Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.

v1

Article

Gold Nanoparticle Mediated Multi-Modal CT Imaging of

Hsp70 Membrane Positive Tumors

Melanie A. Kimm1, Maxim Shevtsov2,3,4, Caroline Werner2, Wolfgang Sievert2, Wu Zhiyuan2,
Oliver Schoppe2,5, Bjoern H. Menze2,5, Ernst J. Rummeny1, Roland Proksa6, Olga Bystrova4,
Marina Martynova4, Gabriele Multhoff2,§.* and Stefan Stangl2,§,*
1Department of Diagnostic and Interventional Radiology, Klinikum rechts der Isar der Technischen
Universität München, Munich, Germany; [email protected]; [email protected];
2Central Institute for Translational Cancer Research (TranslaTUM), Klinikum rechts der Isar der Technischen
Universität München, Munich, Germany; [email protected]; [email protected];
[email protected]; [email protected]; [email protected]
3Pavlov First Saint Petersburg State Medical University, St. Petersburg, Russia
4Institute of Cytology of the Russian Academy of Sciences (RAS), Russia, St. Petersburg
5Institute for Advanced Studies, Department of Informatics, Technical University of Munich, Munich,
Germany: [email protected]; [email protected];
6Philips GmbH Innovative Technologies, Research Laboratories Hamburg, Germany,
[email protected]

*Correspondence: Dr. rer. nat. Stefan Stangl and Prof. Dr. rer. nat. Gabriele Multhoff

Dr. rer. nat. Stefan Stangl

TranslaTUM, Klinikum rechts der Isar

Technische Universität München

Einsteinstr. 25

D-81675 Munich

Germany

[email protected]

Phone: +49 89 4140-6013

Fax: +49 89 4140-4299

&

Prof. Dr. rer. nat. Gabriele Multhoff

TranslaTUM, Klinikum rechts der Isar

Technische Universität München

Einsteinstr. 25

D-81675 Munich

Germany

[email protected]

Phone: +49 89 4140-4514

Fax: +49 89 4140-4299

© 2020 by the author(s). Distributed under a Creative Commons CC BY license.

Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

2 of 16

Received: date; Accepted: date; Published: date

Abstract: Imaging techniques such as computed tomographies (CT) play a major role in clinical
imaging and diagnosis of malignant lesions. In recent years, spectral CT has emerged in the field of
computed tomographies, utilizing detailed information from extracted spectral parameters of the
specimen. Metal nanoparticle platforms enable effective payload delivery for this technique. Due to
the possibility of surface modification, metal nanoparticles are predestined to facilitate molecular
tumor targeting.

In this work, we demonstrate the feasibility of anti-plasma membrane Heat shock protein 70
functionalized gold nanoparticles (AuNPs) for tumor specific multimodal imaging. Membrane-
associated Hsp70 is exclusively presented on the plasma membrane of malignant cells of multiple
tumor entities, but not on corresponding normal tissue cells, predestining this target epitope for
tumor-selective in vivo targeting.

In vitro microscopical analysis revealed the presence of cmHsp70.1-AuNP in the cytosol of tumor
cell lines, being internalized via the endosomal-lysosomal pathway. In tumor bearing mice the
biodistribution as well as the intratumorally enrichment of AuNP were examined 24h after i.v.
application, in vivo. In parallel to spectral CT analysis, histological analysis confirmed the presence
of tumor cells. In contrast to control NP, a significant enrichment of cmHsp70.1-AuNPs has been
detected selectively in tumors of different preclinical mouse models. Furthermore, the
biodistribution of AuNP, following i.v. injection, was analyzed by a machine-learning approach on
digitalized slides.

In summary, utilizing mHsp70 on tumor cells for guidance of cmHsp70.1 antibody functionalized
nanoparticles enables sufficient enrichment and uniform distribution of AuNPs in mHsp70-
expressing tumor cells, adequate for various microscopical imaging techniques and spectral-CT-
based tumor delineation, in vivo.

Keywords: gold nanoparticle, heat shock protein 70, molecular imaging, biomarker, spectral-CT, k-
edge imaging

1. Introduction

Detection of all malignant tumor cells in a patients´ body is a prerequisite for a successful treatment.
In established clinical routine, combined positron emission tomography (PET)/computed
tomography (CT) imaging is commonly used for

tumors exceeding 0.5–1 cm3. For standard clinical PET imaging, 18F-Glucose is often used as a PET
tracer. However, glucose-based PET/CT imaging faces disadvantages, such as false-positive and/or
false-negative signals, e.g. triggered by the fact that only metabolically active, but not resting cells
can be visualized, the low tumor-to-background contrast and the relatively low resolution of this
technique [1]. With improved settings, a spatial resolution of 2 mm is technically feasible, as
demonstrated in patients with prostate cancer [2].

The introduction of gold nanoparticle-based contrast agents added a new value to imaging
techniques. Functionalization of novel metal-based nanoparticles with tumor-specific antibodies [3]
and their utilization in imaging techniques such as photoacoustic signal transduction or CT combine
the advantages of molecular, tumor specific imaging with the unique attributes of Au-NPs in clinical
imaging, particularly in spectral-CT. This technology allows spectral imaging as it employs a photon-
counting detector which registers the interactions of individual photons which enables to create an
energy spectrum which can subsequently be converted into a color image. In addition, a non-
spectral attenuation image can be acquired at the same time. This allows for precise spatial
information of the spectral image with high resolution. Notably, signals delivered by gold can be
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

3 of 16

efficiently isolated. With the emergence of clinical SPECT scanners, the need of tumor-specific
contrast agents has further increased.

For in vivo application, it is essential that the applied nanoparticles are nontoxic, biodegradable or
inert and transportable in the blood and lymph system. Biocompatible camouflage of the NP surface
is a prerequisite to avoid immediate and quantitative uptake by macrophages. Small nanosized gold
particles ( 100nm) demonstrated to be beneficial for utilization in clinical applications [4-6]. Apart
from the formulation of the NPs, tumor imaging with nanoparticle-based contrast agents can be
further improved and specified by functionalization with antibodies targeting tumor-specific,
membrane-associated biomarkers. For improved signal-to-background ratio and high tumor speci-
ficity, candidate markers should be selectively expressed on tumor cells while being absent on
healthy tissue. Membrane Hsp70 has been found to fulfill these criteria exhibiting remarkable tumor
specific targeting capabilities [7,8]. In the course of cancer treatment, standard regimens, such as
radiotherapy or chemotherapy, have been shown to increase the membrane expression density of
Hsp70 on tumor cells [9], promoting membrane Hsp70 activated tumor targeting in combination with
standard therapies. Furthermore, an upregulated Hsp70 membrane density could be detected on
relapse tumors and metastases, compared to primary tumors. In multiple studies, the malignancy of
tumors is correlated with the Hsp70 status in the cytoplasm and on the plasma membrane [7,10]. For
a specific in vivo tumor targeting mediated by membrane-associated Hsp70, we developed the Hsp70
membrane-specific antibody cmHsp70.1 [8,11]. To utilize the beneficial features of targeting
membrane Hsp70 with the imaging capabilities of noble metals as a contrast agent, we developed a
gold nanoparticle formulation, functionalized with cmHsp70.1 monoclonal antibody to target
membrane-bound Hsp70 on tumor cells, in vivo. In previous studies we could demonstrate the rapid
and specific binding, uptake and internalization of cmHsp70.1-AuNPs into tumor cells in vitro,
leading to a high intracellular accumulation. Furthermore, following incubation of viable tumor cells
with cmHsp70.1-AuNP, no toxicity was observed up to a concentration of 10 µ g/ml [12].

2. Results

2.1 Functionalization of gold nanoparticles with cmHsp70.1

Figure 1. Antibody conjugation of gold nanoparticles and characterization. (A) Coupling

reaction of maleimide-activated AuNP and sulfhydryl-activated monoclonal antibodies. (B) Size
distribution of the differently functionalized AuNP, given in size by number-histograms.
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

4 of 16

For the coupling of mouse IgG1 isotype- or cmHsp70.1 antibodies to gold nanoparticles, sulfhydryl
groups were added to the N-terminal residue of cmHsp70.1mAb or mouse IgG1 control antibody
using SATA. Amine residues at the gold nanoparticles (AuNP) were coupled to maleimide and
further to a PEG spacer to facilitate steric flexibility of the antibodies (Figure 1A, top panel). A
covalent cross-link of sulfhydryl-activated antibodies and maleimide-activated gold nanoparticles
was established by coincubation at pH 6.5–7.5 (Figure 1A, bottom panel). The size of the mean
hydrodynamic diameter of unconjugated gold nanoparticles (AuNP) was determined to 45±14 nm
(Figure 1B, top panel). The size of cmHsp70.1 antibody (cmHsp70.1-AuNP)-conjugated AuNP was
54±11nm (Figure 1B, middle panel), 59±18 nm for IgG1 isotype-matched control antibody-conjugated
gold nanoparticles (IgG1-AuNP)
(Figure 1B, bottom panel). No self-
aggregation was observed in aliquotes
of the conjugated as well as the
unconjugated gold nanoparticles in
PBS at 37°C during 24 hours; After 4
weeks at 4°C, starting self-aggregation
of the particles was observed. An
exemplary size distribution histogram
is given in Figure S1.

2.2 Uptake and internalization of

functionalized AuNP in tumor cells, in
vitro

To verify the binding capacities and

the specific uptake of cmHsp70.1-
AuNP in comparison to control NPs
(AuNP, IgG1-AuNP) in vitro, we
performed binding tests on viable,
membrane Hsp70-positive tumor
cells. To determine the density of the
target antigen, the cell lines 4T1 and
CT26 were analyzed for their
membrane and cytosolic expression of
Hsp70. The Hsp70 high expressing cell
line 4T1 showed a membrane Hsp70-
positive phenotype on 67±9%,
whereas CT26 cells yielded a
positivity of 43±6% of the cells (Figure
2A). For determination of the total
Hsp70 density in the cell lines, an in-
cell ELISA technique was established Figure 2. Quantification of Hsp70 in target tumor cell
for cells grown in chamber slides. The lines. (A) Plasma-membrane bound Hsp70 of 4T1 (left)
Hsp70 mean signal intensity levels in and CT26 (right) cell lines, as determined by flow
4T1 and CT26 cell lines were 150.11*103
cytometry. (B) Quantitative staining of total Hsp70 in
± 24.92*103 a.u. and 89,39*103 ±
17.19*103 a.u., respectively (Figure 2B). 4T1 and CT26 cells (upper panel) and quantification
These data were verified by a thereof (lower panel), as determined by in cell ELISA
sandwich-ELISA of cell lysates technique. (C) Quantification of total Hsp70 on whole
derived from 10*106 cells of each cell cell lysates of 4T1 and CT26 cell lines, as determined by
line, resulting in Hsp70 contents of
Hsp70 sandwich ELISA.
4.28±1.74 ng/ml and 1.17±0.72 ng/ml
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

5 of 16

for 4T1 and CT26 cells,

respectively (Figure 2C).
Subsequently, both cell lines
were incubated with the three
different types of AuNPs for
24 hours at 37°C, to mimic the
uptake in vivo. In both cell
lines, the content of gold
nanoparticles was visualized
by brightfield and electron
microscopy (Figure 3).
Compared to blank AuNP
(Figure 3A, left panel) and
IgG1-AuNP (Figure 3A,
middle panel), which showed
minor cytosolic uptake,
cmHsp70.1-AuNP displayed
the strongest accumulation in
both, 4T1 and CT26 cells
(Figure 3A, right panel). In
Electron Microscopical
imaging, the cmHsp70.1-
AuNP have been found to
accumulate in intracellular
vesicles within the
internalizing cells of both
investigated cell lines after
24h of incubation. A
representative image of Figure 3. Uptake of AuNP in tumor cells. (A) Intracellular
cmHsp70.1-AuNPs in 4T1
cells is given in Figure 3B. accumulation of blank AuNP (left), mouse IgG1 coupled AuNP
(middle) and cmHsp70.1 functionalized AuNP (right) in 4T1
(upper panel) and CT26 (lower panel) cells. (B) TEM image of
intracellular accumulations in 4T1 cells. Scale bar, 1 µ m.

2.3 Accumulation of
functionalized AuNP in tumors, in vivo

To investigate the specificity and sensitivity of Hsp70.1-conjugated AuNP to target tumors in vivo,
syngenic tumor models in Balb/c mice were established. Animals were inoculated with orthotopic
4T1 and subcutaneous CT26 tumors, respectively. When tumor lesions reached a size of 200 mm3,
two times 2.5 mg of AuNP of each group (AuNP, IgG1-AuNP and cmHsp70.1-AuNP) were injected
i.v. consecutively at an interval
of 24 hours (Figure 4). 24 hours
after the second injection, mice
were euthanized and fixated.
After fixation, mice were
either imaged using spectral-
CT or directly applied to
histological analysis. In the
Figure 4. Timeline of the in vivo experiments. following, organs and tumors
of the animals were
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

6 of 16

histologically analyzed. In parallel, single cell suspension of tumors of both models have been
analyzed for the plasma membrane Hsp70 status. To characterize the tumor models regarding the
main features which determine the accumulation of molecular functionalized contrast agents in vivo,
tumors were histologically analyzed for the target antigen content (Hsp70) in the cytoplasm as well
as on the plasma membrane. Their vascularization status (CD31) and the presence of tumor
infiltrating macrophages (F4/80) have been investigated as well. Both tumor types, 4T1 o.t. and CT26
s.c., displayed similar vascularization and infiltration of macrophages, indicating comparable effects
on the NP input through these routes. Immunohistological Hsp70 staining revealed strong expression
in both tumor models,
featuring cytosolic as
well as nuclear Hsp70
expression. However,
4T1 tumors showed a
more patterned
architecture of the Hsp70
density. To investigate
the membrane Hsp70
status of the tumors in
vivo, a single cell
suspension of the freshly
dissected tumors was
investigated. With
76±7% and 67±13%
membrane Hsp70
positive viable tumor
cells, 4T1 tumors and
CT26 tumors,
respectively, showed
slightly higher
expression, compared to
the in vitro cultured cells.
However, the increased
width of the cytometric
data, as given in
histograms, indicates
increased heterogeneity
in the membrane Hsp70
expression pattern in in
vivo grown tumors. Figure 5. Hsp70.1 tumor characterization. Orthotopical 4T1 (upper
(Figure 5). panel) and subcutaneous CT26 (lower panel) tumors were analyzed
with regard to membrane (FACS, upper inlay) - and overall (IHC)
To investigate the
Hsp70 expression, as well as their content of vessels (IHC, CD31) and
feasibility of cmHsp70.1-
AuNPs as a contrast the infiltration of macrophages (IHC, F4/80). H&E was used as
agent, in a next step we overview stain (lower inlay). Scale bars, 200µ m (Hsp70), 100µ m
analyzed a first cohort of (CD31, MΦ).
tumor bearing mice by
spectral CT. In both tumor models, we additionally investigated the biodistribution of the differently
functionalized AuNP in tumors and organs. Spectral CT technique allows for the simultaneous
detection of AuNPs and the anatomical structures due to a combination of K-edge analysis and
conventional CT. For this pilot study, mice with CT26 tumors underwent spectral CT imaging post
mortem. Each mouse had received one type of AuNP. Interestingly, in all animals we were able to
detect AuNP in the tumors. In any observed tumor, the highest density of nanoparticles was traced
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

7 of 16

back in the tumor periphery. The

mouse treated with IgG1-AuNP i.v.
presented the lowest content of
AuNP in the tumor (3.3 µ g Au/mm3
tumor) with a low accumulation of
particles within the tumor center
(Figure 6A, B). In mice treated with
unconjugated AuNP, high amounts
of NP were mainly found at the
periphery of the tumor (4.3 µ g
Au/mm3 tumor), with considerably
less AuNP content in the tumor
center. In comparison, the highest
content of AuNP was found in mice
injected with cmHsp70.1-AuNP i.v.
(4.4 µ g Au/mm3 tumor). In this
group, the NP were located in the
tumor periphery as well as in the
tumor center (Figure 6C, D). In
spectral CT based biodistribution
analysis, the content of
Nanoparticles was also investigated
in the organs of the mice. We
detected high accumulation in the Figure 6. Tumor detection using spectral-CT. Upper Row:
spleen, and to a lesser extent in the axial view, bottom row: sagital view. AuNP amounts
liver and lung. The other organs did (mg/ml) are pseudo-coloured from black (6.5 mg/ml) over
not exhibit concentrations high
red (10 mg/ml) to white (13.5 mg/ml).
enough to generate a signal in
spectral CT measurements. The gold
signal of spectral CT measurements was further verified by histological analysis of the tumors (Figure
7). Sections were treated with silver enhancement to visualize the gold nanoparticles by light
microscopy (Figure 7A). As revealed by spectral-CT, histological analysis confirmed the varying
intratumoral distribution of the three groups of AuNP. In tumors of mice which were treated with
cmHsp70.1-AuNP, a more homologous distribution of NP was observed throughout the tumor
volume, compared to the blank AuNP or IgG1-AuNP. Notably, the enrichment of AuNP following
i.v. injection of blank AuNP was to a large extent due to introduction by F4/80 positive cells of the
monocyte / macrophage lineage, which phagocytosed the non-camouflaged nanoparticles and
subsequently located into the tumor periphery (Figure 7B, left). In contrast, next to the payload
introduction via tumor-infiltrating macrophages, cmHsp70.1-AuNP were also found in large
amounts in F4/80 negative cells of the tumor. Consequently, an extended accumulation of the
cmHsp70.1-AuNP was also detected in the tumor center region, which showed less content of AuNP
containing, recently infiltrated macrophages (Figure 7B, right). IgG1-AuNP showed the lowest
accumulation in all regions of the tumor. These NP, besides their lack of targeting capacity, exhibited
equal biocompatibility to cmHsp70.1-AuNP, due to the coating by murine antibodies of the isotype
IgG1. Consequently, this type of NP resultet in the lowest intratumoral accumulation yield.
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

8 of 16

To obtain the in vivo

biodistribution
characteristics of
cmHsp70.1-AuNP
featuring a diameter of
about 50 nm, 24 hours
following systemic
application, we
analyzed silver
enhanced sections of
tumors, liver, spleen,
kidneys, heart, lungs
and intestines of
tumor bearing Balb/c
mice (Figure 8).

For improved
accuracy of the
analysis, we used a
machine-learning
approach. The
algorithm for the
analysis of sectrions
was trained on 28
slides in total and was
applied for the
analysis of 13 slides.
In comparison to the
organs, the highest
accumulation of
cmHsp70.1-AuNP
was found in the
tumors, with 85.98*103
± 4.94*103 positive
pixel per mm2, which
equals to 2.7% positive
pixels per section
(Figure 8B). In
accordance to findings
of Yang et al [13], the
majority of the NP Figure 7. AuNP uptake and distribution in CT26 tumors. (A) Silver
accumulated in spleen enhancement of AuNP. Upper row: blank AuNP, middle row: IgG1-
(14.04*10 ± 1.37*10
3 3
AuNP, bottom row: cmHsp70.1-AuNP. Left: ROI at tumor rim area,
pixel/mm2) (2.13% right: ROI set to tumor center. Scale bars, 100 µ m. (B) double staining of
signal/section), liver
Macrophages (F4/80, brown) and silver enhancement of AuNP (black).
(13.34*103 ± 0.7*103
pixel/mm2) (0.11% Arrow: single-positive cells for AuNP. Scale bars: 5 µ m.
signal/section) and
intestine (9.48*103 ± 1.37*103 pixel/mm2) (0.34% signal/section), followed by lungs (6.28*103 ± 1.30*103
pixel/mm2) (0.09% signal/section). In muscle, as represented by heart tissue, the accumulation of
AuNP was below detection limit. Kidney was rarely affected, indicating an enterohepatic secretion
of the AuNP. Next to the analysis of overall entry of AuNP in the organs and tumors, we utilized the
machine-learning approach to analyze the size distribution of nanoparticle agglumeration within the
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

9 of 16

organs. The majority of positively stained events in liver and spleen resultet in a size range up to
10µ m. Accumulation of events in this size range might be suggested to be due to incorporation of the
AuNP by Macrophages and Kupffer cells, as indicated also in Figure 7B. The majority of AuNP in
kidney, lung and intestine yielded in agglumerates of about 1µ m in size. In the tumor, Hsp70.1-AuNP
spots of 25µ m2 were dominant followed by spots of 1µ m2 (Figure 8C). As tumor-associated
macrophages exhibit large cytoplasmatic volumes, the accumulation of AuNP above 10µ m2 suggest
the appearence within this cell type. However, aggregates of 1µ m2 size point to tumor cells with
smaller cytoplasmic space, as tumor cells. This finding is also supported by microscopical inalysis of
in vitro grown tumor cells with the majority of NP accumulating in cellular organelles of about 1µ m,
following endocytosis (Figure 3).

Figure 8. Biodistribution of cmHsp70.1-AuNP. (A) Silver enhancement staining of liver, spleen,

kidney, heart, lung and intestine. Scale bar, 100 µ m. (B) Pixel analysis of silver enhancement
stainings in tumors and organs, given in positive pixel/mm2 tissue. (C) size distribution of silver
enhanced cmHsp70.1-AuNP signals in different organs, following i.v. injection.

3. Discussion
The discovery of suitable contrast agents for the visualization of tumors is one of the most
important areas in clinical research. In tumor imaging, CT imaging is the most commonly used
technique featuring fast scanning speed with high spatial resolution of a large portion of the body.
Functional molecular imaging can be accomplished with positron-electron-tomography (PET), often
used in combination with CT. However, certain limitations affect the quality of tumor visualization.
PET imaging using 18F-FDG is highly dependent on a high tumor cell metabolism compared to the
surrounding normal tissue. Furthermore, PET tracer need to exert ionizing effects, which in turn
increases the patient's risk to accumulate DNA mutations. An approach to overcome the limiting
effects of PET/CT led to the development of spectral-CT analyzers [5,14], which use the k-edge
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

10 of 16

discrimination of elements allowing for their specific identification inside the body [15]. Another
advantage of the spectral segmentation is the possibility to utilize multiple contrast agents within one
imaging session which reduces x-ray exposition of the patient.
One further limitation is the spacial detection limit leading to a potential miss of small lesions.
Therefore, the usage of tumor-specific markers is beneficial.
Herein, the major stress-inducible member of the HSP70 family, Hsp70 (HspA1A) was used as
a target for the tumor-specific uptake of functionalized gold nanoparticles in different tumor entities.
Hsp70 is expressed on a variety of tumor entities, whereas normal cells lack an Hsp70 membrane
expression [8]. In previous studies, we observed that membrane associated Hsp70 is rapidly
internalized, and thereby mediating an efficient uptake of Hsp70 binding probes into the cytoplasm,
such as fluorescence-labeled cmHsp70.1 antibody [11] and tumor penetrating peptide (TPP) [16,17].
Since the binding epitope of cmHsp70.1 antibody is identical in mouse and human tumor cells
[8], murine CT26 colon and mammary 4T1 tumor cell lines were used in the present study.
The application of gold nanoparticles for imaging in vivo is a promising new approach in the
field of in vivo imaging [6,18,19]. The uptake of antibody-conjugated nanoparticles into tumor cells
is dependent on different factors. The expression density of receptors, which are expressed on the
target cells, the distribution of the target epitopes throughout the cells [20], the affinity of the tracer
to the membrane epitope and the speed of internalization [16,21].
In order to monitor the quality of functionalized gold nanoparticles in vitro and in vivo, different
imaging modalities were applied. Dynamic light scattering is a widely used technique to determine
the hydrodynamic diameter of AuNP in the nm–μm range.
Finally, the quantification of gold nanoparticles within tumor cells is of relevance to estimate the
amount of nanoparticles that is necessary for noninvasive in vivo imaging of tumors in mice and
humans as well as for their use as therapeutic agent. In our in vivo/ex vivo setup, we were able to
detect aggregates of gold nanoparticles in tumors and organs in perinuclear areas of around 1–2 μm
in diameter, as well as larger aggregates of around 25µ m in diameter. The here presented machine-
learning approach has proven to be beneficial for the analysis of AuNP distribution. Next to
calculating the quantity of gold signal within tumor and organs, we were able to separate the signal
into groups which allows for an easy analysis of AuNP within different cell types. This is important
for precise prediction of biodistribution and possible toxicity of AuNP.
Toxic side effects of gold nanoparticles following i.v. application was investigated in previous
studies. No negative side effects such as loss in body weight or organic dysfunctions were observed
in these experiments up to a concentration of 500 μg/mL [22]. Concordantly, in the in vivo experiments
of this study we did not observe any sign of toxicity at the injected concentration of 5 mg AuNP per
mouse. However, additional pharmacological and toxicological studies are needed to prove safety.
On the basis of the specific and quantitative uptake of cmHsp70.1-AuNP in Hsp70-positive
tumor cells and its imaging properties, our approach hints to a possible beneficial use in radiation
therapy. Numerous studies have reported on the radiation enhancing effect that gold nanoparticles
within the tumor tissue [23,24].
A promising finding was achieved for radiotherapy enhancement of gold nanoparticles in
preclinical mammary carcinoma studies. The results showed an 86% 1-year survival for an
combinatorial therapy of irradiation in presence of NPs, compared to 20% for radiotherapy alone
[25]. Probable mechanisms involved in radiosensitization are, besides changes in the cell cycle or an
elevated reactive oxygen species, the production and the release of secondary Auger electrons by
gold in very close proximity to the nucleus [24].
In summary, we demonstrate that the functionalization of gold nanoparticles with cmHsp70.1
antibody is a highly promising approach for in vivo tumor targeting. In these preclinical studies, the
accumulation of the investigated tumors was sufficient for visualization in spectral-CT allowing for
3D reconstructions and quantifications.
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

11 of 16

4. Materials and Methods

Antibody coupling of Gold Nanoparticles
Coupling of Hsp70-specific antibody (cmHsp70.1, multimmune, Germany) or an isotype-
matched control IgG1 antibody (Sigma Aldrich, St. Luis, Mo, USA) to gold nanoparticles (Nanopartz,
Loveland, CO, USA) was done as described before [12]. Briefly, PEG-amine coated spherical gold
nanoparticles of 30 nm diameter were maleimide activated (Pierce, Thermo Fischer Scientific,
Rockford, IL, USA) and incubated over night with sulfhydryl-activated antibodies. Vacant binding
sites were blocked with bovine serum albumin (BSA) (Sigma Aldrich, St. Louis, MO, USA). Antibody-
coupled gold nanoparticles or BSA-blocked control gold nanoparticles were analyzed and used for
experiments within 24 hours. Nanoparticles were analyzed (size, aggregation) by dynamic light
scattering (Zetasizer NanoS, Malvern Instruments, Malvern, UK). Measurements were done in
triplets and mean values calculated.

Characterization of AuNP
For coupling, spherical gold nanoparticles with a diameter of 30 nm were used (Nanopartz,
Loveland, CO, USA). After washing of unbound antibodies, vacant binding sites on the gold
nanoparticles were saturated by bovine serum albumin (Sigma-Aldrich, St Louis, MO, USA).
Antibody-conjugated gold nanoparticles or only bovine serum albumin-blocked negative control
AuNP were used for experiments directly after coupling. For characterization of the nanoparticles
and checking for aggregation, particles were analyzed for their size by dynamic light scattering
(Zetasizer NanoS; Malvern Instruments, Malvern, UK). Only nanoparticles with a single peak around
40 nm were used for experiments.

Cell culture
Murine colon carcinoma cell line CT26 (CT26.WT; American type culture collection (ATCC)
#CRL-2638) and the mouse mammary carcinoma cell line 4T1 (ATCC #CRL-2539) were cultured in
Roswell Park Memorial Institute 1640 medium supplemented with 10% (v/v) heat-inactivated fetal
calf serum, 2 mM l-glutamine, 1 mM sodium pyruvate and antibiotics (100 IU/mL penicillin and 100
μg/mL streptomycin). Cells were incubated at 37°C in 95% humidity and 5% (v/v) CO2 and cultivated
twice a week.

Assessment of Hsp70 content of the tumor cells

The Hsp70 membrane phenotype of the cells was assessed by flow cytometry. Single cell
suspension of tumor cell lines was incubated with fluorescein-isothiocyanate (FITC)-conjugated
cmHsp70.1 mAb for 30min on ice. As controls, conjugation of an IgG1 isotype-matched antibody was
done. After washing and adding propidium-iodide (PI) for life and dead discrimination, binding of
antibodies was measured using a FACSCalibur instrument (BD Biosciences). Data were analyzed
using CellQuest Pro software. Only PI negative, viable cells were analyzed. To determine the
membrane Hsp70 status of tumors grown in vivo, single cell suspension of freshly dissected tumors
was generated by combined chopping and trypsin treatment of the tumors. Here, anti-mouse CD45
APC antibody was added to cmHsp70.1-FITC to distinguish tumor cells from mononuclear blood
cells and infiltrated macrophages.
In cell ELISA was performed as described before [16]. Briefly, cells grown in chamber slides were
fixed with DAKO Fix & Perm kit, cellular membranes were permeabilized and incubated with
cmHsp70.1-FITC antibody. Following microscopy with comparable settings, quantification of
fluorescence signal was determined on the mean signal intensity values by ImageJ image analysis
software.
Quantification of the Hsp70 content was confirmed by an Hsp70 sandwich ELISA as described
elsewhere [26]. Shortly, after determination of the cell count, cells were lysed by incubation in Tris-
HCl based buffer containing 1% Triton-X100 and SDS. After centrifugation, the Hsp70 concentration
in the supernatant was measured by total Hsp70 ELISA kit (R&D systems, USA) as described by the
manufacturer. Each supernatant sample was measured in duplicates.
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

12 of 16

Animals
All animal procedures and their care were conducted in conformity with national and
international guidelines (EU 2010/63) with approval from the local authorities of the Government of
Upper Bavaria and St. Petersburg, Russia and supervised by respective Animal Care and Use
Committees. Animals were housed in standard animal rooms in individually ventilated cage systems
(IVS Techniplast) under specific pathogen-free conditions with free access to water and standard
laboratory chow ad libitum. In total, 6 female Balb/c mice (aged 10-14 weeks, Charles River
Laboratories, Sulzfeld, Germany) were used. Induction of subcutaneous tumors was done under
inhalation anesthesia (1.8% isoflurane with medical O2) by injection of 5*105 CT26 cells
subcutaneously in the neck area, or 5*105 4T1 cells orthotopically in the 4th mammary fat pad. When
tumors reached a size of 200 mm3, gold nanoparticles were intravenously injected using standard
procedures. In total, 5 mg unconjugated, IgG1- or cmHsp70.1-conjugated gold nanoparticles,
suspended in phosphate buffered saline, were injected in a consecutive pattern of two times 2.5 mg
per day within 24h. Another 24 hours after the second application, mice were euthanized under deep
anesthesia, fixed in 4% neutral-buffered formalin for 5 days and stored in 70% ethanol for further
analysis.

Bright field Microscopy

For light microscopy, cells were grown in 8-well chamber slides (NUNC-Nalgene; Thermo
Fisher Scientific) at a concentration of 10,000 per well. Upon adherence, cells were incubated with
gold nanoparticles at a non-toxic concentration of 1 μg/mL. Cellular uptake of AuNP or quantum
dots of the same size were analyzed with a Zeiss Observer Z1.

Transmission Electron Microscopy

Cells were co-incubated with functionalized and non-conjugated nanoparticles (at a
concentration 100 ug/ml) for 24 hours. Following incubation cells were washed with PBS, fixed for 1
hour at 4 C in 2.5% glutaraldehyde in 0.1 M cacodylate buffer (pH=7.4), postfixed in 1% aqueous
OsO4 (for 1 hour), dehydrated, and embedded in Araldite-Epon mixture. Sections were assessed
employing Zeiss Libra 120 electron microscope (Carl Zeiss, Germany).

Histology and Immunohistochemistry

Tumors and organs were either dissected following whole-body fixation of mice, or fixed in 3.7%
neutral-buffered formaldehyde and embedded in paraffin. 2µ m sections of the organs and tumors
were prepared, and the morphology visualized by standard H&E staining. For
immunohistochemistry the activity of the endogenous peroxidase was blocked with 1% hydrogen
peroxide and 0.1% sodium azide. After antigen retrieval in citric acid buffer (pH 6) at 100°C, sections
were incubated with anti Hsp70 antibody cmHsp70.1, followed by HRP-labelled secondary rabbit
anti-mouse antibody (Dako, Germany). Diaminobenzidine (Dako) was used as a chromogen. Sections
were counterstained with 1% Mayer’s hematoxylin. To visualize the gold-nanoparticles by light
microscopy, silver-enhancement staining was used according to the manufacturers´protocol (Sigma-
Aldrich, Darmstadt, Germany), followed by counterstaining of the nuclei with 0.1% Nuclear Fast Red
solution. Slides were digitalized with a digital slide scanner (AT2, Leica, Germany).

Spectral-CT imaging and image acquisition

Spectral CT images were acquired at a Philips spectral CT scanner, as described before [14].
Briefly, a preclinical spectral photon-counting CT system (Philips Healthcare, Haifa, Israel) was used
to obtain axial scans over 360° at a beam voltage of 100kVp. For optimal discrimination of the signals
of AuNP, a threshold was set at the k-edge energy of gold.

Spectral CT image-based Gold quantification

Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

13 of 16

Osirix® MD v10.0.5 software (Pixmeo SARL, Bern, Switzerland) was used for analysis. Mean
background was measured from several ROIs in the spleen. Gold amounts were calculated over the
whole tumor volume (µ g gold/mm3).

Deep learning-based quantification of AuNP histology

The bio-distribution of silver-stained AuNP in histological slides was assessed with the help of
a deep neural network. Regions of interest in each slide were defined via manual delineation. Slides
were subdivided into smaller, slightly overlapping patches of 1,000 x 1,000 pixels to meet memory
constraints. In a first preliminary step, AuNP concentrations were segmented with the help of color-
channel-specific dynamic thresholding and morphological operations (opening and closing with
fixed kernels). Parameters were manually tuned to optimize results for the majority of slides. In a
second step, the best resulting binary masks were manually selected and partially manually corrected.
In a third step, a deep neural network was trained to segment AuNP concentrations on this basis. In
a final step, the trained network was used to derive segmentations of AuNP concentrations for all
slide patches. A k-fold rotation of data splits was applied so that the network yields segmentations
on data samples that were not used for training, ensuring that the segmentation represents the result
of the learned task rather than replicating
the threshold-derived training data. This procedure yielded substantially more accurate and
consistent segmentations of AuNP concentrations as compared to the preliminary thresholding
procedure. Subsequent re-concatenation of slightly overlapping slide patches ruled out boundary
effects and double-counting. In a post-processing step, all individual AuNP concentrations were
identified via connected-component analysis, allowing to assess their individual sizes.
The deep neural network follows a U-net like architecture and consists of 4 levels of en- and
decoding units. Each layer of encoding units doubles the number of feature channels, starting from
16 and ending at 265 at the deepest level. The network was trained for 10 epochs on a training set of
ca. 5,000 patches. Details of architecture, implementation, and training procedure follow a previously
described protocol [27].

5. Conclusions
Herein, we show that the visualization of tumor cells with gold nanoparticles by addressing
membrane Hsp70 is feasible. We present data showing a superior uptake of cmHsp70.1 antibody-
conjugated gold nanoparticles in Hsp70 membrane-positive tumor cells. Inside the tumor cells, these
particles accumulated in the perinuclear region within 24 hours. The Hsp70 specificity was shown
since unconjugated nanoparticles and nanoparticles conjugated with an irrelevant control antibody
were not taken up into Hsp70 membrane-positive tumor cells. Furthermore, Hsp70 knockout tumor
cells that do not express Hsp70 in the cytosol and on the plasmamembrane showed no uptake of the
cmHsp70.1-conjugated nanoparticles. Quantification of the internalized cmHsp70.1-conjugated gold
nanoparticles reveals a high sensitivity for the detection of single cells. Experiments are ongoing to
study the capability of cmHsp70.1 antibody-conjugated gold nanoparticles for spectral CT imaging
of further Hsp70 membrane-positive tumor models and whether these nanoparticles can be used for
therapeutic approaches. In the future, these antibody-conjugated gold nanoparticles may be useful
for spectral computed tomography of tumors for diagnostic purposes and radiotherapeutic
interventions.

Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Figure S1:
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

14 of 16

Figure S1. Size distribution of cmHsp70.1-

AuNP after 2 weeks at 4°C. Size distribution
is given in size by number histogram.

Author Contributions:

S. Stangl, M.A. Kimm and G. Multhoff designed the study. S. Stangl, M.A. Kimm, R. Proksa, M. Shevtsov, C.
Werner, W. Sievert, Wu Zhiyuan, Olga Bystrova and Marina Martynova were responsible for data acquisition.
Analysis and interpretation of data was performed by S. Stangl, M.A. Kimm, O. Schoppe, M. Shevtsov, C. Werner,
W. Sievert, B. Menze and G. Multhoff. All authors were involved in drafting the article or revising it critically
for important intellectual content and all authors approved the final version to be published.

Funding:

The study was supported by the Russian Foundation for Basic Research (RFBR) according to the research project
№20-38-70039 and DFG grant (SFB824/3, STA1520/1-1; ME 3718/5-1), Technische Universität München (TUM)
within the DFG funding program Open Access Publishing. Additional financial support was provided by
TaGoNaX DFG project №336532926.

Conflicts of Interest:

The authors declare no conflict of interest.

References

1. Vanderstraeten, B.; Duthoy, W.; De Gersem, W.; De Neve, W.; Thierens, H. [18F]fluoro-deoxy-glucose
positron emission tomography ([18F]FDG-PET) voxel intensity-based intensity-modulated radiation
therapy (IMRT) for head and neck cancer. Radiotherapy and oncology : journal of the European Society for
Therapeutic Radiology and Oncology 2006, 79, 249-258, doi:10.1016/j.radonc.2006.03.003.
2. Bal, H.; Guerin, L.; Casey, M.E.; Conti, M.; Eriksson, L.; Michel, C.; Fanti, S.; Pettinato, C.; Adler, S.;
Choyke, P. Improving PET spatial resolution and detectability for prostate cancer imaging. Physics in
medicine and biology 2014, 59, 4411-4426, doi:10.1088/0031-9155/59/15/4411.
3. Cormode, D.P.; Naha, P.C.; Fayad, Z.A. Nanoparticle contrast agents for computed tomography: a
focus on micelles. Contrast media & molecular imaging 2014, 9, 37-52, doi:10.1002/cmmi.1551.
4. Chen, W.H.; Chen, J.X.; Cheng, H.; Chen, C.S.; Yang, J.; Xu, X.D.; Wang, Y.; Zhuo, R.X.; Zhang, X.Z. A
new anti-cancer strategy of damaging mitochondria by pro-apoptotic peptide functionalized gold
nanoparticles. Chemical communications 2013, 49, 6403-6405, doi:10.1039/c3cc43283a.
5. Cormode, D.P.; Roessl, E.; Thran, A.; Skajaa, T.; Gordon, R.E.; Schlomka, J.P.; Fuster, V.; Fisher, E.A.;
Mulder, W.J.; Proksa, R., et al. Atherosclerotic plaque composition: analysis with multicolor CT and
targeted gold nanoparticles. Radiology 2010, 256, 774-782, doi:10.1148/radiol.10092473.
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

15 of 16

6. Shevtsov, M.; Zhou, Y.; Khachatryan, W.; Multhoff, G.; Gao, H. Recent Advances in Gold
Nanoformulations for Cancer Therapy. Curr Drug Metab 2018, 19, 768-780,
doi:10.2174/1389200219666180611080736.
7. Sherman, M.; Multhoff, G. Heat shock proteins in cancer. Annals of the New York Academy of Sciences
2007, 1113, 192-201, doi:10.1196/annals.1391.030.
8. Stangl, S.; Gehrmann, M.; Riegger, J.; Kuhs, K.; Riederer, I.; Sievert, W.; Hube, K.; Mocikat, R.; Dressel,
R.; Kremmer, E., et al. Targeting membrane heat-shock protein 70 (Hsp70) on tumors by cmHsp70.1
antibody. Proceedings of the National Academy of Sciences of the United States of America 2011, 108, 733-738,
doi:10.1073/pnas.1016065108.
9. Gehrmann, M.; Radons, J.; Molls, M.; Multhoff, G. The therapeutic implications of clinically applied
modifiers of heat shock protein 70 (Hsp70) expression by tumor cells. Cell stress & chaperones 2008, 13,
1-10, doi:10.1007/s12192-007-0006-0.
10. Stangl, S.; Tontcheva, N.; Sievert, W.; Shevtsov, M.; Niu, M.; Schmid, T.E.; Pigorsch, S.; Combs, S.E.;
Haller, B.; Balermpas, P., et al. Heat shock protein 70 and tumor-infiltrating NK cells as prognostic
indicators for patients with squamous cell carcinoma of the head and neck after radiochemotherapy: A
multicentre retrospective study of the German Cancer Consortium Radiation Oncology Group (DKTK-
ROG). International journal of cancer 2018, 142, 1911-1925, doi:10.1002/ijc.31213.
11. Stangl, S.; Gehrmann, M.; Dressel, R.; Alves, F.; Dullin, C.; Themelis, G.; Ntziachristos, V.; Staeblein, E.;
Walch, A.; Winkelmann, I., et al. In vivo imaging of CT26 mouse tumours by using cmHsp70.1
monoclonal antibody. J Cell Mol Med 2011, 15, 874-887, doi:10.1111/j.1582-4934.2010.01067.x.
12. Gehrmann, M.K.; Kimm, M.A.; Stangl, S.; Schmid, T.E.; Noel, P.B.; Rummeny, E.J.; Multhoff, G. Imaging
of Hsp70-positive tumors with cmHsp70.1 antibody-conjugated gold nanoparticles. International journal
of nanomedicine 2015, 10, 5687-5700, doi:10.2147/IJN.S87174.
13. Yang, L.; Kuang, H.; Zhang, W.; Aguilar, Z.P.; Wei, H.; Xu, H. Comparisons of the biodistribution and
toxicological examinations after repeated intravenous administration of silver and gold nanoparticles
in mice. Scientific reports 2017, 7, 3303, doi:10.1038/s41598-017-03015-1.
14. Schlomka, J.P.; Roessl, E.; Dorscheid, R.; Dill, S.; Martens, G.; Istel, T.; Baumer, C.; Herrmann, C.;
Steadman, R.; Zeitler, G., et al. Experimental feasibility of multi-energy photon-counting K-edge
imaging in pre-clinical computed tomography. Physics in medicine and biology 2008, 53, 4031-4047,
doi:10.1088/0031-9155/53/15/002.
15. Ashton, J.R.; Clark, D.P.; Moding, E.J.; Ghaghada, K.; Kirsch, D.G.; West, J.L.; Badea, C.T. Dual-energy
micro-CT functional imaging of primary lung cancer in mice using gold and iodine nanoparticle
contrast agents: a validation study. PLoS One 2014, 9, e88129, doi:10.1371/journal.pone.0088129.
16. Stangl, S.; Tei, L.; De Rose, F.; Reder, S.; Martinelli, J.; Sievert, W.; Shevtsov, M.; Ollinger, R.; Rad, R.;
Schwaiger, M., et al. Preclinical Evaluation of the Hsp70 Peptide Tracer TPP-PEG24-DFO[(89)Zr] for
Tumor-Specific PET/CT Imaging. Cancer Res 2018, 78, 6268-6281, doi:10.1158/0008-5472.CAN-18-0707.
17. Stangl, S.; Varga, J.; Freysoldt, B.; Trajkovic-Arsic, M.; Siveke, J.T.; Greten, F.R.; Ntziachristos, V.;
Multhoff, G. Selective in vivo imaging of syngeneic, spontaneous, and xenograft tumors using a novel
tumor cell-specific hsp70 peptide-based probe. Cancer Res 2014, 74, 6903-6912, doi:10.1158/0008-
5472.CAN-14-0413.
18. Cormode, D.P.; Si-Mohamed, S.; Bar-Ness, D.; Sigovan, M.; Naha, P.C.; Balegamire, J.; Lavenne, F.;
Coulon, P.; Roessl, E.; Bartels, M., et al. Multicolor spectral photon-counting computed tomography: in
Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 17 April 2020 doi:10.20944/preprints202004.0297.v1

16 of 16

vivo dual contrast imaging with a high count rate scanner. Scientific reports 2017, 7, 4784,
doi:10.1038/s41598-017-04659-9.
19. Si-Mohamed, S.; Cormode, D.P.; Bar-Ness, D.; Sigovan, M.; Naha, P.C.; Langlois, J.B.; Chalabreysse, L.;
Coulon, P.; Blevis, I.; Roessl, E., et al. Evaluation of spectral photon counting computed tomography K-
edge imaging for determination of gold nanoparticle biodistribution in vivo. Nanoscale 2017, 9, 18246-
18257, doi:10.1039/c7nr01153a.
20. Schubertova, V.; Martinez-Veracoechea, F.J.; Vacha, R. Influence of ligand distribution on uptake
efficiency. Soft Matter 2015, 11, 2726-2730, doi:10.1039/c4sm02815e.
21. Lammers, T.; Kiessling, F.; Hennink, W.E.; Storm, G. Drug targeting to tumors: principles, pitfalls and
(pre-) clinical progress. Journal of controlled release : official journal of the Controlled Release Society 2012,
161, 175-187, doi:10.1016/j.jconrel.2011.09.063.
22. Zhang, X.D.; Wu, H.Y.; Wu, D.; Wang, Y.Y.; Chang, J.H.; Zhai, Z.B.; Meng, A.M.; Liu, P.X.; Zhang, L.A.;
Fan, F.Y. Toxicologic effects of gold nanoparticles in vivo by different administration routes.
International journal of nanomedicine 2010, 5, 771-781, doi:10.2147/IJN.S8428.
23. Dorsey, J.F.; Sun, L.; Joh, D.Y.; Witztum, A.; Kao, G.D.; Alonso-Basanta, M.; Avery, S.; Hahn, S.M.; Al
Zaki, A.; Tsourkas, A. Gold nanoparticles in radiation research: potential applications for imaging and
radiosensitization. Transl Cancer Res 2013, 2, 280-291, doi:10.3978/j.issn.2218-676X.2013.08.09.
24. Muddineti, O.S.; Ghosh, B.; Biswas, S. Current trends in using polymer coated gold nanoparticles for
cancer therapy. Int J Pharm 2015, 484, 252-267, doi:10.1016/j.ijpharm.2015.02.038.
25. Hainfeld, J.F.; Dilmanian, F.A.; Zhong, Z.; Slatkin, D.N.; Kalef-Ezra, J.A.; Smilowitz, H.M. Gold
nanoparticles enhance the radiation therapy of a murine squamous cell carcinoma. Physics in medicine
and biology 2010, 55, 3045-3059, doi:10.1088/0031-9155/55/11/004.
26. Rothammer, A.; Sage, E.K.; Werner, C.; Combs, S.E.; Multhoff, G. Increased heat shock protein 70
(Hsp70) serum levels and low NK cell counts after radiotherapy - potential markers for predicting
breast cancer recurrence? Radiat Oncol 2019, 14, 78, doi:10.1186/s13014-019-1286-0.
27. Pan, C.; Schoppe, O.; Parra-Damas, A.; Cai, R.; Todorov, M.I.; Gondi, G.; von Neubeck, B.; Bogurcu-
Seidel, N.; Seidel, S.; Sleiman, K., et al. Deep Learning Reveals Cancer Metastasis and Therapeutic
Antibody Targeting in the Entire Body. Cell 2019, 179, 1661-1676 e1619, doi:10.1016/j.cell.2019.11.013.