Resource

Visualizing Intracellular Organelle and Cytoskeletal

Interactions at Nanoscale Resolution on Millisecond
Timescales
Graphical Abstract Authors
Yuting Guo, Di Li, Siwei Zhang, ...,
Jennifer Lippincott-Schwartz, Eric Betzig,
Dong Li

Correspondence
lippincottschwartzj@janelia.hhmi.org
(J.L.-S.),
betzige@janelia.hhmi.org (E.B.),
lidong@ibp.ac.cn (D.L.)

In Brief
A new approach for visualizing dynamic
processes within cells offers insight into
membrane-membrane contact
interactions and microtubule function.

Highlights
d Super-resolution live-cell imaging up to 266 fps at 97-nm
resolution

d Hitchhiking interactions among organelles remodel ER and

mitochondrial networks

d ER-mitochondrion contacts promote coalescence of

mitochondrial membranes

d Collision of late endosomes or lysosomes carried along

microtubules split ER tubules

Guo et al., 2018, Cell 175, 1–13

November 15, 2018 ª 2018 Elsevier Inc.
https://doi.org/10.1016/j.cell.2018.09.057
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

Resource

Visualizing Intracellular Organelle

and Cytoskeletal Interactions at Nanoscale
Resolution on Millisecond Timescales
Yuting Guo,1,2,7 Di Li,1,7 Siwei Zhang,1 Yanrui Yang,3,4 Jia-Jia Liu,2,3,4 Xinyu Wang,1 Chong Liu,1,2 Daniel E. Milkie,5
Regan P. Moore,6 U. Serdar Tulu,6,8 Daniel P. Kiehart,6 Junjie Hu,1,2 Jennifer Lippincott-Schwartz,5,* Eric Betzig,5,*
and Dong Li1,2,5,9,*
1National Laboratory of Biomacromolecules, CAS Center for Excellence in Biomacromolecules, Institute of Biophysics, Chinese Academy of

Sciences, Beijing 100101, China

2College of Life Sciences, University of Chinese Academy of Sciences, Beijing 100049, China
3State Key Laboratory of Molecular Developmental Biology, Institute of Genetics and Developmental Biology, Chinese Academy of Science,

Beijing 100101, China

4Center for Excellence in Brain Science and Intelligence Technology, Chinese Academy of Sciences, Shanghai 200031, China
5Janelia Research Campus, Howard Hughes Medical Institute, Ashburn, VA 20147, USA
6Department of Biology, Duke University, Durham, NC 27708, USA
7These authors contributed equally
8Present address: Division of Plastic and Reconstructive Surgery, Department of Surgery, Stanford School of Medicine, Stanford, CA

94305, USA
9Lead Contact

*Correspondence: lippincottschwartzj@janelia.hhmi.org (J.L.-S.), betzige@janelia.hhmi.org (E.B.), lidong@ibp.ac.cn (D.L.)

https://doi.org/10.1016/j.cell.2018.09.057

SUMMARY
In eukaryotic cells, organelles and the cytoskeleton
undergo highly dynamic yet organized interactions
capable of orchestrating complex cellular functions.
Visualizing these interactions requires noninvasive,
long-duration imaging of the intracellular environ-
ment at high spatiotemporal resolution and low
background. To achieve these normally opposing
goals, we developed grazing incidence structured
illumination microscopy (GI-SIM) that is capable of
imaging dynamic events near the basal cell cortex
at 97-nm resolution and 266 frames/s over thou-
sands of time points. We employed multi-color GI-
SIM to characterize the fast dynamic interactions of
diverse organelles and the cytoskeleton, shedding
new light on the complex behaviors of these struc-
tures. Precise measurements of microtubule growth
or shrinkage events helped distinguish among
models of microtubule dynamic instability. Analysis
of endoplasmic reticulum (ER) interactions with other
organelles or microtubules uncovered new ER re-
modeling mechanisms, such as hitchhiking of the
ER on motile organelles. Finally, ER-mitochondria
contact sites were found to promote both mitochon-
drial fission and fusion.
INTRODUCTION
Organelles interact with one another and the cytoskeleton to
synergistically execute various physiological functions (Eisen-
berg-Bord et al., 2016; Phillips and Voeltz, 2016; Prinz, 2014).
Organelle interactions occur at contact sites, where proteins
physically bridge their respective membranes and bind them
together. Such interactions are increasingly recognized to be
pivotal for diverse cellular functions (Bonifacino and Neefjes,
2017; Friedman et al., 2011; Rowland et al., 2014). Among the
various subcellular organelles, the largest is the endoplasmic re-
ticulum (ER), which spans the cytoplasm in an intricate network.
Its importance is underscored by its role in lipid and protein
biosynthesis; protein modification, translocation, and secretion;
and calcium handling. In addition, the cortical ER (cER) (i.e., the
portion lying in close apposition to the basal plasma membrane)
is increasingly viewed as a hub for organelle interaction (Valm
et al., 2017). ER-mitochondrion contact sites have been shown
to be important for calcium signaling, lipid transfer, mitochon-
drial fission (Friedman et al., 2011), and the synchronization of
mitochondrial DNA synthesis (Lewis et al., 2016). Similarly, ER-
endosome contact sites define the position and timing of endo-
somal fission (Rowland et al., 2014), and they help to connect
late endosomes (LEs) to the molecular motors that transport
them (Raiborg et al., 2015).
These discoveries were only possible with the advent of imag-
ing technologies that allowed the dynamic reorganization of or-
ganelles and the cytoskeleton to be studied. Even so, limitations
in resolution, speed, and z-depth of imaging have affected what
is possible to observe. What has been missing is an imaging
technique with high spatiotemporal resolution yet low propensity
for photobleaching and phototoxicity to noninvasively follow
minute and rapid rearrangements of specific organelles like the
ER, over the entirety of their interactions with other compart-
ments. While conventional wide-field fluorescence imaging is
sufficiently fast, it suffers from out-of-focus background, which
can easily obscure events happening on the scale of organelle

Cell 175, 1–13, November 15, 2018 ª 2018 Elsevier Inc. 1

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
(lower panel), including the cortical ER network, with minimal out-of-focus background in the corresponding GI-SIM image (upper panel).
See Figure S1 for a detailed characterization of the irradiation depth and signal-to-background ratio of the different illumination modalities. Scale bar, 5 mm.
contacts. Total internal reflection fluorescence (TIRF) micro-
scopy eliminates this background, but it can only image within

100 nm from the basal membrane (Steyer and Almers, 2001),
which is too small a distance to span many organelles and their
contact sites. Spinning-disk confocal microscopy (SDCM) elim-
inates both out-of-focus background and the restriction to the
basal membrane, but at the price of substantial photobleaching
and reduced speed: in previous SDCM studies of organelle inter-
action dynamics (Friedman et al., 2011; Rowland et al., 2014),
the imaging speed was limited to 2–5 s/frame for an
2-min
duration. Another issue common to all these methods is that
they are limited by diffraction to
200-nm lateral resolution. To
visualize potential organelle contacts with greater clarity, we
must turn to super-resolution (SR) fluorescence microscopy.
Of the various forms of super-resolution, structured illumina-
tion microscopy (SIM) is best suited to combining high resolution
with long-duration high-speed multi-color imaging, due to its low
excitation intensity, use of conventional fluorescent labels, and
the need for only nine two-dimensional (2D) raw images to recon-
struct an super-resolution one. Although wide-field 2D SIM
(Gustafsson, 2000) is susceptible to out-of-focus background
(which quickly leads to noisy and/or artifact-filled reconstruc-
tion), live-cell TIRF-SIM has been used to image microtubule
(MT) dynamics in artificially flattened S2 cells at 112-nm lateral
resolution and 3.7 frames/s (fps) (Kner et al., 2009) and cla-
thrin-mediated endocytosis and actomyosin contraction with
84-nm resolution at 3 s/frame (Li et al., 2015). However, as noted
above, most organelle interactions lie outside the range of TIRF
illumination.
To overcome this depth limitation of TIRF, we developed
grazing incidence structured illumination microscopy (GI-SIM),
where the illumination entering the objective rear pupil is
launched just inside the critical angle for TIRF, creating an illumi-
nation field parallel to the substrate that is comparable in
thickness to the objective depth-of-focus and, therefore, thin
enough to eliminate out-of-focus background yet thick enough
to encompass many organelles and their interactions with one
another or the cytoskeleton. With GI-SIM, we were able to image
2 Cell 175, 1–13, November 15, 2018
Figure 1. Comparison of TIR, GI, and WF
Illumination Modalities for SIM
(A) The evanescent wave light field generated by
total internal reflection (TIR) presents an expo-
nentially decaying axial intensity profile, which
limits its illumination depth to the vicinity of the
plasma membrane (lower panel). As a result, only
fragments of the ER network are visible in the
corresponding TIRF-structured illumination mi-
croscopy (SIM) image (upper panel).
(B) The conventional wide-field (WF) modality illu-
minates the entire cellular volume (lower panel), so
that in-focus structures such as the cortical ER are
obscured by out-of-focus background (OFB, up-
per panel).
(C) Grazing incidence (GI) produces a light sheet
whose thickness matches the depth-of-focus
(DOF) of a high numerical aperture objective to
illuminate the entire volume of the basal cell cortex
the cER network at 97-nm resolution and 266 frames/s for thou-
sands of frames, and thereby we gained new insights into the
mechanisms of MT dynamic instability and tubular ER genera-
tion, and we examined dynamic subdomain segregation along
ER tubules. Furthermore, multi-color GI-SIM imaging enabled
us to investigate the spatiotemporal coordination among organ-
elles, such as the association of mitochondrial fission and fusion
with ER-mitochondrion (ER-Mito) contacts and LE or lysosome
(Lyso) translocation with respect to ER-LE or Lyso contacts.
Finally, we identified instances where organelles were moved
or deformed by tethering onto other moving organelles.
RESULTS
Development and Characterization of GI-SIM
By illuminating the sample-substrate interface through an objec-
tive lens at a highly oblique angle (Figure 1A, bottom), TIRF
microscopy creates an excitation light field that extends only a
subwavelength distance beyond the basal surface of the cell
(Figure 1A, middle) to provide high-contrast, background-free
images that are ideal for SIM reconstruction. However, TIRF-
SIM is incapable of visualizing the entirety of organelles, such
as the ER that extends beyond this distance (Figure 1A, top).
At the other extreme, excitation through the objective at normal
incidence, such as in wide-field microscopy (Figure 1B), illumi-
nates the entire cell and its contents, but at the cost of substan-
tial out-of-focus background that reduces contrast, introduces
noise, and accelerates photobleaching.
To find an optimum illumination depth between these ex-
tremes, we repeatedly imaged COS-7 cells expressing markers
for the ER or MT network, and we plotted the signal-to-back-
ground ratio (SBR) as we gradually reduced the numerical aper-
ture (NA) of the excitation (and hence its angle of incidence at the
interface) from NA = 1.55 to 1.31 (Figure S1). The maximum SBR
occurred at NA = 1.37, or just below the critical angle for TIRF at
the expected refractive index of the cell. Performing a similar
experiment with a 4-mm-diameter bead in water, we determined
that, at this pre-critical NA, the illumination travels nearly parallel
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

A B C

D E F

G H

Figure 2. GI-SIM Reveals Fast Dynamics of Dynamic Instability of MTs and the ER Network
(A) GI-SIM image of MTs at one time point from a video of MT dynamic instability in a COS-7 cell (Video S1, part I).
(B) Comparison of MT end displacement plots analyzed using data obtained at temporal resolutions of 9.2 frames/s (lower curve) and down-sampled by 8-fold to
1.15 frames/s from the same dataset (upper curve). Green, blue, and red curves indicate the growth, shortening, and pause or fluctuation phases, respectively.
(C) Comparison of MT end detection with deconvolved GI (left) and GI-SIM (right) images.
(D and E) The distributions of MT growth (D) and shortening velocity (E) versus displacement.
(F) Catastrophe and rescue transition frequency computed at different imaging speed.
(G) GI-SIM (right) and GI (left) images of the ER network acquired at 266 frames/s in live COS-7 cells expressing mEmerald-KDEL (Video S1, part II).
(H) Tubular ER skeleton image color coded according to the local oscillation frequency and superimposed onto the fluorescence.
(I) Time-lapse GI-SIM images showing time-dependent variations in subdomains along an ER tubule. Cyan and orange arrows indicate the formation and
disappearance of two constriction sites. At 7.5 ms, these subdomains are well resolved in the GI-SIM image (left), but they cannot be observed clearly in the
diffraction-limited GI image (right). Scale bars, 2 mm (A and G) and 1 mm (C, H, and I).

to the interface in a sheet similar in width to the
1-mm depth-of-
focus of the objective (STAR Methods; Figures S1K–S1M). We
therefore term this condition GI illumination. It not only achieves
the twin goals of low background and complete illumination of
organelles near the basal membrane (Figure 1C) but also in-
creases the signal 10-fold compared to TIRF illumination at
NA = 1.55.
GI illumination is conceptually similar to the highly inclined and
laminated optical sheet (HILO) approach used for single-mole-
cule imaging in whole cells (Tokunaga et al., 2008), yet GI con-
fines the light even closer to the critical angle to eliminate nearly
all of out-of-focus excitation. In HILO, the out-of-focus fluores-
cence background can be computationally removed using a
structured illumination pattern at a period of twice the diffraction
limit (sHILO; Fiolka, 2016), but the lateral resolution extension is
then limited to 1.53, rather than 23 in GI-SIM, and the potential
for photobleaching and phototoxicity contributed by the out-of-
focus excitation remains.
GI-SIM Investigation of MT Dynamic Instability
A fast process demanding high spatiotemporal resolution imag-
ing is the dynamic instability of MTs. The spatiotemporal regula-
tion of MTs is important in many aspects of cell biology (Akhma-
nova and Steinmetz, 2008), and dynamic instability is an
important component of this regulation (Mitchison and Kirsch-
ner, 1984). To image MTs at unprecedented speeds and at su-
per-resolution for extended periods of time, we transfected
COS-7 cells with the bright and photostable MT-associated
fusion protein 33 mEmerald-ensconsin (Gierke et al., 2010) (Fig-
ure 2A), which co-localizes well with a direct tubulin marker but
produces far less cytosolic background (Figures S2A and
S2B). GI-SIM imaging at 97-nm resolution (Figures S2E–S2K)
and 9.2 frames/s for hundreds of frames (Video S1, part I), al-
lowed us to quantify MT dynamic instability on a much finer
spatiotemporal scale than what was achieved in previous
studies (Akhmanova and Steinmetz, 2008; Gierke et al., 2010).
For example, sampling the data at full temporal resolution

Cell 175, 1–13, November 15, 2018 3

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
showed brief transitions between MT growth and shortening and
quick fluctuations when the growth or shortening of MT ends
paused temporally (Video S1, part I). These transient processes
cannot be captured if the sampling frequency drops by 8-fold to
1.15 Hz (Figure 2B). Likewise, the high spatial resolution of GI-
SIM ensures detection of densely packed MTs undetectable to
diffraction-limited methods (Figure 2C). Therefore, the loss of
either spatial or temporal resolution was sufficient to degrade
the precision of MT dynamic characterization.
We measured MT growth and shortening rates of 20.5 ±
11.6 mm/min (N = 178 from 15 cells) and 23.8 ± 13.9 mm/min
(N = 143 from 15 cells), respectively, consistent with earlier re-
sults obtained using tubulin or EB1 markers (Matov et al.,
2010). Furthermore, when analyzing these rates as a function
of the displacement of each growth or shortening event (Figures
2D and 2E), we found that long displacing events correlated with
slow rates of growth or shortening. There was a broad range of
rates during short displacing events, leading to diverse fluctua-
tions in MT ends. Transition frequencies of catastrophe and
rescue were measured to be 0.59 ± 0.48 s1 and 0.67 ±
0.51 s1, respectively (Figure 2F, left column), which are fast
enough to require 9.2 frames/s measurement to achieve accu-
rate results (Figure 2F, middle and right columns). The scattered
distribution of catastrophe and rescue transition might result
from the heterogeneous microenvironments around the plus
ends of different MTs, such as the availability of guanosine
triphosphate (GTP)-tubulin dimers. These observations of the
switch-like behavior of MT ends support the dynamic cap model
(Howard and Hyman, 2009). This theory attributes MT dynamic
instability to variations in the length of the GTP-tubulin cap at
MT ends, which depend on the randomness of GTP-tubulin di-
mers binding to or dissociating from proto-filaments.
GI-SIM Investigation of ER Network Dynamics
The ER is notable not only because it is the largest intracellular
organelle but also because it continuously and rapidly remodels
itself (Hu et al., 2011; Nixon-Abell et al., 2016). To investigate ER
dynamics using GI-SIM, we transfected COS-7 cells with the ER
luminal marker mEmerald-KDEL. GI-SIM, at 97-nm resolution,
resolved the ER with much finer detail than GI illumination alone
(Figures 2G and 2I; Video S1, part II). As we reported previously
(Nixon-Abell et al., 2016), continuous sheet-like structures
observed in the peripheral ER under diffraction-limited GI illumi-
nation were found to be riddled with fenestrations when viewed
with GI-SIM (Figure 2G). Furthermore, due to the bright signals
afforded by GI-SIM and the use of interleaved reconstruction
(Ma et al., 2018), we were able to image these fenestrations at
266 frames/s for 4,500 frames, which confirmed that most of
them consisted of tightly packed tubular matrices that constantly
remodeled themselves (Video S1, part II). By tracking all tubules
with a skeletonization algorithm, we created a spatial map of
their oscillations (Figure 2H). While the median amplitude and
frequency were 144 nm peak to peak at 10.1 Hz, there was
considerable spatial heterogeneity in these oscillations (Figures
S2L–S2O), with the fastest oscillations often occurring at tightly
clustered tubules (Figure 2H). This may be because shorter tu-
bules, like shorter strings, are expected to have higher reso-
nance frequencies, or it may reflect the spatial variability of the
4 Cell 175, 1–13, November 15, 2018
energy sources that can drive such oscillations (Nixon-Abell
et al., 2016).
The high spatiotemporal resolution of GI-SIM also revealed
rapid changes in the thickness of the ER lumen along various tu-
bules (Figure 2I), which were not apparent with GI alone. In
particular, we observed constrictions (arrows, Figure 2I) that
formed in as fast as
10 ms, with mEmerald-KDEL displaced
to either side (7.5- to 18.8-ms time points of Figure 2I). The total
fluorescence intensity during these events was constant to
within 5% (Figures S2P and S2Q), suggesting that the lumen
contents were conserved during constriction and displacement.
Similar constrictions and bulges in ER tubules have long been
observed by electron microscopy (Behnke and Moe, 1964;
Friedman and Voeltz, 2011; Nixon-Abell et al., 2016), but here
we see that these features remodel on the timescale of tens of
milliseconds.
Mechanisms of Tubular ER Formation
In mammalian cells, ER network rearrangements strongly
depend on interactions with the MT cytoskeleton. New ER
tubules can be pulled out of the existing ER membrane by asso-
ciating with motor proteins and then extending along MTs (i.e.,
the sliding mechanism) or by attaching to the tips of polymer-
izing MTs (i.e., tip attachment complex [TAC] mechanism)
(Waterman-Storer and Salmon, 1998). We applied GI-SIM to
characterize ER-MT dynamic interactions and investigate
whether cells employ other mechanisms to reshape the ER
network. COS-7 or U2OS cells were co-transfected with 33
mEmerald-ensconsin and mCherry-KDEL to label the MTs
and ER simultaneously. A representative GI-SIM two-color im-
age frame from a movie of the ER and MTs (Figure 3A; Video
S2, part I) shows that ER-reshaping activities, such as ER ring
rearrangement and ER tubule branching, mostly occurred along
MTs. ER tubule-branching activities included instances of the
sliding and TAC mechanisms (Figures 3B and 3C; Video S2,
parts II and III). In addition, however, we observed that new
ER tubules could be pulled out by attaching to the plus ends
of depolymerizing MTs (Figure 3D; Video S2, parts II and III).
Because the plus-end-binding proteins of polymerizing and
depolymerizing MTs are very different (Akhmanova and Stein-
metz, 2008), this depolymerizing TAC (dTAC) mechanism may
be mediated by molecules other than the STIM1 and EB1 me-
diators of the polymerizing TAC (pTAC) mechanism (Grigoriev
et al., 2008).
In addition to these three mechanisms, where growing ER tu-
bules interact directly with MTs, we also discovered a fourth
mechanism wherein new ER tubules were generated by hitchhik-
ing on highly motile organelles, e.g., LEs or Lysos that were in
turn being transported along MTs by molecular motors (Fig-
ure 3E; Video S2, parts II and III). Although the molecules that
tether the ER membrane to the vehicle LE or Lyso are unknown,
this tether may transiently break to either allow the growing ER
tubule to fuse with an existing ER tubule or to retract back to
the branching point in the event that the tether breaks in the mid-
dle of ER tubule extension.
Finally, as a fifth mechanism, we observe that new ER tubule
branches can appear without apparent direct or indirect involve-
ment of MTs. This process appears as de novo budding of the ER
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

A B C

D E F G

Figure 3. Tubular ER Generation Mechanisms

(A) The dependence of ER network (magenta) rearrangement on MTs (green) in a live COS-7 cell (Video S2, part I).
(B–F) Time-lapse images of tubular ER generation by five mechanisms: (B) sliding, (C) pTAC, (D) dTAC, (E) hitchhiking, and (F) de novo budding. In (E), LEs or Lysos
are labeled in yellow. The white arrows indicate the tips of growing ER tubules (Video S2, parts II and III).
(G) The proportion of tubular ER generation events that occur via different mechanisms.
(H) Comparison of the growth velocity of ER tubules generated by the different mechanisms.
Data are shown as mean ± SEM. **p < 0.01 (one-way ANOVA). Scale bars, 5 mm (A) and 2 mm (B–F).

tubule from an existing ER tubule (Figure 3F; Video S2, parts II
and III), although the existence of an unseen mediator cannot
be ruled out.
Next, we characterized the relative proportions (Figure 3G)
and growth rates (Figure 3H) of these five different types of
ER-branching events. We examined 16 cells having 529 events
in which each ER tubule branch extended at least 1 mm. Sliding
and hitchhiking events had similar growth rates of 40.5 ±
16.2 mm/min and 43.0 ± 14.0 mm/min, respectively, presumably
because they both rely on MT-based molecular motors to
generate their driving force. The proportions of sliding and
hitchhiking events were also comparable, and together they
comprised nearly 60% of the total events. In contrast, ER tubule
growth rates were significantly different for the two types of MT
end-associated mechanisms: 8.3 ± 3.3 mm/min for pTAC versus
15.6 ± 3.6 mm/min for dTAC. In addition, pTAC events occurred
more frequently than dTAC ones. These differences further sug-
gest that pTAC and dTAC mechanisms for ER movement rely on
different molecular machineries. Finally, the fifth mechanism,
MT-independent de novo ER budding, occurred comparatively
rarely (8 events from 30 cells) and more slowly (5.5 ± 1.6 mm/min)
than the other four.

Cell 175, 1–13, November 15, 2018 5

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
To investigate the dependence of these ER-branching mech-
anisms on MT dynamics, we treated COS-7 cells with Nocoda-
zole for 20 min (Figure S3). Consistent with previous studies
(Friedman et al., 2010; Lu et al., 2009), most tubular ER
collapsed, and the pTAC and dTAC events, which depend on
dynamic MT polymerization and depolymerization, ceased,
whereas sliding (Figure S3E) and hitchhiking events still occurred
along MTs stabilized by acetylation (Friedman et al., 2010), albeit
less frequently. De novo budding was not affected (Figure S3F;
9 events from 30 cells), proving it is an MT-independent
mechanism.
Tubular ER’s Role in Mitochondrial Fission and Fusion
Recently, ER-Mito contact sites were found to define the posi-
tions of mitochondrial fission (Friedman et al., 2011). Consistent
with this, we observed that ER-Mito contact sites marked con-
stricted loci in mitochondria (arrows in Figures 4A and 4B),
where the majority of fission events occurred (84%, N = 100
from 97 cells, Figure 4F). In addition, we noticed that ER-Mito
contacts were retained through the entire division process
(Video S3, parts I and II) and the two compartments defined
by the ER tubule writhed back and forth (Figure 4C; Video S3,
part II). Interestingly, we found that during the writhing process
the area of one compartment slightly increased while the area of
the other compartment decreased (Figure S4). Moreover, we
observed that, in some fission events (e.g., upper zoom-in
example in Video S3, part I), one compartment of the dividing
mitochondrion was driven to move away from the other
compartment, which preferentially associated with the ER-
Mito contact and remained relatively static. The mitochondrial
membrane at the fission site was elongated into tubular inter-
mediates until division occurred to give two independent mito-
chondria. These observations suggest that ER-Mito contacts
not only mark the sites of mitochondrial division but also play
a role in stabilizing the mitochondrion, thereby facilitating the di-
vision process.
Given that the ER usually participates in mitochondrial fission,
an intriguing question is whether it can also serve to facilitate
mitochondrial fusion. We scrutinized 134 cases of mitochondrial
fusion from 120 cells, and we found that ER-Mito contact was
involved in
59% of these events (Figure 4F). During such
events (Figure 4D; Video S3, parts I and III), one mitochondrion
first contacts the ER tubule lying between two opposing
mitochondria. As the two mitochondria come into contact, the
local fluorescence intensity of the outer membrane protein,
mEmerald-Tomm20, is transiently enhanced at the established
ER-Mito contact site (10 and 12 s in Figure 4D). The disappear-
ance of the enhanced fluorescence indicates the coalescence
of outer membranes, followed by the relaxation of the constric-
tion at the fusion site. We quantified the duration of all the fusion
events by measuring the time from the first contact of the two
mitochondria until complete relaxation of the post-fusion
constriction. We found that mitochondrial fusion events without
ER involvement (Figure 4E; Video S3, part IV) on average took
longer to complete membrane coalescence compared to the
ones with ER-Mito contact (12.5 versus 21.9 s; Figure 4G), indi-
cating that the ER accelerates membrane coalescence after the
initial contact.
6 Cell 175, 1–13, November 15, 2018
ER Interactions Regulate Transport of LEs or Lysos
The movements of intracellular organelles often correlate with
cellular functions (Bonifacino and Neefjes, 2017). Emerging
evidence indicates that the ER plays an important role in
transporting and controlling the local concentration of intracel-
lular organelles, such as LEs or Lysos (Raiborg et al., 2015). To
investigate how the ER contact affects the transport of LEs or
Lysos, we triply transfected COS-7 cells with 33 mEmerald-
enconsin, mCherry-KDEL, and Halo-LAMP1 labeled with the
JF642 ligand to visualize the MTs, ER, and LAMP1-positive
LEs or Lysos by GI-SIM. Consistent with a previous study
(Rowland et al., 2014), we found that >90% of LEs or Lysos
were associated with the ER (Figure 5A; Video S4). We also
found that ER-LE or Lyso interaction was facilitated by corrals
of ER tubules that shrank from a large polygon shape to a
much smaller one that tightly surrounded the LE or Lyso (Fig-
ure 5B). Afterward, the motion of the LE or Lyso was in step
with its surrounding ER polygon (88 to 238 s in Figure 5B;
Video S5, part I).
A recent study found that ER-LE contacts mediate the transfer
of molecular motors onto motor adaptors on the LE membrane,
prompting the translocation of LEs (Raiborg et al., 2015). We
confirmed that if an ER-associated LE or Lyso was close to
MTs, it tended to be motile, especially when the LE or Lyso
was tightly encompassed by an ER polygon (Figure 5C). More-
over, the contact between the ER and the leading edge of the
moving LE or Lyso was preferentially maintained during the
movement (zoom-in examples in Video S4), which prompted
the long-distance transportation of LE or Lyso structures. In
contrast, if the LE or Lyso was not associated with the ER, it
generally underwent diffusive motion rather than directional
transport along MTs, even though the LE or Lyso was already
in contact with MTs (Video S4, part II).
Furthermore, we noted that the diffusive movement was small
for the LEs or Lysos tightly confined by ER corrals. We therefore
asked whether ER interactions assisted in stabilizing LEs or Ly-
sos before they docked onto MTs. To prevent directional trans-
port of LEs or Lysos along MTs, we simultaneously knocked
down the adaptor proteins FYCO1 and RILP (Figures S5A and
S5B), which are responsible for recruiting the molecular motors
kinesin and dynein, respectively, onto LEs or Lysos (Phillips
and Voeltz, 2016). As a result, the role of MTs in affecting LEs
or Lysos’ movements could be eliminated. We found LEs or
Lysos tightly confined by ER corrals had smaller mean square
displacements (MSDs) than those more loosely confined by
the ER (Figure S5C). Therefore, our data suggest that ER-LE or
Lyso interactions play a role in stabilizing LEs or Lysos.
On rare occasions, we observed ER fission events (Fig-
ure 5D; Video S5, part II) wherein an LE or Lyso (white arrow)
moving along an MT track bent an ER tubule in its path into a
highly curved arc (12 and 14 s in Figure 5D) before eventually
rupturing at an adjacent three-way junction (18 s in Figure 5D)
or in the middle of the ER tubule (example 2 in Video S5, part
II). After passage of the LE or Lyso, a free end of the ruptured
ER tubule could either fuse with an adjacent ER tubule (32 and
36 s in Figure 5D), or the two free ends could fuse with each
other to restore the original connectivity (example 2 in Video
S5, part II).
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

Figure 4. The Association of Mitochondrial Fission and Fusion with the Tubular ER
(A) Contacts between mitochondria (green) and the ER (magenta) shown in a live COS-7 cell.
(B) Mitochondrial channel only. White arrows in (A) and (B) indicate constricted loci in mitochondria associated with ER-Mito contact sites, where mitochondrial
fission occurs (Video S3, upper panel of part I).
(C) Time-lapse images of a typical mitochondrial fission event at an ER-Mito contact site (Video S3, part II).
(D) Time-lapse images of a typical mitochondrial fusion event at an ER-Mito contact site (Video S3, lower panel of part I).
(E) Time-lapse images of a typical mitochondrial fusion event without ER-Mito contact (Video S3, part IV). White arrows in (E) indicate that the ER is absent at the
contact site of the two mitochondria.
(F) The proportion of mitochondrial fission or fusion events occurring with and without ER-Mito contacts.
(G) Comparison of the duration of mitochondrial fusion events with and without ER-Mito contacts.
Data are shown as mean ± SEM. **p < 0.01 (one-way ANOVA). Scale bars, 3 mm (A and B) and 1 mm (C–E).

GI-SIM Reveals Hitchhiking Interactions between
Organelles
In the canonical mode of MT-based intracellular transport,
cargo-specific adaptor molecules recruit molecular motors,
which then directly drive the cargo along the MT track. Recently,
a new paradigm termed hitchhiking was discovered in fungi (Gui-
maraes et al., 2015), in which some cargos achieve MT-depen-
dent motility by tethering onto other cargos that have recruited
molecular motors, rather than by directly coupling with molecular
motors themselves. However, it has not been confirmed whether
the hitchhiking paradigm exists in mammalian cells (Salogiannis
and Reck-Peterson, 2017) and, if so, what its functions might be.
Hitchhiking is mediated by membrane contact sites that tether
the hitchhiker to the vehicle organelles. For example, we found
that ER tubules could anchor onto a moving LE or Lyso and a
new ER tubule could be pulled out and elongate as it followed

Cell 175, 1–13, November 15, 2018 7

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

Figure 5. ER Contacts Regulate the Transport of LEs or Lysos

(A) Full field-of-view image of MTs (yellow), ER (magenta), and LEs or Lysos (green) at one time point during their mutual interactions in a COS-7 cell (Video S4).
(B) Time-lapse images illustrate the process of an ER polygon spontaneously shrinking to the size of an LE or Lyso (Video S5, part I).
(C) Time-lapse images illustrate that an LE or Lyso (white arrow), in contact with the ER, continuously moves along the MT track, during which the ER network is
locally deformed.
(D) Example of a moving LE or Lyso separating the ER tubule from a three-way junction, indicated by white arrows (Video S5, part II).
Scale bars, 5 mm (A) and 1 mm (B–D).

8 Cell 175, 1–13, November 15, 2018

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

Figure 6. Hitchhiking Interactions between Intracellular Organelles

(A) Generation of a new ER tubule (magenta, with tip indicated by white arrow) by hitchhiking on a mitochondrion (green) that is moving along an MT (yellow). See
also Video S6, part I.
(B) Mitochondrial fusion (green) mediated by hitchhiking (white arrow) on an LE or Lyso (magenta) moving along an MT (yellow).
See also Video S6, part II. Scale bars, 1 mm (A) and 2 mm (B).

the trajectory of the LE or Lyso (Figure 3E). We also observed that
ER tubules that adhered to moving mitochondria could elongate
(Figure 6A; Video S6, part I). In either case, the leading edge of
the hitchhiking ER tubule was always attached to the rear of
the moving organelle.
During this hitchhiking process, the extending ER tubule may
fuse with existing ER tubules to form a new three-way junction
(Video S6, part I). Alternatively, we observed vehicle organelles
that stopped while pulling a hitchhiking ER tubule, leaving the
tethered end free in the cytoplasm (Figure S6). This explains
why such free-ended ER tubules have been observed previously
(Friedman et al., 2011; Nixon-Abell et al., 2016; Rowland et al.,
2014), although it was generally thought that they were unstable
(Wang et al., 2016). Taken together, these results demonstrate
that ER-organelle contacts can not only influence organelle func-
tion but also play a role in rearranging the ER network.
Recent studies reported that mitochondria and Lyso or vacu-
oles commonly contact each other (Elbaz-Alon et al., 2014;
Wong et al., 2018). We also observed abundant dynamic Mito-
LE or Lyso interactions using GI-SIM (Figure 6B; Video S6,

Cell 175, 1–13, November 15, 2018 9

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
part II), and we found that LEs or Lysos moving in proximity to
mitochondria often resulted in significant changes in the mito-
chondrial morphology. For example, LE or Lysos pushing against
mitochondria led to large invaginations in the latter (Video S6,
part II). The resulting increase in the contact area might possibly
accelerate physiological functions mediated by Mito-LE or Lyso
contacts. Alternatively, mitochondria hitchhiking on moving LEs
or Lysos could be transported by as much as
4 mm (example
1 in Video S6, part II), or the leading LE or Lyso could pull a thin
tubule from a contacting mitochondrion (example 2 in Video S6,
part II). Many of these tubules retracted back to the parent mito-
chondrion, but, if the moving LE or Lyso passed close to a second
mitochondrion, the tubule from the first could create a bridge
leading to the complete fusion of the two mitochondria (Figure 6B;
example 3 in Video S6, part II). Dynamic mitochondrial tubulation
was recently reported to be important for maintaining mitochon-
drial DNA integrity and interchanging mitochondrial material
(Wang et al., 2015), and here we see that LE- or Lyso-mediated
mitochondrial tubulation provides one such mechanism for the
dynamic remodeling of the mitochondrial network.
Plasticity of Mouse Hippocampal Dendritic Spines and
Dorsal Closure in Drosophila
The plasticity of dendritic spines is an essential element of
learning and memory. To study this plasticity at high resolu-
10 Cell 175, 1–13, November 15, 2018
Figure 7. Actin Remodeling during Dendritic
Spine Plasticity and Dorsal Closure
(A) Morphological plasticity of dendritic spines
before (left) and after (right) glycine treatment for a
primary hippocampal neuron (Video S7, part I).
(B) Time-lapse images of dendritic spine expan-
sion coupled with filamentous actin reorganization.
(C) Two time points from a video of the evolution of
diverse actomyosin structures inside amnioserosa
cells during dorsal closure in a Drosophila embryo
(Video S7, part II). Dashed yellow boundaries
indicate regions of contracting or relaxing medi-
oapical actomyosin array inside each amnioserosa
cell.
Scale bars, 2 mm (A), 1 mm (B), and 5 mm (C).
tion, we co-transfected a primary culture
of mouse hippocampal neurons (days
in vitro [DIV] 18) with membrane-bound
GFP and LifeAct-mCherry to visualize
the plasma membrane and filamen-
tous actin (F-actin), respectively. Upon
glycine treatment to induce long-term
potentiation (LTP), a number of dendritic
spines became enlarged (arrows in
Figure 7A), but they also remodeled
dynamically in response to the redistri-
bution of actin within (Video S7, part I).
In one example, actin accumulated in a
narrow spine (0- and 20-s time points
of Figure 7B), which then expanded
into mushroom-like morphology match-
ing the reorganization of F-actin in the
cortical region beneath the plasma membrane (30-s time point
of Figure 7B). When this cortical actin recondensed, the spine
concomitantly shrank (45-s time point of Figure 7B), in agree-
ment with previous studies (Bosch et al., 2014). Therefore,
additional factors are required to stabilize the cortical actin
scaffold to maintain the morphological features of mature
spines.
We also explored the suitability of GI-SIM for visualizing
Drosophila dorsal closure, which is a model of epithelial sheet
morphogenesis that depends on the construction of func-
tional actin structures to complete development. Observing
the actin dynamics at high spatiotemporal resolution is
particularly challenging because of the low signal-to-noise
ratio and the increased distance from the coverslip due to
the presence of the vitelline membrane. Nonetheless, GI-
SIM successfully observed diverse actomyosin fine struc-
tures (filopodia, lamellipodia, and medioapical arrays) and
their concomitant kinematics (Figure 7C). Even with a rela-
tively dim sample, GI-SIM imaged more than 600 frames
over 30 min (Video S7, Part II). As indicated by the dashed
polygons in Figure 7C, we observed that the medioapical
actomyosin array underwent periodic contraction-relaxation
cycles, which presumably results in the well-known oscilla-
tion behavior of amnioserosa during dorsal closure (Kiehart
et al., 1994).
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
DISCUSSION
GI-SIM (Figures S1 and S7) offers a combination of super-reso-
lution, high-speed, multi-color imaging and low photobleaching
and phototoxicity that makes it well suited for studying intracel-
lular dynamics. Compared to TIRF or TIRF-SIM, GI-SIM probes
events
103 further past the sample substrate with
103
greater fluorescence intensity; compared to SDCM, it provides

23 better spatial resolution and
103 faster imaging speed;
and compared to other super-resolution methods, it better
balances the trade-offs between high spatial and temporal
resolution and low photobleaching and phototoxicity in live-cell
imaging over cellular-sized fields of view (Demmerle et al.,
2015; Li et al., 2015; Tortarolo et al., 2018). Additional gains in
resolution and speed beyond 97 nm at 266 frames/s can also
be anticipated from continuing advancements in fluorophore
photophysics (Wang et al., 2017).
Among our findings, we discovered three new mechanisms of
tubular ER generation: hitchhiking, dTAC, and de novo budding.
These operate in parallel to the previously known mechanisms of
sliding and pTAC (Waterman-Storer and Salmon, 1998). As dis-
cussed above, dTAC is likely to employ different molecules from
STIM1 and EB1 used by pTAC. One potential candidate is the
MT depolymerase MCAK, a member of the kinesin-13 family,
which is known to catalyze catastrophe activity at the MT plus
end (Helenius et al., 2006). De novo budding is a comparatively
rare mechanism for remodeling the ER network, and, perhaps
not by coincidence, it is the only one independent of MTs (Fig-
ures 3F and S3F). It therefore might be helpful in extending the
ER to places where it is difficult for MTs to reach, such as den-
dritic spines (Wagner et al., 2011).
Our results also expand the understanding of organelle-
organelle hitchhiking, which allows the ER to indirectly utilize
molecular motors and reshape its structure according to the
movement of the organelles with which it is in contact. This
achieves a more directed ER rearrangement than the sliding
of ER tubules along MTs alone. In addition, we have shown
that hitchhiking interactions can reshape organelles in a vari-
ety of ways, such as an MT-tracked LE or Lyso pulls a thin
membrane tube from an otherwise stationary mitochondrion
(Video S6, part II). Among 97 datasets of two-color imaging
of mitochondria and ER, we found 35 events of such mito-
chondrial membrane tubulation. GI-SIM provides an ideal
tool to further investigate the mechanism of how friction is
generated to stabilize the elongated membrane tube (Simu-
novic et al., 2017).
While ER-Mito contacts are known to facilitate the majority of
mitochondrial fission events (Friedman et al., 2011), here we
found that ER-Mito contacts are also associated with
60% of
all mitochondrial fusion events and that mitochondrial mem-
brane coalescence is faster when mediated by the ER. These
results indicate that ER-Mito contacts, which physically bring
two mitochondria into close proximity and then keep them
closely apposed, may play a role in facilitating mitochondrial
fusion activity. This suggests future experiments to see if ER-
Mito-tethering proteins, such as PDZD8 (Hirabayashi et al.,
2017) and mitofusin 2 (MFN2) (de Brito and Scorrano, 2008),
are involved in regulating the fusion process.
It is well known that the ER network employs fusion machinery
to regulate its complexity, especially in tubule-tubule connec-
tions or polygon ring closures (Wang et al., 2016). In contrast,
there has been little evidence suggesting that the ER undergoes
fission (Friedman and Voeltz, 2011; Wang et al., 2016). Here,
however, we observed that moving LEs or Lysos can collide
with and break ER tubules (Figure 5D; Video S5, part II). Such
breakage may involve the dynamics of atlastin (ATL)-mediated
ER membrane tethering (Wang et al., 2016). In this scenario,
tubular ER breakage would occur at a tethered, but not fused,
tubular junction. Disruption of trans-ATL dimerization during
the GTP hydrolysis cycle would open the gate for passing Lysos,
and reformation of the tethering ATL dimers would close it.
Consistent with this, some ER fission events occur at three-
way junctions (Figure 5D). However, we also observed cases
of ER breakage in the middle of a tubule (example 2 in Video
S5, part II). Therefore, active fission of the ER by LE or Lyso con-
tact may also involve components that are yet to be identified.
Regardless of the nature of the ER breakage, it is likely caused
by mechanical force generated by the organelle collision. Simi-
larly, mechanical force-induced membrane fission was reported
recently in mitochondria (Helle et al., 2017).
Further improvements in GI-SIM technology can be envi-
sioned. Our imaging speed of up to 266 frames/s/color is faster
than needed to study many pairwise organelle interactions. This
raises the interesting possibility of using the recently developed
spectral unmixing methods (Cutrale et al., 2017; Valm et al.,
2017) to achieve hyperspectral super-resolution imaging of the
mutual interactions of six or more organelles simultaneously. In
addition, while GI-SIM is thus far a strictly 2D technique, with
multi-NA excitation (Boulanger et al., 2014) it may be possible
to decipher events in 3D near the basal surface, at the cost of
requiring more raw images.
In conclusion, the GI-SIM method described and demon-
strated here fulfills an unmet need for imaging intracellular
dynamics with ultrahigh spatial and temporal resolution with
minimal invasiveness. It therefore opens a new window for the
study of minute, highly dynamic interactions within cultured
cells, or cells at the periphery of multicellular organisms.
STAR+METHODS
Detailed methods are provided in the online version of this paper
and include the following:
d KEY RESOURCES TABLE
d CONTACT FOR REAGENT AND RESOURCE SHARING
d EXPERIMENTAL MODEL AND SUBJECT DETAILS
B Cell culture
B Drosophila embryo
B Primary hippocampal neurons
d METHOD DETAILS
B GI-SIM system
B Interleaved reconstruction
B Characterization of the depth-of-focus of the GI-SIM
system
B Determining the optimal condition to image intracel-
lular organelles
Cell 175, 1–13, November 15, 2018 11
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
B Characterization of the irradiation depth of the GI illu-
mination mode
B RNAi transfection and real-time PCR
B Nocodazole treatment and imaging
B GI-SIM imaging for Drosophila embryo
d QUANTIFICATION AND STATISTICAL ANALYSIS
B Tubular ER oscillation analysis
B Characterization of MT dynamic instability
B Tubular ER growth analysis
B Association of mitochondrial fission/fusion with ER-
Mito contacts
SUPPLEMENTAL INFORMATION
Supplemental Information includes seven figures and seven videos and can be
found with this article online at https://doi.org/10.1016/j.cell.2018.09.057.
ACKNOWLEDGMENTS
This work was supported by grants to Dong Li from the Chinese Ministry of Sci-
ence and Technology (MOST; 2017YFA0505301 and 2016YFA0500203), the
National Natural Science Foundation of China (NSFC; 91754202 and
31770930), the Chinese Academy of Sciences (CAS; XDB19040101 and
YZ201651), the CAS Pioneer Hundred Talents Program, the Thousand Young
Talents Program of China, and the Joint Program between CAS and Peking
University and to J.-J.L. from MOST (2016YFA0500100) and NSFC
(31530039). J.L.-S. and E.B. are funded by the Howard Hughes Medical Insti-
tute (HHMI).
AUTHOR CONTRIBUTIONS
Dong Li and E.B. conceived the idea. Dong Li designed the experiments and
built the microscope at Janelia under the supervision of E.B. Di Li and S.Z. built
and improved the microscope system at the Institute of Biophysics under the
supervision of Dong Li. J.-J.L. designed the dendritic spine plasticity experi-
ment, and D.P.K. designed the Drosophila dorsal closure experiment. Y.G.,
Di Li, Y.Y., R.P.M., and U.S.T. performed the experiments. D.E.M. wrote instru-
mentation control code. Y.G., Di Li, X.W., and C.L. analyzed the data with the
conceptual advice from Dong Li, E.B., J.L.-S., and J.H. Di Li and Y.G.
composed the figures and videos under the supervision of Dong Li. Dong Li,
E.B., and J.L.-S. wrote the manuscript with input from all the authors. All au-
thors discussed the results and commented on the manuscript.
DECLARATION OF INTERESTS
The GI-SIM technology described herein is covered by Chinese provisional
patent application 201710296751.6 filed by Dong Li and assigned to the Insti-
tute of Biophysics.
Received: May 2, 2018
Revised: July 21, 2018
Accepted: September 26, 2018
Published: October 25, 2018
REFERENCES
Akhmanova, A., and Steinmetz, M.O. (2008). Tracking the ends: a dynamic
protein network controls the fate of microtubule tips. Nat. Rev. Mol. Cell
Biol. 9, 309–322.
Behnke, O., and Moe, H. (1964). AN ELECTRON MICROSCOPE STUDY OF
MATURE AND DIFFERENTIATING PANETH CELLS IN THE RAT, ESPECIALLY
OF THEIR ENDOPLASMIC RETICULUM AND LYSOSOMES. J. Cell Biol. 22,
633–652.
12 Cell 175, 1–13, November 15, 2018
Bonifacino, J.S., and Neefjes, J. (2017). Moving and positioning the endolyso-
somal system. Curr. Opin. Cell Biol. 47, 1–8.
Bosch, M., Castro, J., Saneyoshi, T., Matsuno, H., Sur, M., and Hayashi, Y.
(2014). Structural and molecular remodeling of dendritic spine substructures
during long-term potentiation. Neuron 82, 444–459.
Boulanger, J., Gueudry, C., Münch, D., Cinquin, B., Paul-Gilloteaux, P., Bardin,
S., Guérin, C., Senger, F., Blanchoin, L., and Salamero, J. (2014). Fast high-
resolution 3D total internal reflection fluorescence microscopy by incidence
angle scanning and azimuthal averaging. Proc. Natl. Acad. Sci. USA 111,
17164–17169.
Cutrale, F., Trivedi, V., Trinh, L.A., Chiu, C.L., Choi, J.M., Artiga, M.S., and
Fraser, S.E. (2017). Hyperspectral phasor analysis enables multiplexed 5D
in vivo imaging. Nat. Methods 14, 149–152.
de Brito, O.M., and Scorrano, L. (2008). Mitofusin 2 tethers endoplasmic retic-
ulum to mitochondria. Nature 456, 605–610.
Demmerle, J., Wegel, E., Schermelleh, L., and Dobbie, I.M. (2015). Assessing
resolution in super-resolution imaging. Methods 88, 3–10.
Eisenberg-Bord, M., Shai, N., Schuldiner, M., and Bohnert, M. (2016). A Tether
Is a Tether Is a Tether: Tethering at Membrane Contact Sites. Dev. Cell 39,
395–409.
Elbaz-Alon, Y., Rosenfeld-Gur, E., Shinder, V., Futerman, A.H., Geiger, T., and
Schuldiner, M. (2014). A dynamic interface between vacuoles and mitochon-
dria in yeast. Dev. Cell 30, 95–102.
Fiolka, R. (2016). Clearer view for TIRF and oblique illumination microscopy.
Opt. Express 24, 29556–29567.
Friedman, J.R., and Voeltz, G.K. (2011). The ER in 3D: a multifunctional dy-
namic membrane network. Trends Cell Biol. 21, 709–717.
Friedman, J.R., Webster, B.M., Mastronarde, D.N., Verhey, K.J., and Voeltz,
G.K. (2010). ER sliding dynamics and ER-mitochondrial contacts occur on
acetylated microtubules. J. Cell Biol. 190, 363–375.
Friedman, J.R., Lackner, L.L., West, M., DiBenedetto, J.R., Nunnari, J., and
Voeltz, G.K. (2011). ER tubules mark sites of mitochondrial division. Science
334, 358–362.
Gierke, S., Kumar, P., and Wittmann, T. (2010). Analysis of microtubule poly-
merization dynamics in live cells. In Microtubules: In Vivo, L. Cassimeris and
P. Tran, eds. (San Diego: Elsevier Academic Press Inc), pp. 15–33.
Grigoriev, I., Gouveia, S.M., van der Vaart, B., Demmers, J., Smyth, J.T., Hon-
nappa, S., Splinter, D., Steinmetz, M.O., Putney, J.W., Jr., Hoogenraad, C.C.,
and Akhmanova, A. (2008). STIM1 is a MT-plus-end-tracking protein involved
in remodeling of the ER. Curr. Biol. 18, 177–182.
Grimm, J.B., English, B.P., Chen, J., Slaughter, J.P., Zhang, Z., Revyakin, A.,
Patel, R., Macklin, J.J., Normanno, D., Singer, R.H., et al. (2015). A general
method to improve fluorophores for live-cell and single-molecule microscopy.
Nat. Methods 12, 244–250.
Guimaraes, S.C., Schuster, M., Bielska, E., Dagdas, G., Kilaru, S., Meadows,
B.R.A., Schrader, M., and Steinberg, G. (2015). Peroxisomes, lipid droplets,
and endoplasmic reticulum ‘‘hitchhike’’ on motile early endosomes. J. Cell
Biol. 211, 945–954.
Gustafsson, M.G.L. (2000). Surpassing the lateral resolution limit by a factor of
two using structured illumination microscopy. J. Microsc. 198, 82–87.
Gustafsson, M.G.L., Shao, L., Carlton, P.M., Wang, C.J.R., Golubovskaya,
I.N., Cande, W.Z., Agard, D.A., and Sedat, J.W. (2008). Three-dimensional res-
olution doubling in wide-field fluorescence microscopy by structured illumina-
tion. Biophys. J. 94, 4957–4970.
Helenius, J., Brouhard, G., Kalaidzidis, Y., Diez, S., and Howard, J. (2006). The
depolymerizing kinesin MCAK uses lattice diffusion to rapidly target microtu-
bule ends. Nature 441, 115–119.
Helle, S.C.J., Feng, Q., Aebersold, M.J., Hirt, L., Grüter, R.R., Vahid, A., Sir-
 c
ianni, A., Mostowy, S., Snedeker, J.G., Sari , A., et al. (2017). Mechanical force
induces mitochondrial fission. eLife 6, e30292.
Hirabayashi, Y., Kwon, S.K., Paek, H., Pernice, W.M., Paul, M.A., Lee, J.,
Erfani, P., Raczkowski, A., Petrey, D.S., Pon, L.A., and Polleux, F. (2017).
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057
ER-mitochondria tethering by PDZD8 regulates Ca2+ dynamics in mamma-
lian neurons. Science 358, 623–630.
Howard, J., and Hyman, A.A. (2009). Growth, fluctuation and switching at
microtubule plus ends. Nat. Rev. Mol. Cell Biol. 10, 569–574.
Hu, J., Prinz, W.A., and Rapoport, T.A. (2011). Weaving the web of ER tubules.
Cell 147, 1226–1231.
Kiehart, D.P., Montague, R.A., Rickoll, W.L., Foard, D., and Thomas, G.H.
(1994). High-resolution microscopic methods for the analysis of cellular move-
ments in Drosophila embryos. Methods Cell Biol. 44, 507–532.
Kiehart, D.P., Galbraith, C.G., Edwards, K.A., Rickoll, W.L., and Montague,
R.A. (2000). Multiple forces contribute to cell sheet morphogenesis for dorsal
closure in Drosophila. J. Cell Biol. 149, 471–490.
Kner, P., Chhun, B.B., Griffis, E.R., Winoto, L., and Gustafsson, M.G.L. (2009).
Super-resolution video microscopy of live cells by structured illumination. Nat.
Methods 6, 339–342.
Lewis, S.C., Uchiyama, L.F., and Nunnari, J. (2016). ER-mitochondria contacts
couple mtDNA synthesis with mitochondrial division in human cells. Science
353, aaf5549.
Li, D., Shao, L., Chen, B.C., Zhang, X., Zhang, M., Moses, B., Milkie, D.E.,
Beach, J.R., Hammer, J.A., 3rd, Pasham, M., et al. (2015). ADVANCED
IMAGING. Extended-resolution structured illumination imaging of endocytic
and cytoskeletal dynamics. Science 349, aab3500.
Lu, L., Ladinsky, M.S., and Kirchhausen, T. (2009). Cisternal organization of the
endoplasmic reticulum during mitosis. Mol. Biol. Cell 20, 3471–3480.
Ma, Y., Li, D., Smith, Z.J., Li, D., and Chu, K. (2018). Structured illumination
microscopy with interleaved reconstruction (SIMILR). J. Biophotonics 11,
e201700090.
Matov, A., Applegate, K., Kumar, P., Thoma, C., Krek, W., Danuser, G., and
Wittmann, T. (2010). Analysis of microtubule dynamic instability using a plus-
end growth marker. Nat. Methods 7, 761–768.
Mitchison, T., and Kirschner, M. (1984). Dynamic instability of microtubule
growth. Nature 312, 237–242.
Nixon-Abell, J., Obara, C.J., Weigel, A.V., Li, D., Legant, W.R., Xu, C.S., Pa-
solli, H.A., Harvey, K., Hess, H.F., Betzig, E., et al. (2016). Increased spatiotem-
poral resolution reveals highly dynamic dense tubular matrices in the periph-
eral ER. Science 354, aaf3928.
Phillips, M.J., and Voeltz, G.K. (2016). Structure and function of ER membrane
contact sites with other organelles. Nat. Rev. Mol. Cell Biol. 17, 69–82.
Prinz, W.A. (2014). Bridging the gap: membrane contact sites in signaling,
metabolism, and organelle dynamics. J. Cell Biol. 205, 759–769.
Raiborg, C., Wenzel, E.M., Pedersen, N.M., Olsvik, H., Schink, K.O., Schultz,
S.W., Vietri, M., Nisi, V., Bucci, C., Brech, A., et al. (2015). Repeated ER-endo-
some contacts promote endosome translocation and neurite outgrowth. Na-
ture 520, 234–238.
Richards, B., and Wolf, E. (1959). Electromagnetic diffraction in optical sys-
tems. II. Structure of the image field in an aplanatic system. Proc. R. Soc.
Lond. A Math. Phys. Sci. 253, 358–379.
Rowland, A.A., Chitwood, P.J., Phillips, M.J., and Voeltz, G.K. (2014). ER con-
tact sites define the position and timing of endosome fission. Cell 159,
1027–1041.
Salogiannis, J., and Reck-Peterson, S.L. (2017). Hitchhiking: A Non-Canonical
Mode of Microtubule-Based Transport. Trends Cell Biol. 27, 141–150.
Self, S.A. (1983). Focusing of spherical Gaussian beams. Appl. Opt. 22,
658–661.
Simunovic, M., Manneville, J.B., Renard, H.F., Evergren, E., Raghunathan, K.,
Bhatia, D., Kenworthy, A.K., Voth, G.A., Prost, J., McMahon, H.T., et al. (2017).
Friction Mediates Scission of Tubular Membranes Scaffolded by BAR Pro-
teins. Cell 170, 172–184.e11.
Steyer, J.A., and Almers, W. (2001). A real-time view of life within 100 nm of the
plasma membrane. Nat. Rev. Mol. Cell Biol. 2, 268–275.
Sullivan, W., Ashburner, M., and Hawley, R. (2000). Drosophila Protocols (Cold
Spring Harbor, NY: Cold Spring Harbor Laboratory Press).
Tokunaga, M., Imamoto, N., and Sakata-Sogawa, K. (2008). Highly inclined
thin illumination enables clear single-molecule imaging in cells. Nat. Methods
5, 159–161.
Tortarolo, G., Castello, M., Diaspro, A., Koho, S., and Vicidomini, G. (2018).
Evaluating image resolution in stimulated emission depletion microscopy. Op-
tica 5, 32–35.
Valm, A.M., Cohen, S., Legant, W.R., Melunis, J., Hershberg, U., Wait, E., Co-
hen, A.R., Davidson, M.W., Betzig, E., and Lippincott-Schwartz, J. (2017).
Applying systems-level spectral imaging and analysis to reveal the organelle
interactome. Nature 546, 162–167.
Wagner, W., Brenowitz, S.D., and Hammer, J.A., 3rd. (2011). Myosin-Va trans-
ports the endoplasmic reticulum into the dendritic spines of Purkinje neurons.
Nat. Cell Biol. 13, 40–48.
Wang, C., Du, W., Su, Q.P., Zhu, M., Feng, P., Li, Y., Zhou, Y., Mi, N., Zhu, Y.,
Jiang, D., et al. (2015). Dynamic tubulation of mitochondria drives mitochon-
drial network formation. Cell Res. 25, 1108–1120.
Wang, S., Tukachinsky, H., Romano, F.B., and Rapoport, T.A. (2016). Cooper-
ation of the ER-shaping proteins atlastin, lunapark, and reticulons to generate
a tubular membrane network. eLife 5, 209.
Wang, C., Taki, M., Sato, Y., Fukazawa, A., Higashiyama, T., and Yamagu-
chi, S. (2017). Super-Photostable Phosphole-Based Dye for Multiple-Acqui-
sition Stimulated Emission Depletion Imaging. J. Am. Chem. Soc. 139,
10374–10381.
Waterman-Storer, C.M., and Salmon, E.D. (1998). Endoplasmic reticulum
membrane tubules are distributed by microtubules in living cells using three
distinct mechanisms. Curr. Biol. 8, 798–806.
Wong, Y.C., Ysselstein, D., and Krainc, D. (2018). Mitochondria-lysosome con-
tacts regulate mitochondrial fission via RAB7 GTP hydrolysis. Nature 554,
382–386.
Young, I.T., Zagers, R., van Vliet, L.J., Mullikin, J., Boddeke, F., and Netten, H.
(1993). Depth-of-focus in microscopy. In Proceedings of the 8th Scandinavian
Conference on Image Analysis, B. Braathen, K.A. Høgda, and K. Heia, eds.
(NOBIM, Norwegian Society for Image Processing and Pattern Recognition),
pp. 493–498.
Cell 175, 1–13, November 15, 2018 13
 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

STAR+METHODS

KEY RESOURCES TABLE

REAGENT or RESOURCE SOURCE IDENTIFIER

Bacterial and Virus Strains
mCherry-KDEL A gift from Michael Davidson Addgene plasmid # 55041
mEmerald-KDEL A gift from Michael Davidson Addgene plasmid#53975
mEmerald-Tomm20 A gift from Michael Davidson Addgene plasmid # 54282
LifeAct-mCherry A gift form Dr. Evelyne Coudrier N/A
GFP-Moe-ABD Kiehart et al., 1994 N/A
Membrane-bound GFP (mGFP) A gift from Dr. Connie Cepko Addgene plasmid Cat#22479
Chemicals, Peptides, and Recombinant Proteins
Nocodazole Sigma-Aldrich Cat#M1404-10MG
Dulbecco’s modified Eagle medium (DMEM) GIBCO Cat#10566-016
McCoy’s 5a Medium Modified (MCMM) GIBCO Cat# 30-2007
Fetal Bovine Serum (FBS) GIBCO Cat#10099-141
Critical Commercial Assays
EndoFree Mini Plamid Kit II TIANGEN Cat#DP118-02
QIAquick Gel Extraction Kit(250) QiaGEN Cat.NO.28706
RaPure Total RNA Micro Kit Magen Cat#R4012-02
PrimeScript RT reagent Kit with gDNA Eraser TaKaRa clontech Cat# RR047A
SYBR Fast qPCR Mix TaKaRa clontech Cat#RR430A
Phusion High-Fidelity DNA Polymerase New England Biolabs Cat#M0530L
Experimental Models: Cell Lines
Cercopithecus aethiops: COS-7 ATCC ATCC CRL-1651; RRID: CVCL_0042
Human: cell line U-2 OS ATCC ATCC HTB-96; RRID: CVCL_0042
Experimental Models: Organisms/Strains
Drosophila embryo Daniel P. Kiehart Lab N/A
Mouse: P0 C57BL/6J Charles River Laboratories N/A
China
Oligonucleotides
siRNA targeting sequence: RILP Ribobio China N/A
siRNA targeting sequence: FYCO1 Ribobio China N/A
q-PCR Primers for RILP/FYCO1, see methods This paper N/A
q-PCR Primers for GAPDH: Forward: PrimerBank primerBank ID:378404907c1
GGAGCGAGATCCCTCCAAAAT
q-PCR Primers for GAPDH: Reverse: PrimerBank primerBank ID:378404907c1
GGCTGTTGTCATACTTCTCATGG
Recombinant DNA
3XmEmerald-ensconsin N/A Addgene Cat#53975
cDNA NM_001198608 National Coalition Building Institute N/A
Halo-LAMP1 Grimm et al., 2015 Addgene plasmid#55073
Software and Algorithms
MATLAB 2016a MathWorks https://www.mathworks.com/campaigns/
products/trials.html?prodcode=ML&s_tid=ML_
mod_pers_trial2&elqCampaignId=8140;
RRID: SCR_001622
GraphPad Prism GraphPad Software https://www.graphpad.com/scientific-
software/prism/; RRID: SCR_002798
Fiji Fiji http://fiji.sc/; RRID: SCR_002285

e1 Cell 175, 1–13.e1–e6, November 15, 2018

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dong Li
(lidong@ibp.ac.cn).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell culture
Bothe COS-7 and U2OS cells were purchased from American Type Culture Collection (ATCC). COS-7 cells were derived from the
kidney of Cercopithecus aethiops, and U2OS cells were derived from human female moderately differentiated sarcoma of the tibia.
COS-7 cells were grown in DMEM media (GIBCO) supplemented with 10% fetal bovine serum (GIBCO) and 1% penicillin/strepto-
mycin at 37 C and 5% CO2 until
60 to 80% confluency was reached. U2OS cells were grown in MCMM media (GIBCO) supple-
mented with 10% FBS (GIBCO) and 1% penicillin/streptomycin at 37 C and 5% CO2 until
60 to 80% confluency was reached.
For GI-SIM imaging experiments, 35 mm coverslips were pre-coated with 50 mg/ml collagen for 1 hour, and cells were then seeded
onto the coverslips to achieve
70% confluence prior to transfection. Transfections were executed using Lipofectamine 3000 (In-
vitrogen) according to the manufacturer’s protocol. Cells were imaged 16-36 hours post-transfection in a microscope stage top mi-
cro-incubator (OKO Lab) maintained at 37 C and 5% CO2. Where indicated, the cells transfected with HaloTag plasmids were
labeled with JF646 ligand following the published protocol (Grimm et al., 2015), and the cells were imaged immediately afterward.

Drosophila embryo
The transgenic Drosophila line stably expressing a GFP-tagged Actin Binding Domain of fly Moesin (GFP-Moe-ABD, referred to as
sGMCA) was generated by Dan Kiehart lab (Kiehart et al., 2000). Embryos were prepared as previously described (Kiehart et al.,
1994). Briefly, adult flies laid eggs for 3-4 hours at 25 C which were subsequently aged at 16 C for
24 hours. Embryos were dechor-
ionated in 50% bleach for 1.25 min. Dorsal closure staged embryos were lined up on a grape juice agar pad, secured to a coverslip
using embryo glue (Kiehart et al., 1994; Sullivan et al., 2000) and desiccated on the coverslip in a tightly sealed jar containing Drierite
(W. A. Hammond Drierite Company, Ltd.) for 6-18 minutes depending on the length of time between dechorionation and desiccation.
This step helps flatten the embryos against the coverslip when mounted. Note that the desiccation time is less than what is typically
used to prepare embryos for microinjection of DNA constructs for creating transgenic flies. The desiccated embryos typically hatch,
and when rescued to a food source, development progresses normally.

Primary hippocampal neurons

C57BL/6J mice were purchased from SPF (Beijing) biotechnology Co., Ltd. The hippocampi from P0 female or male C57BL/6J mice
were dissociated with 0.125% trypsin in Hanks’ balanced salt solution without Ca2+ and Mg2+ at 37 C for 15 min, triturated in DMEM
(Hyclone, Logan, HT) with 10% F12 and 10% fetal bovine serum, and plated on poly-D-lysine-coated 25 mm coverslips at a density of
1 3 105 cells/dish. The medium was replaced with serum-free Neurobasal medium supplemented with 2% B27 supplement and
GlutaMAX (GIBCO, Invitrogen, Carlsbad,CA) 4 h after plating. Half of the media was changed every 3 days until use. For LTP induction
and live imaging, neurons were transfected on DIV (days in vitro) 13 with constructs expressing membrane-bound GFP (mGFP, Addg-
ene plasmid #22479) and LifeAct-mCherry (gift from Dr. Evelyne Coudrier, Institute Curie, France), and treated on DIV18 with glycine
(200 mM) in Mg2+-free extracellular solution (in mM: 125 NaCl, 2.5 KCl, 2 CaCl2, 5 HEPES, 33 glucose, 0.2 glycine, 0.02 bicuculline,
and 0.003 strychnine, pH 7.4). All animal experiments were performed in compliance with the guidelines of the Animal Care and Use
Committee of the Institute of Genetics and Developmental Biology, Chinese Academy of Sciences.

METHOD DETAILS

GI-SIM system
GI-SIM systems at Janelia and the Institute of Biophysics were used for the experiments. Both setups (Figure S7A) represent
improved versions of our previous TIRF-SIM system (Li et al., 2015). Briefly, three laser beams of 488 nm (500 mW, Coherent,
Genesis-MX-SLM), 560 nm (500 mW, MPB Communications, 2RU-VFL-P-500-560), and 642 nm (500 mW, MPB Communications,
2RU-VFL-P-500-642) were collinearly combined and passed through an acousto-optic tunable filter (AOTF; AA Quanta Tech,
AOTFnC-400.650). The beam was then expanded to a diameter of 20 mm, and directed into a phase-only modulator composed
of a polarization beam splitter, an achromatic half-wave plate, and a ferroelectric spatial light modulator (SLM; Forth Dimension Dis-
plays, SXGA-3DM). The incident light was diffracted by the grating pattern displayed on the SLM, and then passed through a home-
made azimuthally patterned achromatic half-wave plate (HWP, Bolder Vision Optik), which consisted of three pairs of segments. The
fast axes of each pair of HWP are properly oriented (Figure S7B), so that the linearly polarized diffracted light from the grating pattern
with a given orientation can be rotated into the desired s-polarized light. Therefore, the contrast of the illumination pattern formed
inside the specimen was maximized. In contrast to the liquid crystal device-based polarization rotator that takes
10 ms to accom-
plish each polarization rotation, the azimuthal HWP requires no settling time, which permitted us to achieve the maximum imaging
speed limited only by the readout time of the sCMOS camera. The diffracted light was spatially filtered by a mask that only allowed the

Cell 175, 1–13.e1–e6, November 15, 2018 e2

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

desired ± 1 diffraction orders to pass through. The ± 1 order light spots were then relayed onto the back focal plane of the high
numerical aperture (NA) objective (Olympus APON 100XHOTIRF 1.7 NA, or UAPON 100XOTIRF 1.49 NA for Drosophila embryo spec-
imen). The distance between the two light spots determined the excitation NA (i.e., the incidence angle onto the interface of the spec-
imen and coverslip), which could be adjusted by changing the period of the pattern displayed on the SLM. The pattern generation
algorithm used for patterned activation nonlinear SIM allowed us fine tune the period of the grating pattern to search the GI NA. After
identifying the GI NA, the fluorescent image generated by the applied excitation pattern of each phase and orientation was collected
by the same objective, separated by a dichroic beam splitter (Chroma, ZT405/488/560/647tpc), and focused by a tube lens onto a
sCMOS camera (Hamamatsu, Orca Flash 4.0 v3). The nine raw images consisting of 3-orientation 3 3-phase for each time point were
reconstructed into a super-resolution (SR) image based on the previous algorithm (Gustafsson et al., 2008). In order to suppress the
noise in the high frequency region, we adopted a Gaussian apodization function with standard deviation, s % 45 nm. Changing the
standard deviation allowed flexible adjustment of the cut-off frequency of the reconstructed optical transfer function (OTF) according
to the practical signal-to-noise ratio (SNR) offered by different specimens. This strategy permitted case-by-case quality control to
minimize the reconstruction artifacts, which usually originate from the high frequency region with low SNR.

Interleaved reconstruction
In the conventional reconstruction algorithm, the raw images of each pattern orientation are only employed once to generate a GI-
SIM image. At the next time point, the GI-SIM image is reconstructed from another set of 3 orientation images (upper in Figure S7C).
However, the high resolution information is independently reconstructed for each pattern orientation, and there is no inherent require-
ment for the acquisition order among the 3 pattern orientations. Therefore, regrouping the data stream with either orientation 2 or 3 as
the starting orientation would generate another two GI-SIM images interleaved between two consecutive conventional time points
(lower in Figure S7C). For continuously acquired datasets (i.e., without intervals between frames, e.g., Video S1, part II), we utilized
the interleaved reconstruction strategy (Ma et al., 2018) to improve the GI-SIM frame rate by 3-fold. The advantage of interleaved
reconstruction is that the new temporal information can be incorporated to update the GI-SIM image as soon as it is available in
the raw data of a single pattern orientation. As demonstrated in Figure 1C, the ultra-rapid constrictions and bulges along ER tubules
can be clearly visualized with interleaved reconstruction.

Characterization of the depth-of-focus of the GI-SIM system

The depth-of-focus (DOF) of a microscope system defines the maximally allowable defocus distance (Dz) with respect to the ideal
focal plane of an objective lens, within which the specimen image can be observed without significant blurring. As any fluorescence
microscope (e.g., GI-SIM) is an incoherent imaging system, the fluorescent emissions from different molecules do not interfere with
one another. Therefore, the DOF of a fluorescence microscope is twice the maximum defocus distance, DOF = 2∙Dz (Figure S7D).
To estimate Dz, we consider the optical transfer function (OTF), which specifies the frequency response of the microscope system.
Its amplitude at any given spatial frequency describes the transmission efficiency of the microscope system for information at that
spatial frequency. In general, the amplitude of the OTF is a monotonously decreasing function with its maximum transmission
efficiency (i.e., OTF = 1) at zero-frequency, and minimum efficiency (i.e., OTF = 0) at the frequency defined by the diffraction-limit
(Figure S7E). Therefore, the area under the OTF curve is useful to evaluate its capability of acquiring fluorescence signals. To deter-
mine how defocus influences the OTF, we consider two point sources along the optical axis, one located in the focal plane, z0; the
other one located outside the focal plane, defocused by Dz (Figure S7D). Given that the objective pupil subtends a half angle of a as
dictated by its numerical aperture (NA), the maximal path length difference between the spherical wavefronts of the two point sources
occurs at the maximum detectable angle a. The maximal path length, dDz, is given by Equation (1) (Young et al., 1993):
dDz = 2Dz sin2 ða=2Þ (1)

Using Equation (1) and the vector model of diffraction (Richards and Wolf, 1959), we simulated the OTFs at different path length dif-
ferences (i.e., different amounts of defocus; Figure S7E). As shown, defocus distorts the OTF curve. In particular, when the path
length difference is greater than 66% of the emission wavelength in the immersion medium (dDz > 0.66l/n), a sign reversal occurs
in the OTF (blue and red curves in Figure S7E) (l = the emission wavelength in a vacuum, n = the refractive index of the immersion
medium of the objective lens). Thus, the contrast of some features in the acquired image will be reversed, complicating our under-
standing of the detected modulation pattern in the GI-SIM raw images. To avoid this, the maximally allowable path length difference
for a green fluorescence image with a central emission wavelength of
525 nm is dDz = 0.195 mm. This corresponds to the defocus
amount of Dz = 0.614 mm, so the corresponding DOF is 2∙Dz = 1.228 mm.
However, in this work we have used a more restrictive criterion to define the DOF of GI-SIM: the area under the defocused OTF
curve must be greater than half of the area under the in-focus OTF curve (Figure S7F). This standard ensured that the contrast of
each raw image was sufficient for artifact-free GI-SIM image reconstruction (Figure S7G). Using this restricted criterion, the DOF
of a GI-SIM system equipped with a 1.7 NA objective lens was determined as: DOFGI-SIM = 2∙Dz half area = 0.986 mm z1 mm (Figures
S7F and S7G).

e3 Cell 175, 1–13.e1–e6, November 15, 2018

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

Determining the optimal condition to image intracellular organelles

To determine the optimal excitation NA for GI-SIM and the NA at which the signal-to-background ratio (SBR) is a maximum in raw
images of intracellular organelles, we imaged the ER and MT network of live COS-7 cells expressing mEmerald-KDEL (for the ER) or
3 3 mEmerald-ensconsin (for MTs). We acquired a series of images as we reduced the excitation NA from ultrahigh TIRF 1.55 NA to
0 NA (i.e., wide-field excitation) (Figures S1A–S1D). The ER or MT network in each image were masked using a skeletonization
algorithm (described in the section of ‘‘tubular ER oscillation analysis’’), and the SBR at each NA was calculated as the ratio of
the total fluorescence intensity of the masked region to that of the unmasked region. We found the maximum SBR occurred at

1.37 NA (Figure S1F).
We also characterized the total fluorescence (including ER or MT features and out-of-focus background) versus the excitation NA.
The result could be fitted by a combination of an exponential function for the TIRF illumination region and a linear function for non-
TIRF illumination (Figure S1E). The critical NA for TIRF, defined as the intersection of these two functions, was 1.371 ± 0.004 NA
(mean ± S.D., n = 10 cells).
A tightly confined excitation spot at the objective rear pupil is essential to confine the illumination to the small range of NA neces-
sary to minimize out-of-focus excitation. However, the focal spot of the excitation light at the back focal plane of the objective is
diffraction-limited, so the focal spot is not infinitesimal. Its radius can be calculated by (Self, 1983):
lf
rF = (2)
prL

where l = 0.488 mm is the wavelength of excitation light, f = 180 mm is the focal length of the tube lens of the GI-SIM system, and rL =
3.43 mm is the radius of the collimated beam at the tube lens. Therefore, the radius of the focal spot is rF = 8.15 mm, which corre-
sponds to a small DNA = 0.0045. Consequently, the focal spot of excitation light actually crosses the critical NA for TIRF, but
most of the excitation light within the focal spot is at an excitation NA just below the critical NA for TIRF. At this pre-critical excitation
NA, the excitation light is refracted at the interface of coverslip and cell specimen with a refraction angle > 85 , and therefore forms the
grazing incidence illumination.

Characterization of the irradiation depth of the GI illumination mode

To measure the depth of GI illumination, we imaged a sample consisting of 4 mm-diameter polystyrene fluorescent beads on a cover-
slip at excitation NAs from 1.55 (ultrahigh TIRF) to 0 (wide-field excitation) (Figure S1K). At any given depth z from the coverslip, the
illumination creates within a bead of radius r a fluorescence circle of radius d in the relationship:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
z = r  r 2  d2 (3)

By plotting d as a function of excitation NA, we determined the irradiation depth z at each NA through Equation (3) (Figure S1K). In
particular we found that the irradiation depth at 1.33 NA was very close to the DOF of our GI-SIM system (i.e., the cyan and orange
dashed circles coincided each other, green boxed image in Figure S1K).
Next, we characterized the total fluorescence intensity as a function of the excitation NA (Figure S1L). Similar to the live-cell mea-
surements, the curve could be fitted by the combination of an exponential function for the TIRF region and a linear function for the
non-TIRF region. The intersection of the two curves represents the critical NA for TIRF, which is
1.333 ± 0.002 for a polystyrene
bead submerged in water. Therefore, 1.33 NA is the GI NA for the polystyrene bead specimen.
In summary, the illumination depth at GI NA is well matched to the objective DOF. The extended illumination depth at GI excites
more fluorescent molecules than TIRF, and all these excited molecules, being in focus, contribute to a high contrast raw image.
In particular, compared to the images acquired at the TIRF NA of 1.55, the signal strength of the GI image was generally improved
by > 10-fold (Figures S1B and S1D). GI illumination therefore provides optimal imaging conditions for live-cell imaging.

RNAi transfection and real-time PCR

RILP and FYCO1 were depleted using genOFFTM siRNA silencing kits (Ribobio, China). U2OS cells were plated in a 6 cm dish at a
density of 5 3 105 cells per dish. The cells were first transfected with 20 nM RNAi oligonucleotides and 20 nM Negative Control siRNA
using LipofectamineTM RNAiMax (ThermoFisher Scientific) in GIBCO OPTI-MEM reduced serum media. After 6 hours of RNAi trans-
fection, the cells were washed and the medium was replaced with DMEM supplemented with 10% FBS and 1% penicillin/strepto-
mycin. 48 hours after the first round of transfection, cells were transfected again with mCherry-KDEL and HaloTag-Lamp1 plasmid
DNA using Lipofectamine 3000 (Invitrogen). After 12 hours of plasmid DNA transfection, cells were re-plated onto 25 mm coverslips
overnight before GI-SIM imaging was performed. Total RNA was extracted from different siRNA-treated samples using TRIzol (Invi-
trogen) reagent and cDNA was prepared by reverse transcription using the RevertAid First Strand cDNA Synthesis kit (ThermoFisher
Scientific). The knockdown efficiency of different siRNAs was confirmed using SYBR green mix (Takara) in a BioRad CFX1000 touch
real-time PCR machine. The qRT-PCR primers were:

RILP forward primer, 50 - GAGCGGAATGAACTCAAAGCC-30 ;

RILP reverse primer, 50 -CCTCTGGTGTCCCTAACATCTTG-30 ;

Cell 175, 1–13.e1–e6, November 15, 2018 e4

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

FYCO1 forward primer, 50 -GACTACTGGGATTACTTCTGTGC-30 ;

FYCO1 reverse primer, 50 -TTTCCCCAAGGATGTTCGGAG-30 .

Nocodazole treatment and imaging

COS-7 cells were passaged and transfected as described above. Coverslips cultured with transfected cells were then mounted in a
chamber placed in the microscope stage top micro-incubator. After locating the cells to image, 5 mM Nocodazole (Sigma) was added
for at least 15 min or until the majority of microtubules (MTs) had depolymerized. Time series GI-SIM images were acquired from cells
after treatment to observe the tubular ER growth activity.

GI-SIM imaging for Drosophila embryo

Because the refractive index of the vitelline membrane is close to that of the coverslip (n = 1.55), the excitation light was not refracted
until it encounters the interface between the vitelline membrane and the amnioserosa cells. Consequently, the grazing incidence illu-
mination formed on the vitelline membrane surface facing the amnioserosa cells. This enabled the apical volume of amnioserosa cells
to be observed with optimized imaging conditions, like those used for cultured cells.

QUANTIFICATION AND STATISTICAL ANALYSIS

Tubular ER oscillation analysis

To precisely analyze the tubular ER oscillation frequency and amplitude, COS-7 cells expressing the ER marker mEmerald-KDEL
were visualized continuously at the frame rate of 266 fps by GI-SIM. The datasets were analyzed by the following procedure imple-
mented in MATLAB 2016a. The GI-SIM images of the ER network were pre-processed by adjusting the brightness and contrast and
removing the local background to enhance the tubular ER features. Second, the pre-processed images were binarized using the
extended-minima transform ‘imextendedmin’, and the binary images were skeletonized by the morphological operation ‘skel’. Junc-
tion points were identified from the skeletonized images using the morphological operation ‘branchpoints’. Third, except for junction
points, the normal of every pixel of the skeletonized ER images was drawn according to its local curvature. The intersections of the
normal with the skeletonized structure were tracked for 540 frames to obtain kymographs, which depicted the oscillation of these
intersections. Finally, the maxima and minima of each kymograph were recognized by the ‘findpeaks’ function in MATLAB. The oscil-
lation frequency of every pixel in the ER skeleton was the ratio of the number of paired maxima and minima within the kymograph to
the length of the time course. The oscillation amplitudes were computed as the average distances between paired maxima and
minima within the kymographs.

Characterization of MT dynamic instability

To analyze MT dynamic instability, COS-7 cells expressing the MT marker of 3 3 mEmerald-ensconsin were visualized at the frame
rate of 9.2 fps by GI-SIM. In order to precisely identify the dynamic position of the MT ends, we used a similar procedure as in the
tubular ER oscillation analysis to skeletonize every time point of the MT image. We then tried to identify the positions of the MT ends
using the ‘bwmorph’ function for each frame, but not all MT ends could be detected, and the resulting positions usually contained
some false-positive results due to non-uniform labeling or varied photobleaching rate in different regions of the MTs. Therefore, a
computer-assisted hand clicking program was used to semi-manually identify the undetected MT ends, and remove the false-pos-
itive positions. Because the pixel size of the GI-SIM image was 30 nm, it allowed us to precisely localize the MT end positions using
the local fluorescent profiles around the identified positions. Next, we extracted the fluorescence intensity profiles along and perpen-
dicular to the line of the MT skeleton. Both profiles were centered around the identified MT end position. Therefore, the MT end po-
sitions could be localized with sub-pixel accuracy.
After detecting the time series positions of the MT end, we calculated: (i) the vectors representing the displacement and direction of
the instantaneous movement on a frame-to-frame basis; (ii) the intersection angle between the vector of instantaneous movement
and the extension of the MT skeleton at the end position. Considering that MTs rarely bend more than 60 over a short distance
(Matov et al., 2010), we defined that if the intersection angle was less than 60 and the vector of instantaneous movement pointed
in the extension direction of the MT, then the MT end movement between the two consecutive frames was determined as a growth
event, and its displacement was designated as positive; if the intersection angle was less than 60 but the vector of instantaneous
movement pointed inward, the MT end movement was determined as a shortening event, and its displacement was designated as
negative; otherwise, if the intersection angle was greater than 60 , the MT end movement usually resulted from a lateral shift, then it
was determined as a pause event with its displacement assigned to zero. After determining the sign of each displacement, the dy-
namic displacement curve of the MT end can be plotted as in Figure 2D.
We categorized the phases of growth, shortening, and pause based on the following criteria. (i) If two or more growth events were
identified consecutively, and the total growth displacement was greater than the GI-SIM resolution of 97 nm, then this time interval
was categorized as a growth phase (labeled in green in Figure 2D). (ii) Similarly, if two or more shortening events were identified
consecutively, and the total shortening displacement was greater than the GI-SIM resolution of 97 nm, then this time interval was
categorized as a shortening phase (labeled in blue in Figure 2D). (iii) The residual time intervals were categorized as pause phases,

e5 Cell 175, 1–13.e1–e6, November 15, 2018

 Please cite this article in press as: Guo et al., Visualizing Intracellular Organelle and Cytoskeletal Interactions at Nanoscale Resolution on Milli-
second Timescales, Cell (2018), https://doi.org/10.1016/j.cell.2018.09.057

which usually presented as fluctuations with amplitudes smaller than the GI-SIM resolution (labeled in red in Figure 2D). The velocity
of each growth or shortening phase was calculated as the slope of the linear fitting of the phase. Consistent with the previous
definition (Gierke et al., 2010), the catastrophe frequency was defined as the number of transitions from growth to shortening divided
by the time the MT spent growing. The rescue frequency was defined as the number of transitions from shortening to growth divided
by the time the MT spent shortening.

Tubular ER growth analysis

To analyze the proportion and growth velocity of tubular ER generation events via the five kinds of mechanism, COS-7 or U2OS cells
co-expressing 3 3 mEmerald-ensconsin (MT marker), mCherry-KDEL (ER marker), and HaloTag-Lamp1 labeled with the JF646 ligand
(late endosome/lysosome marker; LE/Lyso), were imaged by GI-SIM every 2 s for 4 min. We selected LEs/Lysos as representative
vehicle organelles to characterize the properties of the hitchhiking mechanism, because almost all LEs/Lysos were in contact with the
ER (Figure 5; Video S4) (Rowland et al., 2014). Tubular ER generation events were counted if the de novo ER tubules branched from
the existing ER and extended more than 1 mm. After finding the tubular ER generation events in ER fluorescence images, we classified
each event by identifying its dependence on MTs and/or LEs/Lysos in three-color GI-SIM time-lapse images. The proportions of the
five kinds of tubular ER generation mechanism were calculated from a total of 529 events in 16 cells.
To calculate the growth velocity for each event, we first measured the time intervals from the initial appearance of each new ER
tubule branch (start time point) until the tubule fused with another tubule or extended to the furthest position before retracting
(end time point). Second, we identified the tip positions in the key frames depicting the growth trajectory between the start and
end time points, and then the total displacement during the ER tubule extension process was calculated as the sum of the sub-dis-
placements between consecutive tip positions. Finally, the growth velocity of each event was calculated as the quotient of displace-
ment divided by time.

Association of mitochondrial fission/fusion with ER-Mito contacts

To identify the association of mitochondrial fission and fusion with ER-Mito contacts, COS-7 cells co-expressing the mitochondrial
marker mEmerald-Tomm20 and the ER marker mCherry-KDEL were imaged by GI-SIM every 1 or 2 s for 2 to 4 min. We screened the
mitochondrial GI-SIM image series to identify mitochondrial fission and fusion events. To determine the association of ER tubules
with dividing or coalescing mitochondria, we scrutinized the behaviors of mitochondria and ER tubules in the merged images at
time points preceding mitochondrial division or coalescence. For each fission event, if the ER tubule was found to cross the dividing
mitochondrion, and the mitochondrion was constricted at the ER tubule crossing site, then the fission event was identified as asso-
ciated with the ER. For each fusion event, if the local movement of one mitochondrion highly correlated with its adjacent ER tubule
(which was already in contact with the mitochondrion), and the two mitochondria coalesced at the ER-Mito contact site, then the
fusion event was identified as associated with the ER. Moreover, for each fusion event, the fusion time was counted as the time in-
terval from the time point at which the outer membranes of the two mitochondria first touched each other in the GI-SIM image to the
time point at which the fusion pore between the two mitochondria expanded to its largest range. Because mEmerald-Tomm20 labels
the outer membranes of mitochondria, the enhanced fluorescence between the two mitochondria indicated that their outer mem-
branes touched each other (e.g., the 10 and 12 s time points in Figure 4D), but had not coalesced. Otherwise, if the membranes
had already coalesced, the fluorescence enhancement in the outer membrane would disappear (e.g., the 14 and 18 s time points
in Figure 4D).

Cell 175, 1–13.e1–e6, November 15, 2018 e6

Supplemental Figures

(legend on next page)

Figure S1. Characterization of GI Illumination Mode, Related to Figure 1
(A–F) Characterization of the GI NA and generated fluorescence strength at different excitation NAs with live COS-7 cells. SIM images of the ER network (A, B) and
MT cytoskeleton (C, D) acquired at excitation NAs of 1.55, 1.45, 1.39, 1.37, 1.35, 1.33 and 0 (i.e., conventional widefield modality). The images in (A) and (C) were
normalized individually to their own maximum intensity; thus, the visibility of fine structures and signal-to-background ratio can be compared between different
excitation NA conditions. The images in (B) and (D) correspond to the images in (A) and (C) respectively, but are normalized to the maximum intensity value of the
image acquired at GI NA of 1.37. Therefore, the signal strength, I, can be compared between different excitation NA conditions. Scale bar: 5 mm in (A)-(D). (E) Total
fluorescence intensity in logarithmic coordinates versus excitation NA. This plot can be fitted by the combination of an exponential function for the TIRF region
and a linear function for the non-TIRF region. The intersection of the fitted curves shows that the critical NA for TIRF is 1.375 for live cell specimen; therefore, the
excitation NA just below the critical NA was identified as the GI NA (1.37 NA). (F) Signal-to-background ratio (SBR) versus excitation NA, which achieves a
maximum at GI NA. This figure illustrates that GI NA is the optimal excitation condition that maximizes the irradiation depth, usable fluorescent signal strength,
and SBR.
(G–J) Comparison of single molecule imaging performance at TIRF, GI and HILO excitation NAs from fixed COS-7 cells transfected with Halo-Calnexin labeled
with PA JF646 ligand. (G) Representative single molecule images acquired at TIRF NAs of 1.55 and 1.45, GI NA of 1.37, and HILO NA of 1.33 from the same cell at
the same illumination intensity and acquisition time. The green circles in each image indicate the positions of single molecule events successfully detected at this
excitation NA; the red circles indicate the positions of single molecule events failed to be detected at this excitation NA, while successfully detected at GI or HILO
NA. Scale bar: 5 mm. (H) Comparison of the intensity profiles and single-to-background (SBR) of two single molecule events (indicated in (G)) acquired at different
excitation NAs. (I) Comparison of the number of detected single molecule events at different excitation NAs. (J) Statistical quantification of single molecule
tracking time (black curve, Mean ± SEM) acquired at different excitation NAs for SBR larger than 0.5; Comparison of the excitation intensities needed to obtain the
same SBR (> 0.5) at different excitation NAs (Magenta curve). Therefore, GI mode provides superior sensitivity, SBR, and tracking duration compared to TIRF and
HILO illumination modalities.
(K–M) Characterization of GI NA and irradiation depth with fluorescent beads immersed in water. (K) Fluorescent bead images acquired at excitation NAs
corresponding to TIRF, GI and widefield illumination. Cyan dashed circles indicate the irradiation diameters achieved at different excitation NAs, and the orange
dashed circles indicate the depth-of-focus (DOF) diameter at the NA of GI-SIM in water. Scale bar: 1 mm. (L) Total fluorescence intensity in logarithmic coordinates
versus excitation NA. Similar to the measurements in live cells, it can be fitted as the combination of an exponential function for the TIRF region and a linear
function for the non-TIRF region. The intersection of the two curves represents the critical NA for TIRF: 1.333 for bead specimens. Therefore, the GI excitation NA
just below the critical NA was identified as 1.33 NA. (M) Irradiation depth in logarithmic coordinates versus excitation NA. This can also be fitted by a combination
of exponential and linear functions. The DOF of GI-SIM (Characterized in the STAR Methods) is indicated by the orange dashed line, which matches the irradiation
depth achieved at GI NA, also highlighted by the green box in (K).
(legend on next page)
Figure S2. Characteristics of MT Dynamic Instability and ER Network Dynamics, Related to Figure 2
(A–D) Comparison of target protein selection for imaging the MT cytoskeleton and ER network. Top: GI-SIM images of live COS-7 cells expressing (A) mCherry-
a-tubulin and (B) 3 3 mEmerald-ensconsin. Both fluorescent marker proteins correctly label the MT cytoskeleton, but ensconsin labeling produces much less
background fluorescence than a-tubulin labeling. Bottom: GI-SIM images of Live COS-7 cells expressing the ER makers (C) mEmerald-Sec61b and (D) mCherry-
KDEL. The reticular nets are identical in (C) and (D), confirming that mCherry-KDEL labels the entire ER network. Scale bar: 3 mm.
(E–K) Characterization of GI-SIM resolution using mEmerald-labeled MTs. (E) Diffraction-limited GI image of MTs. (F) GI-SIM image corresponding to (E). (G)
Comparison of the boxed regions in (E) and (F). (H) Fourier transform of the GI image in (E). (I) Fourier transform of the GI-SIM image in (F). The classic diffraction
limit dictated by the nominal 1.7 NA of the objective lens is indicated by the cyan dashed circle (radius 6.48 mm-1). However, sample information carried by the
conventional GI image is only visible in the low spatial frequency region indicated by the green dashed circle (radius 5.07 mm-1), which means that the
effective detection NA is
1.33 for 525 nm light (the central wavelength of green fluorescence), as is true for most high NA objectives. The magenta circle (radius
10.31 mm-1) indicates the limit of sample information carried by the GI-SIM image, which is equivalent to being acquired by an objective with an effective detection
NA of 2.71 for 525 nm light. (J) Normalized intensity profiles of an isolated MT from the GI (blue) and GI-SIM (orange) images shown in the upper two panels of (G).
(K) Normalized intensity profiles of four MT bundles from the GI (green) and GI-SIM (red) images shown in the lower two panels of (G). Scale bar: 4 mm in (E) and (F);
1 mm in (G).
(L–O) Statistical analysis of tubular ER oscillation. (L) Histogram of oscillation frequency that follows the logarithmic normal distribution with a median of 10.1 Hz.
(M) Histogram of oscillation amplitude that also follows the logarithmic normal distribution with a median of 144 nm. (N) Oscillation frequency versus the length of
ER tubule. Short ER tubules are capable of oscillating faster than long tubules. (O) Comparison of the heterogeneity in ER tubule oscillations characterized at
different imaging speeds. Precise characterization of ER tubule oscillation demands an ultra-high imaging speed, since GI-SIM even at 44 fps is still not enough to
measure the full frequency spectrum accurately.
(P and Q) Variation in total fluorescence intensity of ER tubules during dynamic segregation. (P) Orange and green curves outline the regions used to quantify the
total fluorescence intensity of specific ER tubules. (Q) Fluorescence intensity variation curves of the orange- and green-outlined tubular ER areas. The nearly
constant intensity suggests that the dynamic constrictions observed in the ER tubules result from heterogeneous displacement of KDEL within the ER lumen.
Scale bar: 1 mm.
(legend on next page)
Figure S3. ER Dynamics after Nocodazole Treatment, Related to Figure 3
(A) The ER network and MT cytoskeleton before Nocodazole treatment.
(B) The same cell after Nocodazole treatment for 20 mins.
(C) MT cytoskeleton from (A) before Nocodazole treatment.
(D) MT cytoskeleton image from (B) after Nocodazole treatment.
(E) Time-lapse images of ER tubule generation via the sliding mechanism along a residual stable MT after Nocodazole treatment.
(F) Time-lapse images of ER tubule generation via de novo budding after Nocodazole treatment. COS-7 cells expressed 3 3 mEmerald-ensconsin (MTs, green)
and mCherry-KDEL (ER, magenta). Scale bar: 5 mm in (A)-(D), 2 mm in (E), 1mm in (F).
Figure S4. The Writhing Dynamics of a Dividing Mitochondrion, Related to Figure 4
(A) Two-color time-lapse images show the morphological changes of the two mitochondrial compartments during fission (Video S3, part II).
(B) Segmented images of the dividing mitochondrion. The segmentation algorithm distinguished the two compartments (labeled in cyan and green) defined by the
ER tubule (labeled in red) crossing the fission site.
(C) Quantification of the temporal changes of the areas of the two compartments. The two compartments exhibited complementary area changes during
constriction before fission was accomplished. Scale bar: 1 mm.
Figure S5. The ER-LE or Lyso Contacts Stabilize the Location of LEs or Lysos, Related to Figure 5
(A) After RNAi interference, the mRNA expressions of the adaptor proteins FYCO1 and RILP1 for the kinesin and dynein molecular motors were significantly
decreased. Therefore, the directional translocation of LEs/lysosomes along MTs was largely prevented. Data is shown as mean ± SEM. **p < 0.01.
(B) GI-SIM live image of ER and LE/Lyso contacts in COS-7 cells expressing mCherry-KDEL and HaloTag-Lamp1. The two LEs/Lysos indicated by arrows #1 and
#2 are tightly corralled by the ER, while the other two LEs/Lysos indicated by arrows #3 and #4 are not.
(C) Zoom-in images and trajectories of the four LEs/Lysos indicated in (B). The tightly corralled LEs/Lysos (#1 and #2) exhibit a much smaller mean square
displacement (MSD) than the two that are not. Scale bar: 3 mm in (B); 0.5 mm in (C).
Figure S6. Hitchhiking Interaction Leads to the Stable Existence of Free-Ended ER Tubules, Related to Figure 6
Time-lapse single color GI-SIM images of the ER network (right column) show that free-ended ER tubules (indicated by white arrows) can be found in the
cytoplasm. However, the corresponding two-color GI-SIM images of mitochondria and ER (left column) reveal that some of the free-ended ER tubules actually
hitchhike on mitochondria that remain relatively static. Scale bar: 2 mm.
(legend on next page)
Figure S7. Miscellaneous Information about GI-SIM System, Related to the STAR Methods
(A) Optomechanical configuration of the GI-SIM system. AOTF: acousto-optic tunable filter; M: mirror; HWP: half-wave plate; SLM: spatial light modulator; DM:
dichroic mirror.
(B) The design of azimuthal HWP. The fast axes of each pair of HWP segmentations (indicated by green dashed arrows) are properly orientated, so as to rotate the
horizontally polarized incident light (indicated by red arrows) into the orientation perpendicular to the connection of the two incident light spots (indicated by blue
spots). Therefore, after passing through the azimuthal HWP, the excitation light of the 3 pattern orientations is s-polarized (indicated by blue arrows).
(C) Illustration of conventional and interleaved reconstruction strategies, Related to the STAR Methods. Red, orange and green boxes represent the raw images
acquired for 3 orientations of the illumination pattern. The acquisition order is orientation [1, 2, 3] followed by looping. Upper: the conventional reconstruction
algorithm uses a set of raw images of orientation [1, 2, 3] to reconstruct a SIM image at each specific time point (e.g., TP0, TP1 in upper panel). Lower: the
interleaved reconstruction algorithm regroups the raw images into orientation [1, 2, 3], [2, 3, 1] and [3, 1, 2], which generate another two SIM images (i.e., TP1, TP2
in lower panel) interleaved between the two consecutive time points produced by conventional reconstruction strategy. Therefore, interleaved reconstruction
improves the GI-SIM frame rate by 3-fold, and the new temporal information can be incorporated to update the GI-SIM image as soon as it is available in the raw
images of a single pattern orientation.
(D–G) Influence of the defocus effect on the optical transfer function (OTF) and point spread function (PSF). (D) Schematic of the path length difference, dDz,
between the in-focus and defocused wavefronts emitted from the focus point and the other two points defocused by Dz. (E) Distortions of the OTF
induced by different path length differences. In particular, when the path length difference is greater than 66% of the emission wavelength in immersion medium
dDz > 0.66l/n, a sign reversal of the OTF occurs R 66% (blue and red curves). Thus, the contrast of some features in the acquired image will be reversed. (F) Area
under OTF curve versus the amount of defocus, Dz. When Dz = ± 490 nm, the area under the OTF curve decreases by 50% relative to the in-focus situation,
indicated by a dark dashed line. (G) The PSF (left column) and corresponding OTF (right column) comparison at Dz = 0, 230, and 490 nm. Scale bar: 0.5 mm.