Imaging beyond the proteome

Abstract

Imaging technologies developed in the early 20th century achieved contrast solely by relying on macroscopic and morphological differences between the tissues of interest and the surrounding tissues. Since then, there has been a movement toward imaging at the cellular and molecular level in order to visualize biological processes. This rapidly growing field is known as molecular imaging. In the last decade, many methodologies for imaging proteins have emerged. However, most of these approaches cannot be extended to imaging beyond the proteome. Here, we highlight some of the recently developed technologies that enable imaging of non-proteinaceous molecules in the cell: lipids, signalling molecules, inorganic ions, glycans, nucleic acids, small-molecule metabolites, and protein post-translational modifications such as phosphorylation and methylation.

Introduction

Molecular imaging is a powerful tool that has enabled the visualisation of biomolecules as they function in their native setting.¹ The ability to monitor biological events in real time at the subcellular level has furthered our understanding of many physiological processes, including protein trafficking, protein localisation, and protein–protein interactions.² Arguably, the most widely used set of tools in molecular imaging are fluorescent proteins. The discovery and development of the green fluorescent protein (GFP) by Shimomura, Chalfie, and Tsien, who were awarded the 2008 Nobel Prize in Chemistry for their efforts, has enabled the tagging and imaging of many proteins of interest.³

Although imaging of target proteins using fluorescent protein fusions has revolutionised many areas of biology, extension of this strategy to other components of the cell such as glycans, lipids, and nucleic acids has remained challenging. While proteins comprise the largest fraction of biological molecules in the cell, non-proteinaceous biomolecules also play important roles in cell biology.⁴ Thus, the ability to directly visualise all the components of the cell would allow for a more comprehensive understanding of cellular biochemistry. Here, we discuss the development of emerging technologies that enable imaging beyond the proteome, namely protein post-translational modifications such as glycosylation, phosphorylation, methylation, and lipidation, as well as other classes of biomolecules, including lipids, glycans, nucleic acids, small-molecule metabolites, and inorganic ions.

1. Fluorescent proteins

Traditionally, fluorescent proteins have been genetically fused to the gene that encodes the protein of interest in order to produce a chimeric protein that contains the fluorescent protein at its N- or C-terminus.⁵ Beyond the now routine use of this strategy to image proteins, several groups have applied fluorescent proteins to image lipids, signalling molecules, and post-translational modifications in the cell. Through protein-small molecule interactions, certain proteins can faithfully report on the location of non-proteinaceous biomolecules in the cell, perhaps most notably many classes of lipids.

1.1 Genetically-encoded probes for imaging membrane lipids

Membrane lipids, such as diacylglycerol, phosphoinositides, and phosphatidylserine, are important regulators of cellular homeostasis and many signal transduction pathways.⁶ These lipids are predominantly found in the inner leaflet of the plasma membrane and on the cytosolic face of organelle membranes and are responsible for recruiting a wide variety of proteins and initiating a variety of signalling pathways.⁷ Fluorescent protein fusions to the lipid-binding domains of these recruited proteins have been used to study lipid dynamics in live cells using fluorescence microscopy.^8,9 For example, the pleckstrin homology domains of phospholipase C δ and Akt have been used to image the phosphoinositides PtdIns(4,5)P₂ and PtdIns(3,4,5)P₃, respectively.⁷ Fluorescent protein fusions of other lipid-binding domains have enabled imaging of other phosphoinositides and membrane lipids as well.⁷ This methodology has been instrumental in determining the localisation of these important lipids during various biological processes.

1.2 Genetically-encoded and Förster resonance energy transfer (FRET)-based sensors of signalling molecules

In addition to their contributions toward developing the fluorescent protein toolkit, Tsien and co-workers have also developed a FRET-based reporter system for imaging secondary messengers.³ They were initially interested in visualising cyclic adenosine 3′,5′-monophosphate (cAMP), a small molecule that is involved in intracellular signal transduction. Their first FRET reporter consisted of fluorescein-labeled catalytic subunits of cAMP-dependent protein kinase (PKA) and rhodamine-labeled regulatory subunits of PKA.¹⁰ Upon binding of cAMP, the regulatory subunit of PKA dissociates from the catalytic subunit, thus causing a loss in FRET between the two labels. Despite the utility of this probe, Tsien hoped to replace the small-molecule organic fluorophores with fluorescent proteins in order to enable imaging of other metabolites as well as proteins through genetic fusion.³ Using mutagenesis, they were able to develop GFP variants with enhanced fluorescence as well as fluorescent proteins of different colors. They discovered that the yellow fluorescent protein (YFP) is a good FRET acceptor for the cyan fluorescent protein (CFP). Following this development, Pozzan and co-workers teamed up with the Tsien laboratory to replace the two fluorophores with CFP and YFP.¹¹ In their work, they demonstrated that this fluorescent protein FRET-based reporter can be used to image cAMP in stimulated rat cardiac myocytes.¹² More recently, Dyachok et al. have developed a FRET-based sensor to measure cAMP dynamics in pancreatic beta cells using ratiometric evanescent wave microscopy.^13,14 Other applications of cAMP imaging probes have been comprehensively reviewed elsewhere.¹⁵

Tsien and co-workers also developed fluorescent protein FRET-based sensors of Ca²⁺, one of the most important secondary messengers in cell biology.¹⁶ Dubbed ‘cameleons,’ these indicators consisted of fusions between CFP, the Ca²⁺-binding protein calmodulin (CaM), the CaM-binding peptide M13, and YFP (Fig. 1).^17–20 Increased levels of Ca²⁺ causes CaM to bind to M13, thus changing the distance between the two fluorescent proteins and causing an increase in FRET. More recently, an improved CaM-based FRET indicator was used to detect single action potentials in neurons from brain slices and in vivo, a feat which was not possible with previously existing Ca²⁺ sensors.^21,22 The success of these probes, however, is limited by both sensitivity and impaired targeting efficiency due to the large size of the chimeric probe.²³ Thus, alternative fluorescent protein-based Ca²⁺ sensors have been developed. One class of second-generation probes, named ‘camgaroos,’ is based on a single YFP which is pH-sensitive.¹⁹ Another second-generation Ca²⁺ sensor, named ‘pericam’ because it is a circularly permuted version of YFP, enables ratiometric measurements, or the simultaneous recording of two distinct fluorescent signals that can allow for quantitative readouts.²⁴ A more recent improvement to this probe known as yellow cameleon-nano was used to image subtle changes in Ca²⁺ dynamics.²⁵ Imoto and co-workers have developed a high-affinity Ca²⁺ probe composed of a single GFP fused to CaM and M13 known as GCaMP.²⁶ Others have developed improved versions of GCaMP termed GCaMP2 and GCaMP3. GCaMP2 has been used to image Ca²⁺ in the murine heart²⁷ and to measure synaptic activity in zebrafish.²⁸ GCaMP3, a more recent improvement to GCaMP2, is brighter, has a greater dynamic range, and has a higher affinity for Ca²⁺.²⁹ Recently, the palette of genetically-encoded sensors for Ca²⁺ was expanded to include improved green and red fluorescent protein indicators.³⁰

Fig. 1.

Genetically-encoded fluorescent protein FRET reporters.

Fluorescent protein-based FRET reporters for sensing glutamate were also developed by Tsien and co-workers as well as Frommer and co-workers. Glutamate is the major excitatory neurotransmitter in the brain, and monitoring its levels could provide insight into many neurological processes. Okumoto et al. have developed a FRET-based sensor consisting of CFP fused to both a clamshell-shaped bacterial glutamate periplasmic binding protein (YbeJ) and YFP (Fig. 1).³¹ Hires et al. have improved this sensor by optimizing the linker sequences and glutamate affinities to enhance its signal-to-noise ratio.^32,33

A genetically-encoded fluorescent indicator of intracellular H₂O₂ was developed by Lukyanov and co-workers.³⁴ H₂O₂ is a reactive oxygen species (ROS) that is speculated to be involved in paracrine signalling.³⁵ In their design, they fused a bacterial H₂O₂-sensitive transcription factor (OxyR) to a circularly permuted YFP. Cysteine oxidation of the OxyR portion induces a conformational change that changes the emission profile of the fluorescent protein. This probe has been used to image peroxide production during wound healing in zebrafish.³⁶

1.3 Genetically-encoded and FRET-based sensors of protein post-translational modifications

Post-translational modifications of proteins are important regulators of protein function.³⁷ Genetically-encoded FRET probes based on the initial design pioneered by Tsien and co-workers have been constructed for monitoring post-translational modifications such as phosphorylation, methylation, acetylation, and glycosylation (Fig. 1).^38,39

Protein phosphorylation by protein kinases is key for the activation of numerous signal transduction pathways.⁴⁰ FRET-based probes for phosphorylation, which sense kinase activities, consist of CFP and YFP fused to a consensus substrate for the relevant kinase and a binding domain.^38,41 These sensors have been used to detect protein kinase A,⁴² protein tyrosine kinase,⁴³ Akt/protein kinase B,^44,45 and protein kinase C⁴⁶ activities. Improved versions of these sensors with better specificity and reversibility have enabled visualisation of protein phosphorylation by the tyrosine kinase SrcA during cell mechanotransduction as well as monitoring of protein kinase A activity during insulin signalling.^47,48 More recently, Kapoor and co-workers have examined the dynamics of protein phosphorylation by aurora B kinase, a key mitotic regulator, using FRET-based sensors in live cells.⁴⁹ There have also been reports of genetically-encoded sensors with enhanced sensitivity and optimized signal-to-noise ratios for fluorescence lifetime imaging.^50,51

Histone modifications play important roles in regulating gene expression. Ting and co-workers have developed CFP/YFP reporters of histone phosphorylation and methylation, which are important modifications that regulate transcription of nearby genes.^52–54 More recently, changes in histone acetylation were monitored with a FRET-based reporter.⁵⁵ Similarly, Mahal and co-workers have developed a FRET-based sensor of protein O-GlcNAcylation, a reversible form of glycosylation of intracellular proteins akin to phosphorylation.^56,57 Development of this probe has allowed for dynamic monitoring of O-GlcNAc modifications in live cells.

2. Small-molecule sensors

Small-molecule indicators can be advantageous relative to genetically-encoded ones because they often exhibit greater dynamic ranges and increased sensitivity.⁵⁸ They also tend to exhibit faster response kinetics. However, unlike genetically-encoded sensors, with few exceptions they cannot be localised or specifically targeted to a particular organelle in the cell by fusion to the protein of interest.¹⁶

Small-molecule fluorophores have been developed for sensing metal ions such as calcium, zinc, copper, and iron. Metal ions are interesting targets for imaging because they play critical roles in cell biology.⁵⁹ The design of small-molecule sensors for metal ions is challenging because the probe must be selective for the specified metal over other biologically abundant cations, including those that exist at much higher cellular concentrations such as Na⁺, K⁺, and Mg²⁺. Thus, the design of metal-selective probes requires the successful application of principles of coordination chemistry.

2.1 Fluorescent Ca²⁺ indicators

The first small-molecule fluorescent Ca²⁺ sensor was developed by Tsien.⁶⁰ In this pioneering work, he replaced the two methylene groups of a well-known Ca²⁺ chelator with two benzene rings that enable the molecule to function as a chromophore (BAPTA, Fig. 2). Since then, many low- and high-affinity dyes for Ca²⁺-sensing have been developed that emit UV and visible light. Some of these probes include the fluorescein derivatives Fluo-3, developed by Tsien and co-workers, and Fluo-4, a brighter, more photostable derivative of Fluo-3, developed by Molecular Probes (Eugene, OR) (Fig. 2).⁵⁸ Indo-1 and Fura-2 are ratiometric dyes that are widely considered the standard for quantitative intracellular Ca²⁺ measurements (Fig. 2).⁶¹ Red-shifted Ca²⁺ indicators based on the rhodamine scaffold, including Rhod-2, have also been developed to minimize autofluorescence from biological samples (Fig. 2).^61,62 Tour et al. have combined the concept of a small-molecule sensor of Ca²⁺ and the localisation that the tetracysteine motif and FlAsH/ReAsH reagents provide by developing a biarsenical Ca²⁺ indicator to probe Ca²⁺ levels around specific proteins.⁶³

Fig. 2.

Small-molecule fluorescent Ca²⁺ sensors.

2.2 Small-molecule Zn²⁺ indicators

In the past century, Zn²⁺ has emerged as an important regulator of protein function and signalling.^64–66 It has also been implicated in the pathophysiology of several neurodegenerative disorders, including Alzheimer’s disease.⁶⁷ Thus, the ability to image Zn²⁺ using small-molecule sensors could elucidate metal ion homeostasis in relation to disease.

Several groups have developed Zn²⁺ sensors that were inspired by analogous probes for Ca²⁺ detection. Gee et al. have removed a single acetate arm from the Ca²⁺ chelator of Fluo-3 to create FluoZin-3, which maintains high affinity for Zn²⁺ (Fig. 3).⁶⁸ Lippard and co-workers have developed several classes of fluorescein-based sensors for Zn²⁺ that are modified with a Zn²⁺ chelator on the xanthene core. These include the Zinpyr,^69–73 Zinspy,⁷⁴ and Quinozin⁷⁵ series of probes (Fig. 3). An improvement to the Zinpyr family of dyes, known as ZPP1, exhibits lower background fluorescence and higher fluorescence turn-on when complexed to zinc.⁷⁶ This probe was recently used to image high levels of Zn²⁺ in a mouse model of prostate cancer.⁷⁷ Nagano and co-workers have developed fluorescein-based Zn²⁺ sensors that are functionalised at the phenyl ring of the fluorophore-scaffold.⁷⁸ These include the ZnAF series (Fig. 3). In addition to fluorescein-based reagents, BODIPY,⁷⁹ cyanine,^80,81 and coumarin^82,83 probes for sensing Zn²⁺ in biological systems have also been developed.

Fig. 3.

Small-molecule fluorescent Zn²⁺ sensors.

Several ratiometric sensors of Zn²⁺ that capitalize on different mechanisms have been synthesized for quantitative measurements of Zn²⁺ concentrations in vivo. FuraZin and IndoZin are ratiometric Zn²⁺ sensors derived from the Ca²⁺ probes Fura and Indo, which operate via an internal charge transfer mechanism.⁸⁴ Nagano and co-workers have also developed the ratiometric Zn²⁺ sensors ZnAF–R1 and –R2, which are based on the Fura scaffold (Fig. 3).⁸⁵ Taki et al. have developed ratiometric sensors known as the Zinbo series that exploit excited-state intramolecular proton transfer processes in benzoxazole scaffolds.⁸⁶ Lippard and co-workers have also developed ratiometric Zn²⁺-sensors that rely on a two-fluorophore system in which ester-mediated hydrolysis liberates coumarin as an internal standard.⁸⁷ In addition to fluorescent sensors, Zn²⁺ sensors that rely on alternative imaging modalities such as magnetic resonance imaging (MRI) have been developed.^88–90

2.3 Small-molecule Cu⁺ and Fe²⁺/Fe³⁺ indicators

Oxidative stress is thought to cause neuronal cell damage that can eventually lead to pathologies such as Alzheimer’s disease, Parkinson’s disease, and amyotrophic lateral sclerosis.⁹¹ Redox-active metals such as Cu⁺ and Fe²⁺ have been implicated in causing oxidative stress by producing ROS such as H₂O₂, superoxide (O₂⁻•), and hydroxyl radicals (•OH).⁹² Therefore, there has been much interest in imaging not only the ROS themselves (see sections 1.2 and 2.4) but the metal ions as well.

Fluorescence detection of copper using small-molecule sensors is difficult because Cu⁺ is the major active species within the reducing environment of the cytosol; however, this ion can readily disproportionate to Cu²⁺ and Cu⁰. In addition, both Cu⁺ and Cu²⁺ are capable of quenching the fluorescence of many fluorophores.⁶⁴ Currently, there are a few reports of fluorescent Cu⁺ sensors that have been successfully used for visualizing Cu⁺ in live cells. Yang et al. have developed a sensor known as CTAP-1 (Fig. 4A), which comprises a pyrazoline dye platform appended to an azatetrathiacrown receptor for metal coordination.⁹³ A more water-soluble version of this probe, CTAP-2, was recently described.⁹⁴ Chang and co-workers have developed a Cu⁺ indicator known as CS1 (Fig. 4A) that combines a BODIPY dye and a thioether metal chelator.^95,96 Since then, a ratiometric fluorescent sensor based on this dye has been described, as well as a mitochondrial targeted version that allows for imaging of mitochondrial pools of copper.^97,98 A second-generation probe with different substituents on the boron center of the BODIPY dye known as CS3 was used in combination with X-ray fluorescence microscopy to image labile copper pools in neurons.⁹⁹ Recently, a near-infrared copper sensor, CS790, was reported for imaging copper in a murine model of Wilson disease.¹⁰⁰ Chang and co-workers have also developed an MRI-based sensor for copper.^101,102

Fig. 4.

Small-molecule fluorescent Cu⁺ and Fe²⁺/Fe³⁺ sensors.

Like copper, the major forms of iron in the cell, Fe²⁺ and Fe³⁺, also exhibit significant quenching capabilities, and designing fluorogenic sensors of iron is thus difficult as well. As a result, all of the sensors for detecting Fe²⁺/Fe³⁺ in cellular systems are turn-off probes. For example, calcein (Fig. 4B) is a commercially available fluorophore that uses a metal chelator for turn-off Fe²⁺ detection.⁶¹ Phen Green SK (Fig. 4B) is another commercial probe that can detect both Fe²⁺ and Fe³⁺.⁶¹ One drawback of these probes, however, is that they are only partially selective for iron and can detect other metal ions as well.¹⁰³ The most specific sensors for Fe³⁺ rely on receptors derived from siderophores, which are small, high-affinity iron chelating compounds secreted by bacteria, that can be conjugated directly onto fluorophores.^61,104,105 A few turn-on sensors for Fe³⁺ have been reported; however, their utility for imaging in cells has not yet been established.^106,107

2.4 Small-molecule ROS and RNS indicators

As discussed in Section 2.3, there is mounting evidence that ROS are linked to several neurodegenerative diseases.⁹¹ Reactive nitrogen species (RNS) are also thought to cause neuropathological disorders.¹⁰⁸ More recently, ROS and RNS have been implicated as signalling molecules during normal physiological processes.^35,109 The canonical ROS and RNS involved in these processes are H₂O₂ and NO, respectively. Due to space limitation, herein we will discuss only small-molecule fluorescence indicators of H₂O₂ and NO because in-depth reviews on sensors for other ROS and RNS as well as metal-based fluorogenic probes for NO have been written.^110–113

The traditional indicators for sensing H₂O₂ include dihydrofluorescein (DHFC, Fig. 5A)¹¹⁴ or dihydrorhodamine (Fig. 5A);¹¹⁵ however, these dyes suffer from auto-oxidation by the excitation light as well as reactivity with other ROS.^116,117 Chang et al. have developed a class of boronate-based fluorescent probes for sensing H₂O₂, which includes their first generation probe Peroxy-fluor-1 (PF1, Fig. 5A).¹¹⁸ In the presence of H₂O₂, the boronic esters are chemoselectively deprotected to yield a fluorescent molecule. This concept has also been applied to blue- and red-shifted analogs, Peroxyxanthone-1 (PX1) and Peroxyresorufin-1 (PR1).¹¹⁹ Second-generation probes that are more sensitive and are capable of detecting H₂O₂ at physiologically relevant concentrations have also been developed. These include Peroxycrimson-1 (PC1) and Peroxygreen-1 (PG1) (Fig. 5A).¹²⁰ Further, sensors of additional colors and FRET-based ratiometric probes for sensing H₂O₂ have been recently described.^121,122 Hydrogen peroxide sensors that can be targeted to specific organelles for visualizing ROS within subcellular locations have also been developed. These include probes for visualizing H₂O₂ in the mitochondria, nucleus, and endoplasmic reticulum.^123–125 Finally, new hydrogen peroxide sensors have been used to elucidate that redox signalling through Nox2 is essential for the normal growth and proliferation of neural stem cells.¹²⁶ These sensors were also used to show that H₂O₂ uptake is mediated by the water channel aquaporin to modulate downstream signalling events.¹²⁷

Fig. 5.

Small-molecule fluorescent ROS/RNS sensors. (A) Fluorogenic H₂O₂ probes. (B) Fluorogenic NO probes.

Most sensors for NO utilize an o-phenylenediamine scaffold that is oxidized to the corresponding aryl triazole in the presence of NO and air (Fig. 5B).¹⁰⁹ In this design, the electron-rich diamine quenches the fluorescence of the dye by photoinduced electron transfer. Reaction with NO and O₂ generates an electron-poor triazole that triggers the fluorescence turn-on of the dye. Nagano and co-workers have developed a wide variety of probes based on this design that contain fluorescein (DAF-2),¹²⁸ rhodamine (DAR-4),¹²⁹ BODIPY (DAMBO and DAMBOO),^130,131 and cyanine (DAC-P, Fig. 5B).¹³² This group has also developed fluorescent sensors of NO with increased sensitivity due to longer intracellular retention times.¹³³ Recently, a variation of the o-phenylenediamine scaffold was developed that is based on a quenched rhodamine B spirolactam (Fig. 5B).¹³⁴

3. Chemical reporter strategy

For biomolecules such as glycans, certain protein post-translational modifications, and nucleic acids, detection via genetically-encoded reporters such as fluorescent proteins or small-molecule sensors has largely not been possible. Thus, alternative strategies have been developed.

3.1 Lectins and antibodies for imaging glycans

Glycans are interesting targets for imaging because they play key roles in many dynamic biological processes. For example, cell-surface glycans are known to mediate cell–cell adhesion and communication, as well as host-pathogen interactions.¹³⁵ Other examples include embryonic development,¹³⁶ leukocyte homing,¹³⁷ and cancer cell metastasis.¹³⁸ Traditionally, glycans have been detected using antibodies or carbohydrate-binding proteins known as lectins.¹³⁹ Though lectins have been widely used for the detection and enrichment of glycoconjugates,^140,141 these glycan-binding proteins generally have low affinities for their targets and require multivalency for high-avidity binding.¹⁴² Other drawbacks include toxicity and tissue impermeability.¹⁴³ For these reasons, lectins have limited capabilities for in vivo imaging; however, they have been used routinely for imaging glycans ex vivo on cultured cells and on tissue samples.^144,145

Like lectins, antibodies have limited use for in vivo imaging; however, there is one report of their use for glycan-specific imaging in mice. In this study, Licha et al. imaged a peripheral lymph node endothelial glycan termed sulfoadhesin using the MECA-79 antibody.¹⁴⁶ Though there are a number of monoclonal antibodies against distinct epitopes, it is difficult to generate antibodies against certain epitopes because the synthesis of many glycan structures can be very cumbersome.¹⁴⁷ Thus, applications of antibodies toward in vivo imaging are also limited. Furthermore, like lectins, antibodies also suffer from poor tissue access.¹⁴⁸

3.2 Bioorthogonal chemical reporter strategy for imaging glycans

Though lectins and antibodies have been used to image glycans, they are not ideal for imaging dynamic changes in the glycome, the ensemble of glycans displayed on the cell surface. We have developed an approach for imaging the glycome that enables the visualization of glycan dynamics in vivo.¹⁴⁹ Our method is a two-step strategy in which the first step involves the metabolic incorporation of an unnatural monosaccharide into an organism’s glycome.¹⁵⁰ This substrate analog contains a reactive group known as a “chemical reporter.’ In the second step, the chemical reporter is labelled with an imaging probe via a chemoselective reaction.

The requirements for such a chemical reaction are quite rigorous. First, the chemical reporter and its reaction partner must react in a physiological environment (pH 6–8, 37 °C), while simultaneously remaining inert to all of the surrounding functionality in the biological environment. This task is challenging because there are many nucleophilic and electrophilic functional groups found in cells and living organisms. Second, the reaction must not produce any toxic byproducts or cause harm to the biological sample. Finally, the reporter and its complementary probe must have good bioavailability and form a stable bond. Chemical reactions that meet this collection of criteria are known as “bioorthogonal.”¹⁵¹

Currently, only a handful of reactions possess the quality of bioorthogonality. These reactions include the condensation of ketones and aldehydes with hydrazide or alkoxyamine probes, the Staudinger ligation of triarylphosphines and azides, and Cu-free click chemistry between cyclooctynes and azides (Fig. 6A–B, D). These reactions have been used to label not only proteins but glycans and lipids as well.

Fig. 6.

Bioorthogonal chemical reactions. (A) Ketone and aldehyde condensation with hydrazides or alkoxyamines. (B) Staudinger ligation between azides and triarylphosphines. (C) Copper-catalysed azidealkyne 1,3-dipolar cycloaddition (CuAAC). (D) Cu-free click chemistry.

Ketones and aldehydes are not truly bioorthogonal because keto and aldehydic metabolites are abundant within cells and in biological fluids. In addition, the pH optima of hydrazone and oxime formation with hydrazide and alkoxyamine groups is 3–4, which precludes their use in vivo.¹⁵² Ketones and aldehydes have been utilised to label proteins and glycans on cell surfaces and in the extracellular environment. These mild electrophiles can form reversible Schiff bases with primary amines such as lysine side chains; however, in water, the equilibrium favors the carbonyl. In contrast, the formation of hydrazones and oximes using hydrazide and aminooxy groups, respectively, are favored in physiological conditions and are quite stable (Fig. 6A). Dawson and co-workers have greatly accelerated both oxime and hydrazone formation via the use of a nucleophilic aniline catalyst.^153,154

We have recognized the utility of ketones and aldehydes as chemical reporters and have used them to label glycans.¹⁵⁵ In this work, Mahal et al. reported that an unnatural keto analog of N-acetylmannosamine (ManNAc), N-levulinoylmannosamine (ManLev), can be metabolized by mammalian cells to the corresponding keto sialic acid (SiaLev) and incorporated within cell-surface glycans (Fig. 7A–B).¹⁵⁵ Sadamoto et al. have introduced ketones into bacterial cell walls and labeled the chemical reporters with a hydrazide fluorophore probe.¹⁵⁶

Fig. 7.

Bioorthogonal chemical reporter strategy. (A) Metabolic oligosaccharide engineering. (B) Unnatural sugars.

More recently, the azide has emerged as the chemical reporter of choice. This functional group is unique in that it is truly bioorthogonal to the biological milieu. Only one naturally occurring azido-metabolite, which was isolated from unialgal cultures, has been reported to date.¹⁵⁷ Moreover, azides can undergo highly selective reactions such as the Staudinger ligation with triarylphosphines, the copper-catalysed azide-alkyne 1,3-dipolar cycloaddition (CuAAC), and Cu-free click chemistry with strained alkynes (Fig. 6B–D).¹⁵⁸

The Staudinger ligation is based on the classic Staudinger reduction, which was developed by Hermann Staudinger in 1919. In this work, Staudinger reported that azides react with triarylphosphines under mild conditions to yield aza-ylide intermediates.¹⁵⁹ In the presence of water or an appropriate electrophile, this intermediate can be hydrolyzed or trapped to yield an amine and the corresponding phosphine oxide.¹⁶⁰ We modified the classic Staudinger reaction by introducing an electrophilic trap onto the phosphine.¹⁶¹ Termed the Staudinger ligation, this reaction ultimately results in the formation of a covalent amide bond between the azide and phosphine along with oxidation of the phosphine (Fig. 6B). This reaction also proceeds at physiological pH with no apparent toxicity to live cells and whole organisms.¹⁶²

We have shown that an azide-containing analog of ManNAc, termed N-azidoacetylmannosamine (ManNAz), can also be metabolically incorporated into cell-surface glycans as the corresponding azido sialic acid (SiaNAz) (Fig. 7).¹⁶¹ The Staudinger ligation has been used to image these labelled glycans on live cells using phosphine fluorophore conjugates¹⁶³ as well as bioluminescent and fluorogenic phosphines.^164,165 Furthermore, we and others have shown that this reaction can be used to label glycans within living animals.^162,166,167

Paulson and co-workers have shown that natural sialic acids can be oxidized to the corresponding aldehyde using sodium periodate and then subsequently imaged using oxime chemistry with alkoxyamine probes.¹⁶⁸ In addition to metabolic labelling of sialic acid-containing glycans,^169,170 other sectors of the glycome can be metabolically labelled using azido analogs of N-acetylglucosamine (β-O-GlcNAcylated proteins),¹⁷¹ N-acetylgalactosamine (mucin-type O-linked glycans and β-O-GlcNAcylated proteins),^172,173 and fucose (fucosylated glycans) (Fig. 7B).^174,175 Extending the scope of this methodology, which has thus far focused on labelling individual monosaccharides within the larger glycan structure, a recent report described the targeting of the disaccharide LacNAc (Galβ-1,4-GlcNAc).¹⁷⁶

Azides can also be exploited as 1,3-dipoles in a [3+2] cycloaddition with alkynes to form triazole products (Fig. 6C).¹⁵⁸ Sharpless and co-workers and Meldal and co-workers independently showed that the normally sluggish reaction between azides and acetylenes can be greatly accelerated with the addition of a copper catalyst.^177,178 The copper-catalysed azide-alkyne 1,3-dipolar cycloaddition (CuAAC) is one of the most popular reactions from a family of “click reactions,” defined by their highly selective and modular approach to forming chemical bonds.¹⁷⁹ CuAAC is much faster than the Staudinger ligation; however, the necessary copper catalyst is cytotoxic, although new biocompatible Cu(i) catalysts have been described to reduce toxicity to live cells.^{150,180–183} CuAAC has been used for imaging glycans using alkynyl ManNAc analogs as well as azido and alkynyl fucose analogs in fixed cells (Fig. 7B).^175,184,185 In addition, CuAAC was recently used to image lipopolysacccharide (LPS) in bacterial membranes by targeting 3-deoxy-d-mannooctulosonic acid, a specific and essential component of the inner core of LPS.¹⁸⁶

We have eliminated the need for a toxic copper catalyst by activation of the alkyne via an alternative mechanism—ring strain. This reaction is accordingly known as Cu-free click chemistry. Agard et al. first demonstrated this concept using a cyclooctyne reagent (OCT) whose kinetics were on par with the Staudinger ligation (Fig. 6D and 8).¹⁸⁷ In order to accelerate the strain-promoted cycloaddition, Baskin et al. introduced fluorine atoms in an attempt to lower the LUMO energy of the alkyne (MOFO, DIFO, DIFO2, DIFO3; Fig. 8).^188,189 This reaction has been used to image glycans on live cells,¹⁸⁸ the nematode C. elegans,¹⁹⁰ and zebrafish embryos.^191–194 We have also shown that Cu-free click chemistry can be used to label glycans within mice.¹⁹⁵ Second-generation cyclooctynes have been developed, including more water-soluble probes (ALO and DIMAC, Fig. 8)^189,196 as well as faster reagents based on a dibenzocyclooctynyl scaffold that have been utilized for imaging glycans on cells (DIBO, S-DIBO, DIBAC, BARAC, Photo-DIBO, TMDIBO, keto-DIBO; Fig. 8),^197–203 a coumarin-based fluorogenic cyclooctyne,²⁰⁴ and thiacycloalkynes with improved stability.²⁰⁵ Several theoretical studies have been published to explain the enhanced reactivity of a number of these cyclooctynes.^206–208

Fig. 8.

Cyclooctyne reagents for Cu-free click chemistry.

In addition to Cu-free click chemistry using strained alkynes, several new chemistries using strained alkenes have emerged for labelling proteins in live cells and animals.²⁰⁹ These include reactions of strained alkenes with azides,²¹⁰ tetrazines,^211,212 nitrile oxides,²¹³ and nitrile imines,²¹⁴ as well as photoinducible tetrazole–alkene cycloadditions, known as photo-click chemistry.²¹⁵ Recently, a new bioorthogonal reaction between highly strained quadricyclanes and Ni(bis)thiolenes was described that is selective and proceeds readily in aqueous environments.²¹⁶ Although these reactions have only been described for targeting proteins thus far, their selectivity and versatility will hopefully encourage their use in imaging beyond the proteome. Applications using multiple bioorthogonal chemistries would ultimately enable simultaneous visualisation of different classes of biomolecules within the same cell.

3.3 Bioorthogonal chemical reporter strategy for imaging lipids

As secondary metabolites and protein post-translational modifications, lipids, like glycans, are not amenable to imaging via genetically-encoded reporters. Lipids are attractive targets for imaging because they play important roles in cellular processes such as signal transduction and membrane fusion.²¹⁷ In addition, lipidation of proteins regulates the signalling properties, subcellular localisation, trafficking, and activity of many proteins.²¹⁸ One approach to imaging lipids in live cells has been the application of the bioorthogonal chemical reporter strategy.²¹⁹

Phospholipids are the major constituents of cell membranes. Those containing the head group choline are important structural components of membranes and play critical roles in cell signalling.²²⁰ Despite the crucial roles of these lipids, little is known about their localisation and trafficking.²²¹ Thus, Jao et al. have utilized the chemical reporter strategy in order to image their dynamics in cells and tissues. In order to achieve this goal, they synthesized a choline analog that contains a terminal alkyne moiety, which can be detected by using CuAAC (Fig. 9). Using this technology, they were able to examine the kinetics of phospholipid turnover and the distribution of the lipids in cells and in mice.²²¹ Similarly, Schultz and co-workers have shown that phosphatidic acid analogs that contain either terminal alkynes or cyclooctynes within the lipid tail can be used to image these lipids in cells (Fig. 9).²²²

Fig. 9.

Chemical reporters for lipids.

Lipidation of proteins via fatty-acylation and prenylation has been known to greatly affect the specificity and efficiency of signal transduction as well as protein–protein interactions.²¹⁸ Several groups have applied the bioorthogonal chemical reporter strategy to image these protein post-translational modifications.^223–225 Zhao and co-workers have used azido analogs of farnesylated proteins in order to profile proteins that are post-translationally modified (Fig. 9).^226,227 Alkynyl farnesol reporters have also been described.^228,229 Several groups have demonstrated that fatty-acylation of eukaryotic proteins by attachment of S-palmitoyl groups and N-myristoyl to cysteine and N-terminal glycine residues, respectively, can be imaged or profiled by employing azido and alkynyl analogs of these lipids and the Staudinger ligation or CuAAC (Fig. 9).^230–236 More recently, a chemical reporter of cholesterylated proteins has also been described.²³⁷

3.4 Bioorthogonal chemical reporter strategy for imaging nucleic acids

Nucleic acids are traditionally detected by intercalating dyes (e.g., 4′6-diamidino-2-phenylindole), in situ hybridization techniques, and metabolic incorporation of nucleoside analogs that are either radiolabelled or detected via antibodies. Hybridization approaches include molecular beacons, which have been used for visualizing both DNA and RNA.^238,239 Development of these fluorescent oligonucleotide probes requires the incorporation of fluorescent nucleotides, which can be synthesized using different chemistries.²⁴⁰ For example, Ju and co-workers have synthesized fluorescent oligonucleotides via both CuAAC and the Staudinger ligation for the purposes of DNA sequencing using fluorescence detection.^241,242 Excess fluorophore conjugates of oligonucleotide probes must be washed away to eliminate background signal, so fluorogenic approaches are highly sought after as well. Toward this end, Cai et al. have developed a fluorogenic peptide nucleic acid-based probe that is activated by the Staudinger ligation upon DNA hybridization.²⁴³

Other techniques include metabolic incorporation of deoxynucleoside analogs such as [³H]thymidine ([³H]T) and 5-bromo-2′-deoxyuridine (BrdU) (Fig. 10). [³H]T and BrdU are detected by autoradiography and immunohistochemistry, respectively. Though these analogs have been useful for probing DNA synthesis, both of these techniques are not ideal. Autoradiography can be cumbersome, and BrdU immunostaining can compromise the integrity of the tissue samples.²⁴⁴ Mitchison and co-workers have synthesized an alkynyl analog of thymidine, 5-ethynyl-2′-deoxyuridine (EdU, Fig. 10), which is readily incorporated into DNA during DNA replication.²⁴⁵ This chemical reporter can then be detected with CuAAC and has enabled visualization of DNA synthesis in cells and in tissue sections from mice. In addition, this methodology causes much less degradation of biological specimens relative to traditional BrdU immunostaining. Arabinosylated EdU analogs (F-ara-EdU, Fig. 10) and alkynyl cytosine analogs have been reported that were used for imaging DNA in cells and zebrafish.^246,247 More recently, an alkynyl analog of uridine, 5-ethynyluridine (EU, Fig. 10) has been used to image RNA synthesis in vivo using an analogous strategy.²⁴⁸

Fig. 10.

Metabolic labels for nucleic acids.

4. Label-free imaging methods

4.1 Imaging mass spectrometry

All of the technologies discussed thus far require labels or contrast agents in order to visualise the biological molecules of interest. Imaging mass spectrometry (IMS) is an emerging label-free technique that allows for the visualisation of endogenous proteins, lipids, and small molecules.²⁴⁹ IMS can be used to create 2D and 3D images of the proteomic or small-molecule content of cells or tissues by combining the measurement capability of mass spectrometers with a surface sampling process.²⁵⁰ Currently, while there exist many different modes of ionisation, two popular ionisation methods are used to generate images.²⁵¹ These techniques include matrix-assisted laser desorption ionisation (MALDI) and secondary ion mass spectrometry (SIMS) IMS. Though IMS has been used for imaging both proteins and other biological molecules, for the purposes of this review, we will focus exclusively on imaging small molecules.

Of these two IMS techniques, MALDI IMS has seen more widespread use for analysing biological samples due to its large mass-to-charge ratio (m/z) range (up to 100 kDa), which is useful for analysing proteomes.²⁵² Imaging with MALDI mass spectrometry, however, is only capable of achieving spatial resolutions of about 20 µm while SIMS can achieve better than 100 nm with specialised instruments.²⁵³ On the other hand, SIMS has been documented to detect molecules of masses up to only 1 kDa.²⁵⁰ While many groups have taken advantage of these imaging technologies, a number of laboratories continue to pursue the development of new and improved methodologies, including those that can achieve ionisation under ambient conditions.^254,255

For analysis by MALDI IMS, biological samples are prepared in a similar manner to protein or peptide samples. Typically, a thin cryosection of tissue is covered with an organic matrix. The sample is then scanned by the mass spectrometer, and spectra are collected over a predefined area. The resulting dataset contains an array of spectra including different signals of measured mass and intensity. The intensity of signal, or combinations of signals, can then be plotted to generate an image over the sample surface.²⁵⁶ This technique has been used to image many different molecules, including phospholipids.²⁵⁶ As discussed in Section 3.3, these lipids are involved in cell signalling pathways and have been implicated in diseases such as lipid storage disorders as well as Alzheimer’s disease and Down syndrome.²⁵⁷ McLean et al., Jackson et al., and Rujoi et al. have used MALDI IMS to visualise phospholipids in rat brains and mammalian lens tissue.^258–261 A growing number of researchers are applying IMS to the diagnosis of diseases such as cancer by looking at differences in lipid profiles between diseased and healthy tissue.^262–265 In addition to lipids, Dorrestein and co-workers have applied MALDI IMS to imaging secondary metabolites produced by marine cyanobacteria and sponges as well as Bacillus subtilis and Streptomyces coelicolor.^266–268

SIMS uses an accelerated primary ion beam to bombard the sample, a process that generates secondary ions that are then detected by a mass spectrometer.²⁵³ Based on instrument design, there are two different approaches to SIMS. The first method, time-of-flight (TOF) SIMS, uses a primary ion beam that is pulsed, and the resulting secondary ions are detected by a TOF mass spectrometer. The second approach, termed dynamic SIMS, uses a continuous primary ion beam, and preselected ions are detected by a magnetic sector mass spectrometer. Specialised instruments have been developed to improve both the sensitivity and spatial resolution of SIMS (vide infra).²⁵³ TOF-SIMS has been used to image lipids in cells and tissues.^269–271 TOF-SIMS has also been used to image highly curved membranes during Tetrahymena mating (Fig. 11).²⁷² In this study, Ostrowski et al. demonstrated that the fusion region contains elevated amounts of phosphatidylethanolamine, a high-curvature lipid.

Fig. 11.

Tetrahymena thermophila mating results in lipid heterogeneity at membrane fusion sites. (A) SIMS image for m/z 184, representative of the phosphocholine headgroup, demonstrating the absence of phosphatidylcholine at the conjugation junction. (B) SIMS image for m/z 126, representative of the phosphoethanolamine headgroup, demonstrating the presence of phosphatidylethanolamine at the conjugation junction. SIMS images were acquired using a 200 nm beam spot size (B250 nm lateral resolution). Scale bar = 50 µm. From Science 2004, 305, 71. Reprinted with permission from AAAS.

In the last decade, advancements in instrumentation have greatly improved the utility of SIMS for imaging biological samples. These developments include nanoSIMS, which enables image resolution of a few tens of nanometers.²⁵³ This technique has been applied in tracer studies examining ¹³C-labeled free fatty acid transport across cell membranes.²⁷³ Boxer and co-workers have used nanoSIMS to image lipid domains within supported lipid bilayers with a lateral resolution of 100 nm.²⁷⁴ NanoSIMS has also been used to demonstrate that deep-sea anaerobic methane-oxidizing archaea fix nitrogen and cyanide and share the products with sulfate-reducing bacterial symbionts.²⁷⁵ Lechene et al. have developed multiple-isotope imaging mass spectrometry (MIMS), in which multiple isotopes are detected simultaneously, enabling both imaging and quantification of labeled molecules within subcellular compartments.²⁷⁶ This technique was utilised to test the idea that stem cells retain the older template DNA to insure lifetime genetic stability, a concept known as the immortal strand hypothesis.²⁷⁷ In this study, the authors showed that the DNA strands are randomly segregated during cell proliferation. Another technical advance includes cluster time-of-flight (TOF)-SIMS, which has been shown to improve the yield of secondary ions produced.²⁷⁸ This technique utilises heavier primary ions that have improved efficiency for production of secondary ions, thus allowing easier identification of biomolecules. Many groups have demonstrated that cluster TOF-SIMS imaging can be used to image lipids in rodent brains and tissues.^279–283 Finally, Heeren and co-workers have developed matrix-enhanced SIMS (ME-SIMS), which combines the high spatial resolution of SIMS with the sample preparation of MALDI IMS.²⁸⁴ Like cluster TOF-SIMS, ME-SIMS also increases the molecular ion yield.

4.2 Raman and coherent anti-Stokes Raman scattering (CARS) microscopy

Raman and coherent anti-Stokes Raman scattering (CARS) microscopy are emerging label-free methods for cellular imaging.²⁸⁵ These two techniques rely on optical signals and have been used to image proteins, lipids, and nucleic acids.^286,287 Xie and co-workers have recently demonstrated that stimulated emission can also be used as a contrast mechanism for microscopy.²⁸⁸ They have used this technique to image chromoproteins, small-molecule drug distributions in vivo, and haemoglobin. For the purposes of this review, we will focus exclusively on Raman and CARS applications for imaging small molecules.

Raman microscopy relies on a phenomenon known as Raman scattering. This effect occurs when the wavelength of scattered light shifts slightly from the original wavelength of light due to excitation of the light-scattering molecules to a vibrationally excited state.²⁸⁶ Because the shift in wavelength depends strongly on the molecule’s structure and chemical environment, the resulting spectra can be used to identify specific molecules in the sample. In Raman microscopy, spectra are obtained from each position in the sample to construct a 2D or 3D image of molecular distribution based on signal intensities. This imaging technique has been used to visualize nucleic acids^289,290 and lipid bodies²⁹¹ in live cells. Raman microscopy has also been used to discern malignant tissue from healthy tissue for cancer diagnostic purposes.^292–294 Raman microscopy, however, suffers from several limitations.²⁹⁵ First, the Raman effect is very weak, and consequently, data acquisition times are long. Second, creation of images requires high-powered lasers and long integration times per pixel, impeding video-rate microscopy. Xie and co-workers have greatly improved the sensitivity of Raman microscopy by developing a new technique known as stimulated Raman scattering (SRS) microscopy, which has been used to visualize lipids in live cells and tissues from mice.²⁹⁶ Furthermore, in vivo imaging was enabled by enhancement of the signal and by increasing the imaging speed to video-rate.²⁹⁷ Recently, Freudiger et al. have improved the spectral selectivity of SRS, thus allowing the molecular distinction between related molecules by using spectrally tailored excitation.²⁹⁸

Stronger vibrational signals can be obtained using CARS microscopy. In CARS, light beams with two different optical frequencies interact with the sample. A strong CARS signal is produced when the difference of the two frequencies matches the vibrational frequency of the molecules. Because CARS is orders of magnitude more sensitive than Raman microscopy, it allows video-rate vibrational imaging. The Reintjes group was the first to use CARS as a contrast mechanism for microsopy.²⁹⁹ The technique was popularized by Xie and co-workers almost two decades later when they developed a greatly improved method that allowed for higher sensitivity, higher spatial resolution, and three-dimensional sectioning capabilities.³⁰⁰

With these improvements, CARS has been used extensively for many in vitro and in vivo imaging applications. Wurpel et al. and Potma et al. have used CARS to image lipid vesicles and lipids on supported bilayers.^301,302 Xie and co-workers have also used CARS to image lipids in live cells.^303–305 Others have imaged lipids in infection models,^306,307 mouse brains,³⁰⁸ and whole organisms such as C. elegans (Fig. 12).^309–311 Improvements to CARS include video-rate microscopy applications of imaging lipids in the skin of live mice³¹² and frequency modulation CARS, which significantly increases the detection sensitivity of CARS by decreasing the nonresonant background.³¹³

Fig. 12.

CARS imaging of lipid storage in C. elegans. CARS microscopy reveals a dramatic difference in lipid droplet densities from wild-type C. elegans (A) and a feeding-defective mutant, pha-3 (B). From Proc Natl Acad Sci 2007, 104, 14658. Reprinted with permission from National Academy of Sciences, U.S.A.

Conclusions

Recent advances in molecular imaging techniques have been made possible by fundamental advances in disparate areas of chemistry, spanning from metal-coordination chemistry to biochemistry to spectroscopy. These new methods have enabled the visualisation of virtually every major class of molecule inside living cells. As these methods are further improved and gain wider usage, it is our hope that they will continue to shed light on many biological processes.

Biographies

Pamela V. Chang

Pamela V. Chang received a BS in Chemistry from the Massachusetts Institute of Technology (MIT) in 2004. At MIT she worked in the laboratory of Professor Stephen J. Lippard. Upon graduating from MIT, she pursued graduate studies at the University of California, Berkeley, with Professor Carolyn R. Bertozzi, and completed her PhD in Chemistry in 2010. As a graduate student, Pamela’s research interests focused on developing chemical tools for imaging glycans in vivo. Pamela is currently a Jane Coffin Childs postdoctoral fellow at Yale University in the Department of Immunobiology, under the guidance of Professor Ruslan Medzhitov.

Carolyn R. Bertozzi

Carolyn R. Bertozzi is the T.Z. and Irmgard Chu Distinguished Professor of Chemistry and Professor of Molecular and Cell Biology at University of California, Berkeley and an Investigator of the Howard Hughes Medical Institute. She completed her undergraduate degree in Chemistry from Harvard University in 1988 and her PhD in Chemistry from UC Berkeley in 1993. After postdoctoral work at the University of California, San Francisco, in the field of cellular immunology, she joined the UC Berkeley faculty in 1996. Prof. Bertozzi’s research interests span the disciplines of chemistry and biology with an emphasis on glycobiology, microbiology, and nanomaterials.