Review Article
PALM and STORM: Into large fields and high-throughput microscopy
with sCMOS detectors
Pedro Almada, Siân Culley, Ricardo Henriques ⇑
Quantitative Imaging and NanoBiophysics Group, MRC Laboratory for Molecular Cell Biology and Department of Cell and Developmental Biology, University College London,
Gower Street, WC1E 6BT London, United Kingdom
a r t i c l e
i n f o
Article history:
Received 11 March 2015
Received in revised form 28 May 2015
Accepted 3 June 2015
Available online xxxx
Keywords:
Single-molecule localization
Hardware
sCMOS
Homogenization
a b s t r a c t
Single Molecule Localization Microscopy (SMLM) techniques such as Photo-Activation Localization
Microscopy (PALM) and Stochastic Optical Reconstruction Microscopy (STORM) enable fluorescence
microscopy super-resolution: the overcoming of the resolution barrier imposed by the diffraction of light.
These techniques are based on acquiring hundreds or thousands of images of single molecules, locating
them and reconstructing a higher-resolution image from the high-precision localizations. These methods
generally imply a considerable trade-off between imaging speed and resolution, limiting their applicabil-
ity to high-throughput workflows. Recent advancements in scientific Complementary Metal-Oxide
Semiconductor (sCMOS) camera sensors and localization algorithms reduce the temporal requirements
for SMLM, pushing it toward high-throughput microscopy. Here we outline the decisions researchers face
when considering how to adapt hardware on a new system for sCMOS sensors with high-throughput in
mind.
� 2015 Elsevier Inc. All rights reserved.
1. Introduction to super-resolution microscopy
In past decades, the use of fluorescence microscopy has allowed
modern cell biology labs to achieve considerable milestones in our
knowledge of cell structure and organization [1]. However, infor-
mation obtained from fluorescence microscopy has been limited
by its resolution. A conventional optical microscope can only
resolve structures down to approximately 300 nm, depending on
numerical aperture and wavelength of light used. This is due to
the diffraction of light as it passes through the microscope’s optical
elements [2,3]. For greater resolution, electron microscopy (EM)
takes advantage of the <1 nm wavelength of electrons, resolving
sub-nanometer features [4]. Many biological molecules, structures
and processes exist on scales outside the resolution limit of con-
ventional fluorescence microscopy. For example, microtubules
have an outer diameter of 25 nm and a 12 nm inner diameter
[5]; chromatin fibers have a width of 30 nm [6]; the prokaryotic
ribosome has a diameter of c.a. 20 nm [7]. EM can resolve these
features but cannot accurately distinguish between molecular spe-
cies, whereas fluorescence microscopy can selectively label a wide
variety of biological molecules. Additionally, EM sample prepara-
tion is generally difficult, artifact prone, lengthy and strictly lim-
ited to fixed (dead) samples.
The field of super-resolution (SR) has emerged in recent years as
a solution possessing the benefits of fluorescence microscopy but
with resolution improved by an order of magnitude. Three major
SR methods are now experiencing widespread adoption and com-
mercial success: Stimulated Emission Depletion microscopy (STED)
[8], Structured Illumination Microscopy (SIM) [9] and Single
Molecule Localization Microscopy (SMLM) [10–12]. SMLM meth-
ods have relatively simple hardware requirements compared to
STED or SIM. STED is based on a confocal laser scanning system
with the requisite addition of a powerful second laser and special-
ized optics to generate a donut-shaped beam able to deplete fluo-
rophore emission in the periphery of the illuminating spot, thus
reducing the size of the emission spot and improving resolution
[8]. SIM also requires specialized hardware to shape the excitation
light such as motorized grids or spatial light modulators; this
method radially increases the high-frequencies of the acquired
images in a sequential manner and depends on computational pro-
cesses to merge this information into a super-resolution image
[13]. With SMLM however, the difference to a normal widefield
microscope mostly falls on a careful choice of its hardware compo-
nents and can be achieved on most commonly available widefield
http://dx.doi.org/10.1016/j.ymeth.2015.06.004
1046-2023/� 2015 Elsevier Inc. All rights reserved.
Abbreviations:
SR,
super-resolution;
SMLM,
Single-Molecule
Localization
Microscopy;
sCMOS,
scientific
Complementary
Metal-Oxide
Semiconductor;
EMCCD, Electron-Multiplying Charge Coupled Device; SNR, signal-to-noise ratio;
FOV, field of view; TIRF, Total Internal Reflection Fluorescence; FPS, frames per
second; HILO, highly inclined and laminated optical sheet; BFP, Back Focal Plane;
HT, high throughput; PSF, point spread function; ROI, Region of Interest.
⇑ Corresponding author.
E-mail address: r.henriques@ucl.ac.uk (R. Henriques).
Methods xxx (2015) xxx–xxx
Contents lists available at ScienceDirect
Methods
journal homepage: www.elsevier.com/locate/ymeth
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
microscopes with simple or no modifications [14]. For this reason
it remains a highly attractive option for researchers considering
SR methods. With recent developments in sCMOS camera technol-
ogy and localization algorithms, SMLM can now reach speeds at
full field of view (FOV) that are showing great promise for
high-throughput (HT) imaging. But there are challenges in imple-
menting this technology for researchers new to the field, as the lit-
erature assumes a certain practical body of knowledge. Here we
walk the reader through their options and challenges when consid-
ering to adapt their imaging systems or assembling hardware to
perform SMLM with a focus on large field of view, high throughput,
high speed imaging utilizing sCMOS cameras. As throughput can
also be increased by reducing the number of frames required for
a single reconstruction, we also discuss recently developed SMLM
algorithms capable of extracting meaningful results from fewer
frames. In particular, we direct the reader to such algorithms with
publicly available code.
1.1. Single Molecule Localization Microscopy (SMLM)
In a standard fluorescence microscope, the majority of excited
fluorophores will emit light near-simultaneously. If they are sepa-
rated by a distance smaller than the diffraction-limited resolution
of the microscope, then they cannot be accurately distinguished
from each other. But if they are at a distance greater than the res-
olution, they can be localized in an image with a precision which is
greater than the diffraction limited spatial resolution of the micro-
scope system; this precision generally scales with the square root
of the number of photons collected from the fluorophore [15].
Most current SMLM methods are based on ensuring that only a
small randomly changing subset of the fluorophores emits in each
frame, thus allowing images to be captured where individually
emitting fluorophores can be resolved. If only a sparse subset of
molecules are emitting in each frame, it is likely most are sepa-
rated from each other by a distance greater than the resolution
limit. By imaging multiple different fractions of the fluorophore
population over time a reconstructed image of the sample can be
made from all the localizations which will reveal the entire
structure.
Two
major
methods
exist:
Photoactivation
Localization
Microscopy (PALM [10,11]) and Stochastic Optical Reconstruction
Microscopy (STORM [12,16]). The major distinctions between the
two methods are the type of labels used, and how their fluores-
cence emission is modulated. PALM utilizes fluorophores whose
emitting stage can be directly controlled such as photoactivatable
proteins including PA-GFP, Dronpa and mEos2. A typical PALM
imaging cycle will require activation of a small, random subset of
fluorescent proteins by illumination with activating laser light.
This subset of molecules is then imaged and subsequently
bleached with a high power excitation laser, reducing their proba-
bility of appearing in further acquired frames. The cycle is then
repeated starting with the activation of another random subset
of proteins. This cycle is repeated many times until a sufficient
number of localizations can be made in order to reconstruct the
structure of interest [10,11].
STORM mostly uses fluorophores which stochastically turn
their emission on and off, which is usually achieved with a combi-
nation of using organic dyes, such as Alexa- and Cy-dyes, and ade-
quate buffer and illumination conditions [17]. These dyes are
typically capable of emitting an order of magnitude more photons
in a given time frame compared to the genetically encoded pho-
toactivatable proteins used in PALM [16]. As discussed above, the
lower number of photons collected in PALM will subsequently
limit resolution [15]. The inherent photochemistry of fluorophores
commonly used in STORM allows them to easily enter reversible
long-lived dark states (tens of milliseconds, to several seconds, to
several hours in an oxygen-free environment) which allows their
emission to be modulated [16,18–20]. These states are distinct
from bleaching, which is an irreversible process. One of the sim-
plest STORM methods, ‘‘direct STORM’’ (dSTORM) is based on
inducing switching behavior by exciting the fluorophore in the
presence of a buffer solution containing thiols (such as mercap-
toethylamine (MEA) [21], beta-mercaptoethanol (BME), or recently
TCEP [22]) and/or an oxygen scavenging system (ROXS [23]).
Comprehensive evaluations of the various fluorophores and buffers
used in dSTORM are available elsewhere [16,24]. Dyes can also be
induced to re-enter an activated state through excitation with UV
or near UV light such as that provided by a 405 nm laser commonly
present on most microscopes [16].
The basic hardware requirements for SMLM are a widefield or
Total Internal Reflection Fluorescence (TIRF) microscope, a sensi-
tive camera and a sufficiently powerful illumination source, gener-
ally lasers (as opposed to lamps or Light Emitting Diodes (LEDs)
commonly used in conventional widefield microscopy) [3]. In
PALM two lasers are usually used: an activating laser and an exci-
tation/bleaching laser, both of which must match the spectral
properties of the fluorophore of interest. The overwhelming major-
ity of photoactivatable fluorophores can be activated with UV or
near UV laser light such as that from 405 nm lasers [25].
However, should the protein activate spontaneously, only an exci-
tation laser is required as in the ‘‘PALM with independently run-
ning acquisition’’ (PALMIRA) method [26]. In dSTORM, although a
UV or near UV laser can be used to reactivate the fluorophores,
modulation of the photoswitching behavior can be achieved solely
by adjusting the excitation laser power. This means that the only
major hardware difference between PALM and STORM systems is
the power of the lasers, as higher laser intensities are usually
required in STORM than PALM [16].
1.2. Speed and throughput limitations of HT super-resolution
microscopy
For HT microscopy, both reducing acquisition time and increas-
ing FOV will increase the throughput. Any reduction of acquisition
time or the need to repeat steps will considerably accelerate an HT
project given that a typical workflow will require repeating the
same actions hundreds, if not thousands of times. When designing
an HT project that might require super-resolution, the choice of SR
method has to take into account throughput, ease of sample prepa-
ration and instrument complexity.
Of the three major techniques (Table 1), SIM imposes the fewest
requirements on sample preparation making it highly attractive for
HT. Although SIM requires a high SNR in its acquired images to
avoid artifacts in reconstructions, it remains the fastest SR tech-
nique and commercial instruments already make use of large
FOV sCMOS cameras. However, SIM also remains the most
resolution-limited technique, being capable of only a doubling of
resolution
with
currently
available
instruments,
limiting
its
Table 1
Qualitative comparison of the three most common super-resolution methods. The
resolutions mentioned refer to what is routinely achievable with biological samples;
frame-rate compares best-case scenarios to acquire large FOV areas with the above
mentioned resolutions; instrument complexity is an evaluation of how hard large
FOV systems are to implement in modern labs, either manually or with commercial
systems.
SIM
STED
SMLM
Resolution (nm)
100
70–60
30–20
Frame-rate
10 hz
30 hz
0.1 hz
Instrument complexity
High
Very high
Low
Sample requirements
Very low
High
Medium
2
P. Almada et al. / Methods xxx (2015) xxx–xxx
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
applicability. STED has theoretically unlimited resolution but is
slowed by the need to scan a spot over the area to be imaged;
reducing this area increases the available speed at which SR images
can be acquired. For example, Westphal et al. [27] achieved 28
frames per second imaging of synaptic vesicles at 60 nm resolution
by restricting the FOV to a 2.8 � 1.5 lm region. One way to
increase the speed of STED is to parallelize and scan several spots
over a large area. Recently, Bergermann et al. [28] scanned 2500
spots over a 20 lm diameter region resulting in STED images of
vimentin with 68 nm resolution. Theoretically, the whole FOV of
an objective can be imaged in this way. However, such parallelized
systems are restricted to the few labs who can assemble them and
furthermore, STED microscopy in general imposes considerable
limitations on dye choice, is limited two 2 colors for multicolor
imaging and most labs find it hard to improve the resolution to
below 60 nm on biological samples.
Compared to SIM and STED, SMLM is considerably slower. Each
reconstructed image requires thousands of frames to be acquired
in order to collect enough photons for a sufficient number of fluo-
rophore localizations (commonly millions to billions) to accurately
represent cellular structures, since only a few molecules (generally
hundreds) should be emitting in each frame. However, it requires
less specialist hardware than STED or SIM, multicolor imaging is
straightforward, and resolutions of 30 nm in biological samples
can be achieved with some regularity, which makes it a more
attractive technique for HT projects involving fine structural
details. Indeed, SMLM has already been used in HT studies:
Holden et al. [29] used HT PALM with 35 nm resolution to reveal
the in vivo 3D organization of FtsZ in the bacterial Z ring. SMLM
is also ideally suited for imaging signaling events, a common target
for screens and HT projects. For example, Soares et al. [30] used HT
dSTORM with 19 nm localization precisions to reveal signaling ter-
ritories controlling T Cell activation.
The overwhelming majority of publicly available SMLM analysis
software assumes that individual fluorophores do not spatiotem-
porally overlap extensively, and so the speed at which frames
can be acquired depends on the amount of time a molecule spends
on the emitting state. The shorter this ‘emitting state’ lasts, the fas-
ter the possible frame-rate is [16,31]. The lifetime of this state can
be reduced by increasing the excitation laser intensity, or by using
specialist buffers in the case of STORM [16]. The limiting step then
becomes the read-out speed of the camera, one of the advantages
of sCMOS cameras, and how many frames need to be acquired
per reconstruction. While reducing the exposure time reduces
the number of collected photons per molecule per frame, it also
reduces the number of background photons collected, reducing
the impact of low exposure times on resolution. This has been
recently demonstrated and quantified by Lin et al. [32] for Alexa
647, which show speeds of 1600 FPS (for a cropped sCMOS FOV)
can achieve resolutions comparable to slower acquisition speeds
and reaching satisfactory localization densities in a matter of sec-
onds. This resulted in a 16-fold reduction in acquisition time when
compared to the full FOV speed at 100 FPS (Table 2). However, the
throughput was limited as, to achieve such speeds, the FOV had to
be reduced to a square area 256 times smaller than the full FOV.
Table 2 details the relationship between cropping the FOV of an
sCMOS camera and its resultant throughput; from this it is evident
that trading off area for speed does not increase throughput. In
addition, each fold-decrease in area will require moving parts to
cover an area equivalent to the whole FOV, slowing down an HT
workflow. As such, using the whole FOV of the sCMOS is a more
attractive proposition for HT SMLM.
More importantly, new algorithms have been and are currently
under development that are capable of handling overlapping fluo-
rophores, allowing for use of fluorophores with less favorable
blinking statistics in high-speed SMLM (see Section 2.6.2), as well
as reducing the total number of frames needed to acquire. The total
time to acquire frames for a single reconstruction will depend on
the density of recovered fluorophores and current algorithms
require tens of thousands of fluorophores: even at 1600 FPS a
30,000 frame acquisition will take around 18 s, clearly slower than
SIM or STED. One advantage for HT, however, is 3D SMLM. 3D
methods for SMLM extract the extra dimension from data from a
single focus position (see Section 2.3.1), which means extra infor-
mation can be extracted at no cost to the acquisition time, some-
thing neither SIM nor STED are capable of.
Developments in sensor technology, dye chemistry, buffer con-
ditions and algorithms should all lead to ever shorter acquisition
times, making SMLM comparable to the speeds of SIM and STED
but with enhanced resolution and simplified sample and instru-
ment requirements. Below we will guide users on the considera-
tions of assembling the hardware necessary to take advantage of
sCMOS cameras for large FOV high-throughput SMLM.
2. Basis of assembling a high-throughput SMLM system
The main focus of the hardware considerations described in this
review is designing a HT SMLM system which takes advantages of
sCMOS cameras: namely their large field of view and short expo-
sure times enabling higher throughput.
2.1. Excitation path
The localization precision of individual fluorophores is highly
perturbed by the unwanted background intensity of the image,
as differentiating between photons coming from the fluorophore
to be localized and photons from other sources becomes more
challenging [15]. The ratio between the true signal from the fluo-
rophore of interest relative to the background photons and noise
is generally defined as the signal-to-noise ratio (SNR). The higher
the contribution from the background, the lower the SNR and
therefore the precision. To reduce the contribution of unwanted
background,
most
SMLM
applications
use
Total
Internal
Reflection Fluorescence (TIRF), a technique that significantly con-
strains the excitation depth at the surface of a coverslip [33,34].
Table 2
Calculating expected total throughput after reducing the FOV by either cropping whole lines (l) or choosing square regions of interest (s). sCMOS technology only yields a speed
increase for each cropped line, not columns so the speed increase from cropping square regions is the same as when reducing the corresponding number of lines. Fold area
decrease (l) = 20482/nlines*2048; fold area decrease (s) = 20482/nlines
2
; throughput decrease (l) = (nlines*2048)/20482; throughput decrease (s) = (nlines
2
)/20482; speed increase = FPS/
100; total throughput (TT) = throughput decrease * speed increase.
nlines
FPS
Fold area decrease (l)
Fold area decrease (s)
Throughput decrease (l)
Throughput decrease (s)
Speed increase
TT (l)
TT (S)
2048
100
1
1
1
1
1
1
1
1024
200
2
4
0.5
0.25
2
1
0.5
512
400
4
16
0.25
0.0625
4
1
0.25
256
800
8
64
0.125
0.015625
8
1
0.125
128
1600
16
256
0.0625
0.00390625
16
1
0.0625
64
3200
332
1024
0.03125
0.00097656
32
1
0.03125
P. Almada et al. / Methods xxx (2015) xxx–xxx
3
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
This is achieved through total internal reflection at the interface
between two media with different refractive indices (i.e. the glass
coverslip and the sample in aqueous media), where an evanescent
wave is created whose intensity decays exponentially with dis-
tance from the interface. As a result, in TIRF microscopy there is
selective excitation of molecules very close to the coverslip up to
a depth of 60–100 nm [35], reducing out-of-focus fluorescence
and enhancing the SNR [10,12,21]. For total internal reflection to
occur, the excitation light must be incident on the interface at
the ‘critical angle’ or higher [36]; this angle is defined by the differ-
ence between the refractive indices at the interface. The numerical
aperture (NA) of the objective determines the maximum angle at
which light can be incident on the sample, and as such TIRF typi-
cally requires an objective with an NA of 1.33 or higher [36]. In
practice, TIRF objectives have higher NAs as these are capable of
readily exceeding the critical angle. An added advantage is that
high NA objectives collect more light and therefore provide better
SNR at the same magnification [35]. Given that most commercially
available TIRF systems automatically fulfill the criteria for PALM or
STORM imaging (laser illumination, sensitive cameras, high NA
objectives), these can be and are used for SMLM. A similar tech-
nique called highly inclined and laminated optical sheet (HILO)
microscopy makes use of a highly inclined light beam to generate
a thin optical sheet that penetrates the sample at a shallow angle.
HILO has been used effectively in single-molecule fluorescence
nanoscopy of the cell nucleus with high signal/background ratios
[37]. Table 3 offers a qualitative comparison of using either wide-
field illumination, TIRF or HILO.
Objective lenses for large FOV imaging should ideally have cor-
rections for ‘‘flatness of field’’, as uncorrected lenses will have sig-
nificant aberrations and intensity variations between the center
and the edges of the FOV. Objectives with such FOV corrections
are labeled ‘‘Plan’’ and should have an intensity difference of no
more than 20% between the center and the edge of the FOV [38].
However, high NA objectives used for TIRF, as discussed above,
often do not correct for flatness of field as effectively as lower
NA objectives. This results in TIRF being limited to a smaller FOV
at the center which is actually in focus and has minimal chromatic
and spherical aberrations; uncorrected areas will have severely
reduced SNR [39]. While manufacturers may claim to have Plan
corrected objectives with NA > 1.4, this correction may be only
up
to
a
quality
standard
that
may
vary
between
brands.
Interested buyers should require the manufacturer to disclose
the extent of the correction. Thus, while sCMOS cameras enable a
much larger FOV to be acquired for high-throughput imaging, this
will actually be limited by the objective corrections as the FOV may
be restricted to the Plan corrected central region of the objective.
2.1.1. Laser launching
Since SMLM generally requires high illumination intensities,
input of the laser beams into the microscope objective should be
done with maximum possible efficiency. There are two immediate
choices: whether to use mirrors to launch the laser directly (in free
space) to the microscope, or by launching through an optical fiber
(Fig. 1). While launching directly minimizes optical light losses,
fiber launching has a few advantages: it provides long-term tem-
porally stable illumination, it simplifies TIRF/HILO imaging, and
allows for simple methods of making the illumination homogenous
across the FOV. Two types of fibers are generally used for micro-
scopy: single- and multimode. Multimode fibers offer several
advantages over single mode fibers: their large core size means
they are easy to align and there is negligible loss of laser power.
They also maintain alignment over longer periods of time.
However,
while
the
output
of
a
single
mode
fiber
is
a
near-perfect Gaussian profile, a multimode fiber will introduce
speckling and a non-uniform profile [40]. It is this property, how-
ever, that allows them to be used for homogenizing the illumina-
tion over the FOV (Section 2.1.2).
The simplest method for TIRF/HILO imaging with epifluores-
cence microscopes requires the laser to be focused onto a spot
on the back-focal-plane of the objective (BFP) (Fig. 1); the laser will
exit the objective collimated. If the spot is translated on the BFP,
the angle of the laser beam at the exit of the objective will change,
enabling TIRF illumination [35,41]. For HILO [37] an extra mirror is
used to adjust the tilt of the spot on the BFP, adjusting the depth of
the sheet on the imaged area [42]. Placing a fiber coupler on an XYZ
translation
stage
(available
from
Thorlabs,
Inc.
or
Newport
Corporation) allows the laser spot to easily be focused and trans-
lated on the BFP. Free space launching requires a method to adjust
the position (and tilt for HILO) of the spot on the BFP: for more
information on how to achieve this, see [36,43,44].
Of note, most laser light will be linearly polarized and only
molecules which are aligned with the polarization axis of the laser
will be excited efficiently. In order to excite all molecules present
in the sample, the laser light should be converted from linearly
polarized to circularly polarized [39], which can be achieved by
placing a quarter-wave plate in the light path [45]. Another point
to consider is that back-reflections of the laser onto the laser cavity
will destabilize the laser. Optical isolators using polarizing optics
and reduce this problem but a simpler method is to use angled pol-
ished fiber connectors (FC/APC), the reflections of which will not go
back through the optical path and into the laser cavity.
2.1.2. Large FOV illumination
New sCMOS camera technologies are capable of imaging a
2048 � 2048 pixel FOV at 100 frames per second (FPS). This is in
comparison to EMCCD cameras, traditionally used for single mole-
cule imaging due to their high sensitivity [46], which can image a
512 � 512 pixel FOV at a maximum speed of 56 FPS. For example,
Soares et al. [30] used an EMCCD to image 1–6 cells for each
512 � 512 pixel FOV. Using the full 2048 � 2048 chip of an
sCMOS camera would yield a 15-fold larger area and, therefore,
number of cells imaged per FOV. With the doubling of full-frame
speed, this corresponds to a 30-fold increase in throughput, a clear
advantage of sCMOS cameras for HT SMLM. The large FOV afforded
by sCMOS cameras thus makes it important that images are sam-
pled correctly and that the whole FOV is illuminated appropriately.
The total magnification of the system will be the product of the
objective magnification (which takes into account the magnifica-
tion of the tube lens) and any extra magnification present in the
system. The size of a pixel in the projected image will be the phys-
ical camera pixel size divided by the total magnification. The ideal
pixel size for a widefield image should be about 2.3� smaller than
the diffraction-limited resolution of the microscope (according to
the Nyquist criterion) [39]. For a 1.4 NA objective imaging a mole-
cule with 525 nm emission, the diffraction-limited resolution is
230 nm and hence the Nyquist pixel size is 100 nm [39]. As a
result,
whilst
an
EMCCD
camera
(with
a
pixel
size
of
Table 3
Qualitative comparison of the performance of the three mentioned illumination
methods; system complexity refers to complexity when implementing in custom
systems; efficiency refers to power loss before the entering objective; sample
penetration refers to how deep in the sample imaging can be performed; restrictions
refers to whether the technique imposes restrictions on the usable refractive index of
the sample medium and/or NA of the objective.
TIRF
HILO
Widefield
System complexity
Moderate
Moderate
Low
Background reduction
Best
Good
None
Efficiency
Reasonable
Reasonable
Best
Sample penetration
<�100 nm
<�50 lm
<�100 lm
Restrictions
Media; NA
Media; NA
None
4
P. Almada et al. / Methods xxx (2015) xxx–xxx
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
16 � 16 lm2) typically requires a 100x objective to ensure ideal
pixel sampling, an sCMOS camera (pixel size 6.5 � 6.5 lm2) only
requires 63� magnification to satisfy the Nyquist criterion, which
allows for imaging of a larger FOV.
When the illuminating laser spot is focused onto the BFP, as used
in TIRF/HILO imaging, the intensity distribution of the illumination
at the sample will be an image of the beam shape. For most lasers
this is Gaussian, or at least non-uniform [41]. This non-uniform
power distribution results in different fluorophore dynamics in dif-
ferent parts of the image. To increase illumination homogeneity
across the FOV, most SMLM systems illuminate a much larger field
than the one being imaged. However, this reduces the laser power
density, requiring the laser power to be increased concomitantly.
Inhomogeneities across the FOV can also arise from the laser
launching method, as launching using multimode fibers introduces
intense ‘speckles’ in the illumination field [40]. For large FOV
imaging it is important to achieve a large, homogenous illumination
field whilst maintaining suitable on-sample laser power.
Particularly interesting methods for homogenizing illumination
involve manipulating the light path as it travels through a
multimode fiber, and such methods have the advantages of negli-
gible power loss and an output suitable for TIRF imaging. While the
output of a multimode fiber is inherently uniform (‘‘flat-top’’),
intense speckles and interference fringes can appear as a result
of the coherence of laser light [47,48]. Simple methods for reducing
these inhomogeneities involve vibrating, agitating and/or bending
a multimode fiber at a relatively small radius (but not small
enough to be damaged [49,50]). Provided that the vibration fre-
quency is 10� higher than the acquisition rate of the camera, dif-
ferent outputs of the fiber will blur together and create a
uniform illumination field [50]. As such, piezoelectric elements
are typically required to vibrate the fiber [50,51], although in prin-
ciple any vibration source can be used. For example, beam homog-
enization has been achieved by coiling a fiber around a laser
cooling fan [52]. There are also some fibers with a square output,
optimizing the laser illumination to the square area of camera sen-
sors. These fibers are not as easily available as fibers with circular
output, however.
In order to implement a vibrating fiber system, a careful choice
of multimode fiber and vibrating system is required. For example,
Fig. 1. Schematic of the hardware options available. Pictured are the all the major options when assembling a high-speed SMLM microscope. Different colored arrows
represent the possible next step in the optical path and numbers in brackets are the corresponding sections in the text. laser combiner: lasers can be steered with mirrors (M),
combined using dichroic mirrors (DM) and either modulated directly or through an AOTF. Of note, the AOTF will output a diffracted beam (DB) which is directed to a beam
dump (BD). Also, correcting laser divergence using two lenses with the same focal length (L1 and L2) improves transmittance through fibers and cleanup filters (CF) will
minimize background excitation. Beam homogenizers: after the combiner, there are several options to homogenize the beam: a rotation motor (RM) with a wedge (W) or
diffuser (D), or a combination of both can be used with a multimode fiber (F); an engineered diffuser (ED) can be used along with a lens to collimate (L5) and another lens (L6)
to focus on the BFP inside the microscope (MIC); a set of two microlenses (Koehler integrator) can be used before L5 in order to homogenize the FOV and optionally with a
rotating diffuser; a refractive beam shaper (RBS) can be used for TIRF injection and its input can be from a single-mode fiber or directly from the laser combiner (F); or by
simply vibrating a multimode fiber (F) with a Piezo element (P) or any other source of high-frequency vibrations. Fiber launch: when using a fiber, the laser beam needs to be
focused on the fiber coupler (FC) using a lens with an NA equal to the fiber’s (LNL). TIRF injection: TIRF illumination can be achieved by, for example, mounting a mirror, beam
expander (divergent lens L3 with collimating lens L5) and L6 on a translating stage (ST) to move the focus of the laser on the BFP. Using a quarter-wave plate (WP) is advised
to circularly polarize the laser light. For HILO, a few extra lenses are required: see [42]. Emission detection path: the sample’s emission can be split to two cameras, two sides
of a single camera or a combination of both. Emission is separated from the excitation using a DM and most microscopes will have a tube lens (TL) placed immediately after.
An optical relay is placed two focal lengths away from the TL and consists of two lenses (RL1 and RL2). The emission light can then be split using a DM or a 50/50 beam splitter
(BS) (only a DM is shown in the example) which splits the intensity of the light in half, in two directions. The DM or BS can be placed between RL1 and RL2 to split to two
cameras or between RL2 and the camera to split to two different halves of the chip. The shown example with a single mirror will introduce focus shifts in the two images,
which enables 3D imaging. Alternatively, 3D imaging can be accomplished by placing a 10 m focal length cylindrical lens between the objective and the TL (CL1) or a 1 m focal
length lens between RL2 and the camera (CL2).
P. Almada et al. / Methods xxx (2015) xxx–xxx
5
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
fibers with large core sizes transmit more modes and are more sen-
sitive to vibrations [45]. The NA of the fiber will also limit the
achievable
collimation
quality
of
the
laser
after
the
fiber-collimating lens. The focal length of this lens and the NA of
the fiber will offer a trade-off between size of the beam and its col-
limation. Of note for multimode fibers, the NA of the lens that
focuses the beam onto the fiber should match the NA of the fiber
[53]. If the lens NA is too low, there will not be enough speckles
generated to be mixed and homogenize, and the beam output will
be more Gaussian; if the lens NA is too high the difference from the
fiber NA will introduce a power loss proportional to that difference.
Also, as the frequency at which the fiber is vibrated limits the max-
imum imaging frequency over which the FOV remains uniform,
piezoelectric elements with resonant frequencies of 20 kHz and
40 kHz (Premier Farnell plc, RS Components, Ltd.) will limit the
maximum imaging frequency to 2 kHz and 4 kHz, respectively
[50]. However, such imaging frequencies should be sufficient for
most SMLM applications. If the amplitudes of vibrations from
piezoelectric elements are not sufficient, several piezos may be
combined. To reduce speckles, the light path can also be manipu-
lated prior to multimode fiber injection, for example by inserting
a spinning glass wedge [47,48] or rotating plastic petri dish [54]
into the free space light path; these methods can be combined with
a vibrating fiber system if the amplitude of vibrations from piezos
are insufficient. Again, spinning or rotation should be performed at
a higher frequency than the acquisition rate.
There are other optical methods for homogenizing illumination
available. These include refractive beam shapers which output an
arbitrarily
defined
illumination
profile
[55]
(piShaper
from
AdlOptica Optical Systems GmbH, refractive beam shapers from
Newport Corporation) when given a defined input beam, and var-
ious diffractive beam shapers such as Koehler Integrators [56–58]
(available from RPC Photonics or SÜSS MicroTec AG) and engi-
neered diffusers [59,60]. However, the latter two methods are
not suitable for TIRF/HILO as they rely on the creation of multiple
laser spots. The refractive beam shapers are very simple to inte-
grate but are expensive and require high-quality beam shapes to
avoid losses. A qualitative evaluation of beam homogenization
methods is provided in Table 4.
2.1.3. Laser choice and modulation for SMLM
To increase the speed of PALM acquisition, cycling of the
405 nm activation laser with the excitation laser required very fast
laser modulation [17]. Modern laser microscopes typically use
compact solid-state or diode lasers, which have high stability and
offer high-frequency direct modulation of laser intensity [39]. If
the
laser
cannot
be
modulated
fast
enough,
Acousto-Optic
Tunable Filters (AOTF) or Acousto-Optic Modulators (AOM) can
be used (Fig. 1) [39]: AOMs are used to modulate intensity of a sin-
gle laser, whereas AOTFs can modulate the intensity of several
lasers in a wavelength selective manner [39]. If the laser can be
directly modulated, an AOTF might not be necessary but electronic
controllers
might
be
required
to
modulate
the
lasers
at
high-frequency,
such
as
those
available
from
National
Instruments
Corp.,
Measurement
Computing
Corp.
or
the
open-source Arduino platform. If fast dSTORM imaging is to be per-
formed, the 405 nm laser can be continuously present with the
excitation laser, making fast modulation unnecessary. However,
fast dSTORM imaging requires powerful lasers (laser intensity at
the sample in the order of 1–100 kW/cm2 [32]) since excitation
intensity is inversely proportional to ‘‘on-times’’ [16]. For example,
Huang et al. [46] use a 500 mW 642 nm laser to illuminate rela-
tively small FOV’s to be able to achieve power densities of 5–
18 kW/cm2. Such lasers are not easily available or cheap, however.
Telecommunication laser companies do offer models that can be
useful for SMLM; Section 3 mentions a few such companies. High
power lasers do however require additional safety considerations
and precautions to avoid damage to components. Users should
be intimately familiar with the safety recommendations that come
with their laser model and sufficient information should be avail-
able from the local safety officer. When imaging for long periods
these lasers also warm the objective considerably which can dam-
age the coatings of the optics and degrade the cement used for the
optical components in the objective. The authors have seen this
first-hand after imaging for one hour with all lasers turned on at
full power. Taking regular breaks to cool the components should
help with longevity.
Another important property of the laser is its beam shape, or
the transverse electromagnetic mode (TEM) of the beam. The sim-
plest profile corresponds to a Gaussian distribution, referred to in
specifications as TEM00 [45]. Beams with a TEM00 profile are
preferable for SMLM as they are simpler to homogenize than other
modes, and their single peak allows for better definition of the cor-
rect illumination angle in TIRF/HILO imaging. The TEM00 profile
can also be achieved by transmitting any laser light through a
single-mode fiber [45]. However, the disadvantages of using a
single-mode fiber as opposed to a multimode fiber are that extra
optics will be required to homogenize the output, and that cor-
rectly aligning single-mode fibers is substantially harder with typ-
ical power losses of 30–40% [39]. Whilst most SMLM systems
simply magnify TEM00 illumination to a much larger area than
the FOV to achieve uniformity, for large FOV illumination the beam
homogenization techniques described in Section 2.1.2 are recom-
mended to prevent losing power density at the sample. Beam
homogenization also allows for lasers with lower beam quality to
be used, reducing costs.
2.2. Detection optics and frame
2.2.1. Focus drift correction
Regardless of imaging speed, when imaging for several thou-
sands of frames over tens of minutes, sample drift becomes a prob-
lem. This is due to different thermal responses between different
materials in the microscope body. While lateral drift can be cor-
rected post-acquisition by tracking fiducial markers (such as fluo-
rescent beads) included in the sample, axial drift results in the
sample moving out of focus, rendering the acquisition unusable.
Axial drift can be addressed via correction or prevention with hard-
ware and/or software solutions. Software solutions rely on motor-
ized focusing systems and a microscopy acquisition package that
can stop the acquisition at regular intervals, take images a few
microns above and below the current focus, and determine the
position of the previous focus. Most commercial acquisition pack-
ages offer this, as well as the free and open-source MicroManager
microscopy control software [61]. The most common hardware
solution is based on integrating an infrared LED light into the
Table 4
Qualitative evaluation of homogenization methods. Summary table of all four
methods outlined above, highlighting the uniformity of the output, how much laser
power is lost by the implementation, whether TIRF is possible, how expensive it is in
relation to the other methods and how complex the method is to implement on most
modern SMLM systems. MMF – multimode fiber, diff. – diffuser.
Method
Uniformity
Power
loss
TIRF
Cost
Complexity
Refractive beam
shaper
Excellent
Low
Yes
$$$
Low
Vibrated MMF
Moderate
Low
Yes
$$
Moderate
MMF + rotating
wedge/diff.
Very good
Moderate
Yes
$$
Moderate
Engineered diffuser
Moderate
Moderate
No
$
Low
Koehler integrator
Very good
High
No
$$
High
6
P. Almada et al. / Methods xxx (2015) xxx–xxx
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
microscope light path such that some light is reflected from the
interface of the coverslip and sample media. Changes in this signal
will indicate when the focus is drifting, and the microscope can
then compensate for this by moving the focus by the same amount
as the measured drift [62]. Such systems work best when
water-based media are used at the interface. This solution is avail-
able from the ‘‘big four’’ microscope hardware manufacturers
(Table 5) and is offered as an upgrade to modern microscope bod-
ies. Of note, having informally tested several of these systems, we
find the Nikon Perfect Focus System to be the best in both simplic-
ity and performance. There are also add-on solutions for older or
homebuilt systems such as the Applied Scientific Instrumentation
CRISP system or the open-source hardware pgFocus system [63].
2.2.2. Accelerating high throughput and multicolor microscopy
High throughput microscopy relies on fast and automated
acquisition. The automation is achieved by motorizing the sample
stage, illumination and emission capture. Given that a HT micro-
scopy workflow involves repeating the same mechanical move-
ments several hundreds of times, any speed saved on movement
significantly increases throughput. All four of the major micro-
scope manufacturers offer fully automated microscopes with dif-
ferent automation options for the same model. At the heart of
the HT microscope is the sample stage and most suppliers offer fast
XY translation stages. More importantly, the default motorized
focusing systems in most bodies are relatively slow; this can be
overcome by either adding a piezo translator to move the objec-
tive, adding a stage-top piezo Z translator or by purchasing an inte-
grated XYZ piezo stage. Prior Scientific Instruments Ltd., Applied
Scientific Instrumentation (ASI), Mad City Labs (MCL) Inc., Physik
Instrumente (PI) GmbH & Co. KG are just some of the manufactur-
ers who offer some or all of these options. A stage-top Z translator
is a popular option since it can work with multiple objectives (the
piezo objective translator can only be mounted on a single objec-
tive at a time), it can be mounted on stages which are already pre-
sent in most microscopes and will have a much larger travel range
compared to a XYZ piezo translator.
If multicolor illumination is required, this can also be another
bottleneck for speed. In most widefield microscopes separation of
different wavelengths is achieved by moving a revolving turret of
filter cubes, which include excitation, emission and dichroic filters.
These filter cubes typically take a few hundreds of milliseconds to
move, which accumulates considerably during the HT microscopy
workflow. As laser illumination is used in SMLM, laser selection
can be achieved with direct modulation or with an AOTF (delays
of milliseconds) and so mechanical changing of excitation filters
is unnecessary; a single dichroic which only reflects the laser
wavelengths in use is sufficient. However, an excitation clean-up
filter in front of the laser itself is advised (Fig. 1) as this will remove
any additional unwanted wavelengths emitted by the laser.
Although emission filters may also be unnecessary when using
spectrally distinct fluorophores, we recommend that an emission
clean-up filter be present before the camera to remove both stray
laser reflections and unspecific emission. If emission filters are
required (for example when there is fluorophore crosstalk or aut-
ofluorescence), these can be mounted in motorized filter wheels
with switching times of a few tens of milliseconds. These are avail-
able from major microscope brand distributors or some of the pre-
viously mentioned companies (Prior, ASI). For good quality TIRF,
the dichroic mirror should be free from distortion to avoid corrupt-
ing the laser focus. Thicker dichroic mirrors suffer less surface
stress from being clamped in holders, and thick dichroics glued
to holders are available from Chroma Technology Corp.
2.3. Detection path for high-speed SMLM
A further method for accelerating multicolor fast SMLM is by
using two cameras simultaneously instead of one, or using two
sides of the same camera sensor, simultaneously acquiring two dif-
ferent emission spectra from the sample (Fig. 1). Since the acquisi-
tion for each color is made in parallel and not sequentially, this
results in a doubling of the acquisition speed. Although it should
be noted that splitting between two cameras or between different
parts of the same camera decreases lateral resolution due to the
lower number of photons in each camera (or part of the camera)
for localization. Instead of multicolor imaging, these two views
can be used for biplane 3D imaging (2.3.1) [64]. Splitting the emit-
ted light into two cameras is greatly simplified if the user has
access to the infinity path, which is the collimated light path exit-
ing the BFP of an (infinity corrected) objective [39]. However, mod-
ern commercial microscopes do not offer easy access to place a
second dichroic and a camera in the limited geometry of the infin-
ity path. An exception is the Rapid Automated Modular Microscope
system from ASI imaging, which offers complete access to the opti-
cal path.
2.3.1. 3D SMLM
SMLM microscopes can be designed to increase axial (Z) resolu-
tion as well as lateral resolution, and techniques for 3D SMLM are
typically based on manipulating or knowing the shape of the point
spread function (PSF) of the microscope. The PSF describes how a
point source of light is imaged by an optical system, and due to
diffraction of light by the microscope optics the image is blurred
in X, Y and Z; all images made with the microscope can be consid-
ered a convolution of the PSF with the real structure of the sample
[39]. The PSF size is approximately 3� larger in the Z dimension
than in XY, and thus the axial resolution is accordingly poorer than
the lateral resolution [39].
Three common methods whereby 3D SMLM images can be
reconstructed are biplane imaging [64], astigmatic [65] and
double-helix (DH) PSF [66] methods. Biplane imaging takes advan-
tage of the observation that the lateral dimensions of the PSF
increase with axial distance from the focus. By taking two images
at different known focal positions and measuring the ratio in the
Table 5
Comparison of relevant features of modern microscope bodies for custom SMLM systems. We find the ASI RAMM highly versatile and cost-efficient, but their focus system
requires some habituation. Access to optics refers to how easily one can modify the excitation and emission paths; dual turret refers to the possibility of adding more than one
dichroic cube changer.
Ti-E
DMi8
IX83
Axio observer
ASI RAMM
Focus correction
Performance
Best
n/a
Good
Good
Good
Ease of use
Simple
n/a
Reasonable
Reasonable
Learning curve
Access to optics
Good
Good
Good
Poor
Best
Dual turret
Yes
No
Yes
No
Yes
Eyepieces
Yes
Yes
Yes
Yes
No
Cost
$$$
$$$
$$$
$$$
$$
P. Almada et al. / Methods xxx (2015) xxx–xxx
7
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
differences in the lateral PSF shape, one can determine the relative
position of the object in Z [64]. However, as emission light must be
split between two cameras, lateral resolution is decreased (as dis-
cussed above). Astigmatism and DH are based on shaping the PSF
in such a way that its shape will have information on the axial
position of the molecule encoded in it, and this position can be
extracted post-acquisition from the lateral shape of the PSF
[65,67]. Both of these methods require additional optics in the
detection path to shape the light: DH requires a spatial light mod-
ulator (and often an optical relay) [66] whereas astigmatism only
requires a f = 1 m cylindrical lens in front of the camera [68]. If
the camera is inaccessible, astigmatism can be achieved by adding
a f = 10 m cylindrical lens to the infinity path of the microscope (for
example, by inserting the lens into a slider such as that present in
the Nikon Ti-E [69]). In both astigmatism and DH the distribution
of light over a larger area results in decreased lateral resolution.
For comparisons of these methods see Badieirostami et al. [68],
Liu and Lidke [70], Mlodzianoski et al. [71] and Table 6.
2.3.2. Optical relays
In general, it is not simple to split the emission light between
two cameras in a commercial microscope. Commercial options
are available, which adapt the regular camera port of microscopes
to add two cameras (Andor Technology Ltd. TuCam, Cairn Research
Ltd Twincam) or to split the light in two sides of the same camera
chip (Cairn Research Ltd. Optosplit II). When using an sCMOS cam-
era the large chip allows splitting without requiring a second cam-
era; while this increases acquisition speed, it sacrifices FOV which
is not desirable for high-throughput microscopy.
The commercial splitters are essentially optical relays, which
can also be assembled in home-built systems [64,72]. Optical
relays consist of two lenses: one is placed after the tube lens to col-
limate the light, and the other placed in front of the camera to
focus the image onto the sensor (Fig. 1). If these two lenses have
different focal lengths their ratio will introduce an extra magnifica-
tion and so care should be taken when calculating the final magni-
fication of the system (Section 2.1.2). The splitting itself is achieved
with a dichroic or a 50/50 beam splitter between the two relay
lenses to avoid distortions (Fig. 1). When placing the splitting optic
between the second lens and the camera, distortion can be mini-
mized by focusing onto the camera with a long focal length lens,
such as in the original biplane paper [64].
2.3.3. Camera speed considerations
Electronic cameras’ sensors acquire a signal by collecting elec-
trons that are generated when a photon hits the sensor. During
read-out these electrons are then ‘‘read’’, meaning that the signal
is quantified for each pixel and transmitted to the computer.
Camera acquisition time is generally limited by the time it takes
to read the signal for each exposure. Read-out also introduces some
variability into the signal, which manifests as noise in the image
(read noise).
Most commercial TIRF microscopes have an EMCCD camera
installed, and the read-out speed of an EMCCD can be increased
by enabling the Frame Transfer function, which allows the camera
to start acquiring a second image while the first is still being read,
or by cropping the detector to a smaller FOV.
However, for fast high-throughput SMLM, sCMOS cameras are
ideal as they possess larger chips and are inherently faster than
CCD cameras. When the amplification noise of EMCCD cameras is
taken into account, sCMOS cameras also have higher quantum effi-
ciencies (the fraction of photons effectively captured by the cam-
era) than EMCCD cameras [73]. As such, sCMOS cameras can
surpass the performance of an EMCCD for SMLM. The mechanism
through which sCMOS cameras are read out is via a rolling shutter
type exposure [74], which results in different parts of a single
exposure being imaged at different times. While this can be prob-
lematic when imaging objects moving faster than the acquisition
rate of the camera, this should not hinder SMLM imaging of fixed
samples or live-cell samples where any movement is on a suffi-
ciently slow timescale.
If even faster imaging is required at the expense of FOV, the
sCMOS sensor can be cropped as for EMCCDs. However, as each
row in an sCMOS sensor has to have all its pixels read, cropping
the sensor would only increase speed for each complete row of pix-
els not read by the camera. This is in contrast to EMCCD cameras,
where any cropping will always increase speed.
2.4. Acquisition computer
2.4.1. Software control of hardware
All controllable hardware associated with the microscope needs
to be synchronized during acquisition. Most commercial micro-
scopes offer a software package which is capable of controlling
the hardware mentioned so far, but their flexibility will vary. The
open source MicroManager software [75] can be used to control
almost all models of lasers, cameras and microscopy hardware. It
also provides a very flexible beanshell based scripting language
which permits a user to control all aspects of data acquisition, plu-
gins for HT microscopy and fast multi-dimensional acquisition.
When using two sCMOS cameras, care needs to be taken to
ensure that they are synchronized, otherwise the images will have
parts that are time shifted in relation to each other due to the roll-
ing shutter. Synchronization can be achieved by triggering the start
of
each
acquisition
at
exactly
the
same
time
using
a
transistor-transistor logic (TTL) signal. This can be done by sending
a signal from one camera to the other or by sending a signal from
an external source to both cameras, however triggering from one
camera to the other depends on the triggering properties of the
cameras and may not be achievable. External triggering of both
cameras is generally preferable for tight synchronization but
requires a digital I/O device (such as those available from
National Instruments Corp., Measurement Computing Corp. or
the open-source Arduino platform) that is capable of digital trig-
gering at the frame rate of acquisition. The software also needs
to accept two cameras, or else two independent computers have
to be used. MicroManager can be configured to work with two
cameras in software triggering mode and also supports digital I/O
devices.
2.4.2. Data storage considerations for sCMOS cameras
The most sensitive (72% quantum efficiency) sCMOS cameras
acquire 2048 � 2048 pixel 16-bit images at 100 frames per second.
This corresponds to having to transfer �840 megabytes of data per
second (MB/s) to the computer, and a typical SMLM stack of 10,000
to 50,000 frames will require 42–84 gigabytes of memory.
Acquisition software such as MicroManager can stream all of this
data to either RAM or the hard drives. RAM is fast enough to write
the data but generally limited in size and more costly; a hard-disk
drive (HDD) is too slow to write but has sufficient storage space
Table 6
Qualitative comparison of the mentioned 3D SMLM methods (CL – cylindrical lens;
DH-PSF – double helix PSF). Complexity refers to the ease of implementation in
custom SMLM systems. Depth of coverage evaluates the axial extent where
localization precisions are acceptable; low SNR performance refers to how well the
method performs in situations of high background and/or low signal.
CL
DH-PSF
Biplane
Cost
Low
Moderate
High
Complexity
Low
High
Reasonable
Depth of coverage
Poor
Best
Good
Low SNR performance
Good
Low
Best
8
P. Almada et al. / Methods xxx (2015) xxx–xxx
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
and is cost-effective. Alternatively, Solid State Drives (SSDs) are
storage drives which are much faster than HDDs (hundreds of
MB/s versus tens of MB/s, performance varying per model) but
are also more costly. SSDs (or HDDs) can be multiplexed, where x
drives can work in parallel in RAID0 mode to increase speed by a
factor of x, providing more space than RAM at a lower cost.
However, a typical SMLM experiment will generally require several
tens or hundreds of sets of acquisitions, and HT microscopy could
require thousands, and hence it is evident that data storage is a
thorny problem in SMLM. Having a direct fiber-optic network con-
nection to a server with large storage capacity is an obvious advan-
tage. Any new system being considered, commercial or home built,
must have the requisite IT infrastructure to handle the expected
volume of data.
2.5. Setting up a space for building a system
In addition to the above hardware and IT infrastructure consid-
erations for constructing a SMLM system, the local environment of
the microscope is also important. As SMLM systems require the
imaging of single molecules, any high-frequency vibrations present
in the system will result in a blurring of the PSF by the amplitude of
the vibration, degrading the localization precision. Such vibrations
may be present in a system but not visible until imaging is per-
formed at a high frame rate.
There are several ways to minimize vibrations in the system.
Firstly, the system can be built on an air-dampened table, which
filters vibrations transmitted from the ground. In general, heavier
and larger tables will filter vibrations more effectively, and damp-
ening specifications of tables are available from manufacturers
(Thorlabs Inc., TMC Mfg., and Newport Corporation). Additionally,
there are greater vibrations in higher floors of buildings and as
such many SMLM systems are housed in the ground floor or base-
ment of buildings. Any equipment which could cause vibrations
should also be on separate tables; computers should be placed
away from the system, preferably on rubber feet, and laser fan
vibrations can be avoided by placing lasers on a separate table
and using optical fibers to deliver beams to the main table.
Alternatively, water cooling systems may be used instead of the
inbuilt fans in lasers and cameras.
Maintaining an appropriate temperature in the microscope
room is also important. As most modern optics and immersion oils
are optimized for a room temperature of 23 Celsius, care should be
taken to ensure that the air-conditioning maintains this tempera-
ture on the microscope. To achieve this, it may be necessary to
lower the set point of the air conditioning to a slightly lower tem-
perature to negate heat generated by components of the micro-
scope. Thermal stability in the microscope room is crucial for
minimizing drift; we have seen in practice that although drift is
noticeable in the first few minutes after placing the sample, after
this period drift becomes negligible in a thermally stable room.
Systems for correcting residual focal drift are discussed in
Section 2.2.1, and any residual lateral drift can be easily corrected
for computationally by including reference markers (such as fluo-
rescent beads or gold nanoparticles) in the sample [12]. Most com-
mercial and open-source SMLM localization software such as
QuickPALM [76] have integrated drift correction algorithms based
on tracking these reference markers. For temperature and environ-
mental control in live cell imaging, additional hardware such as
incubation chambers are required; these are beyond the scope of
this review but are reviewed elsewhere in a non-SMLM context
[77].
The system should also possess suitable shielding from stray
photons, as these will increase background noise and decrease
the SNR. There are black-out boxes available for almost all com-
mercial systems (Pecon GmbH), and for homebuilt systems
darkening the walls will also allow for easy identification of stray
laser reflections and will thus aid alignment.
Electrical stability is also important. Fluctuations in power sup-
plies may affect the performance of lasers and electronic devices
such as AOTFs and piezo stages. Using an Uninterrupted Power
Supply (UPS) unit is always recommended, as it will stabilize the
electrical system associated with the SMLM components and pro-
tect the devices from power spikes and blackouts.
2.6. Analytics for sCMOS camera acquisitions
2.6.1. sCMOS noise patterning problem
While there is freely available software for SMLM analysis such
as QuickPALM [76], ThunderSTORM [78] and RapidSTORM [79],
these all assume acquisition with an EMCCD camera. As such these
programs may not be suitable for analyzing data acquired with
sCMOS cameras, as each pixel in an sCMOS chip has its own noise
properties [46] and hence SMLM localizations may be biased
toward high-noise pixels. Although manufacturers include a cor-
rection on the camera chip for this, the correction methods are
not
published
and
may
not
remove
the
localization
bias.
However, Huan et al. have published a method for correcting this
bias [46], which involves characterizing pixel noise obtained from
several thousands of dark frames, and SMLM software with correc-
tions for noise patterning is provided on their website [80].
2.6.2. Impact of analysis algorithms on acquisition speed
As discussed previously, a limiting factor in SMLM acquisition
time is ensuring that a sufficient number of molecules are localized
to reconstruct the labeled structures. A simple way to decrease
total acquisition time is to increase the density of on-state mole-
cules per frame, reducing the number of frames required to yield
the same total number of localizations. A further advantage is that
lower laser intensities (or even lamp-based illumination [81]), can
be used for higher density acquisitions (Section 1.1). However,
when acquiring at higher densities, fluorophores start to overlap
and cannot be distinguished by traditional SMLM algorithms.
Specialist algorithms have been developed which can handle
high-density datasets [81–85], with some methods capable of
reconstructing from a few hundreds of frames. High-density meth-
ods are particularly relevant for HT microscopy as they also reduce
data processing and storage hardware requirements.
There are currently three prominent high-density algorithms.
Bayesian Localization Microscopy (3B) has been used to perform
live imaging of podosomes at 50 nm resolution with a simple
widefield microscope, requiring only a few seconds of acquisition
time (200 frames). The drawback of this method is that it is com-
putationally intensive (6 h for a 1.5 � 1.5lm2 area), however the
software is freely available [81]. Compressed Sensing STORM is
an algorithm that has been demonstrated to reconstruct 60 nm
resolution images of microtubules from 169 frames [85], and the
Matlab code is available with the paper. As with 3B, this method
is computationally demanding and one group reported taking a
whole day to process a 100 lm2 region [83]. Stochastic Optical
Fluctuation Imaging (SOFI) is capable of doubling the resolution
with a standard widefield microscope (200 frames) [86] and
achieving 80 nm resolution with a TIRF microscope with low laser
powers (5000 frames) [87]. The virtual removal of background and
out-of-focus light in SOFI allows it to generate optically sectioned
z-stacks, and a Matlab SOFI package is freely available SOFI pack-
age [88]. However, SOFI image interpretation is artifact prone as
resultant image intensity is representative of temporal optical fluc-
tuations and not original signal intensity, so gaps can emerge in
areas where the intensity did not fluctuate sufficiently. While these
high density algorithms may require high computational intensity
P. Almada et al. / Methods xxx (2015) xxx–xxx
9
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
or produce some artifacts, the increase in speed afforded by high
density acquisition is still sufficiently worthwhile.
3. Putting it all together
So far we have described several of the design options for a
high-speed high-throughput SMLM system with an sCMOS camera.
Perhaps the most important decision will be whether or not a TIRF
objective is used, as this will limit the size of the uniform FOV and
hence the maximum area of the sCMOS chip that will yield usable
data. TIRF can however still be achieved (with an added degree of
difficulty) using 1.4 NA non-TIRF specific objectives when imaging
samples in water-based media [16,89].
For high-throughput and large FOV imaging, beam homogeniza-
tion is recommended to increase the uniformity of the FOV. A
vibrating fiber system is perhaps the most efficient method of
those described in Section 2.1.2 and is relatively simple to imple-
ment. Refractive beam shapers are also simple and effective optics
for beam homogenization, but are more costly than other methods.
Further options for enhancing the versatility of the system for
various applications have also been discussed, such as splitting
the emission in half (Section 2.3.2). This enables fast multicolor
imaging or biplane imaging for 3D SMLM [68,70,71], although if
this is achieved by using two halves of the same camera then this
reduces the FOV.
There is room for variation on the final system design, depend-
ing on the purpose of the system and the context in which it is to
operate. To demonstrate this we will outline two possible systems
with different properties: (A) an easily assembled system with
mostly commercial parts and (B) a low-cost, high-speed system.
3.1. An easily assembled system with mostly commercial parts
Modern, motorized microscope bodies such as the Nikon Ti-E,
Olympus IX83, Leica DMi8 or Zeiss Axio Observer offer the most
important features required for high-speed or HT SMLM: a hard-
ware drift correction system, motorized focus and well-corrected
optics. MicroManager can be used to control any two cameras
simultaneously (sCMOS or EMCCD), as well as all of the mentioned
stages, laser systems and microscopes in the system described
here.
The Andor TuCam adapter can be used with any of these sys-
tems to allow for imaging in two sCMOS cameras, such as the
Hamamatsu Flash 4.0 or Andor Zyla 4.2. To obtain 100 nm pixels
(to
satisfy
the
Nyquist
sampling
criterion
as
discussed
in
Section 2.1.2) while using the whole FOV and TIRF illumination
in water-based samples, a 60� 1.4 NA objective should be used
with no additional magnification in components such as camera
adapters.
In order to reduce complexity and provide warranty coverage,
complete turn-key laser assemblies which contain several lasers
coupled together can be obtained from several manufacturers
(Omicron-Laserage Laserprodukte GmbH, TOPTICA Photonics AG,
Oxxius SA, Vortran Laser Technology, Inc.). All of these offer a
TEM00 fiber-coupled output and fast modulation, which can be
directly controlled with MicroManager. However, the laser powers
in commercial turn-key systems may be limited compared to cus-
tom laser systems.
The most challenging part of the system to assemble is the laser
launching
and
beam
homogenization
system.
The
simplest
high-efficiency method to homogenize the laser illumination is
by using a refractive beam shaper such as the piShaper apparatus,
which will convert a 6 mm diameter Gaussian beam into a 6 mm
homogenized beam [90]. A single mode fiber possessing a
Gaussian output [45] can be used to inject the laser beam into
the piShaper. Launching through a fiber will also increase the sta-
bility of the system and allow for lasers (and their vibrations) to be
located away from the optical table. As the output of a fiber is
divergent, it must be collimated using a plano-convex lens, and
the resultant beam size depends on the focal length of the lens f
and the numerical aperture of the fiber NA as 2*f*NA [91]. For
example, to obtain a collimated 6 mm beam to input into the
piShaper from a 0.12 NA single mode fiber (S405-XP, ThorLabs),
an
achromatic
25 mm
lens
is
required
(AC127-025-A-ML,
Thorlabs).
Similarly, the 6 mm uniform output of the piShaper must be
expanded to match the 19 mm diagonal size of the Flash 4.0 and
Zyla 4.2 cameras (of note, this is only the case when not changing
the magnification before the camera and when the lens to focus on
the BFP has the same focal length as the tube lens [36]). This can be
achieved using a two-lens Galilean beam expander (which intro-
duces minimal chromatic aberrations [49]) where the ratio of the
two focal lengths equals the desired expansion. For example, to
expand the 6 mm piShaper output to 19 mm, the output can pass
through a �50 mm plano-concave lens (Thorlabs, LC1715-A) and
then
a
175 mm
plano-convex
lens
(Thorlabs,
LA1229-A).
Importantly, the expected on-sample power density from a given
laser system should be calculated by first multiplying the power
ratings with the transmissions from the beam shaping optics and
the
objective
transmission.
This
will
yield
the
expected
on-sample laser power, which must be divided by the area to
image; depending on the laser power, it may be necessary to
reduce the FOV to reach power densities at least in the few
kW/cm2. Lin et al. [32] provide useful guidelines to determine
the necessary powers given a maximum camera frame rate (deter-
mined by the size of the FOV), keeping in mind that for HT it is
always useful to maximize the FOV. The entire laser launching
and homogenization assembly (Fig. 2A) can be mounted on a
motorized or manual XYZ stage assembly (Thorlabs and Newport
offer many options) to focus the spot on the BFP and translate it
in order to achieve TIRF.
Both the Nikon Ti-E and Olympus IX83 bodies offer the possibil-
ity of a second dichroic turret for use with laser illumination: this
allows for a separate fluorescence lamp or LED illuminator to be
used for inspection of the sample through the eyepieces in addition
to the TIRF illumination system. The Leica DMi8 also has a sec-
ondary illumination port which can be selected by manually mov-
ing a mirror; while this system is simpler and cheaper than the
Nikon and Olympus systems, the stability of the manual mirror is
yet to be fully assessed, and switching between a laser and
lamp/LED cannot be automated. As far as we are aware there is
no easy way to couple a custom laser launch into the Zeiss system
without fully replacing the existing lamp/LED system.
3.2. A low-cost, high-speed HT SMLM microscope
The commercial system described above contains several
expensive components; however, low-cost SMLM systems can also
be readily constructed. Holm et al. [92] recently demonstrated a
simple, low cost system using cheap components. For high-speed
HT SMLM most existing microscope bodies can be adapted by cou-
pling an homogenized laser source and installing an sCMOS cam-
era. Although older microscopes will lack focal drift correction, a
combination of imaging at sufficiently high speed, high-density
acquisition and using a thermally stable room will minimize drift;
any residual drift experienced may be corrected for by software
autofocus algorithms (such as those present in MicroManager).
For single-color dSTORM imaging, which may be sufficient for
most researchers, a single laser will suffice. Alexa647 and Cy5 are
generally the best-performing dyes for dSTORM imaging [24] and
lasers of appropriate wavelengths can be procured at a reasonable
10
P. Almada et al. / Methods xxx (2015) xxx–xxx
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
cost
(MPB
Communications
Inc.,
Shanghai
Dream
Lasers
Technology Co., Ltd., Kvant s.r.o., CNI Laser). Modulation of laser
power can be achieved manually using variable neutral density fil-
ters (Thorlabs, Inc. Newport Corporation) or digitally using low
cost electronic prototyping systems such as the Arduino Due.
For beam homogenization, the cheapest approach may be to use
an engineered diffuser, although TIRF is not possible with this
method. The laser beam can be input directly onto the diffuser
with the output then collimated and focused onto the BFP
(Fig. 2B). The diameter of the collimated output is given by f tana,
where f is the focal length of the collimating lens and a is the out-
put angle of the diffuser [93]. For example, to create a homoge-
nized beam of diameter 19 mm to match the camera chip size
(as discussed for the commercial system above), a diffuser with
20� output (ED1-C20-MD, Thorlabs) can be used in conjunction
with a 60 mm lens (LA1134-A, Thorlabs). A second lens can then
be used to focus this beam onto the BFP. We remind readers to take
power density into consideration when choosing a laser and size of
the FOV at the sample (see Section 3.1).
This simple, low-cost implementation provides a system that
can efficiently illuminate the large FOV of a sCMOS camera,
enabling high-speed HT SMLM of the whole FOV.
4. Discussion and future trends in HT SMLM
We have introduced the reader to the hardware options and
design caveats when assembling or adapting a microscope for HT
SMLM. We hope that this information serves as a basis for more
groups to start pushing SMLM into high-throughput. SMLM shows
promise in offering better ways to tease out signaling interactions
as shown recently [30]. In particular, we have explained factors to
be considered when extending SMLM systems for high-speed, large
FOV acquisition, with obvious implications for HT microscopy.
Most of these considerations revolve around taking advantage of
the high speed of sCMOS cameras and their larger chips. These
cameras are relatively new to the scientific market and we expect
manufacturers to already be improving electronics to increase
dynamic range, read out speed and lower noise characteristics.
Their large size also means that microscope optic manufacturers
need to start adapting their products to this new technology.
‘‘Plan’’ classifications for TIRF objectives that do not cover the
whole FOV need to be reviewed and better corrected, high numer-
ical aperture objectives are imperative to take advantage of these
cameras. Regardless, software corrections for aberrations across
the FOV should become common features of these cameras and
we expect these to become widely integrated in commercial and
open-source SMLM software.
More importantly, new reconstruction algorithms are being
developed which require fewer frames, can be computed in a rea-
sonable amount of time and can handle large numbers of overlap-
ping fluorophores. These algorithms are set to revolutionize the
field since sample, hardware and IT infrastructure requirements
are greatly reduced, facilitating HT SMLM as well as live-cell imag-
ing as a routinely feasible proposition for SMLM. Reducing the need
to invest in expensive and specialist hardware will hopefully result
in widespread adoption and greater development of SMLM, which
is a necessary step to move the field to HT. This will not happen,
however, if localization reconstruction algorithms are not made
available in a simple and widely available manner and it is our
belief the future will lie on those who make their source code
freely available and compatible with free image processing pack-
ages for the life sciences such as FIJI [94]. Likewise, a wider under-
standing of the hardware underlying these systems is necessary to
spread adoption of SMLM and foster its development. We hope
that the guidelines laid in this work serve as a starting point for
many labs.
Acknowledgments
We would like to thank Alan Lowe and Pedro Pereira for their
assistance revising the manuscript. PA is supported by a PhD fel-
lowship from the UK’s Biotechnology and Biological Sciences
Research Council, SC’s post-doc fellowship is supported by the
UK’s Medical Research Council.
References
[1] J. Rino, J. Braga, R. Henriques, M. Carmo-Fonseca, Int. J. Dev. Biol. 53 (2009)
1569–1579, http://dx.doi.org/10.1387/ijdb.072351jr.
[2] E. Abbe, Arch. Für Mikroskopische Anat. 9 (1873) 413–418, http://dx.doi.org/
10.1007/BF02956173.
[3] R. Henriques, C. Griffiths, E.H. Rego, M.M. Mhlanga, E. Hesper, Biopolymers 95
(2011) 322–331, http://dx.doi.org/10.1002/bip.21586.
Fig. 2. Example systems. (A) An example system built of mostly commercial parts. TL, Tube lens; F1, plano-concave-50 mm focal length lens (Thorlabs, LC1715-A); F2, plano-
convex 175 mm focal length lens (Thorlabs, LA1229-A); LNL, collimating achromatic 25 mm lens (AC127-025-A-ML, Thorlabs); SMF, single mode fiber (S405-XP, Thorlabs);
XYZ stage, manual translation stage assembly (LNR25D/M and LNR25P2, Thorlabs); MIC, microscope. (B) A low-cost microscope. F1, 60 mm plano-convex focal length lens
(LA1134-A, Thorlabs); ED, 20� engineered diffuser (ED1-C20-MD, Thorlabs).
P. Almada et al. / Methods xxx (2015) xxx–xxx
11
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
[4] H.H. Rose, Sci. Technol. Adv. Mater. 9 (2008) 014107, http://dx.doi.org/
10.1088/0031-8949/9/1/014107.
[5] R.H. Wade, Mol. Biotechnol. 43 (2009) 177–191, http://dx.doi.org/10.1007/
s12033-009-9193-5.
[6] P.J.J. Robinson, L. Fairall, V.A.T. Huynh, D. Rhodes, Proc. Natl. Acad. Sci. U.S.A.
103 (2006) 6506–6511, http://dx.doi.org/10.1073/pnas.0601212103.
[7] K.R. BruceAlberts, A. Johnson, J. Lewis, M. Raff, P. Walter, Molecular Biology of
the Cell, 4th ed., Garland Science, New York, 2002.
[8] S.W. Hell, J. Wichmann, Opt. Lett. 19 (1994) 780–782, http://dx.doi.org/
10.1364/OL.19.000780.
[9] M.G. Gustafsson, J. Microsc. 198 (2000) 82–87, http://dx.doi.org/10.1046/
j.1365-2818.2000.00710.x.
[10] E. Betzig, G.H. Patterson, R. Sougrat, O.W. Lindwasser, S. Olenych, J.S.
Bonifacino,
et
al.,
Science
313
(2006)
1642–1645,
http://dx.doi.org/
10.1126/science.1127344.
[11] S.T. Hess, T.P.K. Girirajan, M.D. Mason, Biophys. J. 91 (2006) 4258–4272, http://
dx.doi.org/10.1529/biophysj.106.091116.
[12] M.J. Rust, M. Bates, X. Zhuang, Nat. Methods 3 (2006) 793–795, http://
dx.doi.org/10.1038/nmeth929.
[13] S. Cox, Dev. Biol. (2014) 1–7, http://dx.doi.org/10.1016/j.ydbio.2014.11.025.
[14] S. Herbert, H. Soares, C. Zimmer, R. Henriques, Microsc. Microanal. 18 (2012)
1419–1429, http://dx.doi.org/10.1017/S1431927612013347.
[15] R.E. Thompson, D.R. Larson, W.W. Webb, Biophys. J. 82 (2002) 2775–2783,
http://dx.doi.org/10.1016/S0006-3495(02)75618-X.
[16] S. van de Linde, A. Löschberger, T. Klein, M. Heidbreder, S. Wolter, M.
Heilemann, et al., Nat. Protoc. 6 (2011) 991–1009, http://dx.doi.org/10.1038/
nprot.2011.336.
[17] R. Henriques, M.M. Mhlanga, Biotechnol. J. 4 (2009) 846–857, http://
dx.doi.org/10.1002/biot.200900024.
[18] M. Bates, T.R. Blosser, X. Zhuang, Phys. Rev. Lett. 94 (2005) 108101, http://
dx.doi.org/10.1103/PhysRevLett. 94.108101.
[19] S. van de Linde, I. Krstic´, T. Prisner, S. Doose, M. Heilemann, M. Sauer,
Photochem. Photobiol. Sci. 10 (2011) 499–506, http://dx.doi.org/10.1039/c0
pp00317d.
[20] G.T. Dempsey, M. Bates, W.E. Kowtoniuk, D.R. Liu, R.Y. Tsien, X. Zhuang, J. Am.
Chem. Soc. 131 (2009) 18192–18193, http://dx.doi.org/10.1021/ja904588g.
[21] M. Heilemann, S. van de Linde, M. Schüttpelz, R. Kasper, B. Seefeldt, A.
Mukherjee, et al., Angew. Chem. Int. Ed. Engl. 47 (2008) 6172–6176, http://
dx.doi.org/10.1002/anie.200802376.
[22] J.C. Vaughan, G.T. Dempsey, E. Sun, X. Zhuang, J. Am. Chem. Soc. 135 (2013)
1197–1200, http://dx.doi.org/10.1021/ja3105279.
[23] J. Vogelsang, R. Kasper, C. Steinhauer, B. Person, M. Heilemann, M. Sauer, et al.,
Angew. Chem. Int. Ed. Engl. 47 (2008) 5465–5469, http://dx.doi.org/10.1002/
anie.200801518.
[24] G.T. Dempsey, J.C. Vaughan, K.H. Chen, M. Bates, X. Zhuang, Nat. Methods 8
(2011) 1027–1036, http://dx.doi.org/10.1038/nmeth.1768.
[25] J. Lippincott-Schwartz, G.H. Patterson, Trends Cell Biol. 19 (2009) 555–565,
http://dx.doi.org/10.1016/j.tcb.2009.09.003.
[26] C. Geisler, A. Schönle, C. von Middendorff, H. Bock, C. Eggeling, A. Egner, et al.,
Appl. Phys. A 88 (2007) 223–226, http://dx.doi.org/10.1007/s00339-007-
4144-0.
[27] V. Westphal, S.O. Rizzoli, M.A. Lauterbach, D. Kamin, R. Jahn, S.W. Hell, Science
320 (2008) 246–249, http://dx.doi.org/10.1126/science.1154228.
[28] F. Bergermann, L. Alber, S.J. Sahl, J. Engelhardt, S.W. Hell, Opt. Express 23
(2015) 211, http://dx.doi.org/10.1364/OE.23.000211.
[29] S.J. Holden, T. Pengo, K.L. Meibom, C. Fernandez, Proc. Natl. Acad. Sci. U.S.A.
111 (2014) 4566–4571, http://dx.doi.org/10.1073/pnas.1313368111.
[30] H. Soares, R. Henriques, M. Sachse, L. Ventimiglia, M.a. Alonso, C. Zimmer,
et
al.,
J.
Exp.
Med.
210
(2013)
2415–2433,
http://dx.doi.org/
10.1084/jem.20130150.
[31] A. Egner, C. Geisler, C. von Middendorff, H. Bock, D. Wenzel, R. Medda, et al.,
Biophys.
J.
93
(2007)
3285–3290,
http://dx.doi.org/10.1529/
biophysj.107.112201.
[32] Y. Lin, J.J. Long, F. Huang, W.C. Duim, S. Kirschbaum, Y. Zhang, et al., PLoS One
10 (2015) e0128135, http://dx.doi.org/10.1371/journal.pone.0128135.
[33] D.
Axelrod, J.
Cell
Biol. 89
(1981)
141–145, http://dx.doi.org/10.1083/
jcb.89.1.141.
[34] K.N. Fish, Total internal reflection fluorescence (TIRF) microscopy, Curr. Protoc.
Cytom
(Chapter
12
(2009)
Unit12.18),
doi:http://dx.doi.org/10.1002/
0471142956.cy1218s50.
[35] A.L. Mattheyses, S.M. Simon, J.Z. Rappoport, J. Cell Sci. 123 (2010) 3621–3628,
http://dx.doi.org/10.1242/jcs.056218.
[36] D.S. Johnson, J.K. Jaiswal, S. Simon, Curr. Protoc. Cytometry (2012) 1–19, http://
dx.doi.org/10.1002/0471142956.cy1229s61.
[37] M. Tokunaga, N. Imamoto, K. Sakata-Sogawa, Nat. Methods 5 (2008) 159–161,
http://dx.doi.org/10.1038/nmeth1171.
[38] ISO 19012-1:2013, Microscopes – designation of microscope objectives – part
1:
flatness
of
field/Plan,
(n.d.).
http://www.iso.org/iso/catalogue_detail.
htm?csnumber=61652 (accessed 2.02.15).
[39] J.B. Pawley, Handbook of biological confocal microscopy, 3rd ed., Springer, US,
Boston, MA, 2006, doi:http://dx.doi.org/10.1007/978-0-387-45524-2.
[40] J.W. Goodman, Speckle Phenomena in Optics: Theory and Applications,
Roberts & Company, 2007. doi:9781936221141.
[41] a.L. Stout, D. Axelrod, Evanescent field excitation of fluorescence by epi-
illumination microscopy, Appl. Opt. 28 (1989) 5237–5242, http://dx.doi.org/
10.1364/AO.28.005237.
[42] J. Sinkó, G. Szabó, M. Erdélyi, Opt. Express 22 (2014) 18940, http://dx.doi.org/
10.1364/OE.22.018940.
[43] J.L. Ross, R. Dixit, Methods Cell Biol. 95 (2010) 521–542, http://dx.doi.org/
10.1016/S0091-679X(10)95026-7.
[44] R. Fiolka, Y. Belyaev, H. Ewers, A. Stemmer, Microsc. Res. Tech. 71 (2008) 45–
50, http://dx.doi.org/10.1002/jemt.20527.
[45] L.M.P., L.S.P., Frank L. Pedrotti, Introduction to optics, 2nd ed., Prentice Hall,
1992.
[46] F. Huang, T.M.P. Hartwich, F.E. Rivera-Molina, Y. Lin, W.C. Duim, J.J. Long, et al.,
Nat. Methods 10 (2013) 653–658, http://dx.doi.org/10.1038/nmeth.2488.
[47] A.L. Mattheyses, K. Shaw, D. Axelrod, Microsc. Res. Tech. 69 (2006) 642–647,
http://dx.doi.org/10.1002/jemt.20334.
[48] R. Hard, R. Zeh, R.D. Allen, J. Cell Sci. 23 (1977) 335–343.
[49] S. Inoué, K.R. Spring, Video Microscopy: The Fundamentals, Springer, 1997.
[50] Y. Fujimaki, H. Taniguchi, Reduction of speckle contrast in multimode fibers
using piezoelectric vibrator, in: A.V. Kudryashov, A.H. Paxton, V.S. Ilchenko, L.
Aschke, K. Washio (eds.), 2014, 89601S, doi:http://dx.doi.org/10.1117/12.
2037431.
[51] W. Ha, S. Lee, Y. Jung, J.K. Kim, K. Oh, Opt. Express 17 (2009) 17536–17546,
http://dx.doi.org/10.1364/OE.17.017536.
[52] B. Flamion, P.M. Bungay, C.C. Gibson, K.R. Spring, Biophys. J. 60 (1991) 1229–
1242, http://dx.doi.org/10.1016/S0006-3495(91)82157-9.
[53] A. Shukla, Elements of Optical Communication and Opto Electronics, 1st ed.,
University Science Press, New Delhi, 2012.
[54] J.R. Kuhn, T.D. Pollard, Biophys. J. 88 (2005) 1387–1402, http://dx.doi.org/
10.1529/biophysj.104.047399.
[55] A. Laskin, A.S. Shcherbakov, V. Molchanov, V. Laskin, O. Makarov, Int. Comm.
Opt. Light Dev. World 8011 (2011) 1–6, http://dx.doi.org/10.1117/12.901944.
[56] S. Herbert, R. Henriques, Cytometry A 81 (2012) 278–280, http://dx.doi.org/
10.1002/cyto.a.21177.
[57] F.a.W. Coumans, E. van der Pol, L.W.M.M. Terstappen, Flat-top illumination
profile in an epifluorescence microscope by dual microlens arrays, Cytometry
A 81 (2012) 324–331, http://dx.doi.org/10.1002/cyto.a.22029.
[58] R. Völkel, K.J. Weible, Laser beam homogenizing: limitations and constraints,
Proc. SPIE, 7102 (2008), 71020J–71020J–12, doi:http://dx.doi.org/10.1117/12.
799400.
[59] P. Bonifazi, F. Difato, P. Massobrio, G.L. Breschi, V. Pasquale, T. Levi, et al., In
vitro large-scale experimental and theoretical studies for the realization of bi-
directional brain-prostheses, Front. Neural Circuits 7 (2013) 40, http://
dx.doi.org/10.3389/fncir.2013.00040.
[60] T.R.M. Sales, Opt. Eng. 42 (2003) 3084, http://dx.doi.org/10.1117/1.1618817.
[61] A. Edelstein, N. Amodaj, K. Hoover, R. Vale, N. Stuurman, Computer control of
microscopes using lManager, Curr. Protoc. Mol. Biol. (Chapter 14 (2010)
Unit14.20), doi:http://dx.doi.org/10.1002/0471142727.mb1420s92.
[62] T. Wöllert, G.M. Langford, Methods Mol. Biol. 586 (2009) 3–21, http://
dx.doi.org/10.1007/978-1-60761-376-3.
[63] PgFocus – BIG-Wiki, (n.d.). http://big.umassmed.edu/wiki/index.php/PgFocus
(accessed 26.01.15).
[64] M.F. Juette, T.J. Gould, M.D. Lessard, M.J. Mlodzianoski, B.S. Nagpure, B.T.
Bennett, et al., Nat. Methods 5 (2008) 527–529, http://dx.doi.org/10.1038/
nmeth.1211.
[65] B. Huang, W. Wang, M. Bates, X. Zhuang, Science 319 (2008) 810–813, http://
dx.doi.org/10.1126/science.1153529.
[66] S.R.P. Pavani, M.A. Thompson, J.S. Biteen, S.J. Lord, N. Liu, R.J. Twieg, et al., Proc.
Natl. Acad. Sci. U.S.A. 106 (2009) 2995–2999, http://dx.doi.org/10.1073/
pnas.0900245106.
[67] M.D. Lew, M.A. Thompson, M. Badieirostami, W.E. Moerner, In vivo three-
dimensional superresolution fluorescence tracking using a double-helix point
spread function, Proc. Soc. Photo Opt. Instrum. Eng. 7571 (2010) 75710Z,
http://dx.doi.org/10.1117/12.842608.
[68] M. Badieirostami, M.D. Lew, Appl. Phys. Lett. 97 (2010) 161103, http://
dx.doi.org/10.1063/1.3499652.
[69] R.
Henriques,
Tutorial_Astigmatism
–
quickpalm,
(n.d.).
https://code.google.com/p/quickpalm/wiki/Tutorial_Astigmatism
(accessed
27.01.15).
[70] S.
Liu,
ChemPhysChem
15
(2014)
696–704,
http://dx.doi.org/10.1002/
cphc.201300758.
[71] M.J. Mlodzianoski, M.F. Juette, G.L. Beane, J. Bewersdorf, Opt. Express 17 (2009)
8264–8277, http://dx.doi.org/10.1364/OE.17.008264.
[72] I. Testa, C.A. Wurm, R. Medda, E. Rothermel, C. Von Middendorf, J. Fölling, et al.,
Biophys. J. 99 (2010) 2686–2694, http://dx.doi.org/10.1016/j.bpj.2010.08.012.
[73] F. Long, S. Zeng, Z.-L. Huang, Opt. Express 20 (2012) 17741, http://dx.doi.org/
10.1364/OE.20.017741.
[74] C.K. Liang, L.W. Chang, H.H. Chen, IEEE Trans. Image Process. 17 (2008) 1323–
1330, http://dx.doi.org/10.1109/TIP.2008.925384.
[75] A.D. Edelstein, J. Biol. Methods 1 (2014) 1–10, http://dx.doi.org/10.14440/
jbm.2014.36.
[76] R. Henriques, M. Lelek, E.F. Fornasiero, F. Valtorta, C. Zimmer, M.M. Mhlanga,
Nat. Methods 7 (2010) 339–340, http://dx.doi.org/10.1038/nmeth0510-339.
[77] M.M. Frigault, J. Lacoste, J.L. Swift, C.M. Brown, J. Cell Sci. 122 (2009) 753–767,
http://dx.doi.org/10.1242/jcs.033837.
[78] M. Ovesny´, P. Krˇízˇek, J. Borkovec, Z. Svindrych, G.M. Hagen, Bioinformatics
(2014), http://dx.doi.org/10.1093/bioinformatics/btu202.
[79] S. Wolter, A. Löschberger, T. Holm, S. Aufmkolk, M.-C. Dabauvalle, S. van de
Linde, et al., Nat. Methods 9 (2012) 1040–1041, http://dx.doi.org/10.1038/
nmeth.2224.
12
P. Almada et al. / Methods xxx (2015) xxx–xxx
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004

----!@#$NewPage!@#$----
[80] ZhuangLab/storm-analysis
GitHub,
(n.d.).
https://github.com/ZhuangLab/
storm-analysis (accessed 10.03.15).
[81] S. Cox, E. Rosten, J. Monypenny, T. Jovanovic-Talisman, D.T. Burnette, J.
Lippincott-Schwartz, et al., Nat. Methods 9 (2011) 195–200, http://dx.doi.org/
10.1038/nmeth.1812.
[82] T. Dertinger, R. Colyer, G. Iyer, S. Weiss, J. Enderlein, Proc. Natl. Acad. Sci. U.S.A.
106 (2009) 22287–22292, http://dx.doi.org/10.1073/pnas.0907866106.
[83] J. Min, C. Vonesch, H. Kirshner, L. Carlini, N. Olivier, S. Holden, et al., FALCON:
fast and unbiased reconstruction of high-density super-resolution microscopy
data, Sci. Rep. 4 (2014) 4577, http://dx.doi.org/10.1038/srep04577.
[84] E.A. Mukamel, H. Babcock, X. Zhuang, Biophys. J. 102 (2012) 2391–2400,
http://dx.doi.org/10.1016/j.bpj.2012.03.070.
[85] L. Zhu, W. Zhang, D. Elnatan, B. Huang, Nat. Methods (2012), http://dx.doi.org/
10.1038/nmeth.1978.
[86] S. Geissbuehler, C. Dellagiacoma, T. Lasser, Biomed. Opt. Express 2 (2011) 408–
420, http://dx.doi.org/10.1364/BOE.2.000408.
[87] S. Geissbuehler, N.L. Bocchio, C. Dellagiacoma, C. Berclaz, M. Leutenegger, T.
Lasser, Opt. Nanoscopy 1 (2012) 4, http://dx.doi.org/10.1186/2192-2853-1-4.
[88] Balanced
super-resolution
optical
fluctuation
imaging,
(n.d.).
http://documents.epfl.ch/users/l/le/leuteneg/www/BalancedSOFI/index.html
(accessed 10.03.15).
[89] T.J. Gould, V.V. Verkhusha, S.T. Hess, Nat. Protoc. 4 (2009) 291–308, http://
dx.doi.org/10.1038/nprot.2008.246.
[90] piShaper – Versatile Beam Shaping Optics, (n.d.), http://www.pishaper.com/
shaper_6_6.php (accessed 10.03.15).
[91] Focusing and Collimating, (n.d.), http://www.newport.com/Focusing-and-
Collimating/141191/1033/content.aspx (accessed 4.03.15).
[92] T. Holm, T. Klein, A. Löschberger, T. Klamp, G. Wiebusch, S. Van De Linde, et al.,
ChemPhysChem
15
(2014)
651–654,
http://dx.doi.org/10.1002/cphc.
201300739.
[93] How to use Engineered Diffusers, (n.d.). http://www.rpcphotonics.com/how_
to.asp (accessed 10.03.15).
[94] J. Schindelin, I. Arganda-Carreras, E. Frise, V. Kaynig, M. Longair, T. Pietzsch,
et al., Nat. Methods 9 (2012) 676–682, http://dx.doi.org/10.1038/nmeth.2019.
P. Almada et al. / Methods xxx (2015) xxx–xxx
13
Please cite this article in press as: P. Almada et al., Methods (2015), http://dx.doi.org/10.1016/j.ymeth.2015.06.004