Multiplexed 129Xe HyperCEST MRI Detection of Genetically Reconstituted Bacterial Protein Nanoparticles in Human Cancer Cells

Gas vesicle nanoparticles (GVs) are gas-containing protein assemblies expressed in bacteria and archaea. Recently, GVs have gained considerable attention for biotechnological applications as genetically encodable contrast agents for MRI and ultrasonography. However, at present, the practical use of GVs is hampered by a lack of robust methodology for their induction into mammalian cells. Here, we demonstrate the genetic reconstitution of protein nanoparticles with characteristic bicone structures similar to natural GVs in a human breast cancer cell line KPL-4 and genetic control of their size and shape through expression of reduced sets of humanized gas vesicle genes cloned into Tol2 transposon vectors, referencing the natural gas vesicle gene clusters of the cyanobacteria planktothrix rubescens/agardhii. We then report the utility of these nanoparticles as multiplexed, sensitive, and genetically encoded contrast agents for hyperpolarized xenon chemical exchange saturation transfer (HyperCEST) MRI.


Introduction
Gas vesicle nanoparticles (GVs) are gas-containing spindle-(or bicone-) shaped protein nanostructures with dimensions ranging from tens to hundreds of nm that are expressed in the cyanobacteria, algae, and Gram-positive bacteria. Since only gas molecules are permeable to the GV protein shell, GV-containing organisms acquire buoyancy by selective permeation of various ambient gases in the GV interior, which facilitates optimal supply of light and nutrition [1][2][3]. While GVs have been studied in the field of microbiology for several decades, they have gained significant attention in recent years for their potential use for antigenic peptide display in vaccination [4] and as genetically encodable contrast agents for ultrasonography [5][6][7] and HyperCEST MRI [8,9]. Among these applications, HyperCEST MRI, which utilizes laser-polarized xenon-129 (HPXe) and chemical exchange saturation transfer (CEST) [10] to yield unprecedented enhanced MRI detection sensitivity, is of particular interest as GVs hold the potential to enable functional molecular imaging that is unfeasible in conventional thermally polarized proton MRI. (In principle, GV concentrations of as low as pM-nM can be detected). In this context, GVs can be thought of as an MRI analog of the green fluorescent protein (GFP) optical imaging reporter, with a genetically encodable nature and multiplexing capability facilitated by ready modulation of their size and shape, similar to the multicolor variant of GFPs [8]. However, despite their attractive features for imaging cellular/molecular processes, the practical in vivo use of GVs as genetically encoded contrast agents is at present hampered by a lack of robust techniques to introduce GVs into mammalian cells, which has been considered challenging due to the complexity of GV gene clusters [11].
GVs are composed of multiple proteins, and the number of genes responsible for GV expression is usually [8][9][10][11][12][13][14] (typically denoted GvpA, B, C, etc.). Among these genes, the principal component proteins are the hydrophobic major protein GvpA and hydrophilic minor protein GvpC; the roles of other accessory GV genes in constituting GV wall structure remain a subject of controversy [3]. In order to optimize GV delivery in vivo, understanding of the minimal number and type of GV gene components required to reconstitute the GV nanostructure in mammalian cells is crucial. In this study, we focus on the unique gene clusters of GVs derived from Planktothrix rubescens/agardhii (praGV). e praGV gene clusters have been studied extensively by Walsby and coworkers [12][13][14], who showed that parts of praGV gene clusters are composed of gvpA and three variants of gvpC named gvpC16, gvpC20, and gvpC28, although it is known that existence of other accessary GV genes have also been demonstrated [3]. Properties of praGVs such as size, shape, and stiffness (pressure threshold for collapse) are known to be modulated depending on the combination of gvpC variants included in their constituent gene clusters [13]. us, we hypothesized that combinatorial expression of such reduced sets of genes in mammalian cells would allow reconstitution of protein nanoparticles with similar properties to GVs in natural organisms which can be functionalized as a contrast agent for HyperCEST MRI in mammalian cells and genetic control of their size and shape.

Synthesis of Humanized GV Genes.
GV genes were searched in Genbank over as many strains as possible derived from Planktothrix rubescens/agardhii to synthesize humanized genes for mammalian expression of GV proteins. e gene gvpA was chosen from the strain pla-9303, gvpC16 from pla-9401, gvpC20 from pla-9401 of Planktothrix rubescens, and gvpC28 from Planktothrix agardhii CYA29. Coding sequences of these genes with codons optimized for expression in mammalian hosts were synthesized (outsourced to Genscript).

Molecular
Cloning. Primers used for gene cloning were purchased from Hokkaido System Science. Coding sequences of GV genes were PCR-amplified with 5′ primers encoding a NheI site and 3′ primers encoding an EcoRI site without termination codons using KOD-plus-Neo (TOYOBO). T2A-fluorescent protein (mKate2, mKO2, and EGFP) fusion genes were also PCR amplified with 5′ primers encoding an EcoRI site and 3′ primers encoding a NotI site. A Tol2 cloning vector (donated by Dr. Akira Takai of RIKEN and described in detail previously [15][16][17]) was also digested in the same way. e PCR products and restriction enzyme digestions were purified by agarose gel electrophoresis followed by processing with the Wizard SV Gel and PCR cleanup system (Promega). Restriction enzymes were purchased from Fermentas. e digested PCR products of GV genes and T2A-fluorescent protein fusion genes were ligated to the vectors using Ligation high Ver.2 ligase (TOYOBO) following the recommended protocol of the manufacturer. Plasmids were prepared from the bacterial liquid culture by using the Pure Yield Plasmid Miniprep System (Promega) and PureLink HiPure Plasmid Midiprep Kit (Invitrogen). e DNA sequences were read by dye terminator cycle sequencing using the BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems).

Transfection and Cell
Culture. Plasmid DNAs containing Tol2 vectors comprising GV genes and fluorescent protein genes described above and transposase mRNA were introduced to KPL-4 cells by lipofection using FuGene HD (Promega) in OptiMEM (Gibco), according to manufacturer protocols. Transfected cells were maintained in a low-glucose DMEM (FUJIFILM Wako Pure Chemical Corporation) with 10% FBS (Gibco) and 1% penicillin-streptomycine (Gibco).

KPL-4 Monoclonal Cell Lines Expressing GV Gene Sets.
KPL-4 cells were transfected with the GV gene constructs, and 1 µg/ml puromycin (Invitrogen) was added to cell cultures to select only the transfected cells. e cells were sorted several times to collect polyclonal cells expressing GV gene sets at high levels (>×10 2 fluorescence intensity compared to control cells) by detecting fluorescence of mKate2, mKO2, and EGFP in correlation with the expression levels of GV genes using a BD FACS Aria III cell sorter (BD Bioscience). Spectral overlaps of each fluorescence were calibrated using the cells expressing each single fluorescent protein gene. Among the polyclonal cells with high gene expression level, single colonies were picked and expanded to establish the monoclonal GV cell lines expressing the chosen GV gene set at a high level. Expression levels of each monoclonal cell line were checked by fluorescence intensity profiles (Supplementary Figure 1).

Cell Proliferation Assay.
Cell proliferations of each monoclonal GV cells and control KPL-4 cells with or without 1 μg/ml Dox were probed using a Cell Counting Kit-8 (WST-8 reagent) (Dojindo Molecular Technologies) [18]. For this reagent, relative absorbance at 450 nm is proportional to the number of viable cells. ∼5000 cells in 100 μl medium described above were prepared by a Countess II cell counter (Invitrogen) and cultured in each well of a 96-well plate (IWAKI) with 10 μl addition of 10 μg/ml Dox or PBS for each GV cell culture and control KPL-4 cell culture. After 72 hours of incubation, 10 μl WST-8 reagent was added to each well and cultures were further incubated for an hour.
en, absorbance of each well at 450 nm was measured by using a microplate reader ( ermo Fischer Scientific). 8 wells were used for each cell culture condition. Means and SEMs of the absorbances of the 8 wells for each condition were calculated. Absorbance was normalized to that of the mean of the "Dox-" condition data for each GV cell culture and control cell culture.

Purification of GVLPs from Mammalian Cells.
Each monoclonal GV cell line was cultured in a 15 cm tissue culture dish (IWAKI) with 30 ml DMEM until it reached 100% confluent growth. Cells were washed with PBS, peeled with 0.25% trypsin-EDTA (Sigma-Aldrich), and collected in 50 ml sample tubes. Cells were centrifuged, collected, and resuspended in PBS. e samples were transferred to a new 15 ml sample tube and centrifuged again, and cell pellets were collected. e cell pellets were suspended in 5 ml pure water at 4°C with 0.4% TritonX-100 (Sigma-Aldrich), 2 μl DNAaseI (Sigma-Aldrich), and a protease inhibitor cocktail (Roche). e samples were incubated at 4°C with gentle mixing for two hours. e same amount of 1.3 M sucrose solution (5 ml) was added to the sample (thus, the final sucrose concentration was 0.65 M), which was then incubated for a further two hours with gentle mixing at 4°C. 0.5 ml PBS was then carefully placed on the surface of the sample. e samples were then centrifuged for over 4 hours at 100 ×g at 4°C with a swing bucket rotary and appropriate balance to keep the rotating samples perpendicular to the gravity axis. After the centrifugation, ∼200 µl samples were collected from the sample surface and the GVLP suspension was used for subsequent analysis.

Transmission Electron Microscopy (TEM).
Purified GVLP samples were prepared for TEM analysis in two ways: (1) Purified GVLP from GV_AC28 cells was fixed with the same amount of 2.5% glutaraldehyde (GA) solution (FUJIFILM Wako Pure Chemical Corporation) and prepared with a standard negative staining protocol as follows. Purified suspension of 3 μl was placed onto the 200 mesh copper grid and incubated for 60 seconds. e extra solution was absorbed with a filter paper, and a water droplet of 3 μl was placed on the grid. e droplet was absorbed again immediately with a filter paper and a 2% uranium acetate droplet of 3 μl was placed on the grid and absorbed 3 times. e dried sample was then used for TEM JEM1011 (JEOL) at an acceleration voltage of 100 kV. (2) GV_AC20 cells and purified GVLPs from GV_AC20 cells and GV_AC16C20 cells were first embedded in 2% agarose gels (∼1 mm 3 ) prior to fixation. After the fixation with half Karnovsky mixture (2.5% GA and 2% paraformaldehyde), the samples were dehydrated by soaking them with 30%, 50%, 70%, 80%, and 90% sequences for 10 minutes each. After that, samples were soaked three times with 100% ethanol. Samples were then soaked with 100% propylene oxide (PO) for 10 minutes and in 1 : 1 and 1 : 2 mixture of PO solvent and Quetol-812 (Nisshin EM) epoxy resin (EPON) for 6 hours and 2 hours, respectively. Furthermore, the samples were soaked in a mixture of EPON and epoxy curing agent DMP-30 for 6 hours. Finally, the samples were embedded in the same EPON and DMP-30 mixture formed with a plastic or silicon mold and cured in the oven at 45°C for 24 hours and at 60°C for 72 hours, respectively. e obtained resin block was trimmed with a glass knife, and ultrathin sections (∼70 nm) were prepared using a Leica EM UC7 ultramicrotome (Leica) and ultra diamond knife 35°(Diatome). e resulting ultrathin sections were scooped onto 200 mesh copper grids and stained with modified Sato's lead solution and 4% uranium acetate solution. e grids were dried and observed by TEM Hitachi-7500 (Hitachi High-Technologies) at an acceleration voltage of 120 kV or TEM JEM1011 (JEOL) at an acceleration voltage of 100 kV.

Dynamic Light Scattering (DLS).
Size distributions of nanoparticles in the GVLP suspensions were measured at 25°C by using a Zetasizer Nano-ZS (Malvern Instruments) with a 633 He/Ne Laser. e distribution of hydrodynamic size of purified GVLPs was estimated from the DLS data (Supplementary Figure 2).

Confocal Microscopy.
e GV_AC28 cells incubated with Dox for over 3 days were plated onto a 35 mm glass bottom dish filled with phenol red-free DMEM (Gibco) with supplements. Fluorescence images were obtained by using an inverted confocal microscope FV-1000 (Olympus) combined with a stage-top incubator INUC-KRi (Tokai-Hit) to control the temperature at 37°C. CO 2 was loaded at 5% into the incubator GM-2000 (Tokai-Hit). GFP and mKate2 were simultaneously excited by 473 nm and 559 nm lasers, respectively. A 60x objective lens (NA 1.40, oil) PLAPON 60XOSC2 (Olympus) and an observation area of 30.705 × 30.705 μm 2 (2048 × 2048 pixels) were used.

MR Measurements.
MR measurements were performed on an Agilent Unity INOVA 400WB high-resolution NMR system (Agilent Technologies). A 9.4 T vertical magnet with a bore width of 89 mm (Oxford Instruments) was used. A multinuclear ( 1 H-129 Xe) NMR Probe (inner φ: 10 mm) was used for spectroscopy, and a self-shielded gradient probe Clear Bore DSI-1117 (inner φ: 34 mm) (Doty Scientific) was used for imaging. Litz volume RF coils tuned to the Larmor frequency of 129 Xe (110.6 MHz) were employed for 129 Xe detection. A gas mixture of 70% xenon (natural abundance, 129 Xe fraction 0.26) and 30% N 2 was hyperpolarized using a home-built continuous flow-type 129 Xe polarizer and delivered to the GV cell suspensions in a 10 φ NMR sample tube by using a diaphragm pump, LABOPORT ® N86 KN.18 (KNF Neuberger GmbH), at a flow rate of 50 mL/min under atmospheric pressure for HyperCEST NMR/MRI measurements. e polarization level obtained was ∼10% using the AW-SEOP-laser system (Aurea Works Corporation) [19]. HyperCEST NMR measurements were performed using a pulse sequence: b 1 -d 1 -b 2 -d 2 -TS-α-t acq , where b 1 and b 2 are the gas bubble periods of 1.0 s, d 1 and d 2 are the delay periods of 0.5 s to make the GV cell suspension static, TS is a radio-frequency (RF) saturation period of 2.0 s, and α is the flip angle of the RF pulse used for observation of dissolved HPXe NMR signal (194 ppm) after saturation, and t acq is the NMR acquisition period. During the TS period, hundred frequency-selective saturation RF pulses (interpulse delay 16 ms) were irradiated. Frequency-dependent saturation spectra were acquired by varying the saturation pulse offset from −83.7 ppm to 278.0 ppm, yielding Z-spectra (i.e., dissolved HPXe signal acquired with off-resonance Contrast Media & Molecular Imaging saturation normalized by that acquired with on-resonance saturation). Z-spectra were acquired from at least four different cell preparations of each GV cell type to ensure reproducibility. Prior to HyperCEST NMR studies, the duration of the saturation and observation pulses (sincshaped 4000 μs and Gauss-shaped 800 μs, respectively) was calibrated to achieve saturation. e strength of the saturation RF pulse was 1.5 μT (30 dB). Chemical shifts were referenced to free gaseous HPXe NMR signal (0 ppm). Imaging was performed using a four-shot balanced steadystate free precession (b-SSFP) sequence (intershot delay 3 s) modified to acquire HyperCEST MR images. Frequency saturation similar to the HyperCEST NMR measurement was conducted for 2 s within the intershot delay. e duration and strength of the saturation RF pulse employed for HyperCEST MRI were Gauss-shaped 1000 μs and 5.9 μT (52 dB), respectively. MRI acquisition parameters were as follows: a 1000 μs Gaussian-shaped RF pulse of flip angle α � 40°with a bandwidth of 2800 Hz centered on the dissolved HPXe resonance; TR/TE � 3.

Results and Discussion
In this work, we established stable expression of protein nanoparticles with characteristic bicone structures similar to natural GVs in human cancer cells and demonstrated genetic modulation of their size and shape. In addition, GVs were shown to be applicable as multiplexed, genetically encoded HyperCEST MRI contrast agents in human cells in vitro.
First, praGV gene sequences available on the online database GenBank were examined (see Supplementary Text).
e genes used in this study are listed in Table 1. ese praGV genes with codons optimized for expression in mammalian hosts were synthesized and cloned into Tol2 transposon vectors [15,16,20] under the control of tetracycline-inducible elements (Tet-On) [17]. Puromycin-resistant genes were also included in the vector to select correctly transfected cells. T2A- [21] fluorescent protein (EGFP, mKO2 [22], and mKate2 [23]) fusion genes were cloned to the 3′ ends of GV genes (according to schemes in Figure 1(a)), for quantitative evaluation of GV gene expression by flow cytometry. T2A sequences were self-excised just after translation to avoid inhibiting GV structure formation, for example, by steric hindrance of linked fluorescent proteins. We offer expression vectors of these humanized praGV genes at the National Bio-Resource Center (https://dnaconda.riken. jp/search/depositor/dep103337.html).
Combinations were chosen in reference to the natural Planktothrix rubescens/agardhii GV gene clusters [13]. In this study, we investigated 3 specific combinations: gvpA + gvpC20 (GV_AC20), gvpA + gvpC28 (GV_AC28), and gvpA + gvpC16 + gvpC20 (GV_AC16C20). Transfected cells expressing the GV gene sets at high levels were selected using puromycin selection and fluorescence-activated cell sorting (FACS). Monoclonal single colonies were picked and expanded for GV_AC20, GV_AC28, and GV_AC16C20 cells. Obtained monoclonal cell pellets are shown in Figure 1(b). Expression levels of the fluorescent proteins of the monoclonal cultures were measured by flow cytometry (Supplementary Figure 1). e expected cellular burden to express these GV genes in mammalian cells was evaluated by a cell proliferation assay using a WST-8 reagent (Materials and Methods). GV_AC28 cells showed a significant decrease of cell proliferation after 72 hours of induced GV gene expression with doxycycline (Dox+) (p � 0.024, t-test) compared with that of noninduced control cells (Dox−) (Figure 1(c)). Other GV cells did not show a significant decrease of cell proliferation (p � 0.86, 0.18, and 0.34, t-test for control, GV_AC20, and GV_AC16C20 cells, respectively).
To confirm expression of foreign protein nanoparticles in the cells, we developed a method of purifying buoyant protein nanoparticles from mammalian cells (Figure 2(a)). e protocol was modified from that previously described [25] to exploit the much lower density of GVs compared to water. e purified putative nanoparticle suspensions were observed by TEM. Characteristic bicone structures were observed when using a standard negative staining protocol (Materials and Methods) in a purified suspension of the GV_AC28 cells (Figure 2(b)). In other cell clusters, distinct structures were not observed. us, we tested another protocol, usually adopted for observation of intracellular structures in cells, which we proposed would help avoid collapse of the nanoparticles during TEM observation. Putative nanoparticle suspensions were first embedded in ∼1 mm 3 2% agarose gels and fixed with glutaraldehyde (GA) (Materials and Methods). Using this protocol, distinct bicone structures were also observed in suspensions purified from GV_AC20 and GV_AC16C20 cells (Figure 2(b)). It is noteworthy that these observed nanostructures were not cellular organelles or any membranous structures, because they were collapsed by the act of osmotic pressure and detergent used in the purification method. We additionally note that the TEM images could not be any salt crystals because the applied staining method specifically contrasted the proteins. We also obtained Dynamic Light Scattering data on the purified nanoparticle suspensions to gain insights into the size distribution of the protein nanoparticles (Supplementary Figure S2). e obtained TEM images and DLS data clearly indicated the expression of characteristic bicone nanostructures in mammalian cells for each of our GV gene constructs, and genetic control of their size and shape was observed, as hypothesized (Supplementary Text). In addition, the cell proliferation assay indicated that cellular burden of these nanoparticle expressions in mammalian cells may depend on the expressed nanoparticle size (Figure 1(c)).
Moreover, we were able to clearly observe the nanostructures in GV_AC28 cells as a negative contrast by fluorescent confocal microscopy due to their large size (above the spatial resolution of confocal microscopy) (Figure 2(c)). We call these nanoparticles expressed in mammalian cells by reduced humanized GV gene sets with characteristic bicone nanostructures similar to natural GVs and GV-like particles (GVLPs) and our proposed methodology of GVLPs expression in mammalian cells with genetic control of their size and shape utilizing humanized praGV genes as FC-SOPRAGA: Facilitated Control of Self-Organized Planktothrix rubescens/agardhii Gas vesicle-like particle Assembly.
To investigate whether the reconstituted GVLPs could be used as multiplexed and genetically encoded HyperCEST MRI contrast agents (as previously demonstrated in prokaryotic cells [8]), we measured the MR signal intensities of hyperpolarized xenon dissolved in cells expressing GVLPs and control KPL-4 cells as a function of saturation frequency offset (i.e., recorded CEST Z-spectra) using a 2 s saturation time. e saturation was performed to detect the chemical shift of GVLP-bound xenon, i.e., the saturation chemical shift at which the bulk dissolved xenon signal was most attenuated by chemical exchange transfer of saturated spins between GVLPs and bulk-dissolved xenon.
e HyperCEST Z-spectrum of GV_AC28 cell samples consistently showed a maximal signal decrease at around 20 ppm saturation offset chemical shift and occasionally  (Figure 3(a)). We believe the two unique peaks of GVLP-bound xenon may be

GV_AC20
GV_AC28 GV_AC16C20 explained by the inhomogeneity of GVLP size in GV_AC28 cells as indicated in TEM images (Figure 2(b) and DLS data (Supplementary Figure 2)). e 20 ppm peak may be attributed to the xenon bound to large vesicles and the minor peak (∼70 ppm) presumably relates to the xenon bound to smaller vesicles [26].
Furthermore, we observed a slight but repeatable saturation peak of GV_AC16C20 cells at around 60 ppm in its Z-spectrum despite a lower number of GV cells (2.0 × 10 7 cells/ml vs. 8.0 × 10 7 cells/ml for GV_AC28 cells). In this case, the background signal in the Z-spectrum was reduced, similar to that previously reported for the β-lactamase HyperCESTreporter gene [11]. erefore, we concluded that the peak at 60 ppm was indeed derived from the saturated GV-bound xenon transfer rather than an experimental artifact (Figure 3(a)). On the other hand, the Z-spectrum of GV_AC20 cells (Supplementary Figure 3) did not show any practical saturation peaks for HyperCEST contrast at 2 s saturation time.
Subsequently, bulk-dissolved xenon NMR signal intensity was measured after applying saturation prepulses at the on-resonance and off-resonance chemical shifts (onresonance signal intensity S on and off-resonance signal intensity S off , respectively) as a function of saturation time, and the %CEST contrast was calculated as defined in the following equation for GV_AC28 and GV_AC16C20 cells (Figure 3(b)): In GV_AC28 cells, chemical shifts of on-resonance and off-resonance saturation prepulses were set to 20 ppm and −20 ppm, respectively, and for GV_AC16C20, 60 ppm and −60 ppm were used. GV_AC28 cells showed 5.5 ± 1.3 and 6.5 ± 0.9% CEST contrast at 1.5 s and 2 s saturation time, respectively. At 3 s saturation time, CESTcontrast maximally reached 8.0 ± 0.7%. For GV_AC16C20 cells, 2.6 ± 1.1 and 3.1 ± 1.5% CEST contrast was observed at 1.5 s and 2 s saturation time, with a maximal 5.8 ± 0.7% contrast at 2.5 s saturation time. Considering the longitudinal relaxation time (T 1 ) of gaseous xenon in vivo (∼30 s, note that the T 1 of dissolved xenon is lower), the saturation time for in vivo HyperCEST MRI should ideally be less than 2 s [27,28]. It is noteworthy that a low saturation time is also desirable for a decreased specific absorption rate, which is an important consideration regarding the application of GVs in human patients. Taking these concerns into account, a HyperCEST MR image of GV_AC28 cells was successfully acquired by applying a 2 s saturation scheme at 20 ppm (Figure 3(c)).

Conclusions
In this study, we showed that expression of reduced GV gene sets derived from Planktothrix rubescens/agardhii in mammalian cells resulted in the formation of GVLPs in the cell and they could be functionalized as a genetically encoded HyperCEST MRI contrast agents. GVLPs have distinct properties from that of natural GVs expressed in bacterial cells, and it should be further characterized in the future. A recent study [29] also demonstrated that gas vesicles can be expressed in mammalian cells and their utility as a genetically encoded ultrasound contrast agent by similar methods under almost the same overall experimental design is also demonstrated, which is consistent with our results. Note that our method of mammalian expression of gas vesicles was conceived independent of the previous study [29] demonstrated by the patent application already open to the public [30]. GV_AC28 construct demonstrated promising CESTbased MRI contrast within a practical saturation time, which has not been reported to date in genetically encoded HyperCEST MRI contrast agents for mammalian cells. Our findings should help facilitate the future development of GV-based HyperCEST MRI reporters for in vivo applications, though realization of in vivo HyperCEST still requires overcoming several technical hurdles, such as continuous or high-volume batch supply of xenon with nonrenewable polarization [31] and tissue-specific targeted delivery of GVs.

Data Availability
All data and materials underlying this study are available upon request to the corresponding author. Expression vectors of humanized praGV genes were deposited to and are available from the BioResource Research Center, RIKEN.

Conflicts of Interest
RIKEN has a pending patent application already open to the public (WO2018043716A1) regarding this work, in which R. M. and T. M. W. are inventors.
Platform" of the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan. is work was supported in part by JSPS KAKENHI under Grant no. JP17K20121.

Supplementary Materials
Supplementary Text: we described how to choose gas vesicle genes for our study in the selection of GV genes section. Note on TEM observation of GV structures is the short technical comments to observe GVLP structures by TEM. e purified GVLP structure and size distribution section is the description on the variability of the size and shape of GVLP measured and observed by TEM and DLS in the study. Supplementary