Cloneable Contrast Across All Biological Length Scales

Abstract

Cloneable contrast in biological microscopy is exemplified by Green Fluorescent Protein (GFP) in fluorescence microscopy. There are no similarly useful cloneable contrast agents that function in biological electron microscopy. This paper reports a cloneable Selenium NanoParticle (cSeNP) that produces molecular contrast in imaging modalities including cellular electron microscopy, fluorescence microscopy, and X-ray computed tomography. This set of imaging modalities can image all biologically relevant length scales, from subcellular structure to whole organisms. The cSeNP is a ~5 nm diameter Selenium nanoparticle that is made and conjugated by a protein. Because the cSeNP is electron dense compared to biological molecules, it has high contrast in biological electron microscopy. DNA encoding the cSeNP protein was concatenated to DNA encoding FtsZ, the procaryotic analog of tubulin. FtsZ is membrane associated throughout the cell cycle and localizes to the cleavage furrow of dividing cells. Escherichia coli cells expressing FtsZ-cSeNP fusion proteins were examined by transmission electron tomography and fluorescence light microscopy. These experiments show cSeNP decorated FtsZ filaments and/or cSeNPs in locations that correlate to known FtsZ locations, with less than 5% of cSeNPs in unexpected locations. X-ray imaging shows contrast attributable to cSeNPs is distinguishable from background in E. coli. The cSeNP, therefore, represents a cloneable imaging contrast agent that facilitates location and correlation of proteins-of-interest across all biological length scales. This is especially useful in biological electron microscopy, where larger-area imaging modalities such as fluorescence microscopy are employed to identify sub-areas containing a protein-of-interest to prepare for electron microscopy study.

Introduction

Cloneable contrast is well established in biological fluorescence microscopy, exemplified by Green Fluorescent Protein (GFP).^1,2 What makes GFP and related fluorescent proteins useful is that the DNA encoding GFP can be concatenated to the DNA encoding any protein of interest (POI). When that DNA is expressed in cells, a GFP-POI fusion protein is created. The physiological location of the POI can then be inferred in fluorescence micrographs based on GFP fluorescence. The widespread adoption and utility of fluorescent proteins was recognized by the 2008 Nobel Prize in Chemistry.³

The ‘cloneable contrast’ paradigm of GFP in light-based microscopy inspires cloneable contrast approaches in other imaging modalities. Despite significant efforts, there are no robust cloneable contrast agents that function in biological electron microscopy. A “GFP for cellular EM” is sometimes referred to by biological electron microscopists as a “holy grail.”⁴

The community of biological electron microscopists is rapidly growing, due to methodological developments such as cryo-preservation, direct electron detectors, and automation of electron microscopes that facilitate higher-throughput and higher-fidelity imaging than was possible in prior decades.⁴ Biological EM is instrumental to the overarching goals of integrative structural biology.^5,6 Whereas ‘classical’ structural biology aimed to determine atomically resolved structures of individual biopolymers (proteins and nucleic acids), integrative structural biology aims to contextualize such structures in their cellular, tissue, and organismal context. The goal of integrative structural biology is empirically informed 3D atomic resolution models of entire cells, or even tissues and organisms, forming the basis for computational simulation of biological processes with chemical levels of detail.

3D electron microscopy (3DEM) produces the highest resolution images of biological specimens. 3DEM comprises 3 distinct methods. At the largest sample volumes, volume Electron Microscopy (vEM) methods now produces 3D reconstruction of organs, exemplified by the drosophila brain and connectome.⁷ Cryo-electron tomography (cryo-ET) produces high fidelity 3D images of sub-areas of cells (most cells are too large to be imaged entirely by cryo-ET).⁸ Single particle analysis of protein images taken by cryo-EM facilitates determination of atomistic models of proteins, and is expected to soon out-pace X-ray crystallography in annual protein data bank depositions.⁹

The cloneable contrast approach we describe is expected to facilitate integrative structural biology and be most useful for localizing POIs in vEM and cryo-ET. The methods of vEM and cryo-ET differ in sample preparation – in vEM, samples are typically chemically fixed and plastic embedded, facilitating serial sectioning of larger volumes.¹⁰ Such fixation and plastic embedding introduces artifacts, such as membrane distortion. In contrast, in cryo-ET, cells are preserved by vitrification (rapid freezing, so that solvent solidifies in a glassy, amorphous state). Cryo-ET imaging of cells produces 3D images of cells at ~2 nm resolution (where resolution is limited by beam damage to the sample, not by microscope optics).⁸

At the ~2 nm resolution of cryo-ET, protein and nucleic acid densities are visible. However, only the largest biomacromolecular complexes, such as ribosomes and filaments, can be reliably identified. Significant machine learning efforts to identify smaller biomacromolecules, based on their observed shapes, are ongoing. However, the theoretical limits on this approach mean that at least 60% of the proteome is too small to identify and therefore localize.¹¹ Due to plastic embedding used in vEM, protein and nucleic acid densities are obscured. This makes proteins and nucleic acids more difficult to identify in vEM than they are in cryo-ET.

To localize most POIs within vEM or cryo-ET data, therefore, most POIs will need added contrast. Four approaches to imparting contrast on POIs in TEM are previously described. These are: (1) targeted inorganic NanoParticle (iNP) conjugates;^12-15 (2) in situ inorganic nanoparticle synthesis on cysteine rich proteins;^16,17 (3) osmium staining of osmiophilic polymers generated by singlet-oxygen generating proteins;^18,19 (4) the use of large proteins with distinctive shapes.^20,21 Three of these approaches deliver contrast in the form of inorganic elements. This is because inorganic elements used for EM contrast are heavier (higher Z) than the carbon/oxygen/nitrogen that comprise most of cell biology. Therefore, densities formed from these elements provide contrast against biological background in TEM, where contrast scales with electron density (and therefore atomic number Z). For this work, we aimed to make SeNPs size controlled at approximately 5 nm in diameter (about 4000 inorganic atoms) so that they could be easily identified against cellular background.

Targeted iNP-conjugates, where a conjugate molecule binds to a POI and localizes an iNP to a POI, represent an approach used since the 1970s to localize proteins in electron micrographs of biological material.²² This approach was classically accomplished with antibody-gold nanoparticle conjugates (e.g., colloidal gold immunohistochemistry). More recently, quantum dots conjugated to multiple copies of a 7.5 kDa protein that binds to a 5.2 kDa peptide tag facilitates a more modern approach dubbed VIPER.^14,22 Another approach involves using cysteine rich proteins, such as metallothionein, to localize inorganic atoms to a protein based on metal coordination to the thiol functional group of cysteine. Each cysteine binds stoichiometrically to one metal ion. Because an iNP must contain about 4000 inorganic atoms (~5 nm diameter iNP), it’s infeasible to rely only on the 1:1 metal:cysteine stoichiometric interaction alone for useful contrast (typical proteins have 300 to 400 amino acids). Therefore, cysteine-rich metal decorated proteins are subsequently enhanced with additional metal deposition in reductive metal deposition processes to form adequately sized inorganic particles.^16,17 A third approach uses small singlet oxygen generating (SOG) proteins to provoke polymerization of an osmiophilic polymer at the site of the POI-SOG fusion protein. Subsequent osmium staining reveals the location of the polymer, from which the approximate location of the POI can be inferred. Because the polymer is diffuse, POIs are located only approximately. A fourth approach uses relatively large and distinctly shaped proteins, such as encapsulins or ferritin. Based on the size and shape of the protein, it can be identified in biological micrographs.^20,21 Each of these 4 broad approaches has shortcomings that hinder their widespread adoption as contrast agents in the biological EM community.

To contextualize the different shortcomings of each of the 4 approaches, we outline 10 criteria for evaluating biological EM contrast approaches in Table 1. Some of the criteria are self-explanatory (e.g., visibility in electron micrographs of cells). Other criteria (e.g., stoichiometric labeling) are elaborated below. Fluorescent proteins, even though they do not provide contrast in TEM, are included in table 1 for comparison of properties to a truly widely used contrast tool in biological microscopy. Table 1 also includes the ‘Cloneable Nanoparticle’ approach described in this paper, which is distinct from any previously described approach.

Table 1.

Ten criteria for protein-localizing EM contrast agents, and how the 6 common approaches satisfy those criteria

Contrast Approach	Ex situ iNP Contrast Agents	Encapsulins	Singlet Oxygen Generators	In situ Metal Deposition	Cloneable Nanoparticles	Fluorescent Proteins
Examples	iNP-conjugates VIPER14 SPOT23	iGEM20 Ferritag21	miniSOG18 APEX1924	Metallo-thionein16,17,25,26	cSeNP (this work)	e.g., GFP
EM Contrast	✔	✔	✔	✔	✔	✖
Fluorescent	✔	✔	✖	✖	✔	✔
X-Ray Contrast	✔	✖	✔	?	✔	✖
In vivo function	Varies	✔	✖	✖	✔	✔
DNA encoded	✖	✔	✔	✔	✔	✔
Benign	?	?	?	?	?	✔
Stoichiometric	✖	✖	✖	✔	✔	✔
Size (protein/ inorganic)	Varies/Varies	25nm/--	~2nm/diffuse	2nm/5nm	6nm/5nm	3nm/--
Tag-to-Target Distance (est)	Varies	50nm	--	2nm	6nm	1.5nm
Organisms	Many	Mammalian Cells	Many	Many	Many	E. coli

The criteria of table 1 which are not self-evident include whether a contrast agent functions in live cells (in vivo), whether a contrast agent is intrinsically stoichiometric, and whether a contrast agent is visible in imaging modalities such as light microscopy and X-ray imaging.

In vivo contrast generation is desirable and is also more difficult to achieve than contrast added after a cell or tissue is fixed. In vivo contrast generation requires chemistries compatible with living cells, whereas contrast added post-fixation allows considerably more chemical flexibility. For cryo-ET imaging, contrast generation in vivo is an absolute requirement, because cells are not chemically fixed before the contrast agent is applied. For vEM, contrast agents can be incorporated in vivo or added post-fixation. In the case of SOGs and metal development of cysteine rich sequences, and very often with iNP-conjugates, the contrast agents are only applied after biological specimens are fixed.

Stoichiometric labeling refers to a 1:1 stoichiometry between a POI and a contrast agent. With perfect stoichiometry there is exactly one instance of the contrast agent for every instance of the POI. Deviations from perfect stoichiometry result in background signals and/or undetected POIs, both of which complicate data analysis. As an example, addition of iNP-conjugates to samples makes stoichiometric labeling difficult to achieve, due to intrinsic equilibrium chemistry and uncertainty in the concentration of a POI. If the POI is present at a concentration approaching an equilibrium constant of the targeting-molecule / POI equilibrium constant, then excess contrast agent must be added to drive POI/targeting molecule binding. If the concentration of the POI is uncertain, then it is likely that there will be more or fewer contrast agents than POIs. Stoichiometric labeling, in other words, allows quantification of the number of POIs in an image. Because they are not intrinsically stoichiometric labels, encapsulins, SOGs, and iNPs, do not facilitate reliable quantification of the number of POIs in an image by quantifying the number of contrast agents present in the image.

Contrast in imaging modalities that function on larger samples than can be accommodated in the EM is surprisingly useful in EM imaging. Biological EM is limited to samples that are a maximum of ~300 nm in thickness, where the highest resolution reconstructions come from samples thinner than this. Because typical bacterial cells have dimensions on the order of 3 μm and eucaryotic cells have dimensions on the order of 10s of μm, only a small fraction of a cell (often less than 10% of the cell volume) can be imaged in a single EM section. POIs typically localize to distinct areas within cells (mitochondria, nucleus, Golgi, periplasm, cellular poles, membranes). Therefore, any single 300 nm thick section of a cell may or may not contain a POI. Therefore, larger area imaging modalities such as fluorescence microscopy can be used to first identify an area that contains a POI, and that area can then be prepared for electron microscopy. This workflow is exemplified in the now-established method of “Correlative Light and Electron Microscopy” (CLEM).²⁷ In this approach, POIs are fused to a fluorescent protein, and the signal from the fluorescent protein enables identification of the sub-area of a sample to prepare for EM imaging. Modern EM sample preparation equipment, such as cryo- focused ion beam / scanning electron microscopes (cryo-FIB-SEM) often include fluorescence microscopes to facilitate identification of an area of interest for EM examination within a larger sample that may be majority comprised of areas that do not contain a POI. ²⁸

Like CLEM, correlative X-ray/EM imaging is an approach²⁹ where the cNP approach produces contrast in both methodologies. We anticipate that correlation of X-ray and EM images will be useful in vEM, where samples for EM may be taken from larger organisms, which may be too thick for fluorescence microscopy. Therefore, we anticipate that cloneable contrast agents that are simultaneously visible in EM and X-ray images will become useful as EM and X-ray correlations become more widely used.

Thus, multimodal contrast agents that are visible in X-ray, fluorescence, and EM imaging modalities may facilitate the study of POIs within large organisms with both lower resolution / large area imaging modalities and high resolution / small area imaging methods (Figure 1, panel A), using a single contrast agent. Note that the ‘achievable resolutions’ depicted for X-ray and EM imaging assume ‘native’ state cells (e.g., vitrified or “living” – not chemically fixed). X-ray modalities achieve higher resolutions for fixed samples and EM modalities can reconstruct thicker samples with fixed samples.

Figure 1.

Panel A shows a schematic of the length scales and resolution of X-Ray, Optical, and Electron based imaging methods. Panel B shows a schematic of the cNP paradigm, where ‘protein of interest’ is abbreviated to ‘POI’. Panel C shows a schematic of the cSeNP in apo and cSeNP forms. Panel D shows an image of bacteria expressing GRLMR, grown on SeO₃²⁻ supplemented agar. The cell colonies take on the red color because they are producing SeNPs of the red allotrope of Se (α-Se). Image taken from reference 26. Panel E shows a schematic of the disproportionation reaction that converts SeNPs to fluorescent metal selenide nanoparticles, created with BioRender.

As highlighted in table 1, iNPs produce contrast in all commonly used biomedical imaging modalities.³⁰ Aiming to combining the superior contrast properties of iNPs with the ease of use of fluorescent proteins, we developed a cNP platform (Figure 1, panel B).³¹ Fundamentally, a cNP is an iNP that is made by a protein. The ‘cloneable nanoparticle protein’ determines the physicochemical properties (elemental composition, size, morphology, crystal structure, allotrope, etc.) of the iNP that it creates. Protein sequence and structure are encoded in DNA, therefore the physicochemical properties of the iNP made by the cloneable nanoparticle protein are also encoded in DNA. Changing the DNA sequence encoding the cNP can change the physicochemical properties of the resulting cNP,^32,33 just as changing the DNA encoding a fluorescent protein can change the color of the protein. Just as ‘cloning’ the DNA encoding a fluorescent protein into a different organism endows the receiving organism with the capability of producing a fluorescent protein, cloning the DNA encoding a cNP into other organisms, endows those organisms with the capability of recombinant production of the cNP.

The DNA encoding a cNP usually encodes a modified inorganic-ion oxidoreductase. The enzymatic activity converts soluble, ionic bio-coordination complexes to insoluble iNP form. The enzyme both selects a soluble inorganic ion from a milieu that can contain a variety of essential, bioavailable inorganic ions and alters the oxidation state of the ion from soluble to insoluble (particulate) form. Examples of enzymes that have this activity include Selenium Reductase, Metalloid Reductase, Tellurite Reductase, Ferroxidase, and Mercuric Reductase.^32,34-37 Redox equivalents are provided by NAD(P)H or oxygen. To effect control over nanoparticle size, morphology, crystal structure and so-forth, the enzymes can be modified with peptides that were selected for binding to iNPs with a targeted/desired size, morphology and crystal structure.³⁸ In many cases, such peptides template the formation of iNPs during their synthesis, producing iNPs/cNPs of similar size/morphology/crystal structure relative to the target/desired size. We recently reviewed the concept.³¹

In the present work, we show that a cSeNP functions as a cloneable contrast agent in electron, X-ray, and visible light-based imaging modalities. The cSeNP is comprised of a homodimeric GS-Se-SG reducing enzyme --Glutathione Reductase-Like Metalloid Reductase (GRLMR)³⁹ -- modified with 2 SeNP binding peptides (LTPHKHHKHLHA) attached (Figure 1, panel C).³³ Expression of GRLMR in cells grown in SeO₃²⁻ supplemented media results in a striking-red color, attributable to the red allotrope of α-SeNPs (Figure 1, panel D).⁴⁰

cSeNPs produce distinctive contrast in X-ray and electron imaging modalities, arising from the higher electron density of Se (Z = 34) relative the C, O, N, S, and P (6 < Z < 16) that dominate biological samples. SeNPs can react with transition metal cations, forming luminescent metal-selenide semiconductor quantum dots (Figure 1, panel E), facilitating fluorescence contrast.^41,42

Overall, the cSeNP that we present here is fundamentally different than previous strategies for identifying POIs in biological EM. The contrast is generated in vivo, which is not possible with cysteine rich protein/iNP development strategies and singlet oxygen generators. The strategy facilitates stoichiometric labeling, which is not straightforward with iNP-conjugates and encapsulin-based approaches. We demonstrate the approach in E. coli in this initial report. The visibility of cSeNP contrast in fluorescence and X-ray images means that this is the first cloneable contrast agent that produces contrast across all biological length scales. Because there are more mature and validated cloneable contrast strategies for fluorescence microscopy and CT imaging, the cloneable contrast facilitated by cSeNPs is most likely to be of interest to biological EM practitioners, both to identify POIs in electron micrographs and to find POIs in samples that are too thick for EM.

Methods

M1 - Whole cell in vivo SeNP formation expressing the cSeNP-FtsZ chimera

T7 Expression lysY/I^q Competent E. coli cells containing a pD441-CH expression vector encoding the N- or C-term tagged cSeNP-FtsZ chimera were grown in a shaking incubator at 225 – 275 rpm for 16 – 24 hours at 37 °C in 2 mL Miller’s Luria Broth (LB) containing kanamycin (25 μg/mL) and chloramphenicol (34 μg/mL) (Kan/Cam) antibiotics. After cells reach confluency, the culture was diluted 1/100^th by volume into 5 mL fresh antibiotic containing LB and allowed to grow 30 – 37 °C for an additional 3 – 6 hours to reach log phase. After an apparent optical density of 0.2 – 0.4 was reached, cells were treated with a final concentration of 1 μM – 10 mM SeO₃²⁻ and 1 μM – 1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) to promote nanoparticle formation and expression of the cSeNP-FtsZ construct.

After an induction period (30 – 360 minutes) cells were collected by centrifugation at 3000 xg at 4 °C for 3 minutes and washed 3-5x with ice cold filter sterilized 0.1 M phosphate buffer (PB) pH 6.9 – 7.0. Cells were then fixed in buffered 1.5 – 2.5% glutaraldehyde (25% Electron Microscopy Sciences) for 1 hour at 4 °C. A final wash 3 – 5x in ice cold buffer at 3000 xg at 4 °C for 3 minutes was done before reconstituting the sample into 100 – 200 μL buffer to either plate cells or prepare them for dehydration. Cells were drop-cast onto negatively glow-discharged C-coated 200 mesh Cu grids for 1 – 2 minutes followed by 2 washes with 5 μL filtered UpH₂O for 1 minute each.

M2 - Formation of AgSe and CuSe particles post-fixation using the cSeNP-FtsZ construct

Following SeNP formation, the sample was washed 4x in ice cold filtered 0.1 M phosphate buffer at pH 7 at 3000 xg at 4 °C for 3 minutes each. Samples were fixed using 1.5 – 2.5% buffered glutaraldehyde for 1 hour. Samples were then washed 3x with ice cold filtered 0.1 M acetate pH 5.6 at 3000 xg at 4 °C for 3 minutes each. Samples were then incubated with either AgNO₃ or Cu(OAc)₂ at a final concentration of 1 mM for 1 – 4 hours at 4 °C. Samples were washed 3x in acetate buffer and prepared for embedment. Controls did not contain Se supplement.

In another preparation of CuSeNPs, a “ramp up” protocol was performed. A starter culture was inoculated with 10 μM IPTG and 1 μM SeO₃²⁻ for an overnight low level induction. After subsequent dilution and growth to OD 0.4, we induced with 100 μM IPTG and 2 mM SeO₃²⁻ for 3 hours. For disproportionation reactions performed post-fixation, the sample was first fixed as described above, washed 3x with filtered 0.1 M acetate pH 5.6 and immediately soaked with 1mM Cu(OAc)₂ overnight at 4 °C. The control was treated identically except the [SeO₃²⁻] was kept at 1 μM.

M3 - Dehydration Protocol

Fixed cells were first encapsulated into 2-3% low melting temperature Agarose by centrifugation. Once firm enough to handle, 1 mm³ cubes were diced and placed in the dehydration solvent. Serial dehydration occurred in ethanol according to the following series with at least two solvent changes during each step: 5 minutes in 30%, 5 minutes in 50%, 10 minutes in 70%, 10 minutes in 80%, 10 minutes in 90%, 10 minutes in 100%, 10 minutes in 100%. Following traditional preparations, ethanol was replaced with propylene oxide for 1 hour with 1 change.

M4 - Embedment procedure

Following traditional embedment, the samples were transferred from propylene oxide to a 1:1 mixture of propylene oxide to catalyzed resin (Luft’s formulation of 10 g Araldite, 7.5 g DDSA, and 1.5% w/w DMP-30) for 1 hour with one solution change. The sample was moved to complete resin for 2 hours with one resin change, after which the sample was moved to molds for curing described below.

In a non-standard procedure, ethanol was used as the transition solvent. Because ethanol is not miscible with the resin, the sample was transferred to a 1:1 mixture of ethanol to uncatalyzed resin for 16 – 18 hours with constant agitation until all ethanol had evaporated to allow enough time for infiltration. The sample was transferred to 100% uncatalyzed resin for an additional 4 – 8 hours with one change, then transferred to complete (catalyzed) resin for 2 hours with one change. The sample was placed into either an Easy-Mold^™ or a BEEM size 3 capsule and set into a prewarmed oven to either cure for 16 hours at 60 °C or for 3 – 5 days at 42 – 45 °C as a low temperature alternative.

M5 - High Pressure Freeze Substitution

Alternatively, live cells were immediately spun down and plated into brass PELCO^® freezer hats, vitrified in a Wohlwend Compact 02 High Pressure Freezer, and transferred to a liquid nitrogen (LN₂) cooled freeze substitution cocktail composed of anhydrous 0.25% glutaraldehyde in ethanol. Samples were slowly brought up to −20 °C over the course of 4 – 5 days by surrounding the samples with dry ice in a Styrofoam box in a −20 °C freezer. The sample was removed from the freezer and placed in a refrigerator for 4 hours at 4 °C followed by 1 hour at room temperature. The samples were carefully washed 3x with molecular biology grade ethanol and removed from their freezer hats followed by embedment described above.

M6 - Formation of ZnSe particles in vivo using the cSeNP-FtsZ construct

To show that the construct could be used to form fluorescent ZnSe particles in live cells, we modified SeNP formation by doing a “fast particle” growth protocol. E. coli containing the cSeNP-FtsZ construct were grown to log phase as described above. At OD 0.4, cells were induced with 250 μM IPTG and 10 mM SeO₃²⁻ for 1 hour at 37 °C and 275 rpm. After, cells were spun down at 3000 xg at 22 °C for 3 minutes and fresh, room temperature LB containing antibiotics was used to replenish the decanted supernatant. Cells were resuspended and allowed to recover for 20 minutes at 22 °C with agitation. Following recovery, Zn(OAc)₂ was added to the media at a final concentration of 1 mM for an additional 2 hours. The control was treated identically to the sample, except no SeO₃²⁻ was added to the media.

Following the Zn addition, cells were washed 1x with 4 °C LB, then 2x 4 °C 0.1 M filtered PB pH 7 at 3000 xg for 3 minutes at 4 °C. The sample was then fixed with buffered 2.5% glutaraldehyde for 1 hour at 4 °C, washed 3x with 1 mL PB, and plated onto TEM grids for analysis.

M7 – Mock Tissue Preparation for microCT Analysis Using cSeNP as a Cellular Marker

For these studies, T7 express E. coli that contained only the cSeNP gene or wild type GRLMR gene were used. Starter cultures were grown at 37 °C 275 rpm overnight in LB with Kan/Cam. The next day, cultures were diluted 100x to 200 mL total volume of LB/Kan/Cam and grown for 2 – 3 hours at 37 °C 225 rpm until OD 0.4 was reached. Bacteria containing the enzyme were induced with 1 mM IPTG and either 10 mM, 2 mM, 1 mM, 500 μM, 250 μM, 100 μM, 10 μM, or 1 μM SeO₃²⁻ and allowed to grown overnight. The next day, samples were washed 3x with 10 mL ice cold, filtered 1x PBS pH 7.4 at 3000 xg for 5 minutes followed by fixation in 10 mL buffered 2.5% glutaraldehyde overnight at 4 °C. For the control, T7 express E. coli that did not contain the cSeNP gene (blank T7s) were treated identically to the samples made above, except these were not grown in either 10 mM or 2 mM SeO₃²⁻ to avoid cell stress. Additionally, the blank T7s were not exposed to kanamycin, as this antibiotic resistance is only present in the cSeNP plasmid.

In these studies, the goal was to simulate a tissue sample. Therefore, to maintain cell stability throughout the CT scan, cells were kept in glutaraldehyde. For data collection, mock tissue was prepared by pelleting cells in 2 mL Eppendorf tubes. From 100 mL of culture, a 500 – 750 mm³ volume of pelleted cells was collected and analyzed.

M8 – Micro-Computed Tomography (MicroCT) Analysis

Quantitative analysis of X-ray attenuation within the sample tubes was performed using Scanco μCT 80 analysis software suite (Scanco μCT 80, Scanco USA Inc. Wayne, PA; Scanco Medical Evaluator 1.1.15.0). A low density volume rendering was performed to segment material within a range of 0-400 Hounsfield units, and a higher density rendering was performed to segment material within a range of 800-1800 Hounsfield units.

Bone mineral density analysis done in the Scanco software suite against the standard hydroxyapatite (HA) phantom was used to quantify low (<400 HU) and high density (>800 HU) materials in a 55 mm³ volume of interest (VOI) of each sample. The volume chosen was within the same Z plane that contained material in both the control and SeNP containing samples (i.e. no air was captured in the analysis). In the 0 – 400 HU range, 60.5% of the control showed scattering intensity with respect to the total VOI, and 4% of the SeNP containing sample attenuated in that range. In the 800 – 1800 HU range, 0% of the control attenuated versus 77.7% of the SeNP sample with respect to the total VOI.

X-ray renderings (Figure 5) were produced by coloring voxels according to attenuation using the Scanco μCT 80 analysis software. Figure M1 in the supplementary material shows the attenuation curve and the HU range used for tomogram construction. We chose a control range of 0 – 400 HU (bracketed with white), representing attenuation from soft tissue, and a nanoparticle range from 800 – 1800 HU (bracketed with red) for the SeNPs.

Figure 5.

X-ray CT imaging of E. coli expressing cSeNPs. Panel A shows the X-ray attenuation (HU) of cells expressing the cSeNP (red) and cells not expressing the cSeNP (black) grown in the indicated concentrations of SeO₃²⁻. Panel B shows the attenuation of cells expressing cSeNP, with the experimental cells grown in 2 mM SeO₃²⁻ and control omitting Se. Panel C shows the attenuation for cells grown in varying concentrations of SeO₃²⁻; concentrations are 10 mM (red), 2 mM (pink), 1 mM (purple), 500 μM (green) and 1 μM (black). Panel D shows CT images of (1) GRLMR expressing cells in 10 mM SeO₃²⁻; (2) GRLMR expressing cells in 2 mM SeO₃²⁻; (3) GRLMR expressing cells; (4) cNP expressing cells in 10 mM SeO₃²⁻; (5) cSeNP expressing cells in 2 mM SeO₃²⁻; (6) cSeNP expressing cells; (7) in vitro synthesized SeNP powder; (8,9) empty tubes. Panel E shows a CT tomogram of sample 5. Se attenuation is represented in red, buffer attenuation in blue; This tube was centrifuged in a fixed-angle rotor, explaining the shape of the cell pellet. Panel F shows a tomogram of tube 6. The blue rendering indicates no Se in these cells. Scalebars in panels E and F are 1 mm.

M9 – Materials

T7 Expression lysY/I^q Competent E. coli cells (Cat no. C3013I) were purchased from New England Biolabs, Inc. Original plasmids containing GRLMR and F-cSeNP-His were designed and purchased from ATUM (DNA Twopointo, Inc.) in their pD441-CH expression vector. Luria Broth was purchased from Fisher Scientific. Kanamycin, Chloramphenicol, and IPTG were purchased from GoldBio. Sodium hydrogen selenite was purchased from Alfa Aesar. Molecular Biology grade ethanol and HPLC grade acetone were purchased from Fisher Scientific. Araldite 502 Kit; Araldite, EMbed 812 Kit; 25% aqueous EM grade glutaraldehyde, and 0.25% aqueous Toluidine Blue O Solution were purchased from Electron Microscopy Sciences.

Sterile snap cap culture tubes (max volume 12 mL) were used for initial growth and induction experiments. 15 mL falcon tubes were used during centrifugation and fixation. 1.5 mL microfuge tubes were used for fixation and encapsulation in agarose. 20 mL glass scintillation vials with screw tops were used during dehydration and embedment. BEEM^® size 3 Embedding Capsules were used during high pressure freeze substitution protocols, and either BEEM^® Flat Embedding Molds or Easy-Molds^™ were used for standard fixation and embedment procedures.

The PELCO easiGlow^™ Glow Discharge system was used to prepare EM grids. Either 200 mesh C-coated (5-6 nm thick) Cu grids or Cu slot grids coated in house-prepared Formvar were used for sample mounting.

M10 - Instrumentation, Equipment, and Software

EM Grid Preparation

200 mesh C-coated Cu grids were prepared for sample mounting by glow discharging in air for 45 seconds using the PELCO easiGlow^™ apparatus.

Ultramicrotomy

The blocks were removed from the oven and roughly trimmed under a dissection scope using injector blades followed by fine trimming using a glass knife made with a LKB Type 7801B KnifeMaker. A Reichert-Jung Ultracut E Microtome was used with either a DiATOME Ultra 35° Diamond Knife or a glass knife to make thin (50 – 100 nm) sections for EM analysis. In some instances, semi-thin (100 – 200 nm) sections were made for better elemental mapping.

Scanning Electron Microscopy (SEM) Imaging

A JEOL JSM-6500 Field Emission Scanning Electron Microscope with an In-Lens thermal field emission electro gun (TFEG) was used for SEM imaging. It is also equipped with an Oxford Instruments energy dispersive X-ray spectrometer (EDS) and uses Oxford Aztec software for qualitative and quantitative elemental analysis. STEM mode was used in bright field to collect all data by mounting an EM grid onto the STEM sample holder.

Transmission Electron Microscopy (TEM) Imaging and Energy Dispersive X-ray Spectroscopy Characterization (EDS)

A JEOL JEM-2100F Transmission Electron Microscope equipped with a Gatan Ultrascan 2K x 2K CCD camera to capture images in TEM mode and diffraction patterns was used. For STEM mode, a JEOL annular dark field (ADF) detector was used. For elemental analysis, an Oxford Instruments SSD EDS detector with Aztec software was used.

Electron Tomography Acquisition and Modeling

All data collection and tomogram reconstruction was done by Eileen O’Toole and Garry Morgan at the University of Colorado in Boulder. Tomograms were reconstructed from a tilt series collected from a 200 – 300 nm thick section using SerialEM and ETomo software. No post-fixation stain was applied and no gold fiducials were added. This means that there is no question that the contrast we are observing is from the intracellular particles and not from salts or fiducials added ex situ.

The tilt series were collected with SerialEM using 2° steps over 120° at 31 kX, 39 kX, and 59 kX magnification corresponding to 6.436 Å, 5.144 Å, and 3.338 Å pixel sizes respectively on a FEI Tecnai F30 FEG-TEM equipped with a Gatan OneView IS (4k) CCD at 300 kV. Tilts were reconstructed with the IMOD software package by inputting the Setup parameters into Etomo. Of particular note is that Etomo was able to align tilts using intracellular particles as the fiducial markers for the seed model. Following tomogram generation, a nonlinear anisotropic diffusion filter was applied to improve contrast, which is not usual but was necessary for our samples since they were not stained with heavy metals. The 3dmod package was used to semi-automatically segment the tomographic reconstructions to reproduce models of the nanoparticles and cell membrane. The script findbeads3d was used to automatically identify and segment nanoparticles. Example input:

$$\begin{gathered} \text{findbeads3d}\text{-}\mathsf{l}\text{CuSeF}\_\text{gF}4\_\text{cell5a}\_\text{rec}.\text{mrc}\_\text{nad}.\text{mrc}\text{-size}10\text{-}\mathrm{o} \\ \text{cell5a}\_\text{findbeads}.\text{mod} \end{gathered}$$

Fluorescence Microscopy

Fluorescence images were acquired on an inverted Olympus IX73 microscope (Olympus IX73) in an epi-fluorescence illumination set-up. Immediately before imaging, the sample was placed on a quartz slide and a coverslip was placed over the sample. The edges of the coverslip “sandwich” were sealed. The sample was placed on to the inverted microscope with the coverslip side facing down and brought into focus using transmitted white light. The sample was excited with a 50 mW 355 nm laser (Co-herent Genesis CX) and the emission was spectrally filtered (585 ± 40 nm band-pass). Photons emitted from the sample were collected through a water-immersion 60x microscope objective (Olympus) at either 1.6X or 2X magnification before passing through a long pass filter (cutoff wavelength: 364 nm). The pixel length at 60X corresponds to 216 nm. Emission was collected onto a 512 x 512 pixel electron-multiplying charge-coupled device camera (Andor iXon 897). Movies were recorded at an acquisition rate of 10 fps. Images were processed using ImageJ version 1.51j8 by importing the movie stacks and Z-projecting the average intensity value.

Results and Discussion

Cellular EM assessment of FtsZ-cSeNP expression in Escherichia coli.

To assess the cSeNP as a contrast agent in cellular electron microscopy, we fused DNA encoding the ‘filamentous temperature sensitive protein Z (FtsZ), a bacterial tubulin homologue, to cSeNP DNA. FtsZ forms protofilaments observed in two forms. In non-dividing cells, FtsZ is found in membrane associated dispersed helical protofilaments, comprised of 30 – 80 FtsZ protomers, extending to lengths of 100 nm. During cell division, protofilaments localize to the cleavage furrow of the dividing cell, forming a Z ring (Figure 2, panel A), which is implicated in generating mechanical force driving cell division.⁴³ Protofilaments on average lie 16 nm from the cellular inner membrane. Expression of cSeNP-FtsZ fusion proteins in E. coli should result in SeNPs organized by FtsZ at different points in the cell cycle as shown in Figure 2, panels A and B.

Figure 2:

Panel A shows a schematic of FtsZ locations in prokaryotic cells through the cell cycle. Panel B shows a schematic of FtsZ and FtsZ-cSeNP assembling into helical protofilaments. Panel C renders a portion of a transmission electron tomogram of putative FtsZ-cSeNP extracellular filaments. Panel D shows a segmentation of the filaments and cSeNPs from the tomogram in panel C, with filaments rendered in yellow and cSeNPs rendered in red. Additional particles are visible in this rendering that are not visiable in Panel C because the segmented rendering includes the entire tomographic volume (where some stray particles were apparent). Scale bar is 25 nm. Panel E shows E.coli cells grown in 2 mM SeO₃²⁻, expressing FtsZ-cSeNP which was fixed and imaged after drop-casting on a TEM grid. Due to FtsZ overexpression, this is an elongated cell with several cleavage furrows. Density at these cleavage furrows is attributed to accumulation of FtsZ-cSeNP. We also observe a longitudinal filament. FtsZ is known to form such filaments when overexpressed. A corresponding control made without SeO₃²⁻ addition reveals none of these densities (Figure S1, panel A). Panel F shows a STEM-EDS elemental map of a dividing cell. Se is shown in red, Phospohous in yellow.

We overexpressed FtsZ-cSeNP fusions in E. coli BL21 cells, as described in the Supplemental Information (Methods M1 and SD1.) The population of FtsZ protomers inside the cell, therefore, is comprised of both genomic FtsZ and recombinant FtsZ-cSeNP. FtsZ overexpression interferes with completion of cell division, resulting in cells up to 20 – 30 μm in length,⁴⁴ whereas wild-type cells are 3 – 5 μm. In such extended cells observations of helical protofilaments forming wave-like patterns along the membrane are known.⁴⁵

Identification of Conditions for Observation of FtsZ-cSeNP Fusions.

Optimization of SeO₃²⁻ supplementation, IPTG concentration, and growth times was accomplished with a high throughput, low resolution approach. In this approach, cells were fixed with glutaraldehyde and drop-cast on continuous carbon TEM grids. Full details are found in the SI, section D1. These experiments identified conditions where FtsZ-cSeNP expression resulted in punctate SeNP formation (Figures S2 – S7). FtsZ filamentation and function in vivo is sensitive to FtsZ expression levels, which are altered in our work. We optimized FtsZ expression by modulating induction levels with an IPTG screen (Figure S4). This work suggested that 100 – 250 μM IPTG was optimal for preserving both cell phenotype and SeNP-based contrast.

We found that solvents typically used in plastic embedding, such as acetone, dissolve SeNPs (Figures S8 – S10). We found that the SeNPs could be stabilized against dissolution, post-fixation, by adding Cu(OAc)₂ to the fixed cells. This treatment transforms amorphous SeNPs to crystalline CuSeNPs (Figure 1, panel E).^42,46 The resulting CuSeNPs resist solvent dissolution (Figures S6 and S11).

We made additional modifications to the conventional fixation and embedment protocols. We omitted the metal stains typically added to plastic sections that react with specific functional groups in biomolecules, such as osmium tetroxide (membrane staining) and uranyl acetate (nonspecific protein, nucleic acid staining). These metal stains were omitted because they may form granules that complicate assignment of cSeNP densities. The resulting tomograms are of lower than typical contrast. Note also that fixation and embedment is known to result in some ultrastructure distortions.⁴⁷

Cells grown under these optimized conditions were examined further. Figure 2 panels C and D show extracellular filaments, presumably from cells induced with 1 mM IPTG and 2 mM SeO₃² that ruptured in the preparation process. Figure 2, panel E shows cells grown in a high FtsZ-cSeNP expression condition (2.0 mM SeO₃²⁻,1 mM IPTG); Apparent in this image is that contrast increases at cleavage furrows, where FtsZ-cSeNP concentration is expected to be high. That contrast is attributed to FtsZ associated cSeNPs. Putative longitudinal FtsZ-cSeNP filaments are also observed, consistent with prior observations of longitudinal FtsZ filaments under high level FtsZ overexpression.^45,48,49 In a control experiment omitting SeO₃²⁻, contrast appears uniform across cells (Figure 1). Figure 2 panel F and figure S2 show elemental maps of Se in a dividing cell, collected by scanning transmission electron microscopy-energy dispersive X-ray spectroscopy (STEM-EDS). These images show that the density at the cleavage furrow is Se-rich, also consistent with organization of SeNPs by FtsZ.

Electron Tomography of cSeNP-FtsZ in E. coli.

We acquired a total of 11 electron tomograms (methods M10) of E. coli embedded in plastic sections. Figure 3 shows renderings and analysis of electron tomographic reconstructions of experimental and control cells. Figure 3, panel A shows 75 of 340 slices through an E. coli cell expressing cSeNP-FtsZ, grown in 2 mM SeO₃²⁻ for 3 hours. Two features are readily apparent: (1) the outer membrane and portions of the inner membrane; (2) punctate densities attributed to cSeNPs.

Figure 3.

Panel A shows a tomogram of E. coli expressing the cSeNP-FtsZ construct grown in 2 mM SeO₃²⁻ supplemented LB for 3 hours following the ramp up method described in ESI M2; 75 of 340 vertical slices comprise this 2D image. Visible are the outer membrane (OM), some remnants of the inner membrane (IM), and many punctate inclusions, (cSeNPs), noted with blue arrows. Panel B shows a magnified of the area within the orange box of panel A, with membrane segmented manually in blue. Panel C shows an isosurface auto-segmentation of the cSeNPs in the orange box. Panel D shows a segmented tomogram, with the membrane in blue and cSeNPs in red. Panel E shows a closer view of the area highlighted by the orange box in panel C. Panels F and G show tomograms of a cell grown in the absence of SeO₃²⁻, but otherwise handled identically. Panel H shows the isosurface auto-segmentation of the area in the green box of panel F. Panel I shows a histogram of particle to membrane distances for the cell in panel A. Panel J shows a tomogram of a dividing cell, with the membrane segmented in blue. Panel K and L show two magnified views of sections of the dividing cell. Areas appearing to be filaments are marked with yellow triangles. Panel M shows segmented filaments in yellow, with arrows pointing to the particles noted in panels K and L. Panel N shows a fully segmented image of the dividing cell. Panels O and P show a side on view of the dividing cell, with the membrane hidden in panel P. Panel Q shows a view of the dividing cell where a small window between the two dividing cells remains. Scale bar = 100 nm for A, D, F, J, N, O, and P. Scale bar = 25 nm for C, H, M, and Q.

Figure 3, Panel A shows a slice through the tomogram, where a few cSeNPs are highlighted. Panel B shows a magnified view of the boxed area panel A, with the outer membrane segmented in blue. The cSeNPs have sufficient contrast for automatic segmentation using the findbeads3d script in the IMOD software package.^50,51 This approach produces the isosurface rendering of cSeNPs shown in panel C as magenta objects. Figure 3, panel D shows a manual membrane and particle segmentation of a 250 nm thick tomogram; figure S1, panel E shows automatic nanoparticle segmentation of the same cell. Figure 3, panel E shows a magnified image of the boxed area of panel D.

Panels F, G, and H show images of a control cell, grown without SeO₃²⁻, but otherwise treated identically to the cell in panel A (including Cu²⁺ exposure). There are no obvious punctate densities in this tomogram. Automatic segmentation does not identify voxels of intensity comparable to those in experimental cells. We previously published tomograms of cells expressing unmodified GRLMR grown in SeO₃²⁻ supplemented media; In this ‘control’ experiment, we observed intracellular SeNPs with diameters ranging between 5 and 60 nm that were randomly distributed.⁵²

FtsZ is a membrane associated protein. Based on prior studies it is expected to be between approximately 16 nm and 35 nm from the inner and outer membranes, respectively, throughout the cell cycle. An analysis of cSeNP locations relative to the outer membrane suggests that 94% of observed cSeNPs are found in expected locations relative to the membrane.

This analysis accounts for the size of the cSeNP protein component (~10 nm in maximum dimension). Because the protein component may increase the distance between FtsZ and the observable cSeNP and because of dehydration artifacts causing part or all of the periplasm to collapse, FtsZ associated cSeNPs are expected to be found between approximately 16 and 45 nm from the outer membrane. Figure 3, panel I shows a histogram of cSeNP – outer membrane distances for the cell in panel A. Of 104 observed cSeNPs, 98 are within 45 nm of the outer membrane. Of those, 37% are closer than 16 nm to the outer membrane. We attribute particles closer than 16 nm to the outer membrane to membrane fixation artifacts, which are well-known with glutaraldehyde fixation. Therefore, over 90% of observed cSeNPs lie in FtsZ associated locations.

Figure 3, panels J through Q, show a variety of renderings of a tomogram of a cell nearing completion of division. Panels K and L show renderings where positive contrast putative FtsZ filaments appear. Panel M shows a rendering of FtsZ filaments (yellow) decorated with cSeNPs. Analysis of 6 additional tomograms, including one additional control, are available as electronic supporting information (Figure S1). All tomograms show similar results.

Overall, electron tomography shows that cSeNP-FtsZ expression results in cSeNPs in locations where FtsZ is expected. Evidence for this includes (1) contrast attributable to cSeNPs in FtsZ locations in high-throughput 2D-samples (Figure 2, panel E); (2) Se elementally mapping to FtsZ locations (Figure 2, panel F); (3) punctate densities in locations expected for FtsZ in 3D cellular reconstructions (Figure 3, panels A-E, J-Q); (4) absence of such contrast features in control samples.

Correlative Light and Electron Microscopy (CLEM) of cSeNP-FtsZ expressing E. coli.

SeNPs can react with transition metal cations such as Zn²⁺ and Cd²⁺, forming fluorescent metal selenide nanoparticles (e.g., quantum dots).^41,42 This suggests the possibility of the cSeNP as a cloneable CLEM contrast agent. We assessed this possibility by growing cSeNP-FtsZ expressing E. coli first in 10 mM SeO₃²⁻ supplementation, followed by a change of media and subsequent growth in ZnOAc supplemented media, as described in the Methods section M6.

Figure 4 shows CLEM images of these cells. Panels A and B show fluorescence and electron images, respectively, of the same cells. Panel C shows an overlay of correlated light and electron images. Panels E, F and G show higher magnification images of panels A, B, and C. Panels D and H show fluorescence and electron images of control cells grown without SeO₃²⁻ but with Zn(OAc)₂.

Figure 4.

CLEM images of E. coli expressing cSeNP-FtsZ fusions. Panels A and B show fluorescence and electron images, respectively of cells grown in 2 mM SeO₃²⁻. Panel C shows an overlay of panels A and B. Panel D shows a fluorescence image of cells grown without SeO₃²⁻. Panels E, F, and G show magnified images of panels A, B and C. Panel H shows an electron image of cells grown without SeO₃²⁻.

Fluorescence images of cSeNP-FtsZ expressing cells treated with Zn(OAc)₂ show brightest emission at the cleavage furrow of dividing cells, consistent with FtsZ-based contrast. The uniform background fluorescence is interpreted as arising from FtsZ-cSeNPs that are dispersed in the cell away from the cleavage furrow. Corresponding electron images also show the highest electron contrast at cleavage furrows. When overlayed, regions of high electron contrast overlap with areas of high fluorescence emission.

The higher magnification electron microscope image in panel F shows E. coli cells with contrast densities at the midpoint of 3 dividing cells and at the poles of 3 other cells. We attribute these densities to cSeNPs localized by FtsZ. The fluorescence images of panel E show a relatively uniform fluorescence across the entire cells, with brighter emission at the midpoints of 2 of the dividing cells and at the pole of one of the other cells. The overlay image of panel G shows regions of high florescence correspond uniformly to the locations of SeNP density. The control fluorescence images in panel D show typical cellular autofluorescence.

Overall, the Zn(OAc)₂ treated cSeNPs produce clear fluorescence emission above cellular autofluorescence, which becomes more intense in areas of expected FtsZ concentration. Combined, this set of experiments suggests that the cSeNP, after Zn treatment, is a promising approach for CLEM contrast. Furthermore, because Zn was added in vivo, cZnSeNPs may be an effective tool for cryo-CLEM based methods. We do note 2 current shortcomings for the use of cZnSeNPs for fluorescence based imaging. First, all biogenic ZnSeNPs have quantum yields of less than 10%, because the highly fluoresencent core-shell quantum dot structures can not yet be synthesized with biogenic approaches. Second, it appears that not all SeNPs convert to become fluorescent ZnSeNPs, meaning that some SeNPs remain ‘dark’ in fluorescence imaging. This is apparent in Figure 4 (more dark inclusions in EM data than bright spots in fluorescence data). This may be due to variations in intracellular Zn regulation.

cSeNPs as genetically encoded X-ray contrast.

X-ray imaging modalities such as Computed Tomography (CT) provide images at length scales of meters with penetration deep into tissues and organisms. X-ray attenuation in CT is quantified with the Hounsfield Unit (HU). The HU scale is defined by the radiodensity of distilled water (0 HU) and air (−1000 HU). For biological tissues, typical HU values range from −700 (lungs) to +1900 (bone), with most soft tissues (fat, muscle, organs) having HU values between −120 and 60. Inorganic materials attenuate X-rays more effectively than biological tissues; Pure copper attenuates X-rays at 14,000 HU. We calculate that pure Se attenuates X-rays at 48,000 HU. In the case of cSeNPs, X-ray attenuation depends on particle concentration (e.g., the measured HU values will be a mixture of biomatter and SeNPs).

X-ray tomograms of E. coli cells expressing the cSeNP or GRLMR grown in varying concentrations of SeO₃²⁻ were acquired in a Scanco micro-CT 80 instrument. CT data is presented in figure 5. Panel A shows that cells expressing the cSeNP in varying growth conditions attenuate X-rays with an efficiency that depends on the SeO₃²⁻ concentration. Attenuation of 700 HU is observed for cells grown in 1 mM SeO₃²⁻. Notably, cells grown in the same concentrations of SeO₃²⁻ but not expressing the cSeNP never attenuate more than 100 HU, providing additional evidence that endogenous SeNP formation is a relatively slow process. Panel B shows a voxel-parsed trace of X-ray attenuation. Here we observe that cells expressing the cSeNP grown in 2 mM SeO₃²⁻ show voxels with minor attenuation peak at 190 HU and a major attenuation peak at 1100 HU. Corresponding controls (not grown in SeO₃²⁻) show only an attenuation peak near 100 HU, comparable to soft tissue. Panel C shows the attenuation of cSeNP expressing cells grown in concentrations of SeO₃²⁻ ranging from 1 μM to 10 mM. Like panel A, attenuation increases with the concentration of SeO₃²⁻ in which the cells were grown. Panel D shows a 9-position microcentrifuge-tube holder. Whiter color in these images corresponds to higher X-ray attenuation and thus Se absorption. Panels E and F show tomograms of the samples in tubes 5 and 6. They are colored according to attenuation, with red representing attenuation near 1100 HU and blue representing attenuation near 100 HU, as detailed in methods M8. Overall, these tomograms show that cells expressing the cSeNP show clearly distinguishable X-ray contrast relative to cells grown in the same concentrations of SeO₃²⁻ that do not express the cSeNP.

Toxicity and other physiological impacts of cSeNPs.

In this work, we supplement cell cultures with SeO₃²⁻ to provide Se equivalents needed for intracellular cSeNP formation. A reasonable question is whether Se supplementation alters underlying observable biology or causes toxic effects?

Se is an essential element. Its main biological functions are found in selenocysteine (sometimes referred to as the 21^st amino acid) containing enzymes. Selenocysteine is found in the active sites of ubiquitous enzymes involved in redox homeostasis such as glutathione peroxidase and thioredoxin reductase. In the absence of Se, redox homeostasis is diminished, and cells grow slowly, if at all. At ‘essential’ concentrations, Se plays an anti-oxidant role. At high concentrations, Se becomes a pro-oxidant and exerts toxic effects (e.g., through free radical generation, which can oxidize DNA and membranes).

In the present work, we optimized SeO₃²⁻ supplementation concentrations. We built on our prior work showing that recombinant GRLMR expression alters selenium homeostasis in E. coli.³⁵ Overexpression of GRLMR makes E. coli significantly more tolerant of SeO₃⁻² supplementation – raising the concentration of SeO₃²⁻ that inhibited 90% of cell growth (IC₉₀) from ~2 mM to ~20 mM. GRLMR expression also resulted in a requirement to supplement cultures with ~10 μM SeO₃²⁻ to sustain normal cell growth.³⁵

SeNPs are known to form spontaneously inside wild-type E. coli cells grown in Se supplementation.⁵³ This suggests an alternative interpretation of our observations is that some or all observed SeNPs were made by endogenous processes of E. coli—not by the recombinantly expressed cSeNP-FtsZ proteins. Our data in aggregate suggests that the SeNPs observed inside cells are not due to endogenous SeNP formation inside E. coli, but can rather be attributed to recombinant cSeNP production as outlined below.

First, we note that endogenous SeNP formation appears to be a slow process compared to the enzymatic process we developed. Previous reports of SeNPs in wild-type E. coli require 72 hours of growth in Se supplemented media resulting in 100 nm or larger SeNPs. In our work, we observe 5 nm diameter cSeNPs after a maximum of 4 hours of growth in Se supplemented media, suggesting that the process we are observing is quite distinct from endogenous SeNP formation. Other evidence suggesting that endogenous SeNP formation is minimal in our experimental approach is in figure S4 (IPTG induction screen) and Figure 5a (CT of cells). Further, in figure 5, panel A, cells grown in 2 mM Se show dramatically different X-ray contrast (SeNP accumulation) depending upon whether the cSeNP is expressed. X-ray contrast scales with Se concentration when the cSeNP is expressed, and X-Ray contrast does not scale in this way when the cSeNP is not expressed.

The overall conclusion from our prior work is that recombinant GRLMR overexpression alters Se homeostasis, such that the E. coli require Se supplementation to thrive, and the cells also tolerate an order of magnitude more SeO₃²⁻ before growth is significantly impaired. Our study of intracellular SeNPs in E. coli leads to a hypothesis that SeNPs are a general feature of Se homeostasis in E. coli. This hypothesis is directly analogous to how cells handle excess of other essential inorganic elements, such as iron and phosphorous. Excess iron is stored in a bio-unavailable form in ferritin; Similarly excess phosphate is stored in bio-inert inorganic polyphosphate nanoparticles.⁵⁴ We suggest that excess ionic (bioavailable) Se may likewise be segregated and stored in bio-inert SeNPs – and that Se toxicity is observed when concentrations of Se overwhelm the cells innate capability for storage of Se in SeNPs.

In the present work, the majority of data was collected on E. coli supplemented with 2 mM Se when expressing GRLMR. In a few cases (e.g., experiments in figure 4), we used up to 10 mM Se supplementation – however in these higher SeO₃²⁻ conditions, we limited exposure to Se to 1 hour, and followed with ‘recovery’ in Se deprived media to minimize impacts of this high concentration of Se.

In the Se supplementation conditions used in the present work, we observe no signs of acute Se toxicity. Cellular growth rates are normal. Cell morphologies are as expected (and altered significantly from normal due to the overexpression of FtsZ). Further, we don’t observe stress granules or other signs of cellular stress. While there may be more subtle effects of Se supplementation on underlying biology than we’ve observed so-far, our observations in aggregate suggest that Se supplementation is not significantly altering underlying biology.

Downstream Use Cases:

We show in this work the use-case of the cSeNP for revealing FtsZ locations in an E. coli model system. We expect that the cSeNP will function widely in E. coli for revealing subcellular protein locations at electron microscope resolutions. We are presently assessing the cSeNP in other model systems (Drosophila, Caulobacter, and some in vitro reconstituted systems). Understanding the interplay of Se supplementation and GRLMR overexpression on Se homeostasis in these systems is a key consideration. For widespread adoption, the cSeNP will need to function in all kinds of cells, analogous to the widely used fluorescent proteins. At this stage, we do not suggest that the cSeNP is ‘perfect’ and it is expected to cause some perturbance on underlying biology in many use cases, although we do not observe significant perturbance in the present study.

The “Cloneable labeling” approaches that find widespread success (even Nobel prize recognition in 2008) – namely the Cloneable Fluorophores exemplified by GFP – are imperfect labels. Even for GFP, there is evidence of some toxicity and immunogenicity.⁵⁵ GFP has long been known to dimerize with itself, provoking undesirable aggregation of GFP labeled proteins. GFP can also impact the subcellular localization of GFP tagged proteins. For instance, one study showed that C- vs. N-terminal labeling of proteins with GFP impacted the subcellular localization of about 10% of proteins in yeast.⁵⁶

Given the need for cloneable contrast, especially in cellular electron microscopy, even imperfect tags are (1) likely to find community adoption and (2) form a basis for improved iterations/ generations of cloneable nanoparticles that reduce observed artefacts. (E.g., recent iterations of GFP are much brighter emitters and largely eliminate the dimerization problems).

Conclusions

Expression of the cSeNP in an E. coli model system produces molecular contrast in electron and fluorescence microscopies, as well as cellular contrast in X-ray imaging (where resolutions are not molecular). This paper demonstrates a ‘proof of concept’ for cSeNPs as multi-modal cloneable contrast agents, using E. coli, a workhorse laboratory model-organism. What is true in E. coli, typically translates to more complex model and experimental systems. As a general outlook, showing generality in E. coli (beyond just FtsZ labeling) and translation into other model organisms represent necessary future steps to demonstrate widespread utility.

Supplementary Material

Supplementary data including tomograms, additional elemental analysis, descriptions of optimization of [SeO₃²⁻], resin embedment, and SeNP dissolution details. Tomograms were deposited in the Electron Microscopy Data Bank, under accession numbers [pending].