Super-Resolution Microscopy Reveals Nanoscale Arrangement of PD-L1 Immune Checkpoint.

Introduction
Programmed death ligand 1 (PD-L1) is the
major ligand of programmed
death 1 (PD-1), a coinhibitory receptor expressed on the surface of
activated T cells.
−

PD-1/PD-L1 pathway acts as a “don’t
find/kill me” signal to T cells and makes them ignore cancer
cells.
,
High expression of PD-L1 in human cancers,
such as breast cancer, colorectal cancer, gastric cancer, nonsmall
cell lung cancer, and testicular cancer,
,
allows them
to evade immune surveillance and survive, thus leading to poor prognosis. The discovery of the PD-L1/PD-1 immune checkpoint
has brought a paradigm shift in cancer immunotherapy. Extensive drugs
targeting the PD-L1/PD-1 immune checkpoint, including monoclonal antibodies
and small molecule inhibitors, have been developed for cancer treatment.
−

Traditional protein characterization approaches, such as flow
cytometry, Western blot, and
immunohistochemistry, describe the expression
of PD-L1 at multicellular or tissue levels. At the single-cell level,
conventional light microscopy cannot obtain the nanoscale insights
of immune checkpoints due to its resolution limitations of 200 nm,
thus calling for in situ, specific, quantitative, and high-resolution
methods to investigate the spatial arrangement of PD-L1 at the molecular
level.
Super-resolution microscopy (SRM), especially single-molecule
localization
microscopy like stochastic optical reconstruction microscopy (STORM),
,
has greatly revolutionized our understanding of subcellular structures.
−

Membrane proteins have long been popular targets for SRMs due to
their fundamental biological function and distribution beyond the
diffraction limit. With the application of SRM, membrane proteins
were found to distribute in distinct forms on the membrane. A substantial
proportion of membrane proteins, such as the glucose transporter GLUT1, epidermal growth factor receptor (EGFR), peroxisomal proteins, and activated T cell antigen receptor, have been shown to be organized in clusters, which are considered
as signals of cellular activation. By
contrast, some membrane proteins, such as CD86 and the resting T cell antigen receptor, distribute as monomers on the membrane. Herein, based on
STORM and diversified analytical methods, we aim to quantitatively
investigate the nanoscale organization of PD-L1 in MDA-MB-231 breast
cancer cells.

Results

STORM Quantitatively Reveals the Organization of PD-L1 on the
Membrane in MDA-MB-231 Cells
We first utilized 3D-STORM to
investigate the spatial organization of PD-L1 on the membrane in MDA-MB-231
cells using a dye-tagged primary antibody targeting the extracellular
segment of PD-L1. We thus found that PD-L1 exhibited a dispersive
and extensive distribution on the membrane (Figure
a). The axial section showed a clear bilayer
structure of PD-L1 (apical and basal membranes), while the diffraction-limited
conventional epifluorescence image (Figure
b) could not resolve clear distributions.
Here, we chose the apical surface of the cell for further quantitative
analysis. By overlaying the centers of localized single PD-L1 “clusters”
in STORM images, the normalized distributions showed a full width
at half-maximum (Fwhm) of ∼25 nm in both the horizontal and
vertical directions (Figure
c), comparable to the resolution of the STORM system.

STORM data contains not only the cluster size information
but also
the blinking information, which can be used to estimate how many molecules
(protein copy number) assemble to a cluster, especially when the cluster
size goes below the STORM resolution. The higher the blinking number of a “cluster,” the
more protein copy number it contains. Therefore, we compared the blinking
numbers of the PD-L1 “cluster” on the cell surface and
the isolated single “background” antibodies on the coverslip,
which serves as a monomeric calibration reference. The histograms
of blinking counts and cluster areas per cluster (Figure
d) showed a similar distribution.
Moreover, the average counts of blinking per cluster (Figure
e) were nearly identical between
PD-L1 on the cell membrane (16.8 ± 1.2) and isolated single “background”
antibodies on the coverslip (17.0 ± 2.8). These results clearly
indicate that PD-L1 predominantly exists as individual molecules across
the membrane of MDA-MB-231 cells.
Appropriate analysis of STORM
data can help discover hidden information.
Here, we applied Voronoï polygon analysis
,
to assess the randomness of PD-L1 on the apical surface. The normalized
histogram of Voronoï polygon areas (A/<A>) was fitted
with
a γ distribution, and the peak position of the fitted curve
for PD-L1 (0.74 ± 0.02) was not significantly different from
that of simulated random points (0.73 ± 0.01) (Figure
f), indicating that PD-L1 was
randomly distributed on the membrane. By contrast, the obtained characteristic
value was 0.9 for the N-terminus of β-spectrin in expanded human
erythrocytes, which was known to distribute
uniformly as a triangular lattice across the cell (Figure S1). Furthermore, we applied Voronoï polygon
and Ripley’s K function
−

to compare the spatial distribution
of PD-L1 on the apical and basal cell surfaces. The peak positions
of the fitted Voronoï polygon area curves were both close to
0.73, and the Ripley’s K functions exhibited similar peak positions,
heights, and overall shapes (Figure S2),
quantitatively demonstrating that PD-L1 was distributed in a consistent
manner on both surfaces.

STORM Quantitatively Unveils Changes in PD-L1 Distribution upon
Drug Treatments
The expression of PD-L1 in cancer cells could
be modulated by multiple drugs as previously studied based on other
methods, such as flow cytometry, Western
blot, and immunohistochemistry at the multicellular or tissue level. Here,
with STORM SRM, we aim to quantitatively analyze variations in the
PD-L1 level on the cell surface upon drug regulation at the molecular
scale.
Interferon-γ (IFN-γ)-mediated transcriptional
up-regulation of PD-L1 has been well understood through Western blot and flow cytometry. In the present work, STORM results showed that the density of PD-L1
dramatically increased after the treatment of IFN-γ (Figures
a,b and S3a), consistent with our Western blot data (Figure
f). Meanwhile, 2-deoxyglucose
(2-DG), a glucose analog that inhibited glycolysis, was reported to
reverse immunosuppression in triple-negative breast cancer. We also found a decrease in the density of PD-L1
on the membrane in response to 2-DG (Figures
c and S3b). These
results validated the applicability of STORM in detecting variation
of PD-L1 expression at the single-cell level. Simvastatin, an oral
antihyperlipidemic drug, was recently reported to possess anticancer
activity. STORM SRM showed that simvastatin
downregulated the density of PD-L1 at the surface of MDA-MB-231 cells
(Figures
d and S3c). The statistical data of PD-L1 densities
were 9.1 ± 2.3, 15.1 ± 2.4, 5.8 ± 1.4, and 5.6 ±
1.6 clusters/μm2 for control, IFN-γ, 2-DG,
and simvastatin groups (Figure
e), respectively, showing good accordance with Western blot
results (Figure
f).
Furthermore, STORM revealed no significant difference
in the average
blinking counts among these groups, suggesting that PD-L1 maintained
a monomeric state upon drug treatment (Figure
g). The normalized distributions of PD-L1
clusters had a Fwhm of ∼25 nm in both the horizontal and vertical
directions in all groups (Figures
a–d and S3a–c). No significant differences were observed in the peak position
of Ripley’s K function between groups, whereas differences
in peak height further confirmed variations in cluster density at
similar average blinking counts per cluster (Figure S3d). These results together quantitatively demonstrated that
the three types of drugs regulated the density of PD-L1 on the membrane
without affecting its monomeric state. Together, the STORM SRM provided
valuable nanoscale information on the PD-L1 distribution, including
the density and monomeric state.

PD-L1 Has a High Lateral Mobility and Clustering Ability on
the Membrane
To characterize the lateral motility of PD-L1
on the membrane, we first performed a fluorescence recovery after
photobleaching (FRAP) experiment in living cells labeled by the dye-tagged
primary extracellular segment PD-L1 antibody. The fluorescence intensity
recovered to 60–80% of the original intensity within 20 min
after the photobleaching of the selected region with 11 μm diameter
(Figure
a,b and Movie S1), indicating a high lateral diffusivity
of PD-L1 on the membrane of MDA-MB-231 cells.
The bivalent property of the intact IgG 2nd antibody
allows it
to interact with two primary antibodies, resulting in extensive cross-linking
of the target protein (Figure
c). Therefore, we used an IgG
2nd antibody to cross-link the PD-L1 molecules labeled by dye-tagged
primary antibody in living cells. The time-lapse results showed that
after adding the 2nd antibody, the distribution of PD-L1 transformed
from uniformity to aggregated clusters within 30 min (Figure
d and Movie S2), indicating that due to its high lateral mobility, PD-L1
has a high clustering ability on the membrane of MDA-MB-231 cells.
It should be noted that “clustering” refers to the spatial
aggregation of molecules, passively driven by external factors such
as 2nd antibody cross-linking used here. This is distinct from “polymerization,”
which typically involves the active formation of stable, bonded structures
determined by internal factors such as structural domains and binding
sites. We further performed STORM imaging on cells with PD-L1 cross-linked
by the 2nd antibody to visualize and quantify the reorganization of
PD-L1 at nanometer resolution. Results (Figures
e and S4a) showed
a significant increase of the FWHM (∼50 nm) (Figures
f and S4b), the average blinking per cluster (from 16.8 ± 1.2
to 172.6 ± 38.4 in Figure
g), the average cluster areas (from 1446 ± 379 nm2 to 6448 ± 978 nm2 in Figures
h), as well as a significant reduction in
the density of PD-L1 clusters in the membrane upon cross-linking with
the 2nd IgG antibody (from 9.1 ± 2.3 clusters/μm2 to 0.7 ± 0.3 clusters/μm2 in Figure
i). The distinct differences
in blinking counts were also clearly and visually revealed through
the blinking event analysis of representative STORM data sets from
the control group (blinking counts = 17) and the 2nd antibody cross-linking
group (blinking counts = 173) (Figure
a). Furthermore, Ripley’s K function (Figure S4c) revealed that compared with the control
group (without 2nd antibody cross-linking), the Clustered group (with
2nd antibody cross-linking) exhibited a significant larger peak position,
indicating an increase in cluster size; a higher peak, reflecting
enhanced clustering, and a broader curve shape, suggesting more heterogeneous
cluster distribution.
Subsequently, based on our monomeric reference
model, we provide
the distribution of the protein oligomeric state following the methodology
previously described, which showed a
predominant monomeric state in the control group, while a broad oligomeric
state distribution in the 2nd antibody cross-linking group (Figure
b,c), confirming
the 2nd antibody-induced clustering.
Overall, STORM combining
2nd antibody-induced cross-linking and
FRAP collectively demonstrated a high lateral mobility and clustering
capability of PD-L1 on the membrane.

PD-L1 Remains Mobile on the Membrane after Paraformaldehyde
(PFA) Fixation
The noted high mobility of PD-L1 prompted
us to inquire whether PD-L1 retains its mobility after PFA fixation,
which is a widely employed fixing procedure. Previous work by Tanaka
et al. systematically investigated the
efficacy of chemical fixation for blocking the lateral diffusion of
membrane proteins and demonstrated that PFA alone often fails to completely
immobilize them. Consistent with this, our FRAP experiment showed
that PD-L1 was still partially mobile after PFA fixation (Movie S3). Further results applying whole IgG
2nd antibody to cross-link the PD-L1 molecules labeled by dye-tagged
first antibody in PFA-fixed cells revealed persistent large clusters
of PD-L1 (Figure
a,b),
confirming that relying solely on PFA fixation was insufficient to
block the mobility of membrane proteins for super-resolution imaging.
Subsequently, we endeavored to assess the fixation efficacy of various
combinations of fixatives and antibodies. We initially employed direct
immunofluorescence with dye-tagged first antibodies under paraformaldehyde
and glutaraldehyde (PFA+GA) fixation (Figure
a). This strategy was chosen to mitigate
the risk of artificial cross-linking, ultimately resulting in a realized
density of 9.7 ± 1.9 clusters/μm2. Afterward,
we performed indirect immunofluorescence with the first antibody and
whole 2nd antibody under PFA fixation, resulting in a significantly
lower density of 2.1 ± 0.4 clusters/μm2 (Figure
c), due to residual
mobility and 2nd antibody-induced cross-linking. It was known that
the monovalent fragment (Fab) 2nd antibody could avoid cross-linking.
Thus, we further tested other combinations, including Fab 2nd antibody
under PFA fixation, Fab 2nd antibody under PFA+GA fixation and whole
2nd antibody under PFA+GA fixation, with similar densities obtained
(9.1 ± 2.3, 9.3 ± 2.3, and 9.3 ± 2.6 clusters/μm2, respectively, in Figure
c), indicating that PFA+GA could totally block the
lateral moving of PD-L1. Together, these imaging results combining
STORM and 2nd antibody-induced cross-linking quantitatively demonstrated
the inadequacy of using PFA alone for the fixation of PD-L1 in super-resolution
microscopy.

PD-L1 Forms Nanoclusters at Cell–Cell Conjugate Sites
between MDA-MB-231 and Jurkat T Cells
A recent study suggested
that PD-1 and PD-L1 form microclusters at the NK cell immunological
synapse. However, the detailed arrangement
of PD-L1 during interaction with T cells, i.e., the functional form,
remains obscure. Therefore, we cocultured MDA-MB-231 cells with activated
Jurkat T cells. Two regions in the same cell were selected for further
analysis, i.e., the cell–cell conjugate sites where T cells
contacted MDA-MB-231 cells, and other regions without T cells (Figures
a and S5a). Conventional epifluorescence images showed
a higher PD-L1 fluorescence intensity at the cell–cell conjugate
sites than at other regions of the same cell (Figures
a and S5a, inset
images). STORM imaging quantitatively unveiled that PD-L1 at the cell–cell
conjugate sites formed nanoclusters, characterized by higher blinking
counts and larger areas (2745 ± 893 nm2 in cell–cell
conjugates and 1566 ± 273 nm2 in other regions) (Figures
b–g and S5a–d), rather than existing as individual
molecules. The clustering tendency of PD-L1 at the cell–cell
conjugates was also confirmed by Ripley’s K analysis, which
showed a larger peak position and peak height compared to other regions
(Figures
h and S5e). Furthermore, the ratio of average blinking
between the cell–cell conjugates and other regions of the same
cell was >1 (1.5 ± 0.4) (Figure
i), and the oligomeric state distribution
indicated
a higher proportion of oligomers at conjugate sites (Figure
j). Together, these results
demonstrate that a subset of PD-L1 molecules undergo clustering at
cell–cell conjugates, representing a potential functional state.
Such partial clustering of PD-L1 may be involved in the signal amplification
of the PD-1/PD-L1 engagement during immunosuppression.

Discussion
The total level of PD-L1 at tissue or multicellular
scales has
previously been quantified by traditional methods.
−

Yet, detailed
organization information at the molecular scale is essential for us
to understand how it acts as a “don’t eat me”
signal. The emerging SRM has provided a new opportunity to gain insight
into the spatial information on subcellular structures. Compared to
other super-resolution imaging techniques such as SIM
,
and STED,
,
STORM offers the highest resolution,
which is crucial for examining the molecular-level organization of
PD-L1. Although STORM has certain limitations in terms of imaging
speed, this study focuses on static organization rather than dynamic
changes in membrane proteins. Additionally, STORM captures the blinking
behavior of fluorophores, enabling semiquantitative analysis of protein
clustering, which is particularly advantageous for investigating membrane
protein organization. These features make STORM the most suitable
choice for this study. In the present work, we quantitatively resolved
the random distribution of PD-L1 on the membrane in MDA-MB-231 cells
using STORM SRM (Figures
a,b,f and S2). Furthermore, the
monomeric organization of PD-L1 was demonstrated through blinking
information (Figure
d,e).
Traditional methods such as Western blot and flow cytometry
could
detect the variation of PD-L1 level induced by IFN-γ
,
and 2-DG at the multicellular level.
Here, using STORM SRM, it was shown that IFN-γ upregulated the
PD-L1 density on the membrane, while 2-DG reduced it (Figure
a-c), consistent with Western
blot results (Figure
f). Particularly, our STORM data (Figure
b,e,f) demonstrated that regulation of the
PD-L1 level by IFN-γ and 2-DG resulted from the density increase
of PD-L1, rather than the clustering of it. Moreover, we determined
that simvastatin could downregulate the PD-L1 level (Figure
d,e), providing an explanation
for its anticancer activity. These results
verified the capability of STORM in providing molecular-level information,
including density and clustering state, which was inaccessible by
traditional approaches.
Membrane proteins exist in a variety
of motional states, ranging
from fully laterally mobile within the lipid bilayer to being partially
confined or fully immobilized through interactions with the cytoskeleton
or other molecular complexes, and these states are closely linked
to their functions. In the present work,
we observed a high lateral mobility of PD-L1 on the membrane by using
FRAP (Figure
a,b).
This mobility is a prerequisite for its clustering ability, which
was illustrated by 2nd antibody-induced cross-linking based on STORM
SRM (Figures
c–i
and ). A recent study suggested an unanticipated
intrinsic function of PD-L1 that it was concentrated at the rear of
migrating cancer cells and facilitated rear retraction, which also required high lateral mobility of
PD-L1 on the membrane.
The advancement of super-resolution microscopy
has greatly facilitated
the detailed elucidation of the nanoscale organization of membrane
proteins. Notably, a substantial number of these proteins have been
identified to organize into clusters, e.g., dopamine transporters, GLUT1, and CD47. Nevertheless, researchers have tended to underestimate
the importance of sample fixation in their investigations. The commonly
employed fixative, PFA, exhibited limitations in providing optimal
fixation for membrane proteins, as suggested by a single fluorescent
molecule tracking study, indicating the
potential unsuitability for its application in super-resolution imaging.
Our further results combining STORM and 2nd antibody-induced cross-linking
quantitatively showed that PD-L1 remained mobile after PFA fixation,
thus resulting in the formation of artificial clusters (Figure
). These imaging outcomes collectively
suggest the necessity for researchers to exercise caution when using
PFA as a fixation agent, particularly in membrane protein studies.
Besides, our approach, leveraging STORM combined with 2nd antibody-induced
cross-linking, presents a robust and quantitative methodology for
evaluating both the mobility of membrane proteins and the efficacy
of fixatives.
As the crucial immune checkpoint, it is of great
importance for
PD-L1 to execute rapid, sensitive, and specific engagement with its
receptor PD-1 at the surface of T cells. The clustering capability
of PD-L1, as illustrated in our coculture results (Figure
), may play functional roles
in the signal amplification of the PD-1/PD-L1 pathway for T cell immunosuppression.
Protein nanoclusters are ubiquitous in biological systems, with more aggregated proteins exhibiting stronger
signaling transduction efficiency. Additionally,
studies have shown that activated PD-1 receptors on human T cells
are also organized in microclusters. Therefore,
in the absence of T cell recognition, PD-L1 receptors are randomly
distributed. Upon contact with T cells, their rapid lateral mobility
facilitates the formation of nanoclusters, an efficient process that
may support the adaptive responses of cancer cells for immune evasion.
Notably, while our STORM data provide nanoscale insights into the
organization of PD-L1, it is important to consider the potential influence
of the labeling efficiency on the absolute quantification of protein
oligomeric states. Specifically, suboptimal labeling efficiency could
potentially lead to an underestimation of higher-order oligomers.
However, our conclusions regarding the relative changes in PD-L1 clustering
upon secondary antibody cross-linking or in response to T cell engagement
remain robust, as these comparative analyses are not compromised by
uncertainties in absolute labeling efficiency. Future exploration
employing recently developed calibration strategies, such as the use
of coexpressed reference fusion proteins,
,
will be invaluable for achieving absolute quantification and further
refining our understanding of PD-L1’s supramolecular architecture
based on STORM.
Together, using STORM SRM, we quantitatively
revealed the nanoscale
organization of PD-L1 as random monomers on the membranes of MDA-MB-231
cells. STORM can provide molecular-level information as a promising
drug evaluation tool. PD-L1 exhibited a high lateral mobility and
clustering ability of PD-L1. We further suggested the inapplicability
of PFA for membrane protein fixation. Moreover, PD-L1 aggregated into
nanoclusters in the cell–cell conjugate sites with Jurkat T
cells. These findings open new avenues for our understanding of PD-L1
organization and functions in breast cancer cells as well as provide
novel insights in drug screening for cancer immunotherapies. Additionally,
the imaging strategies and data analysis methods we developed offer
new paradigms for the quantitative study of other membrane proteins.

Materials and Methods

Cell Culture and Coculture with Jurkat T Cells
MDA-MB-231
human breast cancer cells were cultured in DMEM with 10% fetal bovine
serum (FBS) and 1% penicillin/streptomycin. The cells were grown in
a 5% CO2 atmosphere at 37 °C. Jurkat T cells were
maintained in RIPM1640 medium with 10% FBS and 1% penicillin/streptomycin
in 5% CO2 at 37 °C. For coculture, Jurkat T cells
were activated by phorbol 12-myristate 13-acetate (PMA, 50 ng/mL)
and ionomycin (1 μg/mL) in the medium for 24 h. Activated Jurkat
T cells were then collected and cocultured with MDA-MB-231 cells for
4 h before being fixed for 30 min using 3% PFA + 0.1% GA.

Sample Preparation
MDA-MB-231 cells were seeded on
12 mm glass coverslips in a 24-well plate at ∼2 × 104 cells per well and cultured for 12 h. For cell fixation,
cells were treated with 3% PFA and 0.1% GA in PBS for at least 30
min. Afterward, the cells were blocked in a blocking buffer (3% w/v
BSA in PBS) for 20 min. For direct immunofluorescence, the cells were
incubated with anti-PD-L1 primary antibody conjugated with Alexa647
(1:200, 41726S, CST, extracellular domain specific) in blocking buffer
for 1 h at room temperature, washed three times with washing buffer
(0.2% BSA in PBS), and then processed for imaging. For indirect immunofluorescence,
the cells were incubated with anti-PD-L1 primary antibody (1:200,
86744S, CST) in blocking buffer for 1 h at room temperature, washed
three times with washing buffer, and then incubated with Alexa647-conjugated
secondary antibody (1:400, A21245, Invitrogen). After three additional
washes with washing buffer and one with PBS, the cells were ready
for further imaging steps. For 2nd antibody-induced cross-linking
experiments, the cells were incubated with anti-PD-L1 primary antibody
conjugated with Alexa647 (1:200, 41726S, CST, extracellular domain
specific) for 1 h in DMEM medium. After washing with PBS, the cells
were incubated with the 2nd antibody to enable clustering for 40 min
at room temperature. Then, the cells were fixed with 3% PFA and 0.1%
GA in PBS for 30 min. Besides, for the drug treatments, the cells
were pretreated with 50 ng/mL IFN-γ for 24 h, 5 μM Simvastatin
for 24 h, or 1 mM 2-DG for 48 h before being harvested for fixation.

Time-Lapse Imaging of 2nd Antibody Cross-Linking in Living Cells
For cross-linking of PD-L1 in MDA-MB-231 cells, the living cells
were incubated with the primary antibody of PD-L1 conjugated with
Alexa647 (41726S, CST) at room temperature for 1 h. Then, cells were
placed in a CO2 incubator (INUB-ZILCSGH-F1, Tokai Hit,
Japan) mounted on an inverted optical microscope (Ti-E, Nikon, Japan)
with a sCMOS instrument (ORCA-flash4.0, Hamamatsu, Japan). As the
imaging began, the 2nd antibody of IgG (goat antirabbit, 1 μg/mL)
was added to the cell. The images were captured every 10 s. The obtained
data were analyzed by ImageJ software.

Fluorescence Recovery after Photobleaching (FRAP) Experiment
The living MDA-MB-231 cells or cells fixed with 4% PFA, or with
3% PFA plus 0.1% GA, were incubated with primary antibody of PD-L1
conjugated with Alexa647 at 37 °C for 1 h. A laser of 647 nm
was introduced into the cell for 1 min to enable photobleaching of
a selected region of the membrane. Afterward, the cell was imaged
for 30 min with an interval of 2 min. The fluorescence intensity of
the cell was calculated by ImageJ software.

STORM Super-Resolution Microscopy
Before being mounted
on rectangular glass slides, the prepared samples on the 12 mm coverslips
were first immersed in a standard STORM imaging buffer (Nano-Microimaging
Biotechnology, China). Then, the imaging process was performed on
a commercial STORM system (STORM Ultra300, Nano-Microimaging Biotechnology,
China) based on an inverted optical microscope (Ti-E, Nikon, Japan)
equipped with an EMCCD (iXon Ultra 897, Andor, U.K.), as described
in our previous work. A 647 nm excitation
laser was introduced on the sample to enable blinking of the dye.
3D-STORM was realized via a cylindrical lens in the imaging path. The images were recorded at 110 frames per 2nd,
and typically recorded ∼40,000 frames per image. 3D-STORM super-resolution
images were reconstructed from the single-molecule images using Insight3
software, as described previously.

Expansion of Human Erythrocytes
Human erythrocytes
were adhered onto a 12 mm coverslip as previously described. The erythrocyte sample was expanded using an
ultrastructure expansion process. Afterward,
the sample was labeled with anti-N-term. β-spectrin primary
antibody (Millipore, ABT185) and corresponding secondary antibody
were used for the following STORM imaging.

Blinking Analysis
For the blinking analysis of the
STORM data, we normalized the total blinking counts in a specific
region of the cell by the number of clusters identified using the
density-based spatial clustering of applications with the noise clustering
(DBSCAN) algorithm through a customized MATLAB program. For DBSCAN,
we consistently used ε = 20 as the neighborhood radius and minpts
= 5 as the minimum number of points required to form a cluster. To
include a calibration standard of monomers, the average blinking count
of isolated single “background” fluorescence-conjugated
primary antibodies on the coverslip was calculated. Besides, for the
determination of blinking counts, we configured parameters to eliminate
repeat counting according to the handbook. Specifically, when fluorescent spots appeared in successive frames
with positional displacement ≤60 nm, they were considered the
same blinking event without cumulative counting; if they appeared
in positions differing by >60 nm across consecutive frames, they
were
registered as distinct blinking events (Figure
a).

Ripley’s K Function Analysis
In order to analyze
the spatial distribution of the STORM data, Ripley’s K function,
specifically its transformed form H­(d), was applied.
,
The region of interest in individual cells was analyzed with a customized
MATLAB program. The clustering behavior was indicated by positive
values of H­(d) in the plot, with the d-value corresponding to the
maximum value of H­(d) (denoted as D
peak) representing the characteristic cluster size, typically between
the radius and diameter of the cluster.
,,,
The peak height reflects
the relative degree of clustering, which depends on both the number
of blinks per cluster and the cluster density. When the average blinking
number per cluster is comparable, a higher cluster density results
in a lower overall aggregation, thereby producing a lower normalized
peak value. The overall shape of the H­(d) curve provides further insights
into the spatial organization: a narrow and sharp peak suggests uniform,
compact clusters of similar size, whereas a broad curve indicates
clustering occurring over multiple spatial scales or more heterogeneous
cluster organization. Variations in these parameters between treatments
can thus reveal changes in PD-L1 cluster size, clustering degree,
cluster density, and spatial arrangement.

Voronoï Polygon Area Analysis
The Voronoï
diagram is constructed using MATLAB based on the center of mass of
the membrane protein clusters obtained from clustering analysis of
STORM images. The area of each Voronoï polygon is divided by
the average area of all polygons (A/<A>), and a histogram was
generated
from the normalized values. The obtained histogram was then fitted
with a γ distribution,
,
which represents the
theoretical distribution expected for a completely random Poisson
process. The x-coordinate corresponding to the peak
of the fitted curve (denoted as X
peak)
reflects the spatial distribution state of the points. For randomly
distributed ideal points, the maximum value of the γ distribution
corresponds to an x-axis coordinate of 0.72. When
the points tend to be uniformly distributed, the corresponding x-axis coordinates tend to be close to 1. The simulation
of the random distribution should prevent two molecular centers of
mass from being too close to each other due to the resolution of the
STORM images. Considering the size of the STORM image point clusters,
we set a radius of 15 nm for randomly distributed points. As a result,
the maximum value of the γ distribution of the Voronoï
polygon of randomly distributed points corrected by STROM imaging
features will be slightly higher than 0.72 in the x-axis coordinates.

Western Blot
Western blot experiments were done as
we previously described, using primary
antibody of PD-L1 (86744S, CST, extracellular domain specific) (1:200).
Mouse anti-β-actin (1:20,000, 60,008–1–lg, Proteintech,
USA) was used as a reference. The immunoblot result was visualized
in a Tanon 5200 MultiImage System.

Quantification and Statistical Analysis
All data are
presented as the mean ± the standard deviation (SD) from at least
three independent experiments. A comparison between the two groups
was performed by unpaired Student’s t-test
using GraphPad Prism 9 software. Statistical significance was defined
as *p < 0.05, **p < 0.01,
***p < 0.001, ****p < 0.0001,
and ns, not significant (p > 0.05).

Supplementary Material