royalsocietypublishing.org/journal/rsob
Review
Cite this article: Mendes A, Heil HS, Coelho
S, Leterrier C, Henriques R. 2022 Mapping
molecular complexes with super-resolution
microscopy and single-particle analysis. Open
Biol. 12: 220079.
https://doi.org/10.1098/rsob.220079
Received: 16 March 2022
Accepted: 7 July 2022
Subject Area:
biophysics/biotechnology/cellular biology/
structural biology
Keywords:
structural biology, super-resolution microscopy,
single-particle analysis
Author for correspondence:
Ricardo Henriques
e-mail: rjhenriques@igc.gulbenkian.pt
†The authors contributed equally.
Special Feature: Advances in Quantitative
Bioimaging. Guest edited by Ricardo
Henriques, Ilaria Testa, Christophe Leterrier and
Aubrey Weigel.
Mapping molecular complexes with
super-resolution microscopy and
single-particle analysis
Afonso Mendes1,†, Hannah S. Heil1,†, Simao Coelho1, Christophe Leterrier2 and
Ricardo Henriques1,3
1Instituto Gulbenkian de Ciência, Oeiras, Portugal
2Aix Marseille Université, CNRS, INP UMR7051, NeuroCyto, Marseille, France
3MRC Laboratory for Molecular Cell Biology, University College London, London, UK
AM, 0000-0001-7324-555X; HSH, 0000-0003-4279-7022; SC, 0000-0002-0763-3172;
CL, 0000-0002-2957-2032; RH, 0000-0002-2043-5234
Understanding the structure of supramolecular complexes provides insight into
their functional capabilities and how they can be modulated in the context of
disease. Super-resolution microscopy (SRM) excels in performing this task by
resolving ultrastructural details at the nanoscale with molecular specificity.
However, technical limitations, such as underlabelling, preclude its ability to
provide complete structures. Single-particle analysis (SPA) overcomes this
limitation by combining information from multiple images of identical struc-
tures and producing an averaged model, effectively enhancing the resolution
and coverage of image reconstructions. This review highlights important
studies using SRM–SPA, demonstrating how it broadens our knowledge by
elucidating features of key biological structures with unprecedented detail.
1. Introduction
Organized assemblies containing several copies of one or more molecular entities
are a hallmark of the living world, and their existence underlies every biological
process. Studying their structure and assembly dynamics is crucial to understand-
ing their formation, function and resulting activity. Typically, these structural
assemblies are characterized by specific molecular interactions, recruitment,
conformational changes and catalysis of subsequent reactions. These complex
supramolecular constructs range from cellular components to viruses and display
structural redundancy across different levels of complexity (figure 1a). For
example, the human immunodeficiency virus type 1 (HIV-1) capsid, which
houses and directs the intracellular trafficking of the viral genome towards
the nucleus during infection [5], is an approximately 40 MDa supramolecular
complex containing at least 1500 copies of the HIV-1 capsid protein (CA) [1].
Notably, its formation involves assembling the CA monomers into approximately
12 pentameric and 250 hexameric complexes and their subsequent congregation
to form a final and more complex structure [2,3] (figure 1b).
Among the experimental tools to study biological structures, microscopy
stands out by allowing their direct observation. In particular, fluorescence
microscopy allows observations with molecular specificity. The spatial resolution
achievable with conventional light-based microscopy is limited to approximately
half of the wavelength (approx. 200 nm) [6]. Electron microscopy (EM) is cur-
rently the microscopy modality that achieves the greatest resolutions, being the
only suitable option to resolve structures below a certain size. However, it pro-
vides limited information on the identity of molecular components and the
precise location of a substantial population of molecules within very small com-
plexes. The advent of super-resolution microscopy (SRM) has enabled nano-sized
© 2022 The Authors. Published by the Royal Society under the terms of the Creative Commons Attribution
License http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the original
author and source are credited.
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
structures to be resolved while retaining the molecular contrast
provided by molecule-specific fluorescent labelling.
Despite the potential of SRM to resolve small molecular
structures, factors such as low labelling densities and localiz-
ation uncertainty hinder its ability to map supramolecular
complexes with high precision and generate complete models.
Single-particle analysis (SPA) is an analytical method that com-
bines information from several views of a structure of interest,
termed particle, and produces an averaged model (figure 1c).
It was first developed in the field of cryo-EM [7–10] but has
recently been applied successfully in SRM. Despite fundamental
differences between EM and SRM, SPA can overcome important
SRM limitations, such as under-labelling due to stochastic
fluorophore excitation. SPA is primarily used to map biological
structures by generating image reconstructions with a higher
signal-to-noise ratio (SNR) than the original images used.
Here, we provide an overview of SPA applications in SRM,
discuss current limitations and consider future developments.
2. Single-particle analysis overcomes critical
single-molecule localization microscopy
limitations
Current SRM technologies can be classified into two broad
categories based on the principle underlying their ability
to achieve sub-diffraction resolutions. One type includes
methods that modulate the light’s path by exploring different
illumination
schemes,
such
as
structured
illumination
microscopy (SIM) [11] and stimulated emission depletion
(STED) microscopy [12]. The second category includes modal-
ities that exploit the intrinsic properties of fluorophores,
namely the ability to control their stochastic ‘on/off’ switching
(i.e. ‘blinking’) and their excitation/emission spectra (i.e.
photoconversion). This last category contains methods such as
photoactivated localization microscopy (PALM) [13], direct
stochastic
optical
reconstruction
microscopy
(dSTORM)
[14,15] and point accumulation for imaging in nanoscale top-
ology (PAINT) [16,17], which enable the reconstruction
imaged structures from the localization coordinates of single
and isolated fluorophores. These are known as single-molecule
localization microscopy (SMLM) modalities.
At the interface of these two categories lies minimal photon
fluxes (MINFLUX), which uses a doughnut-shaped excitation
light beam to detect the presence of a fluorophore and then tri-
angulates its exact localization by refining the beam’s position
until the doughnut’s zero centre matches the fluorophore’s pos-
ition, thus, minimizing the flux of photons [18]. A more detailed
explanation of the principles, advantages and disadvantages of
each SRM modality is available in [19].
Despite its ability to reveal the localization of individual mol-
ecules within supramolecular complexes, SMLM has notable
limitations such as labelling artefacts, target-to-label distances
(i.e. linking error) and lack of temporal information. For example,
single proteins are typically imaged using antibody-based
A
alignment
selection
averaging
model
HSV-1
~170 nm
VACV
~400 nm
centriole
~250 nm
clathrin-coated pits
~100 nm
NPC
~120 nm
monomers
hexamers
capsid
HIV-1 particle
scale
(a)
(b)
(c)
Figure 1. Assembly of supramolecular complexes. (a) Supramolecular complexes such as clathrin-coated pits, the NPC, the HSV-1, the centriole and the VACV
comprise highly ordered structures across a size range of several hundred nanometers. (b) Structural redundancy occurs at different scales within the same biological
assembly. For example, the HIV-1 capsid is composed of at least 1500 copies of the HIV-1 capsid protein (CA) [1], which assemble into approximately 250 hexameric
CA complexes before congregating to form the fully assembled capsid [2,3]. (c) Supramolecular complexes can be mapped by combining SRM with SPA. For this,
several image sections containing super-resolved views of single particles are chosen. A filtering step removes non-matching objects based on predefined criteria. The
remaining image sections are aligned, and an averaged view of the object’s architecture is generated, which is then used to infer a model of the supramolecular
structure. (NPC and HIV-1 CA structures were created using Mol* [4]; PDB IDs: 7N9F (NPC), 2M8N (CA monomer), 6OBH (CA hexamer) and 3J3Y (HIV-1 capsid)).
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
2
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
labelling (i.e. immunolabelling) or fusions between fluorescent
tags and the proteins of interest. In immunolabelling, the anti-
body’s epitope is separated from the fluorophore by a short
linker, resulting in a slight displacement of the fluorescent signal
relative to the actual binding site in the target molecule. In turn,
fusions with fluorescent tags minimize this linking error. Still,
they can disturb the target proteins’ function, an effect often
related to steric hindrance or alterations in their spatial distri-
bution and binding affinities (e.g. [20]). Another caveat of
SMLM is related to the density and coverage of the labelling.
SMLM stochastically samples the distribution of labels in a speci-
men [15]. Thus, there is no guarantee that all the fluorophores in a
sample are acquired or excited proportionally. Furthermore, in
immunolabelling, different regions within a supramolecularcom-
plex might not be equally accessible to the labelling antibodies,
resulting in heterogeneous or incomplete coverage. In addition,
immunolabelling is performed using chemically fixed structures.
Although this allows for a snapshot of critical cellular activities
and subsequent nanometer characterization, the spatial-temporal
progression of the cellular activities is confined to the time of
fixation and may contain fixation-induced artefacts.
SPA
is
a computational
approach
that
significantly
enhances structural studies using microscopy. In SPA, particles
are compared, and goodness-of-fit metrics such as cross-corre-
lation [21,22] evaluate the particle before averaging. The typical
procedure of SPA comprises (i) image acquisition, (ii) particle
detection and segmentation, (iii) selection, (iv) alignment, (v)
averaging and (vi) model generation. An example pipeline
showcasing a dSTORM dataset containing the nuclear envel-
ope of a Xenopus oocyte labelled for the nuclear pore
complex (NPC) protein gp210 [23] (data available at https://
doi.org/10.5281/zenodo.5068525) is shown in figure 2.
3. Mapping supramolecular structures at
the nanoscale with super-resolution
microscopy–single-particle analysis
The studies referenced in the following sections highlight the
potential and limitations of SRM and SPA in mapping the
architecture of biological molecular assemblies.
3.1. The nuclear pore complex
The NPC mediates and regulates the translocation of molecules
across the nuclear envelope. It is one of the largest supramole-
cular complexes in the eukaryotic cell, with a diameter of
approximately 125 nm and a height of approximately 70 nm.
Its cylindrical structure with eightfold symmetry is composed
of multiple copies of at least 30 different proteins called nucleo-
porins (Nups) that form a cytoplasmic ring, a nuclear ring and
a central channel with a diameter of 35–50 nm [25–27]
(figure 1a). Molecules crossing via the NPC need to be at
least slightly smaller than its central channel. Thus, knowing
the exact diameter of the NPC is crucial to understand which
molecules or molecular complexes can travel across it. This
becomes particularly relevant in the context of infection with
pathogens trying to reach the cell’s nucleus, such as HIV-1
[28]. In addition, owing to key structural features such as
large size, radial symmetry and low variability between
individual entities, NPCs became a preferential biological
structure to evaluate SRM–SPA methods [29]. Furthermore,
while averaged reconstructions of NPC densities were made
using EM–SPA [30], the organization of its components
within the complex remained unknown.
Schermelleh et al. [31] used three-dimensional SIM to
resolve single NPCs with molecular specificity, but the spatial
resolution was insufficient to reveal fine structural details.
Löschberger et al. [32] tackled this problem using dSTORM
to image NPCs in nuclear envelopes isolated from Xenopus
laevis (frog) oocytes. They focused on gp210, a protein that
forms homodimers and then multi-mers that surround and
anchor the NPC on the luminal side of the nuclear pore mem-
brane [33]. Previous experiments using biochemical tools
suggested a model where at least eight gp210 dimers were
present in each NPC [34]. Löschberger et al. [32] resolved
an eightfold symmetrical arrangement of gp210 and calcu-
lated an average diameter for the complex of 161 ± 17 nm.
They generated images from 426 individual rings (approx.
160 000 localizations) that were automatically detected and
aligned to confirm the statistically observed eightfold sym-
metry. The final averaged image displayed the eightfold
symmetry of the gp210 ring previously observed and a diam-
eter of 164 ± 7 nm. Attempts to distinguish gp210 monomers
1 . . . n
model
alignment +
averaging
selection
particle detection and segmentation
super-resolved
image
SMLM
WF imaging
(a)
(b)
(c)
(d)
(e)
(g)
( f )
Figure 2. Example SPA framework using dSTORM data of the NPC protein gp210. (a–c) A super-resolved image is acquired before the analysis. (d) Particles
corresponding to the structure of interest are manually or automatically detected from a super-resolved image, generating an image library containing several
segmented particles. (e) The particle population is filtered to remove unwanted objects according to specific criteria. (f ) Each particle is aligned and used in
an iterative averaging process [24], resulting in a final model (g) with enhanced structural accuracy. Scale bars represent 2 µm (a–d) and 100 nm (e–g).
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
3
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
and dimers failed because the linkage error (approx. 18 nm)
was too high to achieve the required spatial resolution.
Finally, the authors exploited the high binding affinity of
fluorescently labelled wheat germ agglutinin for N-acetylglu-
cosamine-modified Nups to resolve the inner lining of the
central channel of the NPC. The same analytical approach
was performed by combining 621 rings (approx. 40 000
localizations) and calculating a 41 ± 7 nm pore diameter.
Szymborska et al. [35] extended this type of approach to
other components of the NPC using a distinct SPA routine.
They labelled Nup133 in.whole cells and mapped the pos-
ition of the fluorophores relative to the centre of the nuclear
pore with a precision of 0.1 nm and an accuracy of 0.3 nm.
The authors conducted a thorough characterization of the
NPC’s
structure
by
labelling
several
members
of
the
Nup107-160 complex, a primary component of the NPC. In
doing so, they addressed the existence of three contradictory
models for the orientation of the subcomplex and the radial
position of its components on the nanometer scale [36–38].
The same SPA framework was used to map the relative
positions of the proteins in the complex. Nup-GFP fusion
proteins and dye-coupled anti-GFP nanobodies were used
to minimize the linking error and overcome the inability
to
obtain
antibodies
for
several
Nup107-160
members
working in SRM. Using this methodology, they calculated
similar ring diameters, and the differences between the two
approaches corresponded to the difference in the linker’s
size between the two approaches. These measurements con-
cluded that a head-to-tail arrangement of the Nup107-160
complexes along the nuclear pore’s circumference probably
explained the data. Importantly, they showed that the devel-
oped SPA implementation could be applied to asymmetric
structures without losing precision if a molecular reference
in a second colour channel was used.
Many studies on the nanoarchitecture of NPCs followed,
and this subject remains an intense area of research today.
Some of these studies will be discussed further in light of
the SPA implementations used (e.g. LocMoFit [39]; figure 3a).
3.2. Viruses
Viruses are small multi-protein entities that can replicate
their genome within a host cell. Due to their small size,
viral structures have historically been described using EM
[45]. However, this approach is limited in mapping specific
molecules’ position and molecular identity within viral
supramolecular complexes because they either use labelling
methods that probe only a small subset of protein copies
(e.g. gold nanoparticle immunolabelling) or take advantage
of light microscopy methods that fall outside the SMLM
domain. SRM–SPA overcomes this limitation by using data
from thousands of labelled viral particles to provide averaged
three-dimensional models with molecular specificity.
Moreover, Gray et al. [40] developed VirusMapper. This
SPA framework automatically performs detection, segmenta-
tion,
alignment,
classification
and
averaging
thousands
of
individual
viral
particles
to
generate
high-resolution
reconstructions of viral particles with substructural detail.
Importantly, VirusMapper is a user-friendly and open-source
ImageJ/Fiji implementation. The authors evaluated the analytic
capabilities of the framework using multi-colour SIM and STED
imagesof Vacciniavirus (VACV), the smallpoxvaccine virus [46].
VirusMapper enabled mapping the relative localizations of
several viral proteins within the fully assembled virion.
Importantly, it also allowed the detection of changes in the
arrangement of a particular viral protein (L4) in the viral com-
plex induced by the virion’s fusion with the cell membrane, a
feature that even EM had not been able to demonstrate robustly.
In a follow-up study [47], Gray et al. used VirusMapper and an
extensive collection of virus mutants to investigate the nanoscale
organizationof theVACVentry complex, asupramolecularcom-
plex locatedontheviralmembranethatiscrucialforbindingand
fusion with the membrane of the host cell. They determined that
fusion activity and success correlated with a specific orientation
of the virion and the spatial distribution of viral binding and
fusion proteins (figure 3b).
The Herpes simplex virus type 1 (HSV-1) is a common cause
of cold sores but can also cause blindness and encephalitis
[48,49]. It is highly contagious and establishes a remarkably
persistent and latent infection in sensory ganglia with periodic
reactivation leading to symptomatic or asymptomatic virus
shedding [50]. The structure of the virus, an approximately
200 nm diameter sphere with a 125 nm diameter icosahedral
capsid housing the viral DNA genome, was first determined
by EM/ET [51–53]. The most complex layer of the virus is
called tegument and contains over 20 proteins [54]. Laine
et al. [41] used two-colour dSTORM to map the relative localiz-
ation of several HSV-1 proteins in the viral envelope and
tegument. Using viral proteins fused with fluorescent proteins,
they resolved individual capsids inside infected cells. They
could discriminate between enveloped and non-enveloped
particles and reveal the subcellular localization of viral com-
ponents and their relative amounts in the cytoplasm and the
nucleus. With this information, the authors classified the
resolved structures based on the presence of a capsid, tegument
or envelope, effectively achieving nanoarchitecture-based
characterization of single virions. A SPA framework was devel-
oped and used to map more than 50 viral particles. The
algorithm generates averaged models for individual viral pro-
teins based on their localization and radii distribution. Also,
the thickness of each protein layer is estimated from the popu-
lation’s diameter variability and deviation from spherical
symmetry (figure 3c).
HIV-1 is the causative agent of acquired immunodeficiency
syndrome (AIDS). The HIV-1 genome encodes at least 12 func-
tional proteins [55], but several cellular factors are required for
its successful replication, including members of the endosomal
sorting complexes required for transport (ESCRT) [56–58].
Understanding the nanoscale organization and function
of ESCRT subcomplexes at viral assembly sites is crucial to
eluicidate the mechanisms underlying HIV propagation. Van
Engelenburg et al. [59] used PALM to image ESCRT subcom-
plexes that interact directly with HIV-1 Gag (the main
component of the HIV-1 capsid) and play a role in HIV-1 mem-
brane abscission. They expressed ESCRT proteins fused with
fluorescent proteins in the COS7 cell line and resolved struc-
tural details of viral assembly sites at the plasma membrane.
Furthermore, using a Gag-mEOS2 protein fusion as a reference,
they detected ESCRT subcomplexes forming clusters at the
plasma membrane that were discernible from their correspond-
ing cytosolic pools. Importantly, a two-colour SPA approach
allowed the authors to quantitatively determine that large
fractions of some ESCRT proteins become trapped inside the
HIV-1 Gag lattice of viral-like particles during viral assembly.
Furthermore, the deletion of a specific amino acid motif in
Gag precluded this phenomenon.
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
4
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
(i)
(ii)
Abp1-mMaple
Las17-SNAP
(ii) running average
inferred time
<–7 s
–7 s
–5 s
–3 s
–1 s
1 s
(i) individual
envelope (gD)
outer tegument (VP16)
inner tegument (pUL37)
inner tegument (VP1/2)
segmented and aligned data
orientation separation in
seed selection
model generation
raw data
model
(b)
(a)
(i)
(ii)
(iii)
(iv)
(v)
side view
CR
NR
top view
frontal
0% match
99% match
A4
F17
A4
F17
F17
A4
sagittal
frontal
sagittal
Nucleus
d
c
b
a
f
e
CCDC41 +
SCLT1
CCDC41 +
FBF1
CCDC41 +
Cep164
SCLT1 + FBF1 +
CCDC41 + Cep164
SCLT1 +
FBF1
( f )
(e)
(d)
(c)
centriole
cilia
clathrin-mediated endocytosis
98% match
HSV-1
VACV
nuclear pore complex
Figure 3. Overview of biological structures analysed with SRM–SPA. (a) NPC. Single NPC particles (raw data) were assumed to represent samples of the same
underlying distribution. A particle was found (in the example the particle (iv)) so that it best describes all other sites based on the rank on sum LL of the all-to-all
matrix, where the 50 subset sites were fitted to each other. The initial template is built based on sequential registration in the order of the sum LL rank. The final
fused particle is used to register all sites in the 318-site dataset. This procedure yields an updated fused particle, which is used to register the dataset again. This
process is iterated until convergence. The final average (model) was calculated from 318 particles without any assumption on the underlying geometry or symmetry
in a tilted view (mode, top left). For comparison, the EM density of the NPC with C-termini of Nup96 is indicated in red (model, top right). Top and side views
(model, bottom), where the nucleoplasmic and cytoplasmic rings are shown together (left panel), or separately (middle, right panels). The two proteins per ring per
symmetric unit give rise to tilted elongated average protein distributions in the averages (arrows in model, bottom). Scale bars represent 50 nm. Adapted from [39].
(b) VACV. Segmented and aligned particles are used to generate multi-component models of single virions. For this, more than one seed selection criterion is applied
to a single reference channel (VACV A4 frontal and sagittal pictured). Additional virion components (VACV F17 pictured) are aligned to the reference for generating
virion orientation-based (frontal and sagittal) models of various viral components. A composite image of all sagital models is shown (A17(cyan)/CM(yellow)/A4(red)/
DNA(blue)/L4(green)/F17(magenta)). Scale bar represents 100 nm. Adapted from [40]. (c) HSV-1. Virus images obtained from aligned particles (right, larger insets)
and individual representative particles (left, smaller insets). Scale bar represents 100 nm. Adapted from [41]. (d) Clathrin-CME. (i) Dual-colour side-view super-
resolution images of Las17-SNAP and Abp1-mMaple at individual sites. Images were rotated so endocytosis occurs upward and sorted by the distance of
Abp1 centroid to Las17 at the base. (ii) Averages of Las17 and Abp1 at endocytic sites. For comparison, average outer boundaries of the actin network
(dotted lines) and average plasma membrane profiles (solid line) obtained by correlative light-EM [42] are overlaid for each time point, as inferred from the
images. Scale bar represents 100 nm. Adapted from [43]. (e) Cilia. (i) Two-dimensional STORM images of the ciliary DAs of mTEC cells with CEP164 labelled.
The STORM localizations of an individual structure are fitted to an ellipse, which is then deformed to a circle. The circularized structure is normalized to a
ring with a fixed diameter calculated by averaging the diameter of 31 original structures. Image (ii) shows the resulting averaged structure after alignment.
Scale bar represents 100 nm. Adapted from [24]. (f ) Centriole. RPE-1 C1-GFP cells were immunolabelled for the indicated proteins and imaged first in a
wide-field mode, followed by three-dimensional STORM imaging, and correlative EM analysis. Averaged STORM signals were pseudocoloured, rotated and then
superimposed to generate a horizontal distributional map of the DAPs. Scale bar represents 200 nm. Adapted from [44].
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
5
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
3.3. Clathrin-mediated endocytosis
Clathrin-mediated endocytosis (CME) is critical for many bio-
logical processes, such as signalling, nutrient uptake and
pathogen entry. CME involves the assembly of supramolecu-
lar complexes containing clathrin and many other proteins at
the plasma membrane, leading to the inward budding of an
intracellular vesicle. The vesicle is then pinched off the
plasma membrane and rapidly uncoats to allow its fusion
with endosomes. Because of the high conservation of CMEs
between yeast and humans, yeast models are often used to
study their structures and dynamics. The endocytic machin-
ery comprises sub-diffraction limit structures extensively
characterized using microscopy and SPA.
Berro & Pollard [60] developed a SPA framework that
tracks protein patches located at the endocytic sites and
then
performs
particle
alignment
and
averaging.
This
approach captures in super-resolution the endocytic pathway
as a function of the conformational state. This is done by clas-
sifying different conformational states within the membrane,
thus inferring sequential progress. Among other discoveries,
the authors mapped the average positions of endocytic
proteins
along
membrane
invagination
using
confocal
fluorescence microscopy. Furthermore, Picco et al. [61] devel-
oped
a
similar
SPA
approach
to
characterize
protein
dynamics during vesicle budding. They combined live-cell
imaging data with previous EM structural data and charac-
terized the endocytic process thoroughly. Finally, Mund
et al. [43] developed a SPA framework that uses SRM
images to characterize the dynamics of vesicle budding
during CME. They fluorescently tagged 23 different endocy-
tic proteins and imaged thousands of fixed yeast cells,
collecting data from more than 100 000 endocytic sites.
Using live-cell imaging data as a temporal reference, the
average positions of labelled proteins at endocytic sites
were mapped with unprecedented accuracy, and three-
dimensional models of their distribution at different stages
of endocytosis were generated (figure 3d).
3.4. Cilia
Cilia are small organelles that take the shape of a protrusion
projecting from the cell body. Their type functions as cellu-
lar antennas to integrate environmental signals or create
extracellular flows by continuous beating. They are generated
from the basal body, a cylindrical structure composed of triplet
microtubules arranged with ninefold symmetry [62–65]. The
skeleton of cilia, called the axoneme [66], comprises microtu-
bule doublets arranged with ninefold symmetry around a
central microtubule doublet and is contained by a membrane
contiguous to the plasma membrane. In humans, the impor-
tance of cilia is highlighted by several diseases, called
ciliopathies, which arise from defective ciliary function typi-
cally caused by mutations. These include polycystic kidney
disease, polydactyly and Joubert syndrome (JBTS).
Similar to NPCs, cilia are the subject of several SRM
studies. In particular, the ‘transition zone’ separates the cili-
ary and plasma membranes and is proposed to function as
a gate to the cilium [67]. EM studies on the transition
zone’s structure revealed ‘Y’-shaped densities called Y-links,
with the stem anchored at the microtubule doublets and
the two arms attached to the ciliary membrane [68]. However,
the Y-links’ components and the arrangement of proteins
in the transition zone were unknown. Shi et al. [69] imaged
the transition zone using a two-colour three-dimensional
STORM. They observed several proteins forming rings with
different diameters localized between the axoneme and the
ciliary membrane. Based on the radial, angular and axial dis-
tribution of the ciliary components, they constructed a three-
dimensional map of the transition zone with a resolution of
15–30 nm, demonstrating that the protein rings observed
were consistent with being Y-links. A protein called Smooth-
ened (SMO) formed a ring of discrete clusters specifically
in the transition zone. The authors revealed that certain
JBTS-associated mutations reduced SMO localization in the
transition zone and the ciliary membrane.
This study also showed that the diameter of the rings com-
posing the ciliary transition zone, often with an elliptical shape,
varied from 369 to 494 nm. This reveals an important structural
characteristic of many large cellular organelles called semi-
flexibility. Briefly, semi-flexible structures can display slightly
different sizes and shapes due to elastic deformation or mol-
ecular composition variations while maintaining symmetry
and angular arrangement. This heterogeneity poses a problem
for particle alignment and averaging, reducing the accuracy of
the final reconstructed images. To address this problem, Shi
et al. [24] developed SRM–SPA algorithms that take structural
flexibility as a degree of freedom for image registration. They
work by deforming heterogeneous elliptical structures into
more uniform ones and then aligning and averaging the
deformed structures. The algorithms were evaluated using
simulated and experimental SMLM data of ciliary appendages
from mouse tracheal epithelial cells (figure 3e). They resolved
the ninefold ciliary symmetry and generated multi-colour
three-dimensional models with a higher resolution than
state-of-the-art algorithms based on rigid structure registration
[70–72].
Importantly,
particle
deformations that
can
be
explained with the structural semi-flexibility model can instead
be a consequence of perspective, i.e. the observation of particles
randomly oriented in the sample. The method developed in
this study can correct for this aspect.
Robichaux et al. [73] combined cryo-ET and STORM ima-
ging to map repeating structures in a specialized structure of
the rod sensory cilium called ‘connecting cilium’. Subtomo-
gram averaging provided a structural framework of the
connecting cilium, used as a reference to map the distribution
of specific molecular components resolved with STORM. This
method allowed mapping subcompartments of the connect-
ing cilium with high precision and revealed structural
differences in knockouts for basal body proteins.
3.5. Centriole
The basal ciliary body originates from the centriole, a cellular
structure [74] sharing a similar organization, including micro-
tubule triplets arranged radially with ninefold symmetry
[75]. They display proximal-distal polarity. The proximal
end shows a supramolecular matrix called the pericentriolar
material [76–78], which is important for microtubule nuclea-
tion and centriole duplication, the latter resulting in a mother
and a daughter centriole. The distal end of the mother cen-
triole contains projections called distal appendages (DAs),
which are required for ciliogenesis [79–85], microtubule
anchoring and centriole positioning [86–88]. Several proteins
are distributed around the centriole’s centre, forming 100–
200 nm diameter rings. Due to centrioles’ small size and
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
6
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
high molecular complexity, mapping the exact distribution of
several centriole proteins remains a challenge.
Gartenmann et al. [89] devised a SPA method to resolve
the diameter of centriole rings with unprecedented precision
in Drosophila larval wing discs. Their framework involved
an initial imaging step with a three-dimensional SIM to
resolve the ring of the well-established centriole protein Asl.
Then, several GFP fusions to centriole proteins were imaged
with SMLM and using the centre of the Asl rings as initial
estimates for the centriole’s centroid, localizations within
100 nm were averaged to calculate a new centre. The process
was repeated until the centroid positions converged. Finally,
the average radius of the rings was calculated, and the
distribution of GFP fusion proteins in the centriole was recon-
structed. By taking advantage of the high labelling density of
three-dimensional SIM, the high precision of SMLM and the
resolution enhancements and statistical robustness of SPA,
this framework allowed calculating radii with an accuracy
of ±4–5 nm.
Bowler et al. [44] used a combination of cryo-ET, STORM
imaging and SPA to map the precise location of DA proteins
in three-dimensional (figure 3f ). They revealed the dynamic
nature of DAs by demonstrating with ultrastructural detail
the reorganizations that occur before and during mitosis.
The centriole is also a central component of the centro-
some, a microtubule-organizing structure comprising a pair
of centrioles responsible for the mitotic spindle formation
during cell division. Sieben et al. [72] developed SPARTAN,
a graphical user interface employing a novel SRM–SPA
framework that generates three-dimensional structure recon-
structions from two-channel two-dimensional SMLM data.
In this study, centrosomes from a human cell line were puri-
fied and concentrated in coverslips by centrifugation before
imaging with STORM. A library of densely labelled particles
(10%−20% of the entire population) was created, and the par-
ticles
were
aligned
and
classified
according
to
their
orientation. An averaged model for one labelled protein
was generated, on which models for other proteins could
be mapped. Using this tool, they confirmed the ninefold sym-
metry of Cep164, a crucial centriolar component. They also
revealed unknown features of the human centriole’s architec-
ture, such as closer proximity of the Cep164 N-terminus to
the centriolar wall than previously reported [78].
4. Quantitative structural descriptions
outside the scope of single-particle
analysis
SPA refers to frameworks that register and average particles
to produce a final model with improved SNR. Despite com-
parable analytical methods (e.g. pair-correlation) sharing
key statistical approaches, such as numerical descriptions of
ensemble parameters (e.g. particle diameter), these are not
well suited to produce improved image reconstructions—
unlike SPA.
For example, spatial descriptive statistics such as pair-cor-
relation [90], Ripley’s K-function [91] and density-based
spatial clustering of applications with noise [92] have been
employed to determine clustering states and average cluster
sizes of particle populations. Gunzenhäuser et al. [93] devel-
oped a quantitative method to reveal the functional and
morphological aspects of the HIV-1 assembly based on
SMLM data. They analysed hundreds of HIV-1 Gag clusters
and determined the number of Gag molecules per cluster,
as well as the frequency distribution of the clusters with
respect to their radius and aspect ratio. Malkusch et al. [94]
evaluated the performance of different cluster analysis
methods in the same context. They were able to classify
Gag clusters corresponding to three different stages of viral
assembly. Finally, Floderer et al. [95] expressed assembly
defective Gag mutants in live cells to follow the trajectories
of individual molecules. They calculated the duration and
the energy proportions involved in each assembly stage.
A type of analytical tool that has recently been explored in
microscopy is deep-learning (DL). Examples of its appli-
cations in bioimage analysis are image processing (e.g.
denoising) and particle tracking and segmentation, with
several user-friendly tools available to researchers as open-
source (e.g. [96]). The use of DL in SPA is showcased in
[97]. Here, six DL models were trained and benchmarked
on their ability to recognize the septin ring, a structure arising
in cells treated with the actin polymerization inhibitor cyto-
chalasin D. Models were able to identify hundreds of septin
rings with high accuracy. The segmented rings were then
used for particle averaging, producing models with sensi-
tivity for slight structural deviations induced by a septin
ring-related gene knockout.
Finally, supramolecular complexes can be mapped using
different fluorescent labels in the same protein. For example,
Mennella et al. [77] resolved kendrin/pericentrin, a com-
ponent of the pericentriolar material in human centrioles,
with antibodies against the protein’s N-terminus and a GFP
tag in the C-terminus using SIM. The distance from the differ-
ent fluorescent signal to the centre of the structure allowed to
determine the protein’s orientation in the supramolecular
complex, with the C-terminus positioned closer to the
centre and the N-terminus extending outwards with radial
symmetry. Another example is Leterrier et al. [98], where
the nanoscale organization of the neuronal protein ankG in
a region of the axon called the axon initial segment was deter-
mined using antibodies targeting different domains of the
protein. A known spectrin-binding domain of ankG, which
is closer to the protein’s N-terminus, was colocalized with
spectrin
bands
just
under the
plasma
membrane
and
displayed a marked periodicism. By contrast, downstream
domains closer to the C-terminus progressively lost this
periodicity and were found to be located deeper in the
cytoplasm than the N-terminus by three-dimensional STORM.
5. Current challenges in single-particle
analysis
The studies mentioned in the previous sections showcase
the potential of SRM–SPA to map the nanoarchitecture
details of supramolecular complexes. However, they also
highlight
current
limitations.
In
general,
the
increased
computation resources and times required to perform the
analyses in novel SPA frameworks are minimized by exploit-
ing
different
mathematical
approaches
and
hardware
specifications, such as fast Fourier transforms [99] and
graphical processing units (e.g. [100]). Additionally, struc-
tures occluded by noise in SMLM data can be revealed by
introducing spatial descriptive statistics in the SPA pipelines
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
7
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
[101]. However, more fundamental challenges remain. In par-
ticular, crucial steps in SPA methodologies involve using an
initial template. Particles detected in a dataset can be included
or excluded from the analysis with respect to how well they
match the chosen template. Additionally, particle averaging
involves iterative refinement of the shape of an initial template.
The template chosen is typically arbitrary, often consisting of a
simplified shape recapitulating a priori knowledge on the struc-
tures of interest. For example, a simple spherical shape can be
used as a template if the structure of interest is a mature HIV-1
or HSV-1 particle. Another option is to select a representative
particle from the dataset, which can be a single particle, a sub-
population or even a class average (e.g. VirusMapper [40]). The
problem shared among these approaches is that the final
reconstructed structures are biased towards the template used.
Another important limitation of SPA arises from the
assumption that all particles represent the same underlying
structure. However, this is not necessarily true, as most
particle populations are expected to display some degree of
heterogeneity. For example, some biological structures spora-
dically or transiently manifest as variations of their canonical
structure, such as the NPC variant with ninefold symmetry
[35], semi-flexible ciliary structures [24] and multi-core
HIV-1 particles [102]. These particular structures are under-
represented in the final reconstructions, decreasing their
accuracy and remaining undetected.
5.1. Novel particle registration approaches minimize
template bias
Several studies focused on developing so-called ‘template-free’
SPA methodologies, which use non-arbitrary or data-driven
templates to overcome template bias. For example, Fortun
et al. [103] developed a SPA framework that reconstructs
three-dimensional
supramolecular
assemblies
from
two-
dimensional SRM images of particles in different orientations.
While this framework uses a small number of hand-picked
templates to detect particles, it estimates their orientations
and performs volume reconstructions without resorting to a
priori knowledge, creating an initial model that is iteratively
refined by averaging. This framework allowed mapping the
distribution of Cep63 around the central centriole barrel from
STED images with unprecedented accuracy.
Salas et al. [70] used multi-variate statistical analysis
to classify particles according to their orientation. Repre-
sentative examples of each class were then selected based
on the visual match between the class average and the indi-
vidual
particles
and
used
to
perform
multi-reference
alignments. The algorithm successfully produced high-resol-
ution three-dimensional reconstructions of DNA origami
[104] structures and T4 bacteriophages from SMLM and
simulated data.
Furthermore, Heydarian et al. [100] developed an ‘all-to-
all’ registration approach. Each particle is registered pairwise
against every other particle, producing similarity metrics
and estimates of each pair’s relative orientation and position.
All the possible absolute positions and orientations of each
particle are then calculated using a technique from the field
of computer vision called ‘structure from motion’ [105].
A quality control step is performed to remove outliers
and registration errors by inferring the combinations of rela-
tive parameters from the absolute parameters obtained
previously. These retrodicted parameters are then compared
to those found in the all-to-all registration, and pairs deviating
from their registered counterparts above a specified threshold
are discarded from the next steps. These procedures generate a
data-driven template that is further refined by particle aver-
aging. The algorithm’s performance was first evaluated
using a dataset containing two-dimensional DNA origami
particles imaged with DNA-PAINT and generated image
reconstructions
with
approximately
3 nm
resolution.
In
addition, the algorithm was able to perform using particles
with labelling densities as low as 30%. The algorithm was
also tested using the NPC dataset in [32] and successfully
retrieved the NPC’s eightfold symmetry without requiring
prior knowledge. More recently, this framework was extended
to enable three-dimensional SPA [106].
Another example is Blundell et al. [107], where DL was
used to predict each particle’s pose. Here, a neural network
fits a three-dimensional model against a library of particles
viewed in two-dimensional. The framework was able to dis-
cern the centriole’s expected toroidal structure from the
STORM dataset acquired in [72].
It is important to note that, although ‘template bias’ exists
when using arbitrary models as an initial reference for SPA,
completely discarding prior knowledge on the structures of
interest might not be the most sensible action. Indeed, the
‘template-free’ algorithms showcased so far produce recon-
structions that can be further improved by taking prior
knowledge into account. On this note, Wu et al. [39] devel-
oped LocMoFit. This framework extracts particle geometry
from SMLM data and fits the particles to geometric models
based on maximum-likelihood estimation (MLE). When the
geometry of the structure of interest is known, the user can
decide on the class of geometric models to be used for the
fitting. Importantly, when an underlying geometry cannot
be recovered, the algorithm can perform particle averaging
to generate a data-driven template by combining the all-to-
all registration methodology developed by Heydarian et al.
[100] with MLE. Mund et al. [108] used LocMoFit to extend
their work on CME developed in [43]. Here, they densely
labelled clathrin in the human melanoma cell line SK-MEL-
2 and used three-dimensional SMLM to resolve the endocytic
sites. Then, they performed extensive quantitative descrip-
tions of the structures analysed and were able to infer a
spatial-temporal model of the endocytic process.
5.2. Improving sensitivity to detect structure
heterogeneity
Image classification can be used to enhance SPA’s lack of
sensitivity to particle heterogeneity. This usually involves
calculating similarity metrics (e.g. cross-correlation at different
relative rotations [21,22] for particle pairs and then sorting the
particles into a suitable number of classes). Several SPA frame-
works mentioned so far use classification to group particles
according to their viewing orientation (e.g. [40,72]). The par-
ticles’ relative orientations can be calculated to produce a
three-dimensional model from two-dimensional projections
with this information.
A more underlooked challenge of SPA that benefits from
particle classification is the ability to detect and classify
particles representing different underlying structures. This
becomes particularly difficult when the various structures
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
8
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
are not known a priori and reasonable templates cannot be
provided. Huijben et al. [109] extended the pipeline devel-
oped in [100] to achieve this goal. In this implementation,
the similarity metric obtained from the all-to-all registration
strategy is converted into a dissimilarity metric, and multi-
dimensional scaling [110] is used to translate the values of
each registration pair into spatial coordinates projected in a
multi-dimensional space. The multi-dimensional particles
are then classified using k-means clustering [111] and each
class is averaged to generate the final reconstructions. In par-
ticular, the authors provide a relatively simple strategy to
determine an optimal number of classes without a priori
knowledge. The framework was evaluated using a two-
dimensional DNA origami test dataset imaged with DNA-
PAINT [16,112], containing particles belonging to several
different classes. Under these conditions, the algorithm cor-
rectly classified greater than 95% of the analysed particles.
The averaged reconstructions achieved resolutions between
3.7 and 5.7 nm. Most importantly, the algorithm correctly
classified an NPC with ninefold symmetry in a simulated
dataset where this class was highly under-represented (2%
of the total number of particles). Furthermore, the algorithm
detected classes corresponding to elliptical NPCs in the
Xenopus oocyte NPC dataset mentioned previously [32].
Although the presence of these elliptical structures in the
data was deemed by the authors to be an artefact generated
during sample preparation without any biological signifi-
cance, their successful detection showcases once again the
high sensitivity of the classification framework and its ability
to detect unknown structures. In this context, Sabinina et al.
[113]
used
SPA
to
perform
quantitative
descriptions
and obtain an average model of the NPC. They observed
that the nuclear basket protein TPR was distributed over a
larger volume than expected, suggesting averaging over
different
NPC
conformations.
Accordingly,
they
found
significant variation in descriptive parameters, such as circu-
larity and diameter, which exceeded the expected registration
error. Thus, they developed a classification method that does
not depend on a priori information about the structures of
interest and used it to confirm the presence of different
NPC conformations.
6. Conclusion and outlook
Determining the fine structure of supramolecular complexes
gives crucial insight into their assembly dynamics and
functional capabilities. SRM excels at imaging small bio-
logical structures containing multiple components with
different molecular identities at high resolutions. It success-
fully tackles the challenge of surpassing the diffraction limit
of light by bringing together cutting-edge microscopy tech-
nology, modulation of the physical properties of fluorescent
labels and advanced statistical analysis. Furthermore, the
implementation of SPA in SRM substantially improves the
accuracy of image reconstructions overcoming crucial caveats
of SRM such as under-labelling and labelling heterogeneity.
It combines information from thousands of imaged structures
and produces reconstructions that effectively contain more
localizations than their underlabelled counterparts. This
enables the mapping of supramolecular complexes in three-
dimensional
and
retrieving
architectural
features
with
single-digit nanometer precision. Additionally, the analysis
can be performed in multiple colour channels representing
different labelled molecules present in the same molecular
assembly, which allows using a reference molecule to register
the remaining (e.g. [40]).
Despite the great success of SRM–SPA in mapping
supramolecular complexes, its capabilities have not yet been
fully explored. So far, SPA studies have focused almost
exclusively on generating reconstructions of a unique and
fully assembled structure. However, these same structures
undergo a process of assembly comprising multiple meta-
stable
structures
that
share
crucial
molecular
players.
Resolving the fine structural details of intermediate structures
is crucial to describing the assembly dynamics of the corre-
sponding
supramolecular
complexes.
There
are
several
obstacles to mapping metastable structures. For example,
the degree to which each structure is represented in the
data is a function of its relative stability or frequency, result-
ing in a heterogeneous representation. Combined with the
already
existing
caveats
of
SRM
(e.g.
under-labelling)
increases the chances that less stable structures are under-rep-
resented, potentially to a point where it becomes hard or even
impossible to acquire a sample with enough particles to accu-
rately represent the structure. In addition, structural plasticity
as observed in Shi et al. [24] and Sabinina et al. [113] also con-
tributes to under-representation of a particular structure.
Furthermore, the lack of temporal information due to
fixed-sample imaging precludes the mapping of the sequen-
tial assembly of supramolecular complexes. This problem
was addressed in Berro & Pollard [60], where a ‘brute force’
approach was taken. Briefly, CME was oversampled to
ensure that a suitable number of particles representing each
metastable structure were detected. Then, the assembly pro-
cess’s temporal hierarchy was estimated from the similarity
between metastable structures. A potential solution to this
problem might reside in correlating the fine structural details
resolved with SRM–SPA with structural and temporal infor-
mation extracted from live-cell SRM imaging. While still
being limited by the constraints of live-cell SRM today,
remarkable progress and future development of both fields,
SPA and live-cell SRM, will be the key to elucidating the
dynamics of the structural rearrangement of molecular
complexes within living cells.
Data accessibility. This article has no additional data.
Authors’ contributions. A.M.: conceptualization, visualization and writ-
ing—original draft; H.S.H.: conceptualization, visualization and
writing—original draft; S.C.: writing—review and editing; C.L.: writ-
ing—review and editing; R.H.: conceptualization, supervision and
writing—review and editing.
All authors gave final approval for publication and agreed to be
held accountable for the work performed therein.
Conflict of interest declaration. The authors declare no competing interests.
Funding. This work was supported by the Gulbenkian Foundation
(A.M., H.S.H., S.C. and R.H.) and received funding from the Euro-
pean Research Council (ERC) under the European Union’s Horizon
2020 research and innovation programme (grant agreement no.
101001332 to R.H.), the European Molecular Biology Organization
(EMBO-2020-IG-4734 to R.H. and ALTF 499-2021 to H.S.H), the Well-
come Trust (203276/Z/16/Z to R.H.), and NHMRC (APP1183588 to
S.C.). A.M. would like to acknowledge support from the Integrative
Biology and Biomedicine PhD programme from Instituto Gulbenkian
de Ciência.
Acknowledgements. We thank Marie-Christine Dabauvalle and Georg
Krohne for the nuclear envelope preparation and Xiaoyu Shi for
sharing the ‘Deformed Alignment’ code.
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
9
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
References
1.
Summers BJ, Digianantonio KM, Smaga SS, Huang
PT, Zhou K, Gerber EE, Wang W, Xiong Y. 2019
Modular HIV-1 capsid assemblies reveal diverse
host-capsid recognition mechanisms. Cell Host
Microbe 26, 203–216.e6. (doi:10.1016/j.chom.2019.
07.007)
2.
Ganser BK, Li S, Klishko VY, Finch JT, Sundquist WI.
1999 Assembly and analysis of conical models for
the HIV-1 core. Science 283, 80–83. (doi:10.1126/
science.283.5398.80)
3.
Li S, Hill CP, Sundquist WI, Finch JT. 2000 Image
reconstructions of helical assemblies of the HIV-1 CA
protein. Nature 407, 409–413. (doi:10.1038/
35030177)
4.
Sehnal D et al. 2021 Mol∗Viewer: modern web app
for 3D visualization and analysis of large
biomolecular structures. Nucleic Acids Res. 49,
W431–W437. (doi:10.1093/nar/gkab314)
5.
Campbell EM, Hope TJ. 2015 HIV-1 capsid: the
multifaceted key player in HIV-1 infection. Nat. Rev.
Microbiol. 13, 471–483. (doi:10.1038/nrmicro3503)
6.
Abbe E. 1873 Beiträge zur Theorie des Mikroskops
und der mikroskopischen Wahrnehmung. Archiv für
Mikroskopische Anatomie 9, 413–468. (doi:10.1007/
BF02956173)
7.
Frank J. 2009 Single-particle reconstruction of
biological macromolecules in electron microscopy-30
years. Quart. Rev. Biophys. 42, 139–158. (doi:10.
1017/S0033583509990059)
8.
Kudryashev M, Castaño-Díez D, Stahlberg H. 2012
Limiting factors in single particle cryo electron
tomography. Comput. Struct. Biotechnol. J. 1,
e201207002. (doi:10.5936/csbj.201207002)
9.
Tang G, Peng L, Baldwin PR, Mann DS, Jiang W,
Rees I, Ludtke SJ. 2007 EMAN2: an extensible
image processing suite for electron microscopy.
J. Struct. Biol. 157, 38–46. (doi:10.1016/j.jsb.2006.
05.009)
10. van Heel M et al. 2000 Single-particle electron cryo-
microscopy: towards atomic resolution. Quart. Rev.
Biophys. 33, 307–369. (doi:10.1017/
S0033583500003644)
11. Gustafsson MGL. 2000 Surpassing the lateral
resolution limit by a factor of two using structured
illumination microscopy. SHORT COMMUNICATION.
J. Microsc. 198, 82–87. (doi:10.1046/j.1365-2818.
2000.00710.x)
12. Klar TA, Jakobs S, Dyba M, Egner A, Hell SW. 2000
Fluorescence microscopy with diffraction resolution
barrier broken by stimulated emission. Proc. Natl
Acad. Sci. USA 97, 8206–8210. (doi:10.1073/pnas.
97.15.8206)
13. Betzig E, Patterson GH, Sougrat R, Lindwasser OW,
Olenych S, Bonifacino JS, Davidson MW, Lippincott-
Schwartz J, Hess HF. 2006 Imaging intracellular
fluorescent proteins at nanometer resolution.
Science 313, 1642–1645. (doi:10.1126/science.
1127344)
14. Rust MJ, Bates M, Zhuang X. 2006 Sub-diffraction-
limit imaging by stochastic optical reconstruction
microscopy (STORM). Nat. Methods 3, 793–795.
(doi:10.1038/nmeth929)
15. van de Linde S, Löschberger A, Klein T, Heidbreder
M, Wolter S, Heilemann M, Sauer M. 2011 Direct
stochastic optical reconstruction microscopy with
standard fluorescent probes. Nat. Protocols 6,
991–1009. (doi:10.1038/nprot.2011.336)
16. Schnitzbauer J, Strauss MT, Schlichthaerle T,
Schueder F, Jungmann R. 2017 Super-resolution
microscopy with DNA-PAINT. Nat. Protocols 12,
1198–1228. (doi:10.1038/nprot.2017.024)
17. Sharonov A, Hochstrasser RM. 2006 Wide-field
subdiffraction imaging by accumulated binding of
diffusing probes. Proc. Natl Acad. Sci. USA 103,
18 911–18 916. (doi:10.1073/pnas.0609643104)
18. Balzarotti F, Eilers Y, Gwosch KC, Gynnå AH,
Westphal V, Stefani FD, Elf J, Hell SW. 2017
Nanometer resolution imaging and tracking of
fluorescent molecules with minimal photon fluxes.
Science 355, 606–612. (doi:10.1126/science.
aak9913)
19. Jacquemet G, Carisey AF, Hamidi H, Henriques R,
Leterrier C. 2020 The cell biologist’s guide to super-
resolution microscopy. J. Cell Sci. 133, jcs240713.
(doi:10.1242/jcs.240713)
20. Mamede JI, Griffin J, Gambut S, Hope TJ. 2021 A
new generation of functional tagged proteins for
HIV fluorescence imaging. Viruses 13, 386. (doi:10.
3390/v13030386)
21. Bracewell R. 1999 Pentagram notation for cross
correlation. In The Fourier transform and its
applications (ed. RN Bracewell), pp. 46–243.
New York, NY: McGraw-Hill.
22. Papoulis A. 1962 The Fourier integral and its
applications. New York, NY: McGraw-Hill.
23. Heil HS et al. 2018 Sharpening emitter localization
in front of a tuned mirror. Light 7, 1–8. (doi:10.
1038/s41377-018-0104-z)
24. Shi X, Garcia G, Wang Y, Reiter JF, Huang B. 2019
Deformed alignment of super-resolution images for
semi-flexible structures. PLoS ONE 14, e0212735.
(doi:10.1371/journal.pone.0212735)
25. Elad N, Maimon T, Frenkiel-Krispin D, Lim RY,
Medalia O. 2009 Structural analysis of the nuclear
pore complex by integrated approaches. Curr. Opin.
Struct. Biol. 19, 226–232. (doi:10.1016/j.sbi.2009.
02.009)
26. Scheer U, Dabauvalle MC, Krohne G, Peiman Zahedi
R, Sickmann A. 2005 Nuclear envelopes from
amphibian oocytes - from morphology to protein
inventory. Europ. J. Cell Biol. 84, 151–162. (doi:10.
1016/j.ejcb.2004.12.001)
27. Strambio-De-Castillia C, Niepel M, Rout MP.
2010 The nuclear pore complex: bridging
nuclear transport and gene regulation. Nat.
Rev. Mol. Cell Biol. 11, 490–501. (doi:10.1038/
nrm2928)
28. Zila V et al. 2021 Cone-shaped HIV-1 capsids are
transported through intact nuclear pores. Cell 184,
1032–1046.e18. (doi:10.1016/j.cell.2021.01.025)
29. Thevathasan JV et al. 2019 Nuclear pores as
versatile reference standards for quantitative
superresolution microscopy. Nat. Methods 16,
1045–1053. (doi:10.1038/s41592-019-0574-9)
30. Brohawn SG, Partridge JR, Whittle JRR, Schwartz
TU. 2009 The nuclear pore complex has entered the
atomic age. Structure 17, 1156–1168. (doi:10.1016/
j.str.2009.07.014)
31. Schermelleh L et al. 2008 Subdiffraction multicolor
imaging of the nuclear periphery with 3D structured
illumination microscopy. Science 320, 1332–1336.
(doi:10.1126/science.1156947)
32. Löschberger A, van de Linde S, Dabauvalle MC,
Rieger B, Heilemann M, Krohne G, Sauer M. 2012
Super-resolution imaging visualizes the eightfold
symmetry of gp210 proteins around the nuclear
pore complex and resolves the central channel with
nanometer resolution. J. Cell Sci. 125, 570–575.
(doi:10.1242/jcs.098822)
33. Favreau C, Bastos R, Cartaud J, Courvalin JC,
Mustonen P. 2001 Biochemical characterization of
nuclear pore complex protein gp210 oligomers.
Europ. J. Biochem. 268, 3883–3889. (doi:10.1046/j.
1432-1327.2001.02290.x)
34. Gerace L, Ottaviano Y, Kondor-Koch C. 1982
Identification of a major polypeptide of the nuclear
pore complex. J. Cell Biol. 95, 826–837. (doi:10.
1083/jcb.95.3.826)
35. Szymborska A, de Marco A, Daigle N, Cordes VC,
Briggs JAG, Ellenberg J. 2013 Nuclear pore scaffold
structure analyzed by super-resolution microscopy
and particle averaging. Science 341, 655–658.
(doi:10.1126/science.1240672)
36. Alber F et al. 2007 The molecular architecture of the
nuclear pore complex. Nature 450, 695–701.
(doi:10.1038/nature06405)
37. Brohawn SG, Leksa NC, Spear ED, Rajashankar KR,
Schwartz TU. 2008 Structural evidence for common
ancestry of the nuclear pore complex and vesicle
coats. Science 322, 1369–1373. (doi:10.1126/
science.1165886)
38. Hsia KC, Stavropoulos P, Blobel G, Hoelz A. 2007
Architecture of a coat for the nuclear pore
membrane. Cell 131, 1313–1326. (doi:10.1016/j.
cell.2007.11.038)
39. Wu YL, Hoess P, Tschanz A, Matti U, Mund M, Ries
J. 2021 Maximum-likelihood model fitting for
quantitative analysis of SMLM data. bioRxiv. (doi:10.
1101/2021.08.30.456756)
40. Gray RDM, Beerli C, Pereira PM, Scherer KM, Samolej
J, Bleck CKE, Mercer J, Henriques R. 2016
VirusMapper: open-source nanoscale mapping of
viral architecture through super-resolution
microscopy. Sci. Rep. 6, 1–8. (doi:10.1038/
srep29132)
41. Laine RF, Albecka A, van de Linde S, Rees EJ,
Crump CM, Kaminski CF. 2015 Structural analysis of
herpes simplex virus by optical super-resolution
imaging. Nat. Commun. 6, 5980. (doi:10.1038/
ncomms6980)
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
10
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
42. Kukulski W, Schorb M, Kaksonen M, Briggs JAG.
2012 Plasma membrane reshaping during
endocytosis is revealed by time-resolved electron
tomography. Cell 150, 508–520. (doi:10.1016/j.cell.
2012.05.046)
43. Mund M et al. 2018 Systematic nanoscale analysis
of endocytosis links efficient vesicle formation to
patterned actin nucleation. Cell 174, 884–896.e17.
(doi:10.1016/j.cell.2018.06.032)
44. Bowler M et al. 2019 High-resolution
characterization of centriole distal appendage
morphology and dynamics by correlative STORM
and electron microscopy. Nat. Commun. 10, 1–5.
(doi:10.1038/s41467-018-08216-4)
45. Hourioux C, Brand D, Sizaret PY, Lemiale F, Lebigot
S, Barin F, Roingeard P. 2000 Identification of the
glycoprotein 41TM cytoplasmic tail domains of
human immunodeficiency virus type 1 that interact
with Pr55 Gag particles. AIDS Res. Hum. Retroviruses
16, 1141–1147. (doi:10.1089/088922200414983)
46. Sánchez-Sampedro L, Perdiguero B, Mejías-Pérez E,
García-Arriaza J, di Pilato M, Esteban M. 2015 The
evolution of poxvirus vaccines. Viruses 7,
1726–1803. (doi:10.3390/v7041726)
47. Gray RDM, Albrecht D, Beerli C, Huttunen M, Cohen
GH, White IJ, Burden JJ, Henriques R, Mercer J.
2019 Nanoscale polarization of the entry fusion
complex of vaccinia virus drives efficient fusion. Nat.
Microbiol. 4, 1636–1644. (doi:10.1038/s41564-019-
0488-4)
48. Farooq AV, Shukla D. 2012 Herpes simplex epithelial
and stromal keratitis: an epidemiologic update.
Survey Ophthalmol. 57, 448–462. (doi:10.1016/j.
survophthal.2012.01.005)
49. Whitley RJ, Kimberlin DW. 2005 Herpes simplex:
encephalitis children and adolescents. Seminars
Pediatric Infect. Dis. 16, 17–23. (doi:10.1053/j.spid.
2004.09.007)
50. Nicoll MP, Proença JT, Efstathiou S. 2012 The
molecular basis of herpes simplex virus latency.
FEMS Microbiol. Rev. 36, 684–705. (doi:10.1111/j.
1574-6976.2011.00320.x)
51. Brown JC, Newcomb WW. 2011 Herpesvirus capsid
assembly: insights from structural analysis. Curr.
Opin. Virol. 1, 142–149. (doi:10.1016/j.coviro.2011.
06.003)
52. Grünewald K, Desai P, Winkler DC, Heymann JB,
Belnap DM, Baumeister W, Steven AC. 2003 Three-
dimensional structure of herpes simplex virus from
cryo-electron tomography. Science 302, 1396–1398.
(doi:10.1126/science.1090284)
53. Zhou ZH, Dougherty M, Jakana J, He J, Rixon FJ,
Chiu W. 2000 Seeing the herpesvirus capsid at 8.5Å.
Science 288, 877–880. (doi:10.1126/science.288.
5467.877)
54. Fields BN, Griffin DE, Knipe DM, Doe J, Lamb RA,
Malcolm AM. 2001 Fields virology, 4th edition.
Philadelphia, PA: Lippincott Williams & Wilkins.
55. Seitz R. 2016 Human immunodeficiency virus (HIV).
Transfusion Med. Hemotherapy 43, 203–222.
(doi:10.1159/000445852)
56. Demirov DG, Ono A, Orenstein JM, Freed EO. 2002
Overexpression of the N-terminal domain of TSG101
inhibits HIV-1 budding by blocking late domain
function. Proc. Natl Acad. Sci. USA 99, 955–960.
(doi:10.1073/pnas.032511899)
57. Garrus JE et al. 2001 Tsg101 and the vacuolar
protein sorting pathway are essential for HIV-1
budding. Cell 107, 55–65. (doi:10.1016/S0092-
8674(01)00506-2)
58. Martin-Serrano J, Zang T, Bieniasz PD. 2001 HIV-1
and Ebola virus encode small peptide motifs that
recruit Tsg101 to sites of particle assembly to
facilitate egress. Nat. Med. 7, 1313–1319. (doi:10.
1038/nm1201-1313)
59. van Engelenburg SB, Shtengel G, Sengupta P, Waki
K, Jarnik M, Ablan SD, Freed EO, Hess HF,
Lippincott-Schwartz J. 2014 Distribution of ESCRT
machinery at HIV assembly sites reveals virus
scaffolding of ESCRT subunits. Science 343,
653–656. (doi:10.1126/science.1247786)
60. Berro J, Pollard TD. 2014 Local and global analysis
of endocytic patch dynamics in fission yeast using a
new ‘temporal superresolution’ realignment
method. Mol. Biol. Cell 25, 3501–3514. (doi:10.
1091/mbc.E13-01-0004)
61. Picco A, Mund M, Ries J, Nédélec F, Kaksonen M.
2015 Visualizing the functional architecture of the
endocytic machinery. Elife 4, e04535. (doi:10.7554/
eLife.04535)
62. Dutcher SK. 2014 The awesome power of dikaryons
for studying flagella and basal bodies in
Chlamydomonas reinhardtii. Cytoskeleton 71,
79–94. (doi:10.1002/cm.21157)
63. Garcia G, Reiter JF. 2016 A primer on the mouse
basal body. Cilia 5, 1–9. (doi:10.1186/s13630-016-
0038-0)
64. Pearson CG, Winey M. 2009 Basal body assembly in
ciliates: the power of numbers. Traffic 10, 461–471.
(doi:10.1111/j.1600-0854.2009.00885.x)
65. Vertii A, Hung HF, Hehnly H, Doxsey S. 2016 Human
basal body basics. Cilia 5, 1–7. (doi:10.1186/
s13630-016-0030-8)
66. Anderson RGW. 1972 The three-dimensional
structure of the basal body from the Rhesus monkey
oviduct. J. Cell Biol. 54, 246–265. (doi:10.1083/jcb.
54.2.246)
67. Garcia-Gonzalo FR, Reiter JF. 2012 Scoring a
backstage pass: mechanisms of ciliogenesis and
ciliary access. J. Cell Biol. 197, 697–709. (doi:10.
1083/jcb.201111146)
68. Gilula NB, Satir P. 1972 The ciliary necklace.
J. Cell Biol. 53, 494–509. (doi:10.1083/jcb.
53.2.494)
69. Shi X, Garcia G, van de Weghe JC, McGorty R,
Pazour GJ, Doherty D, Huang B, Reiter JF. 2017
Super-resolution microscopy reveals that disruption
of ciliary transition-zone architecture causes Joubert
syndrome. Nat. Cell Biol. 19, 1178–1188. (doi:10.
1038/ncb3599)
70. Salas D, le Gall A, Fiche JB, Valeri A, Ke Y, Bron P,
Bellot G, Nollmann M. 2017 Angular reconstitution-
based 3D reconstructions of nanomolecular
structures from superresolution light-microscopy
images. Proc. Natl Acad. Sci. USA 114, 9273–9278.
(doi:10.1073/pnas.1704908114)
71. Schnitzbauer J, Wang Y, Zhao S, Bakalar M, Nuwal
T, Chen B, Huang B. 2018 Correlation analysis
framework for localization-based superresolution
microscopy. Proc. Natl Acad. Sci. USA 115,
3219–3224. (doi:10.1073/pnas.1711314115)
72. Sieben C, Banterle N, Douglass KM, Gönczy P,
Manley S. 2018 Multicolor single-particle
reconstruction of protein complexes. Nat. Methods
15, 777–780. (doi:10.1038/s41592-018-0140-x)
73. Robichaux MA, Potter VL, Zhang Z, He F, Liu J,
Schmid MF, Wensel TG. 2019 Defining the layers of
a sensory cilium with STORM and cryoelectron
nanoscopy. Proc. Natl Acad. Sci. USA 116,
23 562–23 572. (doi:10.1073/pnas.1902003116)
74. Bornens M. 2012 The centrosome in cells and
organisms. Science 335, 422–426. (doi:10.1126/
science.1209037)
75. Bauer M, Cubizolles F, Schmidt A, Nigg EA. 2016
Quantitative analysis of human centrosome
architecture by targeted proteomics and
fluorescence imaging. EMBO J. 35, 2152–2166.
(doi:10.15252/embj.201694462)
76. Lawo S, Hasegan M, Gupta GD, Pelletier L. 2012
Subdiffraction imaging of centrosomes reveals
higher-order organizational features of pericentriolar
material. Nat. Cell Biol. 14, 1148–1158. (doi:10.
1038/ncb2591)
77. Mennella V, Keszthelyi B, Mcdonald KL, Chhun B,
Kan F, Rogers GC, Huang B, Agard DA. 2012
Subdiffraction-resolution fluorescence microscopy
reveals a domain of the centrosome critical for
pericentriolar material organization. Nat. Cell Biol.
14, 1159–1168. (doi:10.1038/ncb2597)
78. Sonnen KF, Schermelleh L, Leonhardt H, Nigg EA.
2012 3D-structured illumination microscopy provides
novel insight into architecture of human
centrosomes. Biol. Open 1, 965–976. (doi:10.1242/
bio.20122337)
79. Čajánek L, Nigg EA. 2014 Cep164 triggers
ciliogenesis by recruiting Tau tubulin kinase 2 to the
mother centriole. Proc. Natl Acad. Sci. USA 111,
E2841–E2850. (doi:10.1073/pnas.1401777111)
80. Graser S, Stierhof YD, Lavoie SB, Gassner OS, Lamla
S, le Clech M, Nigg EA. 2007 Cep164, a novel
centriole appendage protein required for primary
cilium formation. J. Cell Biol. 179, 321–330. (doi:10.
1083/jcb.200707181)
81. Izawa I, Goto H, Kasahara K, Inagaki M. 2015
Current topics of functional links between primary
cilia and cell cycle. Cilia 4, 1–3. (doi:10.1186/
s13630-015-0021-1)
82. Sillibourne JE, Hurbain I, Grand-Perret T, Goud B,
Tran P, Bornens M. 2013 Primary ciliogenesis
requires the distal appendage component
Cep123. Biol. Open 2, 535–545. (doi:10.1242/bio.
20134457)
83. Tanos BE, Yang HJ, Soni R, Wang WJ, Macaluso FP,
Asara JM, Tsou MFB. 2013 Centriole distal
appendages promote membrane docking, leading
to cilia initiation. Genes Dev. 27, 163–168. (doi:10.
1101/gad.207043.112)
84. Wei Q et al. 2013 Transition fibre protein FBF1 is
required for the ciliary entry of assembled
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
11
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024 

----!@#$NewPage!@#$----
intraflagellar transport complexes. Nat. Commun. 4,
2750. (doi:10.1038/ncomms3750)
85. Ye X, Zeng H, Ning G, Reiter JF, Liu A. 2014 C2cd3 is
critical for centriolar distal appendage assembly and
ciliary vesicle docking in mammals. Proc. Natl Acad. Sci.
USA 111, 2164–2169. (doi:10.1073/pnas.1318737111)
86. Delgehyr N, Sillibourne J, Bornens M. 2005
Microtubule nucleation and anchoring at the
centrosome are independent processes linked by
ninein function. J. Cell Sci. 118, 1565–1575. (doi:10.
1242/jcs.02302)
87. Tateishi K, Yamazaki Y, Nishida T, Watanabe S,
Kunimoto K, Ishikawa H, Tsukita S. 2013 Two
appendages homologous between basal bodies and
centrioles are formed using distinct Odf2 domains.
J. Cell Biol. 203, 417–425. (doi:10.1083/jcb.
201303071)
88. Mazo G, Soplop N, Wang WJ, Uryu K, Tsou MFB.
2016 Spatial control of primary ciliogenesis by
subdistal appendages alters sensation-associated
properties of cilia. Dev. Cell 39, 424–437. (doi:10.
1016/j.devcel.2016.10.006)
89. Gartenmann L, Wainman A, Qurashi M, Kaufmann
R, Schubert S, Raff JW, Dobbie IM. 2017 A
combined 3D-SIM/SMLM approach allows centriole
proteins to be localized with a precision of ∼4–
5 nm. Curr. Biol. 27, R1054–R1055. (doi:10.1016/j.
cub.2017.08.009)
90. Veatch SL, Cicuta P, Sengupta P, Honerkamp-Smith
A, Holowka D, Baird B. 2008 Critical fluctuations in
plasma membrane vesicles. ACS Chem. Biol. 3,
287–293. (doi:10.1021/cb800012x)
91. Ripley BD. 1976 The second-order analysis of
stationary point processes. J. Appl. Prob. 13,
255–266. (doi:10.2307/3212829)
92. Ester M, Kriegel HP, Sander J, Xu X. 1996 A density-
based algorithm for discovering clusters in large
spatial databases with noise. In Proc. Second
International Conf. on Knowledge Discovery and
Data Mining (KDD-96), pp. 226–231. Portland, OR:
AAAI Press.
93. Gunzenhäuser J, Olivier N, Pengo T, Manley S. 2012
Quantitative super-resolution imaging reveals
protein stoichiometry and nanoscale morphology of
assembling HIV-gag virions. Nano Lett. 12,
4705–4710. (doi:10.1021/nl3021076)
94. Malkusch S, Muranyi W, Müller B, Kräusslich HG,
Heilemann M. 2013 Single-molecule coordinate-
based analysis of the morphology of HIV-1
assembly sites with near-molecular spatial
resolution. Histochem. Cell Biol. 139, 173–179.
(doi:10.1007/s00418-012-1014-4)
95. Floderer C et al. 2018 Single molecule localisation
microscopy reveals how HIV-1 Gag proteins sense
membranevirusassemblysitesinlivinghostCD4Tcells.
Sci. Rep. 8, 1–5. (doi:10.1038/s41598-018-34536-y)
96. von Chamier L et al. 2021 Democratising deep
learning for microscopy with ZeroCostDL4Mic.
Nat. Commun. 12, 1–8. (doi:10.1038/s41467-021-
22518-0)
97. Zehtabian A, Markus Müller P, Goisser M, Obendorf
L, Jänisch L, Hümpfer N, Rentsch J, Ewers H. 2021
Precise measurement of nanoscopic septin ring
structures in deep learning-assisted quantitative
superresolution microscopy. bioRxiv. (doi:10.1101/
2021.12.28.474382)
98. Leterrier C, Potier J, Caillol G, Debarnot C, Rueda
Boroni F, Dargent B. 2015 Nanoscale architecture of
the axon initial segment reveals an organized and
robust scaffold. Cell Rep. 13, 2781–2793. (doi:10.
1016/j.celrep.2015.11.051)
99. Heideman M, Johnson D, Burrus C. 1984 Gauss and
the history of the fast Fourier transform. IEEE ASSP
Magazine 1, 14–21. (doi:10.1109/MASSP.1984.
1162257)
100. Heydarian H, Schueder F, Strauss MT, van
Werkhoven B, Fazel M, Lidke KA, Jungmann R,
Stallinga S, Rieger B. 2018 Template-free 2D particle
fusion in localization microscopy. Nat. Methods 15,
781–784. (doi:10.1038/s41592-018-0136-6)
101. Curd AP et al. 2021 Nanoscale pattern extraction
from relative positions of sparse 3D localizations.
Nano Lett. 21, 1213–1220. (doi:10.1021/acs.
nanolett.0c03332)
102. Briggs JAG. 2003 Structural organization of
authentic, mature HIV-1 virions and cores.
EMBO J. 22, 1707–1715. (doi:10.1093/
emboj/cdg143)
103. Fortun D, Guichard P, Hamel V, Sorzano COS,
Banterle N, Gonczy P, Unser M. 2018 Reconstruction
from multiple particles for 3D isotropic resolution in
fluorescence microscopy. IEEE Trans. Med. Imaging
37, 1235–1246. (doi:10.1109/TMI.2018.2795464)
104. Dey S et al. 2021 DNA origami. Nat. Rev.
Methods Primers 1, 13. (doi:10.1038/s43586-020-
00009-8)
105. Eltner A, Sofia G. 2020 Structure from motion
photogrammetric technique. Dev. Earth Surface
Processes 23, 1–24.
106. Heydarian H et al. 2021 3D particle averaging and
detection of macromolecular symmetry in
localization microscopy. Nat. Commun. 12, 1–9.
(doi:10.1038/s41467-021-22006-5)
107. Blundell B, Sieben C, Manley S, Rosten E, Ch’ng Q,
Cox S. 2021 3D structure from 2D microscopy
images using deep learning. Front. Bioinformatics 1,
62. (doi:10.3389/fbinf.2021.740342)
108. Mund M, Tschanz A, Wu YL, Frey F, Mehl JL,
Kaksonen M, Avinoam O, Schwarz US, Ries J. 2022
Superresolution microscopy reveals partial
preassembly and subsequent bending of the
clathrin coat during endocytosis. bioRxiv. (doi:10.
1101/2021.10.12.463947)
109. Huijben TAPM, Heydarian H, Auer A, Schueder F,
Jungmann R, Stallinga S, Rieger B. 2021 Detecting
structural heterogeneity in single-molecule
localization microscopy data. Nat. Commun. 12,
1–8. (doi:10.1038/s41467-021-24106-8)
110. Cox MAA, Cox TF. 2008 Handbook of data
visualization. Berlin, Germany: Springer.
111. Jain AK, Murty MN, Flynn PJ. 1999 Data clustering.
ACM Comput. Surveys 31, 264–323. (doi:10.1145/
331499.331504)
112. Auer A, Strauss MT, Strauss S, Jungmann R. 2020
NanoTRON: a Picasso module for MLP-based
classification of super-resolution data. Bioinformatics
36, 3620–3622. (doi:10.1093/bioinformatics/btaa154)
113. Sabinina VJ et al. 2021 Three-dimensional
superresolution fluorescence microscopy maps the
variable molecular architecture of the nuclear pore
complex. Mol. Biol. Cell 32, 1523–1533. (doi:10.
1091/mbc.E20-11-0728)
royalsocietypublishing.org/journal/rsob
Open Biol. 12: 220079
12
 Downloaded from https://royalsocietypublishing.org/ on 05 January 2024