked in the Figure
the majority of live cell microscopy experiments
3 graphs. The images taken in the very low light
may eventually opt to utilise sCMOS technology,
regime, i.e. below the “cross-over” point, clearly
particularly to benefit from the 6.5 μm pixel
demonstrate the sensitivity advantage of the “zero
size combined with the larger field of view of
read noise” EMCCD in this range.
the 5.5/4.2 megapixel sensor, there will be some
2
techniques
(e.g.
Photo-Activated
sCMOS.
If uncertain about the right camera to choose for a
particular application it is always good practice to
evaluate a range of detectors to make an informed
decision. This article should at least provide an
overview of the different technologies which are
suitable for life science applications.
exceptionally low light instances in which a backFigure 4: Comparative low light images taken with Andor Neo
sCMOS (1.4 electrons read noise @ 560MHz) vs Andor iXon3 888
back-illuminated EMCCD (< 1e- read noise), under a range of photon
intensity conditions, indicated on the figure 3 plots. sCMOS was 2x2
binned in order to have the same effective pixel pitch (and light
collection area per pixel) as the 13 μm pixel of the EMCCD sensor.
The superior sensitivity of back-illuminated EMCCD under extremely
low light conditions is readily apparent. collection area per pixel) as the
13 μm pixel of the EMCCD sensor.The superior sensitivity of backilluminated EMCCD under extremely low light conditions is readily
apparent.
illuminated EMCCD camera. Consideration of the
Application Decisions
Fluorescence) microscopy or when measuring
low light signal intensities means that EMCCD
calcium flux from smooth muscle cells using the
technology will still be the detector of choice for
Nipkow spinning disk confocal modality (see Figure
a number of demanding applications. For example,
5).
the principal microscopy usage of EMCCDs to date
Furthermore, since sCMOS are not single photon
has been in the field of single molecule biophysics,
sensitive, EMCCD technology is still required for
and this is unlikely to change significantly. An
single photon counting experiments. From the
perspective on just how impacting the pixel size
becomes on the overall sensitivity performance.
Across the range shown, the 2x2 binned sCMOS
appears to exhibit a better SNR than the equivalent
pixel size (non-binned) back-illuminated EMCCD
camera. It is reasonable to ask why a “zero read
noise,” back-illuminated EMCCD camera would
exhibit lower SNR than a >2e- noise (when binned)
front-illuminated sCMOS device.The answer to this
lies in the additional multiplication noise imposed by
the EMCCD signal amplification mechanism.
However, the light range shown does not emphasise
the truly low light signal intensities and associated
provide a SNR advantage. Figure 3(b) shows the
same data, but expanded on the low light intensity
range between zero to 0.5 photons per μm2 sensor
22
ISSUE 36 DECEMBER 2014
example when using TIRF (Total Internal Reflection
Such raw sensitivity performance at extremely
curve for 2x2 binned sCMOS provides a keen
applications whereby EMCCD technology will
illuminated EMCCD will remain indispensable, for
Figure 5: Human Embryonic Kidney (HEK) cells transfected with DNA
plasmids encoding for Clathrin, fused to green fluorescent protein (GFPClathrin) and capsaicin receptors labelled with quantum dots (red)
imaged by Total Internal Reflection Fluorescence (TIRF) Microscopy.
Courtesy of Sanaz Sadegh and Prof. Diego Krapf, Colorado State
University. 1 megapixel field of view.
consideration of the sCMOS vs interline evidence,
Dr.
it seems more likely that interline CCD technology
Technology as an Application Specialist in Life
will eventually be displaced by sCMOS technology,
Science for the past four years. Her role brings
especially so in the field of cell microscopy, but
her in touch with all the latest developments
also in other applications such as high throughput
and innovations in both camera technology and
genome sequencing, high content screening and
microscopy applications. Previous to this Dr.
ophthalmology. One exception remains, whereby
Hanrahan managed a microscopy facility in the
a cooled interline CCD maintains an application
Biochemistry Department at Trinity College,
advantage over a cooled sCMOS camera. This
Dublin where she worked with a variety
relates to long exposure luminescence detection, e.g.
of microscopy set-ups including wide-field
bioluminescence microscopy, chemiluminesence gel
fluorescent microscopy, Point Scanning Confocal
documentation or in-vivo bioluminescent imaging,
Microscopy,
in that exposures greater than 60 secs will yield a
Microscopy.
lower overall noise floor (read noise and dark noise
stemmed from her Ph.D. in Biochemistry, during
combined) for the Sony interline CCD. There are
which she gained experience with numerous
two reasons for this; (a) at frame rates slower than
microscopy techniques used in her graduate
1 frame per sec, the Andor Clara interline readout
research.
Hanrahan
has
and
worked
Spinning
with
Disk
Andor
Confocal
Her interest in microscopy
can be reduced to 1MHz, which reduces the read
23