of 10
View
Online
Export
Citation
RESEARCH ARTICLE
|
NOVEMBER 12 2024
Harmonic imaging for nonlinear detection of acoustic
biomolecules
Rohit Nayak
;
Mengtong Duan
;
Bill Ling
;
Zhiyang Jin
;
Dina Malounda
;
Mikhail G. Shapiro
APL Bioeng.
8, 0461
10 (2024)
https://doi.org/10.1063/5.0214306
Articles Y
ou May Be Interested In
Nonlinear ultrasound imaging of nanoscale acoustic biomolecules
Appl. Phys. Lett.
(February 2017)
Ultrafast amplitude modulation for molecular and hemodynamic ultrasound imaging
Appl. Phys. Lett.
(June 2021)
Gas vesicles: Acoustic biomolecules for ultrasound imaging
J Acoust Soc Am
(May 2017)
19 December 2024 18:00:37
Harmonic imaging for nonlinear detection
of acoustic biomolecules
Cite as: APL Bioeng.
8
, 046110 (2024);
doi: 10.1063/5.0214306
Submitted: 16 April 2024
.
Accepted: 30 October 2024
.
Published Online: 12 November 2024
Rohit
Nayak,
1
Mengtong
Duan,
2
Bill
Ling,
1
Zhiyang
Jin,
3
Dina
Malounda,
1
and Mikhail G.
Shapiro
1,3,4,a)
AFFILIATIONS
1
Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, California 91125, USA
2
Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, California 91125, USA
3
Andrew and Peggy Cherng Department of Medical Engineering, California Institute of Technology, Pasadena, California 91125, USA
4
Howard Hughes Medical Institute, California Institute of Technology, Pasadena, California 91125, USA
a)
Author to whom correspondence should be addressed:
mikhail@caltech.edu
ABSTRACT
Gas vesicles (GVs) based on acoustic reporter genes have emerged as potent contrast agents for cellular and molecular ultrasound imaging.
These air-filled, genetically encoded protein nanostructures can be expressed in a variety of cell types
in vivo
to visualize cell location and
activity or injected systemically to label and monitor tissue function. Distinguishing GV signal from tissue deep inside intact organisms
requires imaging approaches such as amplitude modulation (AM) or collapse-based pulse sequences. However, these approaches have limita-
tions either in sensitivity or require the destruction of GVs, restricting the imaging of dynamic cellular processes. To address these limitations,
we developed harmonic imaging to enhance the sensitivity of nondestructive GV imaging. We hypothesized that harmonic imaging, inte-
grated with AM, could significantly elevate GV detection sensitivity by leveraging the nonlinear acoustic response of GVs. We tested this
hypothesis by imaging tissue-mimicking phantoms embedded with purified GVs, mammalian cells genetically modified to express GVs, and
mice liver
in vivo
post-systemic infusion of GVs. Our findings reveal that harmonic cross-propagating wave AM (HxAM) imaging markedly
surpasses traditional xAM in isolating GVs
nonlinear acoustic signature, demonstrating significant (p
<
0.05) enhancements in imaging per-
formance. HxAM imaging improves detection of GV producing cells up to three folds
in vitro
, enhances
in vivo
imaging performance by over
10 dB, while extending imaging depth by up to 20%. Investigation into the backscattered spectra further elucidates the advantages of har-
monic imaging. These advancements bolster ultrasound
s capability in molecular and cellular imaging, underscoring the potential of har-
monic signals to improve GV detection.
V
C
2024 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution-NonCommercial-
NoDerivs 4.0 International (CC BY-NC-ND) license (
https://creativecommons.org/licenses/by-nc-nd/4.0/
)
.
https://doi.org/10.1063/5.0214306
Ultrasound imaging plays a pivotal role in medical diagnostics,
offering high spatial and temporal resolution for examining organ
anatomy and function. The development of micro- and nanoscale con-
trast agents has extended ultrasound
s reach considerably,
1
while the
introduction of acoustic reporter genes (ARGs) encoding gas vesicle
(GV) proteins has broadened the capabilities of ultrasound imaging at
the molecular and cellular level.
2
Particularly, GVs are air-filled protein
nanostructures, and are the first genetically encodable ultrasound
reporters
3
5
encoded by gene clusters of eight or more genes, origi-
nally evolved in various aquatic bacteria and archaea to regulate buoy-
ancy for optimal sunlight exposure and nutrient uptake.
6,7
GVs are distinctive in their structure: typically

85 nm in diame-
ter and

500nm in length, they are encapsulated by a protein shell
about 3 nm thick that can endure large pressures up to hundreds of
kilopascals without collapsing.
8,9
The shell
sinteriorismarkedly
hydrophobic, preventing water ingress while permitting gas molecules
to freely diffuse in and out.
6,7
The GV shell comprises a primary struc-
tural protein, GvpA
a small (7-kDa) and amphiphilic molecule that
polymerizes to form the shell.
6
This protein assembly is further rein-
forced by a secondary protein, GvpC, which externally fortifies the
shell, enhancing the GVs
mechanical strength.
10
The low density and
high compressibility of GVs enable effective sound wave scattering,
generating substantial ultrasound backscatter.
2
This feature becomes
crucial when GVs are heterologously expressed as ARGs in genetically
engineered cells,
3,4
broadening their application from targeted cellular
imaging to real-time monitoring of biological processes.
11
15
Distinguishing GVs
backscatter from the tissue necessitates a
method that separates these signals. The distinct mechanical behavior
APL Bioeng.
8
, 046110 (2024); doi: 10.1063/5.0214306
8
, 046110-1
V
C
Author(s) 2024
APL Bioengineering
ARTICLE
pubs.aip.org/aip/apb
19 December 2024 18:00:37
of GVs, marked by reversible nonlinear buckling beyond specific pres-
sure thresholds, facilitates the use of amplitude modulation (AM) tech-
niques for isolating their signals from background tissue clutter.
16
18
AM pulse sequences typically involve three sequential transmit pulses,
with one at full amplitude and two at half amplitude. Only the full-
amplitude pulse is designed to be above the pressure threshold for GV
buckling, resulting in a nonlinear pressure response, which is minimal
in tissue. By subtracting the signal elicited by the two half-amplitude
transmissions from the full-amplitude pulse, tissue clutter is mini-
mized, whereas the nonlinear response of the GVs is amplified. This
process effectively enhances the visualization of cells expressing GVs
or systemically injected GVs.
AM based on cross-propagating waves (xAM) has proven espe-
cially effective in detecting GVs while minimizing artifacts arising
from nonlinear wave propagation through GV inclusions.
17
As an
alternative, BURST imaging provides the most sensitive detection of
GV by capturing the unique signals produced by their intentional
acoustic collapse
enhancing detectability by an order of magnitude
compared to AM5.
19
However, the applicability of BURST is limited
in contexts requiring preservation of the GVs, such as dynamic imag-
ing or biosensing. Therefore, enhancing the detection sensitivity of
AM-based imaging for GVs is a critical goal in biomolecular
ultrasound.
Harmonic imaging has been a key approach in improving the
contrast and resolution of ultrasound images. For example, tissue har-
monic imaging is routinely used in diagnostic ultrasonography for
generating images with superior tissue definition, improved signal-to-
noise ratio, and reduced artifacts produced by side lobes, grating lobes,
and reverberation.
20,21
The principle behind harmonic imaging lies in
its use of the higher-frequency harmonics generated by the nonlinear
propagation of the fundamental ultrasound wave. These harmonic
waves typically contain fewer artifacts compared to images produced
using conventional fundamental wave ultrasound. Building on this
approach, the integration of harmonic imaging with specialized pulse
sequences such as AM and pulse inversion has enhanced the detection
of ultrasound contrast agents.
22
The relatively strong harmonic signals
arising from the agents
nonlinear acoustic behavior help set them
apart from the surrounding tissue.
23
In fields such as cardiology and
hepatic imaging, harmonic imaging provides refined views of
myocardial perfusion and endocardial borders, as well as more accu-
rate lesion characterization.
20
In this study, we evaluate the potential of using harmonic signals
to enhance the detection of GVs. While existing research suggests that
GVs can produce harmonic scattering,
2,11,24
this observation has not
been integrated with AM, and previous work suggested that
fundamental-frequency imaging provided the best GV imaging perfor-
mance with conventional parabolic AM (pAM) sequences.
16
We
hypothesized that integrating harmonic imaging with xAM could sig-
nificantly improve GV detection sensitivity due to the cleaner nonlin-
ear background of xAM. To test our hypothesis, we imaged tissue-
mimicking phantoms containing purified GVs,
8
mammalian cells
genetically engineered to express GVs as acoustic reporter genes,
5
and
live mice following systemic infusion and liver uptake of GVs.
25
Furthermore, we compared the performance of xAM and its harmonic
counterpart throughout the study.
For all experiments, we employed a 128-element linear array
ultrasound probe with a nominal bandwidth of 14
22 MHz. To
integrate harmonic signals with xAM imaging, we performed tests
at a transmit frequency of 12.5 MHz, where we expected to observe
second harmonic contributions manifesting at 25 MHz (
Fig. 1
). To
evaluate the effectiveness of this approach, we compared the out-
comes with those obtained from conventional xAM imaging, which
was performed at a baseline transmit frequency of 15.625 MHz
(whose second harmonic is beyond the bandwidth of the trans-
ducer). We ensured that both the 12.5 and 15.625 MHz frequencies
were operated at the same transmit pressure of approximately
400 kPa, facilitating a direct and consistent comparison between the
two imaging modes.
We began by imaging purified
Anabaena flos-aquae
GVs,
stripped of GvpC to enable buckling, in tissue-mimicking phantoms
8
[
Fig. 2(a)
]. We conducted imaging of the samples using xAM, har-
monic xAM (HxAM) using the full received signal, and a variant of
harmonic xAM (HxAM-f) employing a high-pass receive filter set at
17.5 MHz to exclude the fundamental signal. We assessed the efficacy
of imaging techniques at varying concentrations of GVs, determined
by optical density (OD) measurements, which is based on light scatter-
ing. An OD of 1, measured at a wavelength of 500nm (OD
500
), corre-
sponds to a concentration of approximately 184 pM GV particles.
8
FIG. 1.
Cross-wave propagation and amplitude modulation in harmonic imaging. (a) Depiction of a three-pulse sequence utilized for amplitude modulation, w
here the first pulse
has twice the amplitude of the subsequent pulses to induce modulation. (b) The nonlinear scattering behavior of harmonic gas GVs when sonicated above
their buckling thresh-
old, as indicated by the converging cross-waves. (c) Comparison of the receive spectra for ultrasound transmissions at 15.625 and 12.5 MHz; the latte
r captures the harmonic
signal within the bandwidth limitations of the transducer and scanner. The dashed gray line marks the 17.5 MHz frequency threshold used for high-pass
harmonic filtering, as
reported in
Fig. 2
.
APL Bioengineering
ARTICLE
pubs.aip.org/aip/apb
APL Bioeng.
8
, 046110 (2024); doi: 10.1063/5.0214306
8
, 046110-2
V
C
Author(s) 2024
19 December 2024 18:00:37
FIG. 2.
In vitro
imaging of GVs at various concentrations in tissue-mimicking phantoms using xAM and HxAM. (a) Schematic of the ultrasound phantom with GV inclusion
(yellow) in a tissue-mimicking matrix (gray). ROIs are indicated in green for GVs, and in blue for the background. The GV inclusions are positioned at a
depth of 4.5 mm
and have a radius of 1 mm. (b) Representative xAM (top row) and HxAM (middle row) images of the same cross section at different GV concentrations. The bot
tom row
shows HxAM images high-pass filtered at a 17.5 MHz threshold to highlight harmonic contributions. All images are displayed using the same dynamic col
or range. (c)
Magnitude Fourier spectra for GV and background signals in xAM and HxAM images at various GV concentrations. (d) Quantitative analysis of GV and backg
round sig-
nals in xAM and HxAM images as related to GV concentration, including STR and CTR performance metrics. (e) Spectral power from the Fourier spectrum of G
Vand
background signals in xAM, HxAM, and HxAM-filtered images at different GV concentrations. Data from N
¼
5 samples; error bars represent the standard error of the
mean. Signal amplitude presented in decibels (dB rel. raw) are calculated based on the absolute signal recorded directly from the Verasonics scanner
without further
normalization. Statistical significance is indicated by

, and non-significance by
ns
.
APL Bioengineering
ARTICLE
pubs.aip.org/aip/apb
APL Bioeng.
8
, 046110 (2024); doi: 10.1063/5.0214306
8
, 046110-3
V
C
Author(s) 2024
19 December 2024 18:00:37
HxAM yielded considerably enhanced images compared to con-
ventional xAM [
Fig. 2(b)
]. Spectral analysis revealed a distinct second
harmonic signal at 25 MHz, which was solely attributable to the GVs
and was not present in the tissue-mimicking background [
Figs. 2(c)
and
2(d)
]. These attributes translated into a notable increase in signal to
background ratio (SBR) and contrast-to-background ratio (CBR), with
improvements of 4.7 and 5.6 dB, respectively. Furthermore, such
enhancements stemmed from an increase in the GV signal, while back-
ground levels remained the same between the two methods [
Fig. 2(e)
].
HxAM maintained a performance advantage of over 5 dB compared to
xAM regardless of averaging parameters. With just 15 averaging repeats,
HxAM surpassed 50 such repeats of xAM in imaging quality (
supple-
mentary material
Fig. 1). The resulting increase in imaging frame rate
can be especially useful for
in vivo
applications and dynamic imaging.
The filtering in HxAM-f substantially suppressed background signal,
while also reducing GV signal, resulting in overall SBR and CBR similar
to unfiltered HxAM [
Figs. 2(b)
and
2(e)
]. To confirm that the distinct har-
monic signals observed in HxAM can be attributed to the nonlinear buck-
ling behavior of GVs, we also imaged stiff-shelled GVs that do not exhibit
robust buckling behavior
11
due to the presence of GvpC and saw a lack of
second harmonic output (
supplementary material
Fig. 2).
In contrast to HxAM, harmonic imaging using the parabolic focus-
ing of pAM did not significantly increase SBR or CBR compared to fun-
damental imaging (
supplementary material
Fig. 3), consistent with our
previous report.
16
This difference arises from a lower harmonic-to-fun-
damental signal ratio (0.69, compared to 0.98 in xAM), which we attri-
bute to artificially elevated fundamental signal at full transmit amplitude
caused by cumulative nonlinear propagation.
17
In contrast, xAM imag-
ing by design has lower propagation nonlinearity, providing a higher
harmonic-to-fundamental ratio for harmonic imaging and a more faith-
ful representation of scatterer locations in the medium.
To ensure that the enhanced contrast of HxAM does not arise
from differences in focal dimensions, we mapped the beam profiles for
xAM and HxAM transmissions using an acoustic hydrophone. The ele-
vational and lateral profiles of xAM and HxAM imaging showed no
major differences. The full width half maximum (FWHM) values for
the elevational beam profile were 0.866 mm for xAM and 0.855 mm for
HxAM, while for the lateral beam profile, they were 0.806 and
0.785 mm, respectively (
supplementary material
Fig. 4). Although
HxAM has slightly more pronounced side lobes, their amplitude
remains below the GVs
buckling threshold at our transmit pressures,
minimizing the likelihood of artifact generation. The increased side lobes
observed when shifting from 15.625 to 12.5 MHz are primarily due to
the deviation from the transducer
s center frequency (18.5 MHz) and
moving outside its manufacturer-recommended bandwidth (14
22 MHz). This was necessary to demonstrate proof of concept for har-
monic imaging, as the existing bandwidth could not fully accommodate
both fundamental and harmonic frequencies. Transducers are designed
to perform optimally within their specified frequency range, and operat-
ing outside this range can degrade beam quality, leading to increased
side lobes. However, this issue can be addressed by choosing a broader
bandwidth transducer such as L10-4 or L35-16 that can accommodate
both fundamental and harmonic frequencies.
To evaluate the utility of HxAM imaging in applications of GVs as
acoustic reporter genes, we imaged MDA-MB-231 cancer cells genetically
engineered to express GVs
5
[
Fig. 3(a)
]. This cell line is commonly used
to model breast cancer
in vivo
. HxAM improved the detection of cells
across a wide range of concentrations. At the highest densities
3

10
6
and 3

10
7
cell/ml
HxAM increased SBR and CBR relative to xAM
[
Figs. 3(b)
3(d)
]. At lower densities where individual cells are expected
to be separated within the field of view, HxAM facilitated the identifica-
tion of a larger number of signal sources (putative cells) compared to
xAM
identifying 2.4 times to 3.0 times more sources [
Figs. 3(b)
and
3
(e)
]. To ensure that the contrast points contained GVs, we applied ultra-
sound pulses at a high pressure (3.6 MPa) using focused parabolic delays
exceeding the GV
s irreversible collapse threshold (

570 kPa) and docu-
mented a loss of scattering (
supplementary material
Fig. 5).
To assess the efficacy of HxAM
in vivo
, we imaged mice during
intravenous administration of purified GVs, focusing on the liver,
where GVs are taken up and degraded as a part of the organ
sphagoly-
sosomal function.
2,25
To evaluate the detection sensitivity, we injected
100
l
l solutions containing GVs at either a standard OD of 30 or a
minimally detectable concentration of OD 10 [
Fig. 4(a)
]. Furthermore,
to enable imaging of the same exact plane with HxAM and xAM and
eliminate confounds from physiological motion, we euthanized the
animals immediately before imaging. HxAM allowed GV uptake to be
imaged substantially deeper inside the liver tissue at both standard
[
Fig. 4(b)
]andlowGVconcentrations[
Fig. 4(c)
,
supplementary mate-
rial
Fig. 6(a)], extending detection depth by up to 20% [
Fig. 4(d)
]. The
GV-based source of the contrast was confirmed by collapsing the GVs
with high-pressure pulses. This improvement was not due to deeper
penetration of the transmit pulse; the expected difference in attenua-
tion between 15.625 and 12.5 MHz is only 0.78 dB over 5 mm of liver
tissue.
26
Spectral analyses confirmed the presence of harmonic signals
in HxAM [
Fig. 4(e)
,
supplementary material
Figs. 6(b) and 6(c)].
HxAM imaging improved SBR and CTR by 4.32 and 11.1 dB at OD 30
[
Fig. 4(f)
], and 4.4 and 15.8 dB at OD 10 [
Fig. 4(g)
], relative to xAM.
Taken together, the results of this study suggest that HxAM
imaging improves the detection of GVs over xAM in all the main
in vitro
and
in vivo
scenarios. Harmonic frequencies are selectively
amplified in the GV signal without altering the background, leading to
marked improvements in SBR and CBR. We anticipate that this
enhanced sensitivity will facilitate
in vivo
cell imaging, while its specif-
icity for buckling GVs over stiff GVs will contribute to dynamic bio-
sensing.
12,27
The enhanced performance of HxAM is attributed to the
capture of harmonic signals arising from GV scatterers, with minimal
background from nonlinear propagation artifacts (unlike pAM). Even
though the current experiments were performed with imaging depth
up to 8 mm using a high-frequency probe, we anticipate that the imag-
ing depth can be extended by using a lower imaging frequency trans-
ducer, such as used to image GVs in a previous study.
5
HxAM imaging inherits some limitations of xAM imaging,
including a reduced lateral and axial field of view. It is also important
to note that the frame rate of HxAM is identical to that of xAM. We
have successfully used xAM in several previous
in vivo
studies
5,12,17,25
without encountering any frame rate-related issues. However, we
acknowledge the potential impact of motion on multi-pulse techni-
ques, and in future studies, we plan to explore the extension of har-
monic imaging to ultrafast imaging sequences to further reduce any
motion-related challenges. Furthermore, HxAM imaging adds the
requirement for transducers and scanners with relatively high
transmit-receive bandwidth. However, HxAM
simprovedabilityto
detect GVs and GV-expressing cells makes it an attractive imaging
method for biomolecular and cellular ultrasound.
APL Bioengineering
ARTICLE
pubs.aip.org/aip/apb
APL Bioeng.
8
, 046110 (2024); doi: 10.1063/5.0214306
8
, 046110-4
V
C
Author(s) 2024
19 December 2024 18:00:37
METHODS
Ultrasound acquisition sequence
We use a Verasonics Vantage ultrasound scanner, equipped with
an L22-14vX probe, to execute various imaging sequences such as
xAM, pAM, and its respective harmonic versions. This probe features
a linear array composed of 128 elements, each spaced at a 0.10-mm
interval. It has an elevation focus of 8 mm, a 1.5-mm elevation aper-
ture, and operates at a central frequency of 18.5 MHz, offering a band-
width of 67% at

6 dB. For transmission, we use single-cycle
waveforms at frequencies of 15.625 and 12.5 MHz for each active ele-
ment in the array, which ensures that our base frequency is divided by
factors of 4 and 5, respectively, in sync with the system
s 62.5-MHz
sampling rate. For harmonic versions of xAM and pAM, we opt for a
FIG. 3.
Comparative ultrasound imaging
of engineered mammalian cells express-
ing acoustic reporter genes
in vitro
. (a)
Schematic representation: Mammalian
cells (in yellow) embedded within an agar
phantom (gray). (b) xAM and HxAM
Imaging: Images of mammalian cells
genetically engineered to express acoustic
reporter genes, set in agar at various cell
concentrations. All images are displayed
using the same dynamic color range. (c)
and (d) Signal Analysis: Quantitative eval-
uation of nonlinear ultrasound signals
from xAM and HxAM imaging at concen-
trations of 3

10
7
and 3

10
6
cells,
respectively. Analysis includes assess-
ment of gas vesicle (GV) signal from a
specified region of interest (ROI) shown in
(a) and background signal from the same
ROI post-GV acoustic collapse. Bar
graphs show GV signals with performance
metrics like STR and CTR. (e) and (f) Cell
count analysis: Cell counts in HxAM and
xAM images at concentrations of 3

10
5
and 3

10
4
cells, respectively, normal-
ized to HxAM counts at each concentra-
tion. Data based on N
¼
5 samples; error
bars denote standard error of the mean.
(g) Fourier spectral analysis: Magnitude
Fourier spectra derived from xAM and
HxAM imaging of a single cell, which is
indicated by a blue arrow in (b). Signal
amplitude presented in decibels (dB rel.
raw) are calculated based on the absolute
signal recorded directly from the
Verasonics scanner without further nor-
malization. Statistical significance is indi-
cated by

, and non-significance by
ns
.
APL Bioengineering
ARTICLE
pubs.aip.org/aip/apb
APL Bioeng.
8
, 046110 (2024); doi: 10.1063/5.0214306
8
, 046110-5
V
C
Author(s) 2024
19 December 2024 18:00:37
FIG. 4.
In vivo
ultrasound imaging of mice liver after intravenous injection of purified GVs. (a) Schematic illustration of the ultrasound mouse liver experiment.
(a) and (c) Representative
examples of xAM and HxAM images of mice liver, acquired at the same cross section using both imaging techniques at OD
500
30 and 10, respectively. Top and bottom rows correspond
to detection of GVs intact and after acoustic-based collapse, respectively. (d) Axial line plots corresponding to GV and background signal at varyin
g depths of images in (b). The solid
line plots and the corresponding dotted double-sided bands represent mean and the standard error, respectively, estimated across all columns of the
ultrasound images in (b). (e)
Magnitude Fourier spectra associated with GV and background signals in xAM and HxAM images reported in (b). (f) and (g) Quantitative analysis of GV and
background signal in mice
liver using xAM and HxAM imaging, at OD
500
of 30 and 10, respectively. The negative CBR arises from taking the mean of the GV signal across the entire liver in the image, except for
the superficial skin layer. N
¼
5 mice at each GV concentration, and the error bars represent standard error of the mean. Signal amplitude presented in decibels (dB rel. raw) are calcu
-
lated based on the absolute signal recorded directly from the Verasonics scanner without further normalization. Statistical significance is indic
ated by

, and non-significance by
ns
.
APL Bioengineering
ARTICLE
pubs.aip.org/aip/apb
APL Bioeng.
8
, 046110 (2024); doi: 10.1063/5.0214306
8
, 046110-6
V
C
Author(s) 2024
19 December 2024 18:00:37