Quantitative imaging of lipid transport in mammalian cells

Abstract

Eukaryotic cells produce over 1,000 different lipid species that tune organelle membrane properties, control signalling and store energy^1,2. How lipid species are selectively sorted between organelles to maintain specific membrane identities is largely unclear, owing to the difficulty of imaging lipid transport in cells³. Here we measured the retrograde transport and metabolism of individual lipid species in mammalian cells using time-resolved fluorescence imaging of bifunctional lipid probes in combination with ultra-high-resolution mass spectrometry and mathematical modelling. Quantification of lipid flux between organelles revealed that directional, non-vesicular lipid transport is responsible for fast, species-selective lipid sorting, in contrast to the slow, unspecific vesicular membrane trafficking. Using genetic perturbations, we found that coupling between energy-dependent lipid flipping and non-vesicular transport is a mechanism for directional lipid transport. Comparison of metabolic conversion and transport rates showed that non-vesicular transport dominates the organelle distribution of lipids, while species-specific phospholipid metabolism controls neutral lipid accumulation. Our results provide the first quantitative map of retrograde lipid flux in cells⁴. We anticipate that our pipeline for mapping of lipid flux through physical and chemical space in cells will boost our understanding of lipids in cell biology and disease.

Fig. 1. Lipid probe library, imaging and MS workflows, and lipid transport time-course experiments.

a, Schematic of the combined analysis of lipid transport and metabolism. Lipid probes were loaded into the PM using α-methyl-cyclodextrin-mediated exchange reactions, crosslinked and fluorescently labelled for imaging or extracted and analysed by MS to monitor metabolism. b, The bifunctional lipid probes synthesized for this study. Unique structural elements are highlighted in red. c, Lipid delivery to the PM and the selectivity of lipid labelling for PC(Y/18:1). Scale bar, 10 µm. d, Ultra-high mass resolution (resolution_m/z=800 = 420,000) enables baseline separation of peaks spaced by a few millidaltons and their unequivocal assignment to molecular ions of lipids (as annotated; [M-H]⁻/[M+HCO₂]⁻) in total lipid extract. M2: second isotopic peak; PI(33:1)D7: deuterated internal standard. e, Representative images (PC(Y/18:1), 30 min timepoint) showing lipid signal (left) and individual organelles markers (right) by four-colour fluorescence imaging. Scale bars, 10 µm. f, Lipid signal assignment for cells shown in e based on automated image segmentation. Lipid signal images are shown at identical settings and scale in e and f. g, Representative images from time-course experiments show the temporal development of the lipid signal distribution for PC(Y/18:1) (left). The coloured arrows indicate lipid localization in different organelles (green, PM; yellow, ER; cyan, mitochondria (Mito); violet, Golgi apparatus; magenta, endosomes (Endo); grey, lipid droplets). Scale bar, 10 µm. Images were brightness–contrast adjusted to enable comparison of lipid distributions at different timepoints. Right, quantification of temporal development of intracellular lipid distribution for PC(Y/18:1). Kinetics were constructed from five independent timepoints. Data are mean ± s.d. Individual n values are provided as source data. Source Data

Fig. 2. Retrograde lipid transport occurs mainly through non-vesicular routes.

a, Schematic of the analysed cellular lipid-transport pipelines. MCS, membrane-contact site. b, Kinetic models for quantifying lipid transport from fluorescence microscopy and MS data. Non-ves., non-vesicular. c, Kinetics of lipid transport for PC(Y/16:0), PC(Y/18:1), PC(Y/20:4), SM(Y) and the corresponding model 1a fits. Unique structural elements of individual lipids are highlighted in red. Scale bars, 10 µm. Images were brightness–contrast adjusted to enable comparison of lipid distributions at different timepoints. Kinetics were constructed from five independent timepoints. Data are mean ± s.d. Individual n values are provided as source data. d, Comparison of rate constants describing retrograde vesicular transport from the PM to endosomes (models 1a–3a shown). e, Comparison of rate constants describing retrograde non-vesicular transport from the PM to the ER and total transport in the anterograde direction. f, Comparison of rate constants describing retrograde vesicular transport from the PM to endosomes and retrograde non-vesicular transport from the PM to the ER for all analysed lipid probes. The mean ± s.d. of the fitted rate constants was calculated from 100 MC model runs. Pairwise effect-size values are shown for PC species comparison of model 1a rates. Values for all lipids pairs are given in Extended Data Fig. 7. For d and f, Cohen’s d values are indicated; very small (VS), >0.01; small (S), >0.20; medium (M), >0.50; large (L), >0.80; very large (VL), >1.20; huge (H), >2.00. Source Data

Fig. 3. Pharmacological and genetic perturbations confirm non-vesicular transport as the primary retrograde lipid-transport route.

a, Inhibition of vesicular trafficking between the Golgi and the ER (brefeldin A) and of endosome formation and trafficking (wortmannin). b, Transferrin uptake and localization and lipid localization in control cells and drug-treated cells at 4 min and 30 min after lipid loading. Inhibition of vesicular trafficking does not affect retrograde lipid transport. Scale bar, 10 µm. Transferrin images are shown at identical settings; lipid images were brightness–contrast adjusted to enable comparison of lipid distributions at different timepoints. c, Quantification of the lipid distribution between the ER and endolysosomes after treatment with DMSO (grey), brefeldin A (pink) or wortmannin (light blue). Data are mean ± s.d. Individual n values are provided as source data. Int., internalized. d, Schematic of GLTPD1-mediated lipid transfer and kinetic model used to assess the effects of CPTP KO on non-vesicular transport of SM(Y). e, Comparison of SM(Y) time-course experiments in CPTP-KO and U2OS WT cells. Scale bar, 10 µm. f, Rate constants for retrograde and anterograde PM–ER lipid transport for PC(18:1/Y), PE(18:1/Y) and SM(Y). Kinetics were constructed from six independent timepoints (mean ± s.d.) containing 5 field of views each with 5–10 cells. The mean ± s.d. was calculated from 100 MC model runs. Statistical analysis was performed using pairwise effect-size tests; Cohen’s d values are indicated. Source Data

Fig. 4. Genetic perturbation experiments confirm the involvement of flippases in species-specific directional lipid transport.

a, Schematic of lipid trans-bilayer movement (lipid flipping) and non-vesicular lipid transport by lipid transfer proteins. P_i, inorganic phosphate. b, Kinetic model for the exchange of lipids between the PM and the ER. c, Comparison of time-course experiments for PE(18:1/Y), showing that lipid internalization dynamics are slower in HCT116 TMEM30A-KD cells than in HCT116 wild-type cells. The coloured arrows indicate lipid localization in different membrane types (green, PM; yellow, ER). Scale bar, 10 µm. Images were brightness–contrast adjusted to facilitate comparison of intracellular lipid localizations. d–f, Quantification of PE(18:1/Y) (d), PC(Y/16:0) (e) and PC(Y/20:4) (f) internalization kinetics and model fits. Kinetics were constructed from five independent timepoints. Data are mean ± s.d. Individual n values are provided as source data. g, Rate constants and quasi-equilibrium constants for retrograde and anterograde PM–ER lipid transport for PC(Y/16:0), PC(Y/20:4) and PE(18:1/Y). The mean ± s.d. was calculated from 100 MC model runs. Statistical analysis was performed using pairwise effect-size tests; Cohen’s d values are indicated. Source Data

Fig. 5. Lipid metabolism is approximately one order of magnitude slower than lipid transport.

a, Bifunctional lipid retention and turnover. The fraction of initially supplied species as the percentage of all bifunctional (BF) lipids and the fraction of bifunctional lipids of the total lipidome for PC(18:1/Y), PE(18:1/Y) and SM(Y). b, The fraction of initially supplied lipid probe as the percentage of the bifunctional lipidome as a proxy for the speed of lipid metabolism. The solid lines indicate mono-exponential fits. SM(Y) data were not fitted as very little interconversion was observed; instead, a linear interpolation is shown. For a and b, data are mean ± 95% confidence intervals of three biological repeats containing two technical replicates each. c, Comparison of the determined mono-exponential rate constants for the metabolism of individual lipid species. The error bars show the s.e. of the mono-exponential fit. n.d., not determined. d, Comparison of transport and metabolic rate constants shows that lipid transport is at least one order of magnitude faster. Error bars were obtained by error propagation. e, Lipid transport and metabolism rate constants are highly correlated for PC species despite a clear time-scale separation. The plot shows the k_{mono-exponential} calculated from fitting the data to a single exponential decay versus the mean of k_PM–ER from 100 MC runs. The error bars show the s.e. (metabolic rate constants) and s.d. (transport rate constants), calculated from 100 MC model runs. Lin. reg., linear regression.

Extended Data Fig. 1. Biophysical characterization of bifunctional lipid containing model membranes, lipidome assessment after bifunctional lipid loading and lipid imaging signal comparison for all probes.

a, b. Formation of liquid ordered L_o (stained by Bodipy-FL-GM₁; green) and liquid disordered L_d (stained by DiD; red) microdomains is unaffected by replacing 5 % of DOPC content with PC(18:1/Y) in giant unilamellar vesicles GUVs. Scale bars: 40 µm. c. Formation of ganglioside nanodomains leading to faster deexcitation of Bodipy-FL-GM₁ donors via FRET is unaffected by replacing 5 % of POPC content with PC(18:1/Y) in GUVs. Scale bar is 40 µm. d. Comparison of lipidome composition directly after lipid loading bifunctional lipid probes (4 min timepoint) with control lipidome. Arrows indicate supplied lipid type. Bars show the mean of 3 biological repeats containing 2 technical replicates each. e. + UV lipid signal vs -UV lipid signal for all probes, 30 min timepoint shown. Note: The high intensity in -UV conditions for PE(18:1/Y) is explained by the fact that PE can be chemically fixed with formaldehyde due to its primary amine group, which is not the case for the other lipids. Scale bar: 10 µm, all images shown at the same magnification.

Extended Data Fig. 2. Optimization of the lipid imaging protocol.

a. Structure of PC(18:1/Y) used for protocol optimization. b. Representative imaging results using optimized lipid loading, crosslinking and click chemistry conditions. Images from experiment and control samples are adjusted to the same intensity. Scale bar: 20 µm, all images shown at the same magnification. c-e. Characterization of UV illumination in the 96-well plate format used for this study. f-h. Optimization of lipid loading and crosslinking conditions. Dashed red lines indicate chosen conditions. Dashed red lines indicate the selected condition for further experiments. Cells were fixed directly after incubation with the loading solution. For f and h samples were incubated with liposomes for 4 min. f-h. Mean and 68% CI of 3 independent experiments. i. Lipid loading does not compromise cell membrane integrity as demonstrated by exclusion of 4 kDa FTIC-Dextran from cell interior. Lipid loading solution was incubated for 10 min before imaging. j-m. Lipid signal visualization using different Picolyl-Azide dyes. AF594-Picolyl-Azide was used for this study. Scale bars: 20 µm.

Extended Data Fig. 3

a-i Lipid transport time courses for all probes and timepoints. Representative images for lipid transport time course experiments. Scale bars: 10 µm, all images shown at the same magnification. Images are brightness-contrast adjusted to facilitate comparing intracellular lipid localization. The full dataset 3D dataset including marker channels can be accessed on https://lipidimaging.org/. N = 3 for each lipid and timepoint.

Extended Data Fig. 4. High time-resolution time courses for PE (18:1/Y) and PA(18:1/Y) and image analysis pipeline.

a, b. Representative images for lipid transport time course experiments at higher time resolution using PA(18:1/Y) (a) and PE(18:1/Y) n = 3 (b). Scale bars: 10 µm. Images are brightness-contrast adjusted to facilitate comparing intracellular lipid localization. n = 3. c, d. Background subtraction strategy. For most data, background was removed using a predicted noise image derived from control images (+ UV, -lipid). In cases where a AF647-Tom20 antibody was used as a mitochondrial stain, we observed a faint mitochondrial signal in the AF594 (lipid) channel in control conditions. For the corresponding +lipid images we estimated the extent of the bleed-through signal by determining the correlation between the mitochondrial signal in the marker channel and the lipid channel, using these parameters to generate an image for the expected artefactual mitochondrial signal in +lipid images & subtracting it from the raw +lipid image. Scale bars: 20 µm. e. Segmentation of marker channels to generate probability masks and representative result of lipid channel background removal. Scale bar: 20 µm. f. Lipid signal assignment to individual organelles shown in e. Scale bar: 20 µm. g. Representative image (PC(Y/18:1), 30 min timepoint) showing lipid signal and individual organelles markers (PM, endosomes, mitochondria) by four-colour fluorescence imaging (left panels). Scale bar: 10 µm. Right panels: Lipid signal assignment for cells shown based on automated image segmentation.

Extended Data Fig. 5. Ultra-high resolution (UHR) shotgun lipidomics of bifunctional lipid probes.

a. Shotgun UHR mass spectrometry resolves lipid peaks spaced by a few mDA and matches bifunctional precursors and their metabolites in multiple lipid classes. Blue line: Section of the spectrum acquired at the conventional (R_s 120,000) resolution on Q Exactive mass spectrometer; orange line: Same spectrum section acquired at R_s ~ 1 M resolution using optional Booster X2 data processing system and extended (2 s) transients. Vertical lines are peak centroids. b. Mol% profile acquired at two time points (see inset for colour coding) of 23 lipid classes (light bars), of which 7 classes comprise lipids with bifunctional lipid moieties (dark bars) produced from PC Y/20:4. c. PC profile covering 22 species with 5 species containing the bifunctional fatty acid. PCs bearing a bifunctional fatty acid (16:1) are annotated as endogenous lipids having the same number of carbons and double bonds in both FA moieties, albeit having different (+28.0061 Da) masses. d. Bifunctional fatty acids from the source PC(Y/20:4) are, incorporated into different lipid classes e.g., the cellular TAG pool consisting of 33 species with 14 species bearing the bifunctional fatty acid. The molar abundance of PC species containing the bifunctional fatty acid other than PC(Y/20:4) (c) and TAG (d) species increases with time, while the abundance of the starting PC(Y/20:4) decreases. Molar% profiles of native lipid classes (b), but also the species profile within PC and TAG classes (c, d) are not perturbed, indicating that the supplemented bifunctional lipids act as true tracer compounds and do not change the overall lipidome compositions. e. Comparison of bifunctional lipid metabolism with the native, isotope labelled, monounsaturated PC species PC(18:1/16:0[¹³C]). f. Development of the PC species distribution of PC(18:1/16:0[¹³C]) and palmitate-containing bifunctional PC species shows similar persistence of the original species in the labelled lipidome and similar product species forming. Mean and 95% CI of 3 biological repeats containing 2 technical replicates each are shown for b-f.

Extended Data Fig. 6. Kinetic analysis of lipid imaging and lipid MS data.

a. Data processing steps for quantification results from lipid imaging time course experiments exemplary shown for PC(Y/18:1) (see Supplementary Information for details). b. Transport scheme detailing kinetic models 1b-3b. c. Model 1a-3a fits for PC(16:0/Y), PC(18:1/Y), PC(20:4/Y). d. Model 1a-3a fits for PC(Y/16:0), PC(Y/18:1), PC(Y/20:4). e. Model 1a-3a fits for PE(18:1/Y), SM(Y). Datapoints in were constructed from 5 independent time points (Mean and SD). Individual n-numbers can be found in the source data files. Source Data

Extended Data Fig. 7. Results of kinetic analysis and estimation of lipid flow through vesicular and non-vesicular pathways.

a, b. Rate constants for vesicular transport from endosomes to the Golgi and from the Golgi to the ER derived from models 1a-3a. c. Rate constants describing lipid cycling (lipid exchange with the extracellular space, model 2a) and lipid depletion (model 3a). d. Ratio of rate constants describing lipid exchange between the ER and mitochondria (models 1a-3a). Note: Individual rate constants could not be identified from the data, presumably as preceding lipid transport steps were rate-limiting. e-g. Comparison of rate constants describing retrograde vesicular transport from the PM to endosomes and retrograde non-vesicular transport from the PM to the ER for all analysed lipid probes, corresponding models 1a and 1b, 2a and 2b, 3a and 3b shown together. h. Ratio of rate constants describing lipid exchange between the ER and mitochondria (models 1b, 2b, 3b). Note: Individual rate constants could not be identified from the data, presumably as preceding lipid transport steps were rate-limiting. a-h Error bars: mean and SD from 100 MC model runs. i. Fraction of bifunctional lipids transported via the non-vesicular route to the ER and the vesicular route to endosomes during retrograde transport, model 1a rate constants used for simulations. Plots show 100 individual model trajectories as well as the mean and SD of the endpoint distribution from 100 MC model runs. j,k. Effect size comparison for differences between lipid species in retrograde transport from the PM to the ER (k_{PM_ER}) and endosomes (k_{PM_Endo}), respectively, numbers in the heatmaps are Cohen’s d. Cohen’s d values: >0.01 very small (VS), >0.20 small (S), >0.50 medium (M), >0.80 large (L), >1.20 very large (VL), >2.00 Huge (H).

Extended Data Fig. 8. Determination of overall bifunctional lipid content and metabolism of the lipid class level and characterization of TMEM30A KD.

a. Bifunctional lipid incorporation and subsequent depletion over 24 h determined by shotgun lipidomics. Error bars: Mean and 95% CI of 3 biological repeats containing 2 technical replicates each. b-j. Development of bifunctional lipid class distribution over 24 h for all lipid probes. Note that final distributions are not identical, even for closely related species. b-j show the mean and 95% CI n = 3 biological repeats containing 2 technical replicates each. k. Confirmation of TMEM30A KD in HCT116 cells shown by qPCR of GAPDH and TMEM30A in WT and KD cells. l. Quantification cycle (Cq) and Cq normalized to GAPDH (DCq). Mean and SD of 3 biological replicates with 4 technical replicates each.

Extended Data Fig. 9. Analysis of PC species distribution.

a-i. Development of PC species distribution over 24 h for all lipid probes. Note that some species, notably PC(16:1/Y) are only produced from a subset of the initially supplied lipids. For SM(Y), no detectable amount of PC was observed. j. Development of PE species distribution over 24 h after loading PE(18:1/Y). k-m. Development of the regioisomer distribution of the most common PC species PC(16:0,Y) estimated via the MS/MS- fatty acid neutral loss fragments. The bifunctional fatty acid is primarily incorporated at the sn2-position. a-m. Mean and 95% CI, n = 3 biological repeats containing 2 technical replicates each.

Extended Data Fig. 10. Analysis of TAG species distribution and lipid droplet populations.

a-i. Development of TAG species distribution over 24 h for all lipid probes. Note that PA (c) is the only lipid that initially gives rise to a single TAG species, whereas all other probes yield a spectrum of TAGs. Mean and 95% of 3 biological repeats containing 2 technical replicates each. j. Comparison of intensity distribution of individual lipid droplets for PC(18:1/Y) and PC(Y/18:1) and PC(20:4/Y) and PC(Y/20:4). Distribution drawn from 5 fields of view for each lipid.

Extended Data Fig. 11. Dedicated pools of fatty acids are utilized during neutral lipid biogenesis.

a. Lipid droplets (stained with LipidSpot 610, green) exhibit a bright lipid signal (magenta) 4 h after loading PC(Y/16:0), bottom panels which is not observed after loading PC(16:0/Y), top panels. Scale bar: 10 µm. b. Upper panel: Quantification of fluorescence intensity of cellular lipid droplets for all PCs over time. Lower panels: Comparison of intensity distribution of individual lipid droplets for PC(16:0/Y) and PC(Y/16:0), 8 h after loading and statistical analysis of the similarity of the respective distributions for all PC species. Error bars: Mean ± SE of 5 fields of view. c. Mass spectrometric determination of bifunctional lipid content in neutral lipids demonstrates that significantly more TAG than CE is generated after loading PC(16:0/Y) whereas similar amounts of CE and TAG are generated after loading PC(Y/16:0) (left panels). Both species yield complex TAG patterns (right panels) and all TAG species are produced with similar kinetics. Mean and 95% confidence intervals of 3 biological repeats containing 2 technical replicates each. d. A single TAG species is initially produced after supplying PA (18:1/Y), whereas all other species are produced with slower kinetics. Mean and 95% confidence intervals of 3 biological repeats containing 2 technical replicates each. e. Schematic overview of neutral lipid biosynthesis at lipid droplets f. Proposed neutral lipid biosynthetic pathway model featuring dedicated free fatty acid / Acyl-CoA pools.