Leaflet specific phospholipid imaging using genetically encoded proximity sensors

Abstract

The lipid composition of cells varies widely across organelles and between individual membrane leaflets. Transport proteins are thought to generate this heterogeneity, but measuring their functions in vivo has been hampered by limited tools for imaging lipids at relevant spatial resolutions. Here we present fluorogen-activating coincidence encounter sensing (FACES), a chemogenetic tool capable of quantitatively imaging subcellular lipid pools and reporting their transbilayer orientation in living cells. FACES combines bioorthogonal chemistry with genetically encoded fluorogen-activating proteins (FAPs) for reversible proximity sensing of conjugated molecules. We first apply this approach to identify roles for lipid transfer proteins that traffic phosphatidylcholine pools between the ER and mitochondria. We then show that transmembrane domain-containing FAPs can reveal the membrane asymmetry of multiple lipid classes in the trans-Golgi network and be used to investigate the mechanisms that generate it. Lastly, we demonstrate FACES can be applied to measure glycans and other molecule classes.

Introduction

Lipid composition determines the biophysical properties of cell membranes and varies across every scale of biological organization^1–3. In cells, membrane heterogeneity between organelles is generated by a complex network of lipid synthesis and transport pathways that can become concentrated at membrane contact sites; hot zones of lipid transport where opposing organelles are in close proximity, often only tens of nm apart^4,5. While many lipid transfer proteins (LTP) acting at these sites have now been biochemically characterized^6,7, measuring their in vivo functions remains challenging. Within organelles, it has long been recognized that each bilayer leaflet – separated by less than 4 nm – could feature a unique lipidome, which is reflected in the biophysical properties of each leaflet and the topology of protein transmembrane domains⁸. Detailed lipidomic investigation of membrane asymmetry has only been carried out in the PM of erythrocytes, whose lack of intracellular compartments allow for calculation of lipid accessibility to externally added phospholipases^3,9. While asymmetry is an emerging concept in membrane biology, the transbilayer distribution of phospholipids in intracellular membranes and its molecular drivers are still poorly understood.

Investigating membrane chemical heterogeneity remains a challenge due to the paucity of tools for imaging lipids in live cells at spatial resolutions needed to investigate lipid transport or asymmetry. Bioorthogonal ‘click’ chemistry has emerged as one tool for fluorescence labeling of specific lipid classes – including phosphatidylcholine (PC)¹⁰, phosphatidylserine (PS)¹¹, phosphatidylinositol (PI)¹², phosphatidic acid¹³, and glycosphingolipids¹⁴ – that can be metabolically tagged with azide-bearing head group components. Azido (N₃) lipids are commonly conjugated to fluorophores using strain-promoted azide-alkyne cycloaddition (SPAAC) reactions, unveiling their cellular distribution once unreactive dyes are washed out. While powerful, this approach indiscriminately labels all lipids in the cell, and is thus limited by the resolution constraints of light microscopy; it cannot discriminate between lipid pools in organelles at close proximity or in individual leaflets of the same membrane, for example. To overcome some of these limitations, organelle-targeted fluorophores have been applied to image subcellular lipid pools¹⁵, but their specificity is limited to just a few chemical environments in the cell.

We hypothesized that challenges for lipid imaging could be overcome using protein-based lipid sensors, whose localization is precisely controlled through signal sequences or fusion to transmembrane domains. Previously, natural lipid binding proteins derived from microbial toxins, such as perfringolysin O¹⁶ (cholesterol) or equinatoxin II¹⁷ (sphingomyelin, SM) have been expressed as fusions with fluorescent proteins (FP), to detect membranes enriched in the target lipids. In these modalities, lipids are only qualitatively detected by colocalization of a probe to a membrane, which limits their application to large, well-separated compartments. We asked whether fluorogenic systems could provide a more generalizable strategy. Fluorogen-activating proteins (FAPs) can be expressed as fusion constructs with other proteins or targeting sequences, similar to FPs, but are not themselves fluorescent¹⁸. Instead, they interact with fluorogens, small cell-permeable molecules that act as fluorophores only upon specific binding to a FAP. Depending on the FAP used, fluorogen binding can be irreversible or reversible; the former provides brighter fluorescence for conventional imaging, while the latter generates stochastic blinking that can be harnessed for super-resolution imaging¹⁹. Recently, FAPs have been applied towards detecting and controlling protein proximity in living cells through reversible dimerization²⁰, demonstrating that this technology can be harnessed in new applications.

Here, we seek to harness the unique properties of fluorogens through Fluorogen-Activating Coincidence Encounter Sensing (FACES), a method that uses FAPs for proximity-dependent reporting of fluorogen conjugated bioorthogonal metabolites, including lipids. We apply the FACES strategy to investigate the transport of N₃-phospholipids in cells and characterize the role of ER-mitochondrial PC transporters. We then show that transmembrane-anchored FAPs can probe leaflet-specific membrane composition, identifying features and drivers of asymmetry generated during vesicle formation at the trans-Golgi network (TGN). Lastly, we demonstrate that FACES is a broadly applicable tool compatible with non-lipid azido-labeled metabolites by measuring fluctuations in mitochondrial N-acetylhexosamine (HexNAz) levels.

Results

Development and validation of FACES

To implement FACES, we sought to couple fluorogenic dyes to label azido-labeled metabolites that incorporate in native phospholipid metabolism (Fig. 1a). The conjugated fluorogen tags would then only fluoresce if bound to a FAP (Fig. 1b). To accomplish this, we synthesized derivatives of the fluorogen malachite green (MG) with a linker connected to a dibenzocyclooctyne (DBCO) group for copper-free SPAAC reactions, including compound 1a (Fig. 1c; Supplementary Fig. 1). MG non-covalently binds to soluble single-chain antibody scFvs FAP and the complex emits far-red fluorescence. In the FACES workflow (Fig. 1d), cells first express a FAP fused to an FP, like GFP, that is targeted to an organelle or site of interest. Cells are then incubated with azido-labeled metabolites for metabolic incorporation. Immediately before imaging, incorporated metabolites are labeled with fluorogen-DBCO, which remains non-fluorescent unless it is bound to the expressed FAP. Excess dye is removed from cells by washing with culture medium, leveraging the reversibility of FAP binding to minimize detection of unreacted fluorogens. Fluorogenic fluorescence thus acts like an AND-gate: it requires both the target molecule be modified with a MG green molecule, and the corresponding FAP to be in the same compartment or membrane. The abundance of the target molecule is measured with the ratio of fluorogen fluorescence, which is at red shifted wavelengths for MG, to that of the reporter FP (e.g. GFP). This ratiometric read out normalizes the concentration of the FAP receptor itself.

Figure 1. Conceptual development of FACES.

a, 1-azidoethyl-choline (N₃-Cho) is metabolically incorporated into azido-PC lipids distributed throughout all cellular membranes. b, Fluorogens form reversible non-covalent fluorescent complexes with genetically encoded FAPs. c, Structures of the fluorogen malachite green linked to dibenzocyclooctyne (DBCO) for SPAAC labeling. d, Conceptual schematic detailing the FACES method. (1) Cells expressing a FAP-GFP fusion protein targeted to an organelle of interest are metabolically labeled with N₃-Cho, or other bioorthogonal metabolite. (2) After incorporation, metabolites are conjugated via a SPAAC reaction with fluorogen-DBCO (1a or 1b). During staining free dye can interact with the FAP to generate far-red fluorescence. (3) After SPAAC labeling, excess free dye is reversibly disassociated from the FAP complex and removed from cells by washing. (4) Subsequent imaging of the fluorogen-FAP complex is used to detect the specific biomolecule pool. Co-measurement of GFP fluorescence is used to normalize the fluorogen signal to FAP sensor density. e, HeLa cells transfected with either FAP-GFP-KDEL (top two rows) or GFP-KDEL (bottom row) were grown in the presence (top and bottom rows) or absence (middle row) of 50 μM N₃-Cho and labeled with 500 nM 1a. Cells were imaged while in the labeling solution (left) and after washing with culture medium (right). Scale = 20 μm. f, Merged confocal images of transfected HeLa cells post washing. White line indicates the segment used for plotting the fluorescent intensity line profiles of 1a (magenta) and GFP (green) shown below. Scale = 10 μm. g, Quantified fluorescence intensity profiles from confocal images expressed as ratiometric average; n = 1, N = 10 cells. Error bars reflect SEM. h, Linear regression of quantified fluorescence from transfected HeLa cells fed 0, 10, 30, 50 μM N₃-Cho and labeled with 500 nM 1a; n = 1, N = 100 individual ER segments derived from 10 cells. Error bars reflect SD. GFP and 1a were excited using 488 nm and 633 nm lasers, respectively.

We tested the compatibility and fluorogenic properties of 1a with FACES using the FAP clone dH6.2, a lower affinity scFvs variant developed for reversible equilibrium binding to MG²¹. In all experiments, we fused dH6.2 – from here on simply referred to as FAP – to GFP for ratiometric imaging. An N-terminal secretion signal peptide and C-terminal KDEL retention sequence were first used to target FAP-GFP to the endoplasmic reticulum (ER) as the major site of PC synthesis²², while GFP-KDEL, an identical construct but lacking the FAP domain, was used as a control. HeLa cells transfected with either FAP-GFP-KDEL or GFP-KDEL were grown for 24h in the presence or absence of 50 μM 1-azidoethyl-choline (N₃-Cho). At concentrations up to 100 μM, N₃-Cho was incorporated into PC at a frequency up to 0.1 mol % of unlabeled PC, but cells showed no detectable N₃-Cho incorporation into SM (Supplementary Fig. 2). Prior to staining, cells were washed and grown in normal growth media for 30 min to allow for the metabolic consumption of any remaining N₃-Cho. Cells were then transferred into FBS-free growth media containing 500 nM 1a and imaged while in the staining solution (Fig. 1e). Images were collected using identical acquisition and processing settings.

Results from pre-washed cells demonstrated that 1a was cell permeable and generated strong far-red fluorescence in cells expressing FAP-GFP-KDEL (Fig. 1e). No far-red fluorescence was detected in cells expressing GFP-KDEL, indicating that 1a was not activated directly by conjugation to N₃-PC. Next we tested the reversibility of the FAP-fluorogen complex by washing cells with growth medium and re-imaging. Three washes were sufficient to dissociate free 1a from FAP-GFP-KDEL cells in the absence of N₃-Cho, resulting in the loss of far-red fluorescence. In contrast, far-red fluorescence was retained in FAP-GFP-KDEL cells that were fed 50 μM N₃-Cho, which implicated the SPAAC reaction product 1a-PC as the source of FAP-dependent fluorescence (Fig. 1e). Similar experiments with a higher affinity FAP variant retained far-red fluorescence after washing in the absence of N₃-Cho labeling (Supplementary Fig. 3), indicating that the reversible binding to the MG product is essential for substrate-specific detection. In this regard, each experiment necessitates that a no azido control be included to assess the wash out of fluorogen and measure the intrinsic background fluorescence.

We further quantified these results in repeated experiments using confocal laser scanning microscopy (CFLSM) to assess linear fluorescent detection. Cells transfected with FAP-GFP-KDEL were fed 10, 30, or 50 μM N₃-Cho for 24h, while FAP-GFP-KDEL (no N₃-Cho) and GFP-KDEL (+ 50 μM N₃-Cho) conditions were used as controls. All cells were stained with 500 nM 1a and were washed as previously described; the 1a-PC reaction product was confirmed by LC-MS/MS (Supplementary Fig. 4). Fluorescence intensity profiles revealed that far-red fluorescence from 1a was tightly correlated with GFP fluorescence in FAP-GFP-KDEL cells fed 50 μM N₃-Cho (Fig. 1f). In contrast, cells from either control condition, lacking the FAP or N₃-Cho component, had low levels of far-red fluorescence that did not correlate with GFP localization. The average fluorescence intensity from 10 cells, using 10 individual ER segments per cell, was calculated as a ratiometric average relative to GFP (Fig. 1g). There was no intensity difference between either control condition, while signal from FAP-GFP-KDEL cells fed 50 μM N₃-Cho was 10-fold higher. Additionally, 1a-PC fluorescence was linearly dependent on N₃-Cho concentration in the medium (Fig. 1h). Together, these results indicate that FACES is capable of linear fluorescent quantification of N₃-PC, via its SPAAC reaction product 1a-PC, in the ER of living cells.

Imaging suborganellar PC distribution in mitochondria

To expand FACES to other organelles we used the OTC mitochondrial targeting sequence (MTS) to direct FAP-GFP to the mitochondrial matrix. When quantifying the performance of FACES in the mitochondrial matrix using CFLSM, we found that switching the C₆ alkane linker in 1a to a polyethylene glycol (PEG)-2 linker (compound 1b) was beneficial for dissociating unreacted dye from this compartment, reducing background fluorogen fluorescence in the absence of the SPAAC reaction. We suspect that the denser membranes of the IMM and/or presence of the outer mitochondrial membrane as an additional retention barrier, necessitating the more polar fluorogen 1b for optimal performance. Similar to the ER-targeted protein, MTS-FAP-GFP detection of 1b-PC generated fluorescence that linearly correlated with N₃-Cho concentration (Supplementary Fig.5a-c).

Using super-resolution lattice-structured illumination microscopy (lattice SIM²), we further tested whether the matrix soluble FAP could resolve structural features of the IMM. When imaged by lattice SIM², FACES signal from 1b-PC occurred in a banding pattern distinct from the general distribution of GFP in the mitochondrial matrix (Supplementary Fig. 5d). These experiments were repeated with cells counterstained with PKmito²³ to label mitochondrial cristae (Fig. 2a). Fluorescence intensity profiles drawn across individual mitochondria show that 1b-PC correlated well with PKmito, which at times formed alternating peaks with GFP (Fig. 2b). We interpreted this localization pattern as reflecting regions of high cristae density in the IMM that contrast with void areas of the matrix²⁴. This is in line with previous work using a chemically targeted approach to label mitochondrial N₃-PC¹⁵.

Figure 2. Using FACES to analyze PC trafficking between the mitochondrion and the ER.

a, SIM² of 1b-PC detection in mitochondria of MTS-FAP-GFP cells counterstained with the cristae marker PKmito. The boxed region is magnified to the right. White line indicates the segment used for plotting the fluorescent intensity profile. Scale = 4 μm and 2 μm (inset). 1b-PC fluorogenic signal forms a banded pattern that co-localizes with cristae. Additional example shown in Supp. Fig. 5d. b, Fluorescence intensity line profile from (a). c, Airyscan images show total N₃-PC pools labeled with BODIPY-DBCO overlap with both the ER (sec61mCherry) and mitochondria (mitoBFP). Scale = 10 μm. d, Airyscan images of a HeLa cell expressing MTS-FAP-GFP and Sec61b-mCherry (cyan) show that 1b-PC fluorescence detected in mitochondria does not overlap with the ER. Scale = 5 μm. e, Airyscan images of a HeLa cell expressing FAP-GFP-KDEL and mito-BFP (cyan) show that 1a-PC fluorescence detected in the ER does not overlap with mitochondria. Scale = 10 μm. For experiments (a-e) 100 μM N₃-Cho was incorporated for 24 h and 500 nM of either 1a (ER) or 1b (Mito) were used for PC labeling. f, Cartoon schematic depicting two proposed modes of PC transport into mitochondria. g-h, Knockdown of two putative ER-mitochondrial lipid transporters alters the distribution of PC between the two organelles. g and i, Representative LSM images of siRNA transfected HeLa cells stably expressing either MTS-FAP-GFP (g) or FAP-GFP-KDEL (i); Scale = 50 μm. e and g, Quantified fluorescence intensity ratios of fluorogen (1a or 1b) to GFP. Cells were treated with 25 pmol siRNA and 100 μM N₃-Cho for 48 h prior to labeling with 500 nM 1a (ER) or 1b (Mito). SPAAC NC represent cells that were treated with NC siRNA but were not any fed N₃-Cho. Experiments were conducted in triplicate and are color coded. Outlined circles reflect population means; n = 3, N = 100–116 cells per condition. Statistics reflected paired two-tailed T-test of the population means.

Lipid transporters contributing to PC trafficking between ER and mitochondria

PC is the most abundant lipid in mammalian cells and is a component of all organelle membranes, which make disentangling its subcellular transport mechanisms challenging. For instance, N₃-PC pools strongly overlap between the ER and mitochondria when labeled with a constitutively fluorescent dye like BODIPY (BODIPY-DBCO) (Fig. 2c). However, using FACES, organelle specific N₃-PC pools can be selectively illuminated and imaged. Cells expressing MTS-FAP-GFP detect on 1b-PC in mitochondria that does not overlap with the ER marker Sec61b-mCherry (Fig. 2d). Likewise, cells expressing FAP-GFP-KDEL detect 1a-PC in the ER that does not overlap with the mitochondria (Fig. 2e).

Utilizing the ability to quantify PC in mitochondria and ER, we tested how defects in lipid transport differentially affect these pools. For this, we generated stable cell lines expressing either FAP-GFP-KDEL (ER) or MTS-FAP-GFP (mitochondria) for further experiments. PC is synthesized in the ER and is transported to mitochondria by lipid transport proteins (LTPs)^22,25. We used siRNA to silence two LTPs: VPS13A, a bridge-type LTP that interacts with VAPB at ER-mitochondria contact sites and transports bulk phospholipids²⁶, and STARD7, a small shuttle-type LTP that localizes to the mitochondrial intermembrane space and is thought to transport PC between outer and inner membranes^27–29 (Fig. 2f). Silencing resulted in a >50% reduction in both VPS13A and STARD7 protein content in each experiment (Supplementary Fig. 6). In both cases vps13a and stard7 siRNAs reduced 1b-PC fluorescence in mitochondria by approximately 50% relative to negative control (NC) siRNA treated cells (Fig. 2g, h). However, only knockdown of vps13a increased 1a-PC fluorescence in the ER (Fig. 2i, j). This is consistent with the proposed role of VPS13A as an ER to mitochondrial LTP and that of STARD7 as an intra-mitochondrial LTP.

Transbilayer asymmetry of multiple phospholipids generated at the trans-Golgi network

There currently exist few, if any, tools capable of measuring membrane asymmetry in living cells. We asked if the spatial specificity of FACES could provide this capacity when FAPs are fused to the transmembrane domain for leaflet-specific fluorogen detection. A potential site of asymmetry generation in the secretory pathway is at the TGN, where vesicles bud off from the Golgi for trafficking of proteins and lipids to the highly asymmetric PM. We explored asymmetric lipid detection in the TGN by reciprocal tagging of the transmembrane protein TGN38, at N- and C-termini, to selectively orient FAPs toward either the exoplasmic or cytoplasmic facing leaflets (Fig. 3a). Here only the transmembrane domain and short cytosolic tail of TGN38 were used, in conjunction with a flexible glycine-serine linker, in order to bring the FAP into close proximity to the membrane surface. Using this strategy we generated stable HeLa cell lines expressing either FAP-TGN38-GFP (exoplasmic leaflet) or GFP-TGN38-FAP (cytoplasmic leaflet) under the control of a doxycycline inducible promoter. TGN38 cycles to and from the PM via secretory vesicles and early endosomes³⁰, thereby labeling a split population of trans-Golgi cisternae and trans-Golgi network vesiculated compartments³¹.

Figure 3. Imaging lipid asymmetry generated at the TGN with FACES.

a, Illustrated schematic representing the constructs used for reciprocal tagging of TGN38 with a FAP exposed to either the cytosolic or exoplasmic leaflet. b, 3D rendering of z-stacked Airyscan images reveal a lack of 1a-PC labeling on the exoplasmic leaflet of vesicles connected to TGN cisternae. Magnified images from the boxed region are shown to the right. Scale = 5 μm and 2 μm (inset). c-f, Fluorescence intensity ratio of 1a to GFP of the cytoplasmic leaflet (top) and exoplasmic leaflet (bottom) in TGN cisternae (TGN-C) and TGN vesicles (TGN-V) measured using LSM. Bar graphs represent mean population values that are normalized to the background fluorescence in SPAAC NC cells. Full data is available in Supp. Fig. 8. Error bars reflect SD; n = 1, N = 25 cells. Example Airyscan images are provided as z-stacked maximum projections. Magnified insets show labeling of the TGN-V population. Structures of N₃-containing lipid head groups are provided. Scale = 10 μm and 1 μm (inset). g, Colocalization of the secreted SM probe GFP-EqtSM and cytoplasmic 1a-PC in individual TGN vesicles. Here, 488 and 633 nm laser lines were used for imaging to avoid imaging of mCer3. Scale = 10 μm and 1 μm (inset). h, Colocalization of mCherry-GPI and cytoplasmic 1a-PC. Scale = 25 μm and 3 μm (inset). i, Conceptual model representing the proposed asymmetric distribution of PC and SM lipids that is generated alongside TGN vesicle formation.

Since these chimeric constructs lack the native N-terminal domain of TGN38, and could differ based on the reciprocal tagging strategy, it was necessary to investigate the identity of the vesicular population labeled by each construct. The fluorescent protein mCherry, with an N-terminal secretion signal peptide (SP-mCherry), was used as a marker for soluble secretory cargo and transfected in each cell line. Nearly all GFP-positive TGN vesicles colocalized with the mCherry marker in both cell lines (Supplementary Fig. 7a). Time lapse imaging further revealed TGN vesicles containing soluble secretory cargo rapidly leaving TGN cisternae (Supplementary Video 1). Next we used RAB5-BFP and RAB11-BFP as markers for early endosomes and recycling endosomes, respectively. RAB11 did not colocalize with GFP in either cell line (Supplementary Fig. 7c). However, in both cell lines a small subpopulation of GFP-positive TGN vesicles colocalized with RAB5 (Supplementary Fig. 7c). This is in agreement with retrograde trafficking of endogenous TGN38, which travel directly from the PM to the TGN in early endosomes that bypass the recycling endosome compartment³².

Next we tested the ability of FAP-TGN38-GFP and GFP-TGN38-FAP to detect 1a-PC in the TGN using FACES. Cells expressing FAP-TGN38-GFP, oriented toward the exoplasmic leaflet, generated 1a-PC fluorescence in the lumen of TGN cisternae that was either absent, or strongly diminished, in the vesicular population. Using Zeiss Airyscan imaging, lateral segregation between exoplasmic 1a-PC enriched and depleted regions could be observed at sites of vesicular budding from the TGN cisternae (Fig. 3b). In contrast, cells expressing GFP-TGN38-FAP, oriented towards the cytoplasmic leaflet, generated 1a-PC fluorescence in the TGN cisternae with very strong signal arising from the vesicular population. These observations were quantified using CFLSM and FACES by measuring the ratiometric fluorescence of TGN cisternae and the vesicular population (Fig. 3c). 1a-PC was equally distributed between both leaflets of TGN cisternae, with a slight enrichment on the cytoplasmic leaflet. However, the abundance shifted between these compartments, with a strong comparative enrichment in the cytoplasmic leaflet of TGN vesicles (Fig. 3c, Supplementary Fig. 8a). These results are consistent with previous measurements of PC in freeze-fractured yeast cells, which showed a cytoplasmic leaflet enrichment arising in the Golgi³³, although we cannot rule out other mechanisms by which lumenal labeling might be reduced in TGN vesicles.

We also asked if FACES could measure the asymmetry of non-choline lipids at the TGN. Recently, metabolic incorporation of azido-containing L-serine¹¹ (C-L-Ser-N₃) and myo-inositol¹² (Ins-2-N₃) have been shown to label PS and PI lipid classes, respectively. The latter is likely to label phosphorylated PIs¹², albeit at lower abundances. Both PS and PI classes are synthesized in the ER and have been proposed to be trafficked to the cytoplasmic leaflet of the trans-Golgi and TGN through lipid transfer proteins^34,35. Metabolic incorporation of Ins-2-N₃ and C-L-Ser-N₃ probes into N₃-PI and N₃-PS phospholipids was confirmed by high resolution LC-MS/MS (Supplementary Fig. 9). When analyzed with FACES, we observed 1a-PI in Golgi cisternae that was evenly distributed between exoplasmic and cytoplasmic leaflets (Fig. 3d, Supplementary Fig. 8b). In TGN vesicles, the distribution of 1a-PI shifted to the cytoplasmic leaflet, similar to that for PC. For 1a-PS, using C-L-Ser-N₃, there was an absence of signal on both leaflets of Golgi cisternae and a high, exclusively cytoplasmic abundance on TGN vesicles (Fig. 3e, Supplementary Fig. 8c). This is consistent with previous observations that the PS-binding C2 domain of lactadherin, when expressed in the cytoplasm, robustly labels TGN vesicles, but not Golgi cisternae³⁶. As a control for these cytosolic leaflet-enriched phospholipids, we imaged the distribution of surface glycans in cells fed with peracetylated N-azidoacetylmannosamine (Ac₄ManNAz)³⁷, which is metabolized into sialic acid (Sia) within the cell. We measured an exclusively exoplasmic distribution of 1a-Sia in both cisternae and vesicles (Fig. 3f, Supplementary Fig. 8d), consistent with the synthesis of protein and lipid surface glycans in the Golgi lumen³⁸. Lastly, co-staining cells expressing mCer3-TGN38-FAP (Blue FP variant) with equimolar amounts 1a and BODIPY-DBCO, a conventional fluorophore, demonstrated that FACES can spectrally separate organelle-specific lipid pools that overlap with total cellular phospholipid distributions (Supplementary Fig. 10)

Previous analyses of TGN vesicles³⁹ and the PM⁴⁰ suggest that loss of exoplasmic phospholipids could be substituted with SM, which is predominantly synthesized in the TGN and packaged in secretory vesicles. Consistent with this model, we observed that the SM probe GFP-EqtSM¹⁷ colocalized with vesicles enriched with cytoplasmic leaflet 1a-PC (Fig. 3g) that rapidly exit the Golgi (Supplementary Video 2). GFP-EqtSM contains an N-terminal secretion signal peptide and passes through the secretory pathway, so it only binds the lumenal face of Golgi membranes. Similar results were obtained for vesicles labeled with the secretory cargo mCherry-GPI (Fig. 3h, Supplementary Video 3), which also co-traffics with SM¹⁷. Overall, these results support a model in which phospholipids become enriched on the cytoplasmic leaflet during TGN vesicularization, while the exoplasmic leaflet becomes enriched in newly synthesized sphingolipids in the TGN lumen (Fig. 3i). The resulting vesicles move both protein cargoes and asymmetrically-distributed lipids to and from the PM.

PS asymmetry in the TGN is generated through lipid transport and flipping pathways

Genetic evidence in yeast suggests that the asymmetry of PS in the vesiculated TGN could be generated by P4-ATPase phospholipid flippases that translocate PS from the exoplasmic leaflet to cytoplasmic leaflet^41–45. How asymmetry is generated in mammalian TGN has remained obscure and it is unknown to what extent lipid flipping, or transport pathways at ER-TGN contact sites, contribute to PS enrichment in the cytoplasmic leaflet (Fig. 4a). In mammalian cells, ORP10 and ORP11 have been characterized as PS transporters at ER-TGN contact sites that form PI-4-phosphate counter exchange complexes with ORP9 ^46–48. Likewise, mammalian P4-ATPase phospholipid flippases play important roles in maintaining PS asymmetry in the PM and secretory pathway by redistributing lipids to the cytoplasmic leaflet^36,49–51. To ensure that the C-L-Ser-N₃ probe faithfully reports native PS distributions in the cells, we compared the localization of the PS-binding protein C2Lact-mCherry to that of N₃-PS labeled with BODIPY-DBCO, which fluorescently labels intracellular N₃-phospholipids. We found that BODIPY-conjugated N₃-PS labeled the same intracellular membranes as C2Lact-mCherry (Fig. 4b).

Figure 4. Mechanisms contributing to PS asymmetry at the TGN.

a, Cartoon schematic depicting two proposed lipid transport pathways generating PS asymmetry at the TGN. PS can be flipped from the exoplasmic leaflet to the cytoplasmic leaflet of the TGN by P4-ATPase/CDC50a flippases, or transported from the cytoplasmic leaflet of the ER to the cytoplasmic leaflet of the TGN by ORP9/10 LTPs at contact sites with VAPA. b, Colocalization of the PS-binding probe C2Lact-mCherry with N₃-PS lipids labeled with 1 μM BODIPY-DBCO. Magnified insets show colocalization of both markers in vesicles. Fluorescence line intensity profile from magnified inset is provided. Scale = 20 μm and 3 μm (inset). c and d, Quantified ratio of 1a-PS to GFP fluorescence in the cytoplasmic leaflet of the TGN in response to siRNA transfection. c, orp9 and orp10 silencing reduce 1a-PS relative to cells treated with NC siRNA. d, Silencing cdc50a also decreases 1a-PS in the cytoplasmic leaflet. e, Silencing cdc50a increases 1a-PS in the exoplasmic leaflet of the TGN-V relative to NC siRNA treated cells. Cells were incubated with 250 μM C-L-Ser-N₃ and doxycycline (1 mg/mL) during the 48 h siRNA transfection period, followed by labeling with 500 nM 1a. SPAAC NC represents cells that were not fed C-L-Ser-N₃ as a staining control. In all experiments n = 3, N = 25 cells per condition. Statistics reflected paired two-tailed T-test of the population means. f, Representative LSM images from siRNA experiments plotted in (c) and (d) using the cytosolic leaflet reporter GFP-TGN38-FAP. g, Representative LSM images from siRNA experiments plotted in (e) using the exoplasmic leaflet reporter FAP-TGN38-GFP. Scale = 10 μm and 2 μm (inset).

To gain mechanistic insight into how PS asymmetry is generated at the TGN, we used FACES to measure bilayer specific changes in N₃-PS distribution in cells silenced for orp9, orp10, or the essential P4A-ATPase subunit cdc50a^52,53. Silencing of orp9, orp10, and cdc50a target genes was validated by western blot and resulted in >50% reduction in each protein (Supplementary Fig. 11a-d). Silencing either orp9 or orp10 reduced 1a-PS by 39% and 58%, respectively, on the cytosolic leaflet of TGN vesicles compared to those treated with a non-targeted, negative control siRNA (Fig. 4c). In comparison, silencing cdc50a reduced 1a-PS by 40% on the cytosolic leaflet of TGN vesicles (Fig. 4d). Some cells silenced for cdc50a were also positive for annexin V staining, indicating accumulation of extracellular PS in the PM (Supplementary Fig. 11e). We thus tested whether silencing cdc50a would also cause PS accumulation in the exoplasmic leaflet of the TGN lumen. Silencing cdc50a increased exoplasmic 1a-PS content in vesicles (Fig. 4e). In these silencing experiments we did not notice obvious phenotypes in the TGN vesicle population despite the changes 1a-PS (Fig. 4f, g). Together these results highlight how different lipid transport mechanisms contribute to generating and maintaining PS asymmetry in the secretory pathway.

FACES reports on changes to mitochondrial glycosylation dynamics

Based on our detection of surface glycans in the Golgi lumen, we asked whether FACES could be used as a general tool compatible for imaging of azido-sugars in living cells. Tetraacylated N-azidoacetylgalactosamine is widely used as a metabolic reporter for glycoconjugates containing N-acetylhexosamines (HexNAz)⁵⁴. Ac₄GalNAz fed to cells is converted to UDP-N-azidoacetylgalactosamine (UDP-GalNAz) by the hexosamine salvage pathway and is then further epimerized to UDP-N-azidoacetylglucosamine (UDP-GlcNAz) (Fig. 5a). This results in robust labeling of N- and O- linked glycoproteins on the cell surface^54,55 (Fig. 5b), secretory compartments where they are formed, and endosomal compartments where they are recycled. Meanwhile, intracellular protein glycosylation is modulated by nucleocytoplasmic and mitochondrial O-linked β-N-acetylglucosamines (O-GlcNAc), which are added to serine/threonine residues by O-GlcNAc transferase (OGT) and removed by O-GlcNAcase (OGA)⁵⁴ (Fig. 5c). O-GlcNAc cycling has emerged as a regulator of metabolism, signaling, and immune evasion of cancer cells^56–58 and mitochondrial motility in neurons⁵⁹. However, intracellular glycosylations are not readily observed by standard bioorthogonal imaging approaches due to the much more extensive incorporation of Ac₄GalNAz into surface glycans; very little signal from conjugated conventional fluorophores colocalizes with mitochondria, for example (Fig. 5d). Currently, detection of O-GlcNAcylation is performed in fixed cells or bulk extracts using O-GlcNAc sensitive antibodies⁶⁰.

Fig. 5. Intracellular imaging of N-acetylhexosamine sugars with FACES.

a, Illustration depicting the metabolic fate of Ac₄GalNAz and Ac₄GlcNAz incorporation into cellular glycoconjugates containing N-acetylhexosamine sugars. Ac₄GalNAz fed to cells is converted to UDP-GalNAz in the cytosol and further epimerized to UDP-GlcNAz, thereby robustly labeling both UDP-HexNAz pools (UDP-GalNAz + UDP-GlcNAz). Conversely, Ac₄GlcNAz is not efficiently recycled to UDP-GlcNAz. b, UDP-HexNAz is incorporated into diverse extracellular glycoproteins proteins on the cell surface. Structures for an N-linked glycoprotein and O-linked mucin are shown as examples. c, UDP-HexNAz is incorporated into intracellular O-GlcNAcylated proteins. Protein glycosylation is catalyzed by O-GlcNAc Transferase (OGT) and can be reversed by O-GlcNAcase (OGA). OGA is pharmacologically inhibited by Thiamet-G (TMG). d, Bioorthogonal imaging of HeLa cells fed 100 μM GalNAz for 24h and labeled with 5 μM TAMRA-DBCO (cell surface) and 5 μM BODIPY-DBCO (intracellular), relative to the mitochondrial marker Mito-BFP. Scale = 20 μm. e-i, Stable MTS-FAP-GFP cells detect mitochondrial HexNAz. e, Quantified mitochondrial HexNAz incorporation in cells fed either 100 μM Ac₄GalNAz or 100 μM Ac₄GlcNAz for 24h, versus unfed controls (SPAAC NC). Box plots reflect the first, median, and third quartiles, while whiskers reflect min to max values; n = 1, N = 100 cells. Cells were labeled with 500 nM 1b and washed as previously described. Representative images of Ac₄GalNAz fed cells and NC SPAAC are shown on the right; scale = 20 μM. f, Response of mitochondrial HexNAz to OGA inhibition using Thiamet-G (TMG). Cells were treated with 50 μM Thiamet-G (or DMSO control) and fed 100 μM Ac₄GalNAz for 24h. Displayed values are normalized to the average background fluorescence of SPAAC NC cells used to control for 1b labeling. Box plots reflect the first, median, and third quartiles, while whiskers reflect min to max values; n = 1, N = 59 cells. g, Response of mitochondrial HexNAz to glucose availability. Cells were fed 100 μM Ac₄GalNAz and grown in DMEM containing either low (5.5 mM) or high (30 mM) glucose for 24h. Displayed values are normalized to SPAAC NC cells used as control for 1b labeling. Box plots reflect the first, median, and third quartiles, while whiskers reflect min to max values; n = 1, N = 63 cells. Representative LSM images (right) show 1b-HexNAz fluorescent intensities displayed using the FireLUT. Scale = 50 μm.

Using MTS-FAP-GFP expressing cells, we tested the efficacy of FACES in reporting mitochondrial-specific changes in HexNAz. Cells were grown in the presence or absence of either 100 μM Ac₄GalNAz or 100 μM tetraacylated N-azidoacetylglucosamine-tetraacylated (Ac₄GlcNAz) for 24h. Ac₄GlcNAz is not efficiently incorporated by the hexosamine salvage pathway⁵⁴ and thus was used as a less metabolically active surrogate. SPAAC staining with 1b generated fluorescence signal intensities 10-fold greater in cells fed 100 μM Ac₄GalNAz compared to unfed control cells (Fig. 5e). In contrast, 1b fluorescence from cells fed Ac₄GlcNAz, which is less efficiently incorporated into glycosylated proteins, was significantly lower than that for Ac₄GalNAz. This indicates that the FACES signal from 1b-HexNAz is from sugars that have been metabolically incorporated and not due to free azido sugars.

Next we tested whether FACES can report changes to mitochondrial O-GlcNAc levels resulting from pharmacological treatment or nutritional state. We found that the OGA inhibitor Thiemet-G increased 1b-HexNAz fluorescence, suggesting that the fluorogen signal was responsive to the activity of O-GlcNAc modifying enzymes (Fig. 5f). In cancer cells, glucose uptake is positively correlated with O-GlcNAc levels, which in-turn reprogram cellular metabolism tailored towards glucose availability⁶¹. We observed that switching cells from high (30 mM) to low glucose (5.5 mM) levels during Ac₄GalNAz labeling reduced 1b-HexNAz fluorescence by 30% (Fig. 5g). This is similar to glucose-dependent changes in mitochondrial O-GlcNAc levels previously reported in breast cancer cells⁶². Taken together, these data demonstrate that FACES can report changes in mitochondrial HexNAz incorporation as a function of cell metabolism and O-GlcNAcylation in living cells, and support the hypothesis that mitochondrial O-GlcNAcylation is a post-translational modification that is responsive to the cell’s dietary state.

Discussion

Studying cellular roles for non-proteinaceous biomolecules has long been hindered by the lack of robust tools for their imaging in live cells. Here we introduce a new approach, FACES, that combines the chemical specificity of metabolic incorporation with the spatial control of genetic protein tagging for imaging of cellular metabolites. FACES harnesses reversible fluorogenic systems, where fluorescence increases many orders of magnitude upon specific binding to a FAP. It combines this turn-on fluorescence with the specificity of bioorthogonal reactions for molecular sensing. This approach is powerful for the subcellular imaging of phospholipids, their transport between organelles, and orientations across membrane leaflets, but can also be applied to other biomolecules. Metabolic labeling has emerged as a particularly important tool for glycobiology and we demonstrate that FACES is responsive to labeled glycoproteins in mitochondria, which have been proposed to act as regulators of metabolic function⁶³. Future applications of FACES could be further expanded through increases in the repertoire of metabolic labeling, bioorthogonal chemistry, and the spectral plasticity of the fluorogen employed.

Compared to other bioorthogonal imaging modalities, FACES has several key advantages for mechanistic lipid membrane biology. For organelle-defined imaging, protein targeting sequences are more specific and generalizable than corresponding chemically-encoded localization tags. We show that this capacity can be applied to the quantitative characterization of lipid distribution phenotypes affected by trafficking networks across membrane contact sites. Beyond organelle-specific imaging, FAPs can also be fused to specific TMDs to read out bilayer asymmetry. Here, FACES provides effective discrimination beyond what is possible with direct fluorescence imaging, as the distance between two bilayer leaflets is less than 4 nm. Fluorogens show no fluorescence when unbound to their corresponding FAP, so quantitative detection of conjugate products occurs through the resulting fluorescence signal, not a change in its localization as for lipid binding proteins. Because of this capacity, FACES is the only method, to our knowledge, that allows for measurement of membrane asymmetry in live cells.

Applications of FACES could be further expanded through increases in the repertoire of metabolic labeling, bioorthogonal chemistry, and the spectral plasticity of the fluorogen employed. Conjugation of fluorogens to biomolecules is dependent on tagged metabolite incorporation, which are continuously expanding across biological systems⁶⁴ and molecular classes⁶⁵. One concern with any metabolic labeling approach is that tagged molecules might not fully mimic native compounds, which is especially relevant for lipids with minimal headgroups. For several azido lipids, previous studies, as well as the data presented here, suggest that compounds localize across a range of cells as expected based off of comparative analyses with other tools^10,11, though it is challenging to preclude all differences between labeled and unlabeled species. An additional limitation common to bioorthogonal approaches is the time required for conjugation and washing out of unreacted fluorogens (1 h); these might make the analysis of rapid redistributions of the underlying labeled species, like those relevant for lipid signaling⁶⁶, challenging. Tuning fluorogenic reagents for faster conjugations⁶⁷ and washing could decrease the time for processing steps to capture dynamic distributions.

Our experiments demonstrate that FACES is an especially powerful tool for investigating cellular mechanisms by which lipids are transported between organelles and across individual membranes, an intense area of investigation. A plethora of shuttle and bridge-like LTPs that perform lipid exchange at contact sites⁶ and flippases/floppases/scramblases that act on bilayer asymmetry⁶⁸ have been identified and biochemically tested. However, these transporters are characterized by overlapping affinities, apparent redundancies, and close interactions with lipid metabolic enzymes; how they act in concert to shape the membrane heterogeneity that defines eukaryotic cells is still unknown. Here, we identify transporters that contribute to the asymmetry of several labeled phospholipid classes at the TGN. The Golgi has long been proposed as a site of asymmetry generation, given the co-occurrence of secretory cargo sorting to the PM and sphingolipid synthesis⁶⁹. We show that for one phospholipid class, PS, cytosolic LTPs and transbilayer flippases both contribute to the asymmetry observed in TGN vesicles. These secretory compartments are enriched in GPI-anchored secretory cargoes that associate with the exoplasmic leaflet of the PM, suggesting a mechanism by which asymmetry at the PM is generated earlier in the secretory pathway. Lipid asymmetry generated at the TGN could also be relevant for sorting membrane proteins based on the asymmetric TMD sequences that are required for their PM targeting^3,70. As a tool that provides spatially-defined molecular quantitation, FACES could be broadly powerful for untangling how lipid trafficking, generation of membrane asymmetry, and transport of cargoes simultaneously occurs throughout eukaryotic cells.

Methods

Synthesis of fluorogen reagents:

Synthesis schematic of compounds 1a and 1b are shown in Supplementary Fig. 1 and synthetic protocols are detailed in the Supplementary Note. Products were validated by LC-MS and ¹H/¹³C NMR.

Generation of constructs:

FAPs were cloned from pcDNA3.1-KozATG-dH6.2–2XG4S-actin²¹, pcDNA3.1-kappa-myc-dK-2xG4S-TMst⁷¹, and pcDNA3.1-kappa-myc-dL5–2XG4S-mCer3-KDEL⁷² (Addgene #73263, #145766, #73209). These FAP sequences and the pcDNA3.1 backbone were used for generating the FAP-GFP-KDEL, GFP-KDEL, and MTS-FAP-GFP constructs used in transfection experiments. The Ig Kappa secretion signal peptide was used in conjunction with a KDEL sequence for ER-targeted constructs, while the targeting sequence from OTC was used for MTS constructs. For stable expression, FAP-GFP-KDEL and MTS-FAP-GFP coding sequences were subcloned into the sleeping beauty expression vector SBbi-PUR⁷³. FAPs targeted to the TGN used only the transmembrane domain and cytoplasmic tail of TGN38 and included an N-terminal secretion signal peptide from VSV-G. The linker sequences (SGGGGSFGLSGL) and (AAGLAHGSGSG) were used in between the FAP and TGN38 sequences, for N-terminal and C-terminal tagged constructs, respectively. Reciprocally tagged FAP-TGN38-GFP and GFP-TGN38-FAP constructs were assembled into the SBtet-Pur expression vector. The GPI-anchor marker protein was generated by subcloning a SBP-mCherry-GPI sequence, received from the Lippincott-Schwartz lab, into pcDNA3.1. The secreted mCherry construct was generated by removing the SBP and GPI sequences and adding a stop codon. The GFP-EqtSM plasmid was received as a gift from the Burd lab. RAB5-BFP (Addgene #49147) and RAB11-BFP (Addgene #79805) were received from Addgene. Additional Sec61mCherry and Mito-BFP plasmids were received from the Lippincott-Schwartz lab. All constructs were validated by whole plasmid sequencing. All FAP plasmids generated in this study have been deposited in Addgene (#242990, #242991, #242992, #242993, #242994).

Cell culture:

HeLa (ATCC CCL-2) cells were received from Millipore Sigma (93021013-1VL) and grown in Dulbecco’s Modified Eagle Medium (Gibco 11995-065) containing glucose (4.5 g/L), glutamine (4.5 g/L), sodium pyruvate (110 mg/L), phenol red, 10% FBS, and 1% PenStrep, at 37°C and 5% CO₂ atmosphere. For microscopy, cells were seeded into 35 mm glass bottom dishes (MatTek) coated with fibronectin (7.5 μg/mL). For transient expression, cells were transfected with Lipofectamine 3000 in Opti-MEM media for 4 h, and changed into normal growth medium containing azido-containing bioorthogonal metabolite for 24 h prior to SPAAC labeling and imaging. Stable cell lines expressing FAPs were generated using the Sleeping Beauty transposon insertion system. After transfection, cells were grown under selection with puromycin (10 μg/mL) for 2 weeks until all showed fluorescence. Doxycycline (1 μg/mL) was used to induce expression in SBtet-FAP-TGN38-GFP, SBtet-GFP-TGN38-FAP, and SBtet-mCer3-TGN38-FAP cell lines (Fig. 3, Fig. 4, Supplementary Fig. 7, Supplementary Fig. 8, Supplementary Video 1, Supplementary Video 2, Supplementary Video 3). Cells were routinely tested for mycoplasma by MycoStrip (InvivoGen).

Metabolic labeling:

1-azidoethyl-choline (N₃-Cho, Jena BioScience) and resuspended in PBS to stock concentration of 50 mM. Ac₄ManNAz, Ac₄GalNAz and Ac₄GlcNAz (Click Chemistry Tools) were resuspended in DMSO to stock concentration of 100 mM. L-Serine analogue C-L-Ser-N₃ and myo-Inositol analogue Ins-2-N₃ were synthesized as previously reported, verified by ¹H/¹³C NMR and high resolution mass spectrometry, and resuspended in PBS to a stock concentration of 50 mM and 100 mM, respectively. In all cases, metabolic incorporation occurred for 24 h, except for experiments in siRNA experiments Fig. 2 and Fig. 4, which occurred for 48 h. Afterwards, cells were washed once with HBSS buffer, and incubated in fresh growth media for 30 min to allow for metabolic consumption of any remaining bioorthogonal metabolites. For experiments with peracetylated sugars, 1 h incubation was used.

SPAAC conjugation:

Culture media was removed by aspiration and replaced with staining media for SPAAC labeling. The staining media contained 500 nM 1a or 1b diluted in serum-free DMEM clear (Gibco 31053-028), and incubated for 15–20 min at 37°C and 5% CO₂ atmosphere. Cells were washed with 4 volumes of culture media containing 10% FBS, 3 times, at 15 min intervals to remove unreacted dye. For experiments with the mitochondrial-localized FAP, cells were incubated for an additional 45 min, followed by a fourth wash step prior to imaging. BDP-FL--DBCO was purchased from BroadPharm (BP-23473) and 1–2 μM was used for constitutive fluorescent labeling of azido-containing metabolites.

Lipid analysis:

Lipids were extracted with a modified Bligh and Dyer protocol. Briefly, cell pellets from T75 culture flasks were resuspended in 200 μL PBS and extracted with 2 mL 1:2 CHCl₃:MeOH (v/v) and cell debris were removed by centrifugation. Extracts were transferred to new vials and 1 mL chloroform was added, followed by 1 mL H₂O, and shaking. The lower organic phase was collected after centrifugation and the aqueous phase was extracted twice more with 650 μL and pooled. Lipid extracts were then dried under N₂ and resuspended in 100 μL 1:2 CHCl₃:MeOH for thin-layer chromatography (TLC) and 500 μL methanol for LC-MS/MS. For TLC, entire lipid extracts were spotted onto 10×10 cm HPTLC silica gel 60 plates, developed with 14:5:1 CHCl₃:MeOH:NH₄, and imaged with a Typhoon FLA 9500 using 633 nm laser. For high resolution LC-MS/MS (Supplementary Fig. 2a, 2b, 9), 5 μL sample was injected, and HRMS spectra of the total lipid extracts from HeLa cells were recorded on a Q Exactive Orbitrap mass spectrometer (Thermo Fisher Scientific) connected to a Vanquish Flex Analytical UHPLC System (Thermo Fisher Scientific) equipped with a C8 column (100 × 2.1 mm, particle size 1.9 μm). Following HPLC method was used for all runs using 0.1% formic acid in water (Solvent A) and 0.1% formic acid in methanol (Solvent B): 0–1 min 30% B, 1–4 min 30%–90% B, 4–15 min 90–99% B, 15–18 min 99%B, 18–20 min 30% B. Lipid standards (POPC, N₃-DSPC) were supplied by Avanti Polar Lipids. Global polar phospholipid analysis (Supplementary Fig. 2c-d) was performed by ESI-MS/MS at the Kansas State Lipidomics Research Center as previously described^74–76.

Image collection:

Confocal micrographs were acquired on a Zeiss LSM 880 microscope equipped with plan-apochromat 63x/1.4 NA or 20x/0.8 NA objectives. For far-red fluorogen-FAP signals, a 633 nm HeNe laser was used at 1–3% power. FP or other fluorophores were excited with 488 nm Argon (GFP), 405 nm (BFP/mCer3) or 561 nm (mCherry) diode lasers. All images used for intensity quantification were acquired in LSM mode using a QUASAR GaAsP detector using identical settings. The calculated power density for individual experiments are as follows: FAP-GFP-KDEL (Fig. 2g, h), 963 W/cm² (488 nm) and 2600 W/cm² (633 nm) with a pixel dwell time of 1.02 μs; MTS-FAP-GFP (Fig. 2i, j), 375 W/cm² (488 nm) and 2600 W/cm² (633 nm) with a pixel dwell time of 1.02 μs; GFP-TGN38-FAP and FAP-TGN38-GFP (Fig. 3c-d, Fig. 4c, d, f), 535 W/cm² (488 nm) and 2600 W/cm² (633 nm) with a pixel dwell time of 2.05 μs; FAP-TGN38-GFP (Fig. 4e, g), 1,017 W/cm² (488 nm) and 3,800 W/cm² (633 nm) with a pixel dwell time of 2.05 μs. Images highlighting fluorogen distribution (Figs. 1e, 2c, 2d, 2e, 3b, 3c, 3d, 3e, 3f, 3g, 3h, ; Supplementary Figs. 7a, 7c, 10; Supplementary Videos 1, 2, 3) were acquired with an Airyscan detector and processed using default settings in Zen Black; these were not used for quantification due to non-linear processing. SIM imaging of mitochondrial 1b-PC was performed on an Zeiss Elyra 7 Lattice-SIM² system equipped with a plan-apochromat 100x/1.46 NA objective using 642 nm and 488 nm diode lasers (10% power each). Images were acquired in ZEN Black using Weak Live default settings with a pixel dwell time of 843 ms.

Image quantification:

All intensities were calculated using ImageJ on unprocessed images. For proof of concept experiments using transfected cells (Fig. 1), line profiles drawn were across 10 individual ER segments per cell and the total fluorescence ratio between fluorogen and GFP emissions were tabulated across them. For per-cell data (Fig. 1f, 1g, 1h), the 10 segments were averaged and 10 cells were analyzed per condition. Raw per segment fluorescence intensity is shown for all 100 segments (Fig. 1h). For mitochondrial and ER quantification (Fig. 2), fields of cells stably expressing to corresponding FAPs were analyzed by segmenting individual cells and computing the total fluorescence ratio between fluorogen and GFP emission across the entire cell, given the highly specific distribution of the signal for the compartment of interest. For TGN cisternae quantification (Fig. 3, Fig. 4), line segments were drawn across the Golgi apparatus of individual cells so that they did not intersect any TGN vesicles. For TGN vesicle quantification, 5 line segments were drawn across Golgi vesicles that did not intersect Golgi cisternae. For each compartment, the total fluorescence ratio between fluorogen and GFP were calculated across each and averaged together for a single cell value. Values were measured this way for 25 cells per condition (exoplasmic vs. cytosolic for PC, PS, PI, and Sia). Mitochondrial quantification in Fig. 5 was carried out identically as in Fig. 3. All statistical analyses were performed in GraphPad Prism 9.

Gene knockdown:

Silencer Select siRNAs were supplied by Thermo Fisher: STARD7 (s32370; GCACCCACCUUUACCAGUA), VPS13a (s531256; GCAACAGGUCUACUGUAUA), ORP9 (s41693; CCGUAGAUGCAAUAUAUCA), ORP10 (s223235; AAUACCACAUGGAAAUGAA), CDC50a (s31427; CGAGAUUGAUUAUACCGGA), and negative control #1 (4390843). Cells were transfected with Lipofectamine RNAiMAX (ThermoFisher) complexed with 25 pmol siRNA for 48h prior to analysis. 50 pmol siRNA was used for CDC50a. Total protein was extracted using RIPA Lysis and Extraction buffer (ThermoFisher) and measured by Bradford assay (Bio-Rad). A total 20 μg protein per extract was boiled in 1X reducing sample buffer and used for SDS-PAGE (BioRad, 4561083). Validation of gene knockdowns was performed by Western blotting using primary antibodies against STARD7 (Proteintech, 15689) 1:2500, VPS13a (Atlas antibodies, HPA021662) 1:1000, ORP9 (Proteintech, 11879-1-AP) 1:1,000, ORP10 (Proteintech, 15491-1-AP) 1:1000, CDC50a (Invitrogen, PA5-99762) 1:1000, α-Tubulin (ThermoFischer, 32-2500) 1:5000, and Actin (Invitrogen, PA5-78715) 1:5000. HRP-conjugated goat anti-rabbit IgG secondary antibody (Abcam, 6721; 1:10000) and goat anti-mouse secondary antibody (Invitrogen, PI31430; 1:10000) and Pierce ECL western blotting substrate (Thermo Fisher, 32209) were used for detection.

Data Availability

All data are available in the Article, Supplementary Information, and Source Data Files. Plasmids generated in this study have been deposited in Addgene (#242990, #242991, #242992, #242993, #242994).