Xe HyperCEST MRI detection of genetically-reconstituted 1 bacterial protein nanoparticles in human cancer cells

16 Gas vesicle nanoparticles (GVs) are gas-containing protein assemblies expressed in bacteria 17 and archaea. Recently, GVs have gained considerable attention for biotechnological 18 applications as genetically-encodable contrast agents for MRI and ultrasonography. However, 19 at present, the practical use of GVs is hampered by a lack of robust methodology for their 20 induction into mammalian cells. Here, we demonstrate the genetic reconstitution of protein 21 nanoparticles with characteristic bicone structures similar to natural GVs in a human breast 22 cancer cell line KPL-4, and genetic control of their size and shape through expression of 23 reduced sets of humanized gas vesicle genes cloned into Tol2 transposon vectors, referencing 24 the natural gas vesicle gene clusters of the cyanobacteria planktothrix rubescens/agardhii. We 25 then report the utility of these nanoparticles as multiplexed, sensitive and genetically-encoded 26 contrast agents for hyperpolarized xenon chemical exchange saturation transfer (HyperCEST) 27 MRI. 28


Introduction 30 31
Gas vesicle nanoparticles (GVs) are gas-containing spindle-(or bicone-) shaped protein 32 nanostructures with dimensions ranging from tens to hundreds of nm that are expressed in the 33 cyanobacteria, algae and gram-positive bacteria. Since only gas molecules are permeable to the 34 GV protein shell, GV-containing organisms acquire buoyancy by selective permeation of 35 various ambient gases in the GV interior, which facilitates optimal supply of light and nutrition 36 1-3 . While GVs have been studied in the field of microbiology for several decades, they have 37 gained significant attention in recent years for their potential use for antigenic peptide display 38 in vaccination 4 , and as genetically-encodable contrast agents for ultrasonography 5 6 7 and 39 HyperCEST MRI 8,9 . Among these applications, HyperCEST MRI, which utilizes laser-40 polarized xenon-129 (HPXe) and chemical exchange saturation transfer (CEST) 10 to yield 41 unprecedented enhanced MRI detection sensitivity, is of particular interest as GVs hold the 42 potential to enable functional molecular imaging that is unfeasible in conventional thermally-43 polarized proton MRI. (In principle, GV concentrations of as low as pM -nM can be detected.) 44 In this context, GVs can be thought of as an MRI analog of the green fluorescent protein (GFP) 45 optical imaging reporter; with a genetically-encodable nature and multiplexing capability 46 facilitated by ready modulation of their size and shape, similar to the multi-color variant of 47 GFPs 8 . However, despite their attractive features for imaging cellular / molecular processes, 48 the practical in-vivo use of GVs as genetically-encoded contrast agents is at present hampered 49 by a lack of robust techniques to introduce GVs into mammalian cells, which has been 50 considered challenging due to the complexity of GV gene clusters 11 . 51 GVs are composed of multiple proteins, and the number of genes responsible for GV 52 expression is usually 8 -14 (typically denoted GvpA, B, C etc.). Among these genes, the 53 principal component proteins are the hydrophobic major protein GvpA and hydrophilic minor 54 protein GvpC; the roles of other accessary GV genes in constituting GV wall structure remains 55 a subject of controversy 3 . In order to optimize GV delivery in-vivo, understanding of the 56 minimal number and type of GV gene components required to reconstitute the GV 57 nanostructure in mammalian cells is crucial. In this study, we focus on the unique gene clusters 58 of GVs derived from Planktothrix rubescens/agardhii (praGV). The praGV gene clusters have 59 been studied extensively by Walsby and coworkers 12-14 , who showed that parts of praGV gene 60 clusters are composed of gvpA and three variants of gvpC named gvpC16, gvpC20 and gvpC28, 61 though it is known that existence of other accessary GV genes have also been demonstrated 3 . 62 Properties of praGVs such as size, shape and stiffness (pressure threshold for collapse) are 63 known to be modulated depending on the combination of gvpC variants included in their 64 constituent gene clusters 13 . Thus, we hypothesized that combinatorial expression of such 65 reduced sets of genes in mammalian cells would allow reconstitution of protein nanoparticles 66 with similar properties to GVs in natural organisms which can be functionalized as a contrast 67 agent for HyperCEST MRI in mammalian cells and genetic control of their size and shape.

69
Results and Discussion 70 71 In this work, we established stable expression of protein nanoparticles with characteristic 72 bicone structures similar to natural GVs in human cancer cells and demonstrated genetic 73 modulation of their size and shape. In addition, GVs were shown to be applicable as 74 multiplexed, genetically-encoded HyperCEST MRI contrast agents in human cells in-vitro. 75 Firstly, praGV gene sequences available on the online database GenBank were examined 76 (see Supplementary Text). The genes used in this study are listed in Table 1. These praGV 77 genes with codons optimized for expression in mammalian hosts were synthesized, and cloned 78 into Tol2 transposon vectors 15,16 17 under the control of tetracycline-inducible elements (Tet-79 On) 18 . Puromycin-resistant genes were also included in the vector to select correctly 80 transfected cells. T2A 19 -fluorescent protein (EGFP, mKO2 20 and mKate2 21 ) fusion genes 81 were cloned to the 3' ends of GV genes (according to schemes in Figure 1a), for quantitative 82 evaluation of GV gene expression by flow cytometry. T2A sequences were self-excised just 83 after translation to avoid inhibiting GV structure formation, for example by steric hindrance of 84 linked fluorescent proteins. We offer expression vectors of these humanized praGV genes at 85 the National Bio-resource Center (https://dnaconda.riken.jp/search/depositor/dep103337.html).

93
Obtained monoclonal cell pellets are shown in Figure 1b. Expression levels of the fluorescent 94 proteins of the monoclonal cultures were measured by flow cytometry (Supplementary Figure  95 1). The expected cellular burden to express these GV genes in mammalian cells was evaluated 96 by a cell proliferation assay using a WST-8 reagent (Materials and Methods). GV_AC28 cells 97 showed a significant decrease of cell proliferation after 72 hours of induced GV gene 98 expression with doxycycline (Dox+) (p = 0.024, t-test) compared with that of non-induced 99 control cells (Dox−) (Figure 1c). Other GV cells did not show a significant decrease of cell 100 proliferation (p = 0.86, 0.18 and 0.34, t-test for control, GV_AC20 and GV_AC16C20 cells, 101 respectively).

102
To confirm expression of foreign protein nanoparticles in the cells, we developed a 103 method of purifying buoyant protein nanoparticles from mammalian cells (Figure 2a). The 104 protocol was modified from that previously described 23 to exploit the much lower density of 105 GVs compared to water. The purified putative nanoparticle suspensions were observed by TEM.

106
Characteristic bicone structures were observed when using a standard negative staining 107 protocol (Materials and Methods) in a purified suspension of the GV_AC28 cells ( Figure 2b). 108 In other cell clusters, distinct structures were not observed. Thus, we tested another protocol, 109 usually adopted for observation of intracellular structures in cells, which we proposed would 110 help avoid collapse of the nanoparticles during TEM observation. Putative nanoparticle 111 suspensions were first embedded in ~ 1 mm 3 2% agarose gels and fixed with glutaraldehyde 112 (GA) (Materials and Methods). Using this protocol, distinct bicone structures were also 113 observed in suspensions purified from GV_AC20 and GV_AC16C20 cells ( Figure 2b). It is 114 noteworthy that these observed nanostructure were not cellular organelles or any membranous 115 structures, because they were collapsed by the act of osmotic pressure and detergent used in 116 the purification method. We additionally note that the TEM images could not be any salt 117 crystals because applied staining method specifically contrasted the proteins. We also obtained 118 Dynamic Light Scattering data on the purified nanoparticle suspensions to gain insights into 119 the size distribution of the protein nanoparticles (Supplementary Figure S2). The obtained TEM 120 images and DLS data clearly indicated the expression of characteristic bicone nanostructures 121 in mammalian cells for each of our GV gene constructs, and genetic control of their size and 122 shape was observed, as hypothesized (Supplementary Text). In addition, the cell proliferation 123 assay indicated that cellular burden of these nanoparticle expression in mammalian cells may 124 depend on the expressed nanoparticle size ( Figure 1c). 125 Moreover, we were able to clearly observe the nanostructures in GV_AC28 cells as a 126 negative contrast by fluorescent confocal microscopy due to their large size (above the spatial 127 resolution of confocal microscopy) ( Figure 2c). We call these nanoparticles expressed in 128 mammalian cells by reduced humanized GV gene sets with characteristic bicone nanostructures 129 similar to natural GVs GV-like particles (GVLPs) and our proposed methodology of GVLPs 130 expression in mammalian cells with genetic control of their size and shape utilizing humanized 131 praGV genes as FC-SOPRAGA: Facilitated Control of Self-Organized Planktothrix 132 Rubescens/Agardhii Gas vesicle -like particle Assembly. 133 To investigate whether the reconstituted GVLPs could be used as multiplexed and 134 genetically-encoded HyperCEST MRI contrast agents (as previously demonstrated in 135 prokaryotic cells 8 ), we measured the MR signal intensities of hyperpolarized xenon dissolved 136 in cells expressing GVLPs and control KPL-4 cells as a function of saturation frequency offset 137 (i.e. recorded CEST Z-spectra) using a 2 s saturation time. The saturation was performed to 138 detect the chemical shift of GVLP-bound xenon, i.e. the saturation chemical shift at which the 139 bulk dissolved xenon signal was most attenuated by chemical exchange transfer of saturated 140 spins between GVLPs and bulk dissolved xenon. 141 The HyperCEST Z-spectrum of GV_AC28 cell samples consistently showed a maximal 142 signal decrease at around 20 ppm saturation offset chemical shift, and occasionally minor 143 saturation peaks at around 70 ppm, in addition to the bulk dissolved phase HPXe saturation 144 peak at 194 ppm, while the control KPL-4 cells showed only the 194 ppm saturation peak of 145 bulk dissolved xenon with a broad and weak background around 0 ppm (Figure 3a). We believe 146 the two unique peaks of GVLP-bound xenon may be explained by the inhomogeneity of GVLP 147 size in GV_AC28 cells as indicated in TEM images ( Figure 2b) and DLS data (Supplementary 148 Figure 2). The 20 ppm peak may be attributed to the xenon bound to large vesicles and the 149 minor peak (~70 ppm) presumably relates to the xenon bound to smaller vesicles 24 . 150 Furthermore, we observed a slight, but repeatable saturation peak of GV_AC16C20 cells at 151 around 60 ppm in its Z-spectrum despite a lower number of GV cells (2.0 x 10 7 cells / ml vs. 152 8.0 x 10 7 cells / ml for GV_AC28 cells). In this case, the background signal in the Z-spectrum 153 was reduced, similar to that previously reported for the β-lactamase HyperCEST reporter gene 154 11 . Therefore, we concluded that the peak at 60 ppm was indeed derived from the saturated GV-155 bound xenon transfer rather than an experimental artifact (Figure 3a). On the other hand, the 156 Z-spectrum of GV_AC20 cells (Supplementary Figure 3) did not show any practical saturation 157 peaks for HyperCEST contrast at 2 s saturation time.

158
Subsequently, bulk dissolved xenon NMR signal intensity was measured after applying 159 saturation pre-pulses at the on-resonance and off-resonance chemical shifts (on-resonance 160 signal intensity Son and off-resonance signal intensity Soff, respectively) as a function of 161 saturation time and the %CEST contrast was calculated as defined in the equation for 162 GV_AC28 and GV_AC16C20 cells (Figure 3b): % = × 100. 163 In GV_AC28 cells, chemical shifts of on-resonance and off-resonance saturation pre-164 pulses were set to 20 ppm and -20 ppm, respectively, and for GV_AC16C20, 60 ppm and -60 165 ppm were used. GV_AC28 cells showed 5.5 ± 1.3 and 6.5 ± 0.9 % CEST contrast at 1.5 s and 166 2 s saturation time, respectively. At 3 s saturation time, CEST contrast maximally reached 8.0 167 ± 0.7 %. For GV_AC16C20 cells, 2.6 ± 1.1 and 3.1 ± 1.5 % CEST contrast was observed at 168 1.5 s and 2 s saturation time, with a maximal 5.8 ± 0.7 % contrast at 2.5 s saturation time.

169
Considering the longitudinal relaxation time (T1) of gaseous xenon in vivo (~ 30 s, note that 170 the T1 of dissolved xenon is lower), the saturation time for in-vivo HyperCEST MRI should 171 ideally be less than 2 s 25,26 . It is noteworthy that a low saturation time is also desirable for 172 decreased specific absorption rate, which is an important consideration regarding the 173 application of GVs in human patients. Taking these concerns into account, a HyperCEST MR 174 image of GV_AC28 cells was successfully acquired by applying a 2 s saturation scheme at 20 175 ppm (Figure 3c). 176 177

178
In this study, we showed that expression of reduced GV gene sets derived from 179 planktothrix rubescenes/agardhii in mammalian cells resulted in the formation of GVLPs in 180 the cell and they could be functionalized as a genetically- assistance with FACS operation, cell culture and molecular cloning. We also thank Dr. Neil J 283 Stewart for proofreading the manuscript. 284 We acknowledge Kawasaki Medical School and Prof. Junichi Kurebayashi for generously 285 providing KPL-4 cells. This work was supported by the NIMS microstructural characterization 286 platform, a program of the "Nanotechnology Platform" of the Ministry of Education, Culture, 287 Sports, Science and Technology (MEXT), Japan. 288 This work was supported in part by JSPS KAKENHI Grant Number JP17K20121. All data and materials underlying this study are available. Expression vectors of humanized 306 praGV genes were deposited to and are available from the BioResource Research Center, RIKEN. 307 The article was previously posted on bioRxiv (http://biorxiv.org/cgi/content/short/599118v1). 308 309 Table 1   Supplementary Text 540

Selection of GV genes 541
We chose the gvpC28 gene from planktothrix agardhii instead of planktothrix rubescens in this 542 study because the full sequence of the gvpC28 gene was not available on the Genbank database. 543 In addition, because previous studies indicated that a heterologous combination of GV genes 544 could be utilized to induce GVs with altered size and shape into bacterial species 1 , we assumed 545 that the heterologous combination of GV genes derived from planktothrix rubescens and 546 planktothrix agardhii could be used to produce GVLPs for these closely related strains. 547

Notes on TEM observation of GV structures 548
In an initial effort to confirm GVLP expression in the cells, we observed ultra-thin slices of 549 embedded GV_AC20 cell samples by TEM and found the putative GVLPs (indicated white 550 arrow in Supplementary Figure 4), exhibiting the characteristic bicone structure, although the 551 complexity of other intracellular components made it difficult to undeniably confirm GV 552 expression. 553 It is noteworthy that fixation of the GVLP purified from GV_AC28 cells with 2.5 % 554 glutaralaldehyde (GA) prior to drying and staining was found to be essential for observation of 555 distinct GV structures; without GA fixation, we could not observe distinct GV structures in our 556 experiments (Supplementary Figure 5). 557

Purified GVLP structure and size distribution 558
The TEM image of GVLPs purified from GV_AC20 cells in Figure 2b shows that the width 559 and length of GVLPs was heterogeneous and ranged from ~ 200 nm to 300 nm. GVLP shape 560 appeared to be biconical or cylindrical without conical ends, comprising a linear outline. The 561 observed cylinders without conical ends may indicate deficient structure formation (e.g. lack 562 of closed protein shell structure); this would reasonably explain the discrepancy between 563 observed TEM sizes ( Figure 2b) and lower size distribution peak of about 90 nm in DLS data 564 (Supplementary Figure 2). In contrast, TEM images of GVLPs purified from GV_AC28 cells 565 indicated a more rounded outline compared to other GV cells. The GV size ranged from 100 566 nm to 700 nm, and included the largest GVs among all that were generated. The two peaks of 567 GVLPs purified from GV_AC28 cells in Dynamic Light Scattering (DLS) data 568 (Supplementary Figure 2) may corresponded to the relatively broad particle size distribution of 569 GVLP in GV_AC28 cells and/or aggregation of the particles. The TEM data showed that the 570 length of GVLPs purified from GV_AC16C20 cells was around 100 nm or less, with a width 571 of 50 -60 nm. DLS data of GVLPs purified from GV_AC16C20 cells indicated a relatively 572 homogeneous size distribution with a peak at ~ 80 nm, which is in accordance with the TEM 573 data. 574