Chemical Approaches for Measuring and Manipulating Lipids at the Organelle Level

Abstract

As the products of complex and often redundant metabolic pathways, lipids are challenging to measure and perturb using genetic tools. Yet by virtue of being the major constituents of cellular membranes, lipids are highly regulated in space and time. Chemists have stepped into this methodological void, developing an array of techniques for the precise quantification and manipulation of lipids at the subcellular, organelle level. Here, we survey the landscape of these methods. For measuring lipids, we summarize the use of metabolic labeling and click chemistry tagging, photoaffinity labeling, isotopic tagging for Raman microscopy, and chemoenzymatic labeling for tracking lipid production and interorganelle transport. For perturbing lipids, we describe synthetic photocaged lipids and membrane editing approaches using optogenetic enzymes for precise manipulation of lipid signaling. Collectively, these chemical and biochemical tools are revealing phenomena and mechanisms underlying lipid functions at the subcellular level.

New methodologies propel biological research, and the study of lipids is no exception. In this review, we will describe approaches exploiting chemical probes and analytical methods to measure and perturb lipids at the subcellular, organelle level.

Methods for measurement are foundational across science because quantitative observations of phenomena enable underlying mechanisms to be deduced. We will first focus on chemical methods for measuring membrane-embedded lipids, particularly those that reveal functions and properties of lipids with high spatial resolution (i.e., at the subcellular, organelle level). Because lipids are small, the addition of fluorophores significantly affects their physicochemical properties, making direct observation of lipids difficult. Recent strategies to address this issue use minimalist tags with only a few atoms or isotope labels. These approaches include the use of click chemistry tagging and Raman microscopy techniques. We will also discuss a chemoenzymatic mass-tagging method to measure interorganelle lipid flux.

Approaches for inducing perturbations are valuable complements to measurement methods, as they aid in validating underlying mechanisms behind the measured phenomena. A typical example in the life sciences is gene knockout; if knockout cells respond differently, we conclude that the gene contributes to the phenotype. The importance of methods for inducing perturbation is best appreciated by the explosive adoption of genome editing, which has revolutionized life science research. In lipid biology, although knockout or overexpression of lipid-metabolizing enzymes are commonly used to perturb lipid dynamics, these approaches sometimes fail to induce desired modifications because of the rapid and redundant metabolism of many lipids. Thus, because lipids are not genetically encoded, genome editing is not as precise a vehicle for lipid/membrane editing as it is for editing the functions of proteins. Therefore, we will discuss new methods that enable membrane editing, as exemplified by caged lipids and light-controlled enzyme-based systems.

CLICK CHEMISTRY–BASED METABOLIC LABELING OF LIPIDS

One approach to obtain lipid spatial information is genetically encoded lipid-binding probes, reviewed extensively elsewhere (Wills et al. 2018; Hammond et al. 2022). These tools are convenient as they easily reveal organelle-level lipid localizations. However, they can also present some problems, such as low specificity for the target lipid and difficulty in quantification and differentiation of newly synthesized from pre-existing pools of lipids. These inherent problems derive from their use of affinity and therefore require different approaches to overcome.

A complementary strategy involves the use of metabolic labeling with precursors bearing click chemistry handles. Click chemistry refers to a group of reactions that use functional group pairs that can react to form covalent bonds with high yield and selectivity, as exemplified by the copper-catalyzed azide-alkyne cycloaddition (CuAAC) (Fig. 1A; Rostovtsev et al. 2002; Tornøe et al. 2002). Azides and alkynes further possess desirable bioorthogonal features, as they are small, absent from cells, and unreactive with endogenous biomolecules. Although CuAAC presents some cytotoxicity due to the copper catalyst, a copper-free and nontoxic strain-promoted azide-alkyne cycloaddition (SPAAC) circumvents this problem, wherein highly strained cyclooctynes enable tagging of azide-labeled biological molecules in live cells and organisms (Fig. 1A; Agard et al. 2004; Baskin et al. 2007; Chang et al. 2010). Among the many cyclooctynes reported to date, dibenzocyclooctyne (DBCO) (Ning et al. 2008) and bicyclononyne (BCN) (Dommerholt et al. 2010) are currently the optimal and most widely used options.

Figure 1.

Click chemistry and its use in metabolic labeling of lipids. (A) Click chemistry reactions of azides and alkynes to form triazole products. Copper-catalyzed azide-alkyne cycloaddition (CuAAC) is promoted by a copper catalyst, and strain-promoted azide-alkyne cycloaddition (SPAAC) is copper-free and instead promoted by ring strain of the cyclooctyne reagents. (B) Metabolic labeling of phosphatidylcholine (PC) using propargylcholine (ProCho) followed by CuAAC tagging to enable visualization (Jao et al. 2009). Shown: immunoelectron microscopy image. (Nu) nucleus, (NM) nuclear membrane, (ER) endoplasmic reticulum. Scale bar, 100 nm. (Adapted from Jao et al. 2009, The National Academy of Sciences © 2009.) (C) Metabolic labeling of PC with azido-choline (AzCho) and SPAAC tagging with a dibenzocyclooctyne (DBCO) fluorophore. (D) Two-color imaging of temporally distinct newly biosynthesized PC pools using sequential ProCho and AzCho metabolic labeling and CuAAC/SPAAC tagging (Jao et al. 2015). (Adapted from Jao et al. 2015, John Wiley & Sons © 2015.) (E) Organelle-selective PC labeling with AzCho and organelle-targeted DBCO probes (Tamura et al. 2020). Shown: fluorescence micrographs of organelle-selective labeling. Scale bar, 5 µm. (Adapted from Tamura et al. 2020, Springer Nature Limited © 2020.) (F) Serine analogs for metabolic labeling of phosphatidylserine (PS) (Ancajas et al. 2023). Shown: fluorescence microscopy images of localization of labeling derived from both probes at the yeast plasma membrane. Scale bars, 5 µm. (Adapted from Ancajas et al. 2023, American Chemical Society © 2023.) (G) C2-substituted inositol analog for metabolic labeling of phosphatidylinositol (PI) (Ricks et al. 2019). Shown: fluorescence microscopy of azido-myo-inositol (AzIno) labeling of human T-24 cancer cells. Scale bar, 5 µm. (Adapted from Ricks et al. 2019, John Wiley & Sons © 2019.) (H) C4-substituted inositol analog for metabolic labeling of glycophosphatidylinositol (GPI)-anchored proteins (Lu et al. 2015). Shown: fluorescence micrographs after CuAAC tagging with azido-biotin and streptavidin-fluorophore labeling. Scale bar, 10 µm. (Adapted from Lu et al. 2015, John Wiley & Sons © 2015.)

To use bioorthogonal or click chemistry reactions such as SPAAC or CuAAC for tagging lipids, one of the two click reaction partners must be first installed onto the lipid molecule. This first step, metabolic labeling, involves introducing metabolite analogs with clickable groups to the cell culture media, leading to their incorporation into lipids via endogenous metabolic pathways. In a second step, a bioorthogonal or click chemistry tagging reaction allows addition of a larger probe for detection (e.g., a fluorophore or biotin).

The first example of metabolic labeling and click chemistry for visualizing lipids targeted phosphatidylcholine (PC) lipids, the most abundant phospholipids in mammalian cells (Jao et al. 2009). Jao et al. developed an alkyne-containing choline analog, propargylcholine (ProCho), for metabolic incorporation into PC and other choline-containing phospholipids (ether-PC and sphingomyelin [SM]) via de novo biosynthesis pathways. The resultant alkyne-labeled lipids, most of them being alkynyl PC analogs (ProPC), were visualized by CuAAC tagging with an azido-fluorophore conjugate (Fig. 1B). This PC measurement strategy revealed synthesis rates, turnover speed, and subcellular localizations of de novo synthesized PC and SM. Analysis of ProPC by mass spectrometry showed a fatty acyl tail distribution similar to that of natural choline-containing phospholipids, demonstrating that ProCho is a faithful analog of choline-utilizing biosynthetic pathways. These labeled lipids were alternatively analyzed at the ultrastructural level by immunoelectron microscopy via CuAAC tagging with azido-biotin followed by antibiotin immunostaining, showing the versatility of using click chemistry tagging (Fig. 1B). New probe development diversified this approach in subsequent studies, including two-color imaging of choline metabolism with azido-choline (AzCho) and a DBCO-fluorophore conjugate via SPAAC tagging (Fig. 1C,D; Jao et al. 2015) and organelle-selective labeling of PC pools by tagging of AzCho-labeled lipids with a panel of organelle-targeted cyclooctyne-dye conjugates (Fig. 1E; Tamura et al. 2020).

Similar approaches have enabled metabolic labeling of other lipid classes. For example, Ancajas et al. treated yeast with synthetic azido-serine analogs (C-AzSer or N-AzSer) to metabolically label phosphatidylserine (PS) (Fig. 1F; Ancajas et al. 2023). When C-AzSer- and N-AzSer-labeled cells were tagged with a DBCO-fluorophore conjugate by SPAAC, the yeast plasma membrane was labeled. Although mass spectrometry and thin-layer chromatography (TLC) analysis confirmed the production of azido-PS (AzPS) in both instances, most of the labeling of N-AzSer, not C-AzSer, was found in phospholipids downstream of PS, including phosphatidylethanolamine (PE), produced by decarboxylation of PS, and PC, produced by methylation of PE. The results suggest that N-AzPS undergoes metabolism similar to PS and can be a useful tool for tracking the fate of PS, whereas C-AzSer can afford selective PS labeling.

The Best group also developed azido-myo-inositol (AzIno) and achieved metabolic labeling of phosphatidylinositol (PI), another major anionic phospholipid (Fig. 1G; Ricks et al. 2019). Among the many possible hydroxyl groups, the C2 position was chosen for azide introduction to avoid affecting the position required for connection to the glycerol backbone (C1) or phosphorylation of PI at (C3, C4, and C5). In human T-24 cancer cells, metabolic labeling followed by SPAAC tagging with a DBCO-fluorophore conjugate led to an endoplasmic reticulum (ER)-like labeling pattern, consistent with the location of PI synthesis (Kim et al. 2011). Production of azido-PI (AzPI) was confirmed by TLC and mass spectrometry. One downside of AzIno is that capping the C2 position of inositol with the azide prevents its use in the synthesis of the essential sphingolipid inositol phosphorylceramide (IPC) in yeast (Dickson and Lester 1999; Cowart and Obeid 2007). Nevertheless, AzIno is a useful tool for tracking the biological roles of PI.

Azide derivatization of inositol at a different position enables labeling of the glycosylphosphatidylinositol (GPI) anchor, a carboxy-terminal posttranslational modification that tethers secreted proteins to the cell surface (Fig. 1H; Lu et al. 2015). In particular, a C4-substituted AzIno analog afforded the strongest labeling in yeast and several human cancer cell lines, enabling investigation of the trafficking and organization of GPIs and potentially the discovery of new GPI-anchored proteins.

Although other classes of lipids remain to be targeted by bioorthogonal metabolic labeling, the above examples demonstrate the generality of the approach. As lipid structures are not much larger than fluorophores, physicochemical perturbation is substantial, but click chemistry minimizes this perturbation by separating the observation process into labeling and detection steps. Azide or alkyne groups used in the labeling step are small enough to preserve the near-native lipid structures, whereas the versatility of click reactions accommodates a wide variety of detection methods. This strategy is particularly useful if trafficking or other dynamic characteristics of lipids are measured. That said, any deviation from the native structure, even if only by a few atoms, may alter the properties of lipids to some extent, and it is in these scenarios that label-free or isotopic methods, described further below, are warranted. Nevertheless, metabolic labeling and click chemistry tagging will likely continue to play a central role in monitoring the biosynthesis of bulk lipids as described above, and other applications as elaborated below.

CLICK CHEMISTRY–BASED MEASUREMENTS OF SIGNALING LIPID PRODUCTION

Beyond its use for tagging bulk phospholipids following metabolic labeling, click chemistry can also be used in conjunction with chemoenzymatic labeling for monitoring the activity of biosynthetic enzymes that produce rare lipid signaling agents. Notably, we developed a technique to visualize the activity of phospholipase Ds (PLDs), which are important lipid-metabolizing enzymes that hydrolyze PC to generate phosphatidic acid (PA) (Fig. 2A), a pleiotropic lipid second messenger and phospholipid biosynthetic intermediate. The PLD family is a major producer of PA upon stimulation of cell-surface receptors including Gq-coupled G-protein-coupled receptors (GPCRs), receptor tyrosine kinases (RTKs), and integrins. Despite this importance, the spatiotemporal context where PA signaling occurs following different forms of stimulation has been an open question. Our approach, termed imaging phospholipase D activity with clickable alcohols via transphosphatidylation (IMPACT), harnesses a second enzymatic activity of PLDs: transphosphatidylation using a primary alcohol instead of water as the cosubstrate (Fig. 2A; Bumpus and Baskin 2017). Bumpus et al. used this characteristic to have PLDs utilize a variety of alkyne- and azide-tagged primary alcohols (e.g., 3-azido-1-propanol [AzPro]) to generate clickable phosphatidylalcohol lipids (Bumpus and Baskin 2016, 2017). Subsequent SPAAC or CuAAC tagging using BCN-BODIPY or alkyne-quaternary ammonium (Alk-QA), respectively, enabled visualization of intracellular membranes bearing active PLD pools following stimulation and mass spectrometry analysis of the generated lipid reporters of PLD activity (Fig. 2B,C). IMPACT using SPAAC tagging is best suited to quantification of whole-cell PLD activity, as exemplified by its application to FACS-based CRISPR screening for regulators of PLD signaling, because of rapid interorganelle lipid trafficking that can occur during the relatively sluggish SPAAC reaction (Bumpus et al. 2021).

Figure 2.

IMPACT: Clickable chemoenzymatic tools to measure phosphatidic acid (PA) production by phospholipase D (PLD). (A) Two types of reactions catalyzed by PLDs: phosphatidylcholine (PC) hydrolysis to generate PA and PC transphosphatidylation with primary alcohols to generate phosphatidylalcohols. In the two-step IMPACT procedure, the first step involves production of clickable AzPC by transphosphatidylation with clickable primary alcohols such as 3-azidopropanol (AzPro). (B) The second step of IMPACT is a click chemistry tagging reaction, wherein AzPC is tagged with BCN-BODIPY or Alk-QA by strain-promoted azide-alkyne cycloaddition (SPAAC) or copper-catalyzed azide-alkyne cycloaddition (CuAAC), respectively, for imaging or mass spectrometry analysis. (C) Imaging of PLD activity by IMPACT. Results of labeling with BCN-BODIPY are displayed. The weaker fluorescence intensity in the condition with isoform-selective PLD inhibitors reveals that IMPACT reports on activity of both the PLD1 and PLD2 isoforms. Scale bar, 10 µm. (Adapted from Bumpus and Baskin 2017, American Chemical Society © 2017.) (D) Real-time IMPACT (RT-IMPACT) involves transphosphatidylation with an oxoTCO-containing alcohol and click tagging with Tz-BODIPY via the rapid and fluorogenic inverse electron-demand Diels–Alder (IEDDA) cycloaddition, allowing rinse-free real-time imaging. Imaging at early (<10 second) time points of the IEDDA reaction enables observation of the location of PLD-generated lipids prior to translocation to other membranes (e.g., 165 second time point). Scale bar, 10 µm. (Adapted from Liang et al. 2019, The National Academy of Sciences © 2019.)

To overcome this kinetics problem, we developed a real-time IMPACT variant (RT-IMPACT) (Liang et al. 2019), wherein oxo-trans-cyclooctene (oxoTCO) alcohols were used, which can react with fluorogenic tetrazine (Tz) reagents in superfast inverse electron-demand Diels–Alder (IEDDA) click chemistry reactions (Fig. 2D). Critically, the rapid kinetics and fluorescence turn-on during tagging in RT-IMPACT enables no-rinse, real-time visualization of precise sites of PLD signaling and subsequent rapid intracellular phospholipid trafficking that occurs on the second-to-minute timescale. Applications of RT-IMPACT have included deciphering the spatiotemporal properties of certain Gq-coupled GPCR signaling pathways and visualization of bulk phospholipid transport at ER-PM contact sites by lipid transfer proteins (Liang et al. 2022, 2023).

IMPACT generates stable reporter lipids whose levels can be quantified and is thus suited for measuring the activity of the PLD-mediated PA biosynthesis pathway. Such measurements are difficult to perform with more indirect methods such as monitoring PLD expression or quantification of total PA using mass spectrometry or fluorescent PA-binding probes, because PA has multiple biosynthetic sources and exhibits rapid metabolism. Nevertheless, phosphatidylalcohols generated by IMPACT are distinct from PA, so they cannot be used as tracers for subsequent PA transport or metabolism.

SUPERRESOLUTION IMAGING OF ORGANELLES USING CLICK CHEMISTRY AND LIPID PROBES

A limitation of using conventional fluorescence microscopy for monitoring tagged lipids is its resolution, which is not high enough to observe detailed structures of organelle membranes within cells. Thus, superresolution imaging methods are desirable, but due to various technical considerations, specialized probes and/or labeling strategies are required to fully take advantage of these methods.

Superresolution strategies such as PALM/STORM and STED enable low-nanometer resolution but have high “photon budgets.” To get around these problems, Schepartz, Toomre, and co-workers have developed high-density environmentally sensitive (HIDE) dyes, which overcome bleaching problems that plague traditional dyes due to switching between on and off states (Erdmann et al. 2014; Takakura et al. 2017; Thompson et al. 2017; Chu et al. 2020). The dyes are localized to organelles in a two-step manner by adding bioorthogonal organelle-targeted lipid probes followed by click tagging to introduce the dye to the organelle membrane of choice. In this manner, superresolution imaging of organelle membranes over long time periods was achieved.

Because superresolution microscopes are not universally available, and because the above methods do not visualize native lipid molecules but rather involve the indirect observation of artificial hydrophobic membrane probes, alternate strategies that enable direct observation of lipids with nanoscale resolution using conventional confocal microscopes are desirable. Expansion microscopy (ExM) solves this problem. Here, biological samples are physically magnified by covalent anchoring of biomolecules to a swellable polymer network (Chen et al. 2015). For example, a commonly used ExM probe is the bifunctional acryloyl-X SE, which covalently tags primary amine groups on protein lysine side chains via its N-hydroxysuccinimide (NHS) ester and is integrated into the polymer network through its methacryloyl group (Fig. 3A). Therefore, protein epitopes can be covalently retained so that after proteolysis, sample clearing with detergent, polymerization to form a hydrogel, and swelling in water, cell or tissue samples that have been expanded isotropically can be imaged. Although conventional microscopes are used, because samples are expanded, the effective resolution is nanometer-scale, and various ExM methodologies can afford from 4.5× to >10× expansion factors, approaching the resolution of PALM/STORM or STED microscopies. Unfortunately, standard ExM approaches involve permeabilization (i.e., lipid removal), so adjustments are required to extend ExM to visualization of membrane lipids, prompting several related and complementary solutions.

Figure 3.

Application of click chemistry to expansion microscopy and chemoproteomics studies of membrane lipids. (A) Overview of expansion microscopy (Tillberg et al. 2016). Cells are fixed and endogenous molecules are labeled with a reaction between primary amines and the N-hydroxysuccinimide (NHS) ester of acryloyl-X SE. When the polymer network is created by gelation, labeled molecules are anchored via the acryloyl group. Proteinase disruption and dialysis in water enables isotropic expansion of samples and observation. (Adapted from Tillberg et al. 2016, Springer Nature Limited © 2016.) (B) Sphingolipid expansion microscopy (Götz et al. 2020). The amino group of the probe is used for anchoring to the polymer via glutaraldehyde, and the azide is used for fluorescent labeling via strain-promoted azide-alkyne cycloaddition (SPAAC). Pre- and post-expansion images are displayed. Scale bars, 5 µm and 50 µm for pre- and post-expansion, respectively. (Adapted from Götz et al. 2020, Springer Nature Limited © 2020.) (C) Click-expansion microscopy (click-ExM) (Sun et al. 2021). After the metabolic labeling of phosphatidylcholine (PC) with propargylcholine (ProCho), cells are fixed and permeabilized and biotin is attached to the labeled PC via copper-catalyzed azide-alkyne cycloaddition (CuAAC), followed by streptavidin-fluorophore labeling, anchoring, polymerization, and expansion. Shown: post-expansion images are displayed of click-ExM (Alk-Cho) with markers of mitochondrial matrix (Mito-GFP) and mitochondrial outer membrane (Tom20). Scale bars, 10 µm (large image) and 2 µm (magnified images). (Adapted from Sun et al. 2021, Springer Nature Limited © 2021.) (D) Lipid expansion microscopy (LExM). Metabolic labeling of PC with ProCho is followed by fixation, CuAAC tagging with a trifunctional LExM probe, polymerization, and expansion. (E) Pre- and post-expansion (6.8× expansion factor) LExM images, with LExM (green) and an outer mitochondrial membrane marker (OMM, magenta) shown. Dotted (pre-expansion) and solid (post-expansion) lines indicate position for line intensity profiles shown at right (White et al. 2022). Scale bars, 0.4 µm (pre-expansion) and 2.72 µm (post-expansion). (Adapted from White et al. 2022, American Chemical Society © 2022.) (F) General schematic for chemoproteomics methods involving photoaffinity labeling of lipid–protein interactions. (G) Chemoenzymatic in situ synthesis of a photoaffinity labeling (PAL) probe for phosphatidylethanol lipids (Yu et al. 2021). Alkyne- and diazirine-functionalized phosphatidylalcohols were added to the media, enabling phospholipase D (PLD)-mediated production of a PAL lipid analog of phosphatidylethanol via cross-linking IMPACT (XL-IMPACT). UV cross-linking and CuAAC tagging enable visualization or enrichment of phospholipid-interacting proteins. Shown: confocal micrograph of XL-IMPACT fluorescence after streptavidin-fluorophore staining. Scale bar, 20 µm. (Adapted from Yu et al. 2021, American Chemical Society © 2021.)

Götz et al. (2020) developed a functionalized ceramide containing primary amine and azide groups in the fatty acid tail (Fig. 3B). After cell labeling, the probe was incorporated into a hydrogel through its primary amine via glutaraldehyde fixation, followed by SPAAC tagging with a DBCO-fluorophore. Alternatively, metabolic labeling of clickable phospholipids has been leveraged for ExM in so-called click-ExM (Sun et al. 2021). Here, metabolically labeled lipids (e.g., using ProCho) are tagged with azido-biotin via CuAAC, followed by staining with fluorescently labeled streptavidin, gelation, digestion, and expansion (Fig. 3C). Although click-ExM requires mild permeabilization before the click reaction, its expansion capability was about 4.4-fold, which is roughly equivalent to that of conventional ExM. Beyond ProCho, click-ExM is versatile in that it can be applied to labels for various lipids, including alkyne-tagged fatty acids and farnesols.

The methods listed above require at least mild detergent–based permeabilization to ensure a uniform distribution of ExM reagents and isotropic expansion of samples, and this process can disrupt the membrane structure. Therefore, we were motivated to develop an all-small molecule method that requires no permeabilization prior to tagging and fixation. Our approach, lipid ExM (LExM), uses metabolic labeling (e.g., using ProCho or IMPACT with an alkynyl alcohol) and tagging with a trifunctional probe that has full access to the interior of paraformaldehyde-fixed cells, enabling direct anchoring of metabolically labeled phospholipids into a polymer network (Fig. 3D,E; White et al. 2022). The trifunctional probe contains an azide for CuAAC tagging to the lipid, a methacryloyl group for hydrogel anchoring, and a BODIPY fluorophore for visualization. The use of neutral, permeable monomers for the polymerization step enabled post-polymerization hydrolysis to the swellable hydrogel, a step that allows for tunable expansion factors and thus different balances between resolution and labeling density depending on the application.

Thus, several solutions have recently emerged that combine metabolic labeling and bespoke tagging reagents to enable lipid click chemistry-mediated retention and membrane visualization by ExM. In addition, methods using bulk lipid probe dosing have also been developed to achieve superresolution by ExM (Karagiannis et al. 2019; Wen et al. 2020). We envision that these techniques, along with alternate strategies such as the HIDE probes, will find diverse applications to visualize lipids and membranes at superresolution, such as imaging of the dynamics of various systems where membranes come in close apposition (e.g., mitochondria, nuclear envelope, membrane contact sites, multivesicular body formation, etc.).

PHOTOAFFINITY LABELING AND CHEMOPROTEOMICS FOR ANALYZING LIPID–PROTEIN INTERACTIONS

Beyond visualization, click chemistry can be combined with photocrosslinking groups such as diazirines to perform chemoproteomics through photoaffinity labeling (PAL). In the lipid arena, this strategy captures transient lipid–protein interactions, enabling identification of protein-binding partners of lipids and physiological functions of lipid–protein binding events. Here, a bifunctional (i.e., photocrosslinkable and clickable) lipid probe is added to the media or generated chemoenzymatically inside cells. First, UV irradiation induces diazirine breakdown to produce reactive intermediates that form covalent bonds with proteinaceous functional groups. Click chemistry tagging to append an affinity handle such as biotin enables subsequent streptavidin-based affinity enrichment and proteomics analysis to identify protein-binding partners. With specialized biotin reagents, modified peptides can be identified, affording information about residues important for lipid binding (Ge et al. 2022).

Several examples exist using these types of strategies, including synthetic PAL probes for identifying interaction partners of phosphoinositides (Balla 2013; Müller et al. 2020, 2021), sterols (Hulce et al. 2013; Cheng et al. 2021), bacterial lipids (Kavunja et al. 2020), and bulk phospholipids within therapeutic liposomes (Jose and Pucadyil 2020; Pattipeiluhu et al. 2020), as we recently reviewed comprehensively (Fig. 3F; Yu and Baskin 2022). As a complement to synthetic PAL probes, we used a cross-linking IMPACT variant (XL-IMPACT) that harnesses PLD transphosphatidylation activity toward a bifunctional primary alcohol to generate alkyne- and diazirine-functionalized phosphatidylalcohols in situ (Fig. 3G; Yu et al. 2021). These lipids, which resemble phosphatidylethanol, a long-lived lipid generated following alcohol consumption, enabled identification of phosphatidylethanol-binding partners that may play pathophysiologically relevant roles following alcohol exposure. Noncovalent affinity approaches can also be valuable in certain circumstances. For example, Zhang et al. applied a biotinylated long-chain fatty acyl CoA analog directly to cell lysates and revealed that fatty acyl CoA binding to NME1/2 inhibits these metabolic enzymes that regulate ATP and GTP levels, ultimately supporting breast cancer metastasis in mouse models (Zhang et al. 2022). In sum, chemoproteomics methods including covalent PAL and noncovalent affinity-based approaches are useful for revealing lipid–protein interactions, harnessing the power of minimalist functional groups such as alkyne, azide, and/or diazirine.

IMAGING LIPIDS USING RAMAN MICROSCOPY

In exchange for sacrificing the versatility and specificity of click chemistry–based approaches that require large contrast agents for detection by fluorescence microscopy, lipids can also be imaged using Raman microscopy with even smaller tags or sometimes under label-free conditions. Raman scattering is an optical process where incident light interacts with a target molecule to produce scattered light with typically smaller energies depending on the vibrational mode of chemical bonds in the sample (Raman and Krishnan 1928). Therefore, Raman scattering contains information about molecular structure. Recent advances in optics have enabled sensitive measurement of Raman scattering, which has an intensity of only about 10⁻⁶ compared to Rayleigh scattering, enabling single-pixel measurements from small areas in Raman microscopy.

Raman spectra obtained from biological samples are divided into three regions: a fingerprint region (below 1800 cm⁻¹) containing vibrational signals deriving from endogenous biomolecules, a silent region (1800–2800 cm⁻¹) containing virtually no signals from endogenous biomolecules, and a high-wavenumber region (above 2800 cm⁻¹) containing vibrations from carbon–hydrogen bonds or carbon–nitrogen single bonds (Fig. 4A; Shen et al. 2019; Egoshi et al. 2022). Convenient tags suitable for Raman imaging are carbon–carbon and carbon–nitrogen triple bonds, which display sharp spectral peaks in the silent region, enabling multiplex tagging strategies to be used in Raman imaging of biological specimens. Taking advantage of the sharpness of Raman spectra compared to fluorescent spectra, tags capable of simultaneous measurement of more than 20 “colors” (i.e., different vibrational maxima that can be separately detected) have been developed by using engineered xanthenes and polyynes (Yamakoshi et al. 2012; Chen et al. 2014; Wei et al. 2017; Hu et al. 2018).

Figure 4.

Visualization of subcellular lipids by deuterium labeling and Raman microscopy. (A) Representative Raman spectra from HeLa cells (Egoshi et al. 2022). Spectra are divided into three regions: fingerprint, silent, and high wavenumber. 532-nm laser excitation was used. (Adapted from Egoshi et al. 2022, Elsevier B.V. © 2022.) (B) Multiplex imaging of deuterated or bromine-labeled fatty acids (Uematsu et al. 2020). The results of images reconstituted using unmixed signals are displayed for each fatty acid, together with representative spectra. Scale bar, 10 µm. (Adapted from Uematsu et al. 2020, John Wiley & Sons © 2020.) (C) Transition of Raman spectra from fully deuterated palmitic acid (C16:0-d31) (Uematsu and Shimizu 2021). Raw spectra of C16:0-d31 under various conditions were interpolated to generate the figure. These spectra can be used as a reference to estimate the physical properties of subcellular lipids. (D) Quantified physical properties of subcellular environment (Uematsu and Shimizu 2021). Physical properties of lipid droplet (LD) regions were dynamically changed depending on perturbations attempting to alter the lipid composition; these properties were relatively constant in non-LD regions. (Adapted from Uematsu and Shimizu 2021, Springer Nature Limited © 2021.) (E) Raman spectra of biochemically extracted lipids, proteins, and DNA (Shi et al. 2018). HeLa cells were grown in DMEM media containing 70% D₂O. Spectra derived from C–D vibrations of DNA were blue-shifted, whereas those of lipids were red-shifted, compared to those of proteins. (F) Raman imaging from pre- and post-unmixing (Shi et al. 2018). Based on the obtained data in E, spectral unmixing was performed to remove residual bleedthrough signal from each channel. (Adapted from Shi et al. 2018, Springer Nature Limited © 2018.)

Compared to these relatively large tags, deuterium labeling is more suitable for lipids and other small molecules to minimally perturb lipid properties. Because carbon–deuterium vibrations also appear in the silent region, deuterium-labeled lipids can be easily distinguished from endogenous lipids. Several groups have shown the distribution of deuterium-labeled lipids taken up by cells (Weeks et al. 2011; Matthäus et al. 2012). Consequently, methods for multiplex imaging have been developed. For example, Stiebing et al. performed partial two-color imaging using two different labeled fatty acids, palmitic acid d-31 (C16:0-d31) and arachidonic acid d-8 (C20:4-d8) (Stiebing et al. 2014). However, the low flexibility of deuterium labeling hindered further multiplexing.

Uematsu et al. (2020) overcame this problem and achieved quadruplex imaging by carefully selecting the labeling patterns on fatty acids and using a spectral unmixing algorithm (Fig. 4B). This technique revealed that accumulation ratios of incorporated fatty acids in lipid droplets (LDs) of HeLa cells are correlated with the number of double bonds. By applying spectral analysis in a different direction, Uematsu and Shimizu (2021) developed an approach to quantify the physical properties of intracellular lipids (Fig. 4C,D). Based on the distortion of Raman spectra depending on the surrounding membrane properties, they established a method for estimating the physical properties of membrane lipids using C16:0-d31 taken up by cells as a probe. The results revealed that the physical properties of cytoplasmic membrane lipids are maintained by the buffering function of LDs, which have many functions beyond simple energy storage. Similarly, Shen et al. used C16:0-d31 to discover that novel large-scale membrane domains are formed in the ER upon the treatment of saturated fatty acids (Shen et al. 2017). They used the spectral shifting of the probe and estimation of the diffusion coefficients from pulse treatment with the probe to reveal that newly formed domains are phase-separated from the ER, exhibiting solid-like characteristics.

Other methods of introducing deuterium labeling exploit incorporation of deuterated water (D₂O) during de novo lipogenesis. Notably, the perturbation to native metabolism is considerably smaller than methods described above using deuterated fatty acids and approaches to visualize de novo lipogenesis using deuterated glucose (Li and Cheng 2014; Du et al. 2020). When cells are treated with D₂O, deuterium atoms are incorporated into fatty acids during de novo lipogenesis both directly from D₂O and indirectly via NADPH, acetyl-CoA, and malonyl-CoA. Shi et al. developed unmixing methods to separate lipid, protein, and DNA signals from spectra of C–D bonds and enabled simultaneous imaging of de novo lipogenesis and protein biosynthesis in animals without tissue bias (Fig. 4E,F; Shi et al. 2018).

Although Raman microscopy technology is arguably still in a developmental stage, it has the potential to obtain rich information of hyperspectral images (i.e., data containing x, y, and spectral axes) by minimal tagging or sometimes without any tag at all. These properties are well suited for studying lipids, which are small and capable of being tagged while emitting relatively strong Raman signals compared to other biomolecules. Even under label-free conditions, it is possible to infer the degree of unsaturation of the lipid environment by measuring vibrations of C = C stretching and C–H bending (Wu et al. 2011; Li et al. 2017). High-sensitivity and high-speed methods such as stimulated Raman scattering (SRS) (Freudiger et al. 2008), coherent anti-Stokes Raman scattering (CARS) (Evans and Xie 2008), and recently developed superresolution SRS (Jang et al. 2023), which are also compatible with super-multiplexing, are expected to further encourage the use of Raman microscopy for analysis of cellular lipids.

FLUX MEASUREMENT

In addition to imaging where lipids are located and biosynthesized, it is desirable to track their interorganelle transport within cells. A promising method to measure such lipid flux is named mass-tagging-enabled tracking of lipids in cells (METALIC) (John Peter et al. 2022). Its basic concept is to chemoenzymatically introduce two different mass tags onto the same lipid at two different organelle membranes (Fig. 5A–C). The detection of doubly mass-tagged lipids by mass spectrometry indicates that these lipid molecules have existed on these two organelles during the labeling period, providing direct evidence of lipid transport between the two organelles (Fig. 5D). METALIC uses cyclopropane-fatty-acyl phospholipid synthase (CFAse) as one of the two enzymes to perform mass tagging in yeast (Fig. 5B). CFAse is a soluble bacterial enzyme that catalyzes conversion of double bonds in fatty acyl tails of unsaturated phospholipids to a cyclopropane by transfer of a methylene group (CH₂) from the cofactor S-adenosyl methionine (SAM). CFAse can be targeted to any membrane of interest, and in the initial study it was tethered to mitochondria. For the second membrane, the endogenous ER-localized PE methyltransferases (PEMTs) Cho2 and Opi3 were used, which use SAM to convert PE to PC (Fig. 5A).

Figure 5.

METALIC: A method to measure interorganelle lipid transport. (A,B) Chemoenzymatic activities used in METALIC. Cho2/Opi3 and CFAse both take S-adenosylmethionine (SAM) as substrates, producing +9 and +16 heavier molecules in the presence of deuterated SAM (d-SAM). (C,D) The concept of METALIC (John Peter et al. 2022). Two types of enzymes, Cho2/Opi3 on the endoplasmic reticulum (ER) and CFAse on mitochondria, function to execute double mass tagging at different sites on the same molecule of phosphatidylcholine (PC). Detection of PC molecules that are 25 Da heavier in the presence of d-SAM (generated from metabolic labeling with d-methionine) compared to without d-SAM provides direct evidence of lipid transport between these two organelles. Note that Cho2/Opi3 are endogenous membrane proteins that localize on the ER and CFAse is fused to a mitochondria-targeting signal. (E) Representative results of METALIC. The percentage of incorporation in the headgroup, fatty acid tail, and both after d-methionine pulse labeling of yeast are displayed as line plots. Three data points at each time point represent independent clones for each genotype. Red lines indicate the data from wild-type, and gray lines indicate the knockout of Sam5, a major transporter of SAM across the inner mitochondrial membrane. (Adapted from John Peter et al. 2022, Springer Nature Limited © 2022.)

The combination of these enzymes with pulsing cells with deuterated SAM precursors enables temporal control over labeling. Relative to natural SAM, deuterated SAM labeling by CFAse results in +16 (CD₂ addition to a double bond) and PEMTs as +9 (nine deuterium atoms in the three methyl groups of PC relative to natural PC) (Fig. 5D). Therefore, detection of PC molecules by mass spectrometry that are 25 Da heavier than natural PC indicates that these lipids have traveled between the ER (where PEMTs reside) and the mitochondria (where CFAse was targeted). The advantages of METALIC are obvious, as it provides a direct and quantitative measurement of interorganelle phospholipid transport. Its limitations are that in its current iteration, one of the two organelles must be the ER due to reliance on endogenous PEMTs, that directionality of transport cannot be determined, and that it is limited to PE/PC. Nevertheless, METALIC represents a quantum leap forward for tracking interorganelle lipid flux compared to established, low-resolution methods involving biochemical fractionation or in vitro assays, and future versions of this method will likely address these limitations.

PHOTOCAGED LIPIDS

Methods for perturbation are important for verifying hypotheses developed by measurement. Because lipids are not directly genetically encoded and can be rapidly metabolized by often redundant pathways, simply knocking out or overexpressing genes for lipid-metabolizing enzymes typically cannot achieve a desired level of specificity. One approach to perturb lipid signaling with high temporal control is using photocaged lipids. Here, lipid molecules are made biologically inactive, or “caged,” through chemical derivatization with a photocleavable protecting group, allowing precise control over when the lipid becomes active (Höglinger et al. 2014). Examples of caged lipids include caged-PA (Goedhart and Gadella 2004), caged-LPA (Hövelmann et al. 2016), caged-sphingosine (Höglinger et al. 2015), caged-PI(3,4)P₂, and caged-PI(3,4,5)P₃ (Mentel et al. 2011; Walter et al. 2017; Citir et al. 2021; Müller et al. 2021). Studies using these lipid probes established causal relationships that release of signaling lipid molecules can trigger various effects including Ca²⁺ signaling, chemotaxis, and endocytosis, deepening the relationship between lipid signaling and cellular functions.

Although the use of UV light affords some extent of spatial control, additional techniques can better control the activation of caged lipids with subcellular spatial resolution by integration of functional groups that confer organelle targeting. For example, caged-sphingosine with lysosomal (Feng et al. 2019) or mitochondrial (Feng et al. 2018) tags revealed distinct differences in local sphingosine metabolism; sphingosine released on mitochondria was quickly phosphorylated into sphingosine 1-phosphate (S1P) and did not induce calcium spikes, in contrast to global sphingosine release (Fig. 6A). Simon et al. measured differences in metabolic flux of ER- and mitochondrial-released PE, further using dual isotopic labeling to discriminate two different pathways of PC synthesis (i.e., remodeling vs. PEMT) (Fig. 6B; Simon et al. 2023).

Figure 6.

Application of caged lipids and evolution of phospholipase D (PLD) as an optogenetic membrane editor. (A) Structure of organelle-targeted caged sphingosine (Feng et al. 2018, 2019). Structures of mitochondrial, lysosomal, and global uncaging forms of sphingosine are displayed. Calcium spikes observed with the global uncaging probe were not observed with the mitochondrial uncaging probe. (Adapted from Feng et al. 2018, eLife Sciences Publications Limited © 2018.) (B) Structure of organelle-selective caged phosphatidylethanolamine (PE). (C) Principle of optoPLD. CRY2 and CIBN are tagged with PLD and an organelle-targeting tag, respectively. Because CRY2 and CIBN heterodimerize upon blue light irradiation, membrane recruitment of PLD can be controlled with high spatiotemporal precision. (D) OptoPLD activation induces membrane recruitment of a phosphatidic acid (PA)-imaging biosensor (PA probe) (Tei and Baskin 2020). OptoPLD with the indicated organelle targeting tags were expressed in HEK293T cells, and the recruitment of the PA biosensor was observed after blue light irradiation. (PM) Plasma membrane, (TGN) trans-Golgi network, (ER) endoplasmic reticulum. Colocalization of the PA biosensor and optoPLD indicates the high spatial resolution of optoPLD for producing local PA pools. Scale bars, 10 µm. (Adapted from Tei and Baskin 2020, Rockefeller University Press © 2020.) (E) Development of superactive PLDs (superPLDs) via activity-dependent directed evolution (Tei et al. 2023). The activity of superPLDs developed by directed evolution was evaluated by IMPACT and flow cytometry. Horizontal and vertical axes represent the expression level of PLD or superPLD and the IMPACT activity. (F) Optogenetic superPLDs (superPLD^low and superPLD^med) induce more robust membrane recruitment of a PA biosensor than optoPLD based on wild-type PLD (PLD^WT). Constructs were expressed in HEK293T cells, and superPLD was recruited to the plasma membrane by blue light irradiation. Scale bars, 10 µm. (Adapted from Tei et al. 2023, Springer Nature Limited © 2023.)

Schuhmacher et al. (2020) used caged-DAG to quantify the kinetics of lipid signaling. In the initial state, caged-DAG molecules are distributed on the outer leaflet of the PM, but uncaging allows transbilayer flip-flop to expose DAG molecules to the cytoplasmic leaflet and thus cytoplasmic environment. By using this DAG release as the perturbation to the system, lipid–protein affinities were inferred by fitting models to the observed translocation of a genetically encoded DAG-binding probe from the cytosol to the plasma membrane. They further improved the model by introducing a profile likelihood method (Gonzales et al. 2023). The power of their model is shown in its ability to infer quantitative parameters of signaling lipid despite the existence of many parameters that are either impossible or extremely difficult to measure in experimental settings, such as uncaging efficiency, scrambling rate of uncaged DAG, and heterogeneities in cellular states.

In sum, photocaged lipids are useful not only for introducing pulses of lipid signals but also quantitatively obtaining spatiotemporal information about lipid metabolism and signaling, enabling a wide range of applications. We expect that new photocaged lipids will continue to emerge for quantifying different aspects of organelle-specific lipid metabolism and signaling.

MEMBRANE EDITING USING LIGHT-CONTROLLED ENZYMES

Instead of inducing controlled release of lipids via photo-uncaging, alternate strategies for membrane editing involve controlling the spatiotemporal activity of lipid-metabolizing enzymes. We achieved this goal by developing an optogenetic PLD (optoPLD) (Tei and Baskin 2020). Here, a bacterial PLD from Streptomyces sp. PMF was connected to the CRY2-CIBN optogenetics system, wherein CRY2 and CIBN heterodimerize upon blue light irradiation. Specifically, PLD and a genetically encoded organelle-targeting tag were fused to CRY2 and CIBN, respectively, thereby enabling the PLD recruitment and, consequently, PA formation on the organelle membrane of interest with high temporal resolution (Fig. 6C,D). By applying this tool to induce PA signaling at multiple organelles, we showed that PA signaling at the PM, but not on other organelle membranes, attenuates the oncogenic Hippo signaling pathway.

Subsequently, to address the low activity of optoPLD, we used activity-based directed evolution encompassing random mutagenesis of PLD, IMPACT labeling, and FACS sorting in mammalian cells to generate superactive PLDs (superPLDs) (Fig. 6E,F; Tei et al. 2023). We isolated many superPLD mutants with activities up to 100-fold higher than wild-type PLD, and we demonstrated their utility by using them to create next-generation optoPLDs that could control the activation of AMP kinase signaling through the binding of PA to the effector protein LKB1. The increased stability of superPLDs enabled their purification in recombinant form in high yields from Escherichia coli, leading to structural and biochemical studies to understand the sources of increased activity and chemoenzymatic syntheses of several natural and unnatural phospholipids.

One limitation of optoPLD, with either wild-type or superPLDs, is that as signal increases, so does background in the dark, catalyzing PA synthesis at undesirable moments. This problem most likely arises because cytosolic PLD retains enzymatic activity and can catalyze the reaction upon the spontaneous interaction with membranes. Other strategies will be necessary to generate ultralow background optoPLDs optimally suited for examining the acute effects of PA signaling in various sensitive cell types. Ultimately, enzyme-mediated membrane editing and synthetic caged lipid probes are complementary approaches for perturbing lipids in a native context. A major barrier is that neither technique is immediately applicable to other lipid classes, as both require de novo construction of new tools, either synthetic compounds or controllable enzymes, for different lipid targets. Yet both exhibit some element of modularity, with swapping out of targeting elements to allow recruitment/activation at different organelle membranes for diverse applications.

CONCLUDING REMARKS

The nongenetically encoded nature of lipids has presented challenges for their precise study using genetic manipulation but has also created opportunities for chemical methods for measuring and perturbing them at the subcellular, organelle level. In addition to classical biochemical techniques, most notably TLC, HPLC, and mass spectrometry, which are highly quantitative but more challenging to get to work at subcellular levels, a variety of chemical methods outlined here can obtain spatial information and have revealed valuable new insights about organelle-specific lipid pools. Click and bioorthogonal chemistries have been indispensable parts of so many of these measurement and visualization strategies. By contrast, there are fewer methods to induce precise perturbations to lipids, although their numbers are increasing. In particular, we hope that the success of optoPLDs as tools for precise manipulation of PA signaling will inspire development of optogenetic membrane editors targeting other classes of lipids, as these approaches hold promise for elucidating physiological roles for specific lipids and mechanisms underlying cellular lipid homeostasis.