Arabidopsis root lipid droplets are hubs for membrane homeostasis under heat stress, and triterpenoid synthesis and storage

Summary

• Plant lipid droplets (LDs) and their associated proteins have numerous subcellular and physiological functions. While considerable progress has been made for LDs in many tissues, the function and composition of LDs in roots remain largely unexplored.

• We investigated the changes in the number of LDs and the lipidome in heat‐stressed Arabidopsis thaliana roots. Furthermore, we isolated root LDs from the Arabidopsis mutant trigalactosyldiacylglycerol 1‐1 sugar dependent 1‐4 and investigated its proteome and lipidome.

• Heat stress led to a degradation of membrane lipids and an increase in triacylglycerols and LDs, while fatty acid steryl esters decreased, probably acting as precursors for acylated sterol glycosides. A variety of proteins were enriched in root LDs, which are thus far not described as LD proteins. Transient expression of these proteins in many cases confirmed their LD localization, for example of the triterpene biosynthetic enzymes thalianol synthase and marneral synthase. We could furthermore show that the educts and products of these enzymes are enriched in root LDs, too.

• We conclude that root LDs simultaneously act as a sink and source during heat stress‐induced membrane remodeling. Furthermore, root LDs play a pivotal role in triterpene synthesis and storage, thereby highlighting LDs as hubs in specialized metabolism.

Introduction

Lipid droplets (LDs) are subcellular structures that store hydrophobic molecules, such as triacylglycerols (TAGs) and fatty acid steryl esters (SEs) in their core, and are surrounded by a phospholipid monolayer with various associated proteins (Bouchnak et al., ²⁰²³; Guzha et al., ²⁰²³). LDs are formed at the endoplasmic reticulum (ER) and protrude into the cytosol (Scholz et al., ²⁰²²). However, whether they actually fully detach from the ER is not clear. In land plants, LDs are most abundant in seeds, spores and pollen (Guzha et al., ²⁰²³), but are also present in root and leaf tissues (Kelly et al., ²⁰¹³; Pyc et al., ²⁰¹⁷). In seeds and spores, one key role of TAG stored in LDs is to provide energy and carbon for cellular growth in the absence of photosynthesis and without provision of sugars from other organs and tissues (Huang et al., ²⁰⁰⁹; Turesson et al., ²⁰¹⁰; Zienkiewicz & Zienkiewicz, ²⁰²⁰; Niemeyer et al., ²⁰²²; Hembach et al., ²⁰²⁴). Similarly, in pollen tubes, LDs are thought to function as an energy and carbon source (Zienkiewicz et al., ²⁰¹³), although they can also act as a sink for membrane lipid‐derived acyl chains during heat‐induced membrane remodeling (Krawczyk et al., ^2022a). Likewise in leaves, TAG‐filled LDs act as a sink and can accumulate under different abiotic stresses (Mueller et al., ²⁰¹⁵; Gidda et al., ²⁰¹⁶; Doner et al., ²⁰²¹), while membrane lipids are remodeled (Higashi et al., ²⁰¹⁵; Tarazona et al., ²⁰¹⁵; Shiva et al., ²⁰²⁰; Scholz et al., ²⁰²⁴). Furthermore, recent work has implied that leaf LDs play a role in pathogen defense (Hanano et al., ²⁰¹⁵; Shimada et al., ²⁰¹⁵; Fernández‐Santos et al., ²⁰²⁰) and are important for stomatal development (Ge et al., ²⁰²²). Recently, LDs have also been proposed to be potentially associated with the synthesis of furan‐containing fatty acids (Omata et al., ²⁰²⁴).

Given the array of functions of LDs across different plant tissues, it is not surprising that their proteomes are also highly diverse (Horn et al., ²⁰¹³; Brocard et al., ²⁰¹⁷; Kretzschmar et al., ^{2018, 2020}; Fernández‐Santos et al., ²⁰²⁰; Doner et al., ²⁰²¹; Niemeyer et al., ²⁰²²; Hembach et al., ²⁰²⁴; Omata et al., ²⁰²⁴; Scholz et al., ²⁰²⁴) and can display marked changes during stress (Scholz et al., ²⁰²⁴) and developmental processes (e.g. seedling establishment, Kretzschmar et al., ²⁰²⁰; and spore germination in the moss Physcomitrium patens, Hembach et al., ²⁰²⁴). For example, oleosins act as major LD surface proteins in desiccation‐tolerant tissues, such as seeds (Huang, 2018), pollen (Roberts et al., ¹⁹⁹³), spores (Huang et al., ²⁰⁰⁹), and the tubers of yellow nutsedge (Cyperus esculentus; Niemeyer et al., ²⁰²²), but are notably absent in leaves (Brocard et al., ²⁰¹⁷; Scholz et al., ²⁰²⁴). However, oleosins have recently also been found as the major LD proteins in the streptophyte alga Mesotaenium endlicherianum when nutrient stressed but fully hydrated (Dadras et al., ²⁰²³). Even more specific to seeds seem to be the proteins LIPID DROPLET PROTEIN OF SEEDS and SEED LIPID DROPLET PROTEIN (Kretzschmar et al., ²⁰²⁰), with the latter and its interaction partner LIPID DROPLET PLASMA MEMBRANE ADAPTOR (LIPA) mediating a membrane contact site between LDs and the plasma membrane (Krawczyk et al., ^2022b). Other notable plant LD proteins are more ubiquitously found. These include caleosins that are found in seeds, pollen and leaves where they are thought to play a role as a peroxygenase (Blee et al., ²⁰¹⁴; Hanano et al., ²⁰²³). Another example is LIPID DROPLET‐ASSOCIATED PROTEIN 1 to 3 (LDAP1 to 3) and LDAP INTERACTING PROTEIN (LDIP) that are involved in proper LD formation at the ER (Pyc et al., ^{2017, 2021}), as well as PLANT UBX DOMAIN‐CONTAINING PROTEIN 10 (PUX10) and CYCLOARTENOL SYNTHASE 1 (CAS1), that play a role in LD protein degradation (Deruyffelaere et al., ²⁰¹⁸; Kretzschmar et al., ²⁰¹⁸) and sterol synthesis (Babiychuk et al., ²⁰⁰⁸), respectively. Interestingly, certain LDAP isoforms are strongly upregulated under different stresses (Kilian et al., ²⁰⁰⁷; Winter et al., ²⁰⁰⁷) and their overexpression or disruption indicated they play a role in drought stress resistance (Kim et al., ²⁰¹⁶; Laibach et al., ²⁰¹⁸).

Overall, the study of LD proteins and their physiological functions to date has proven crucial to our understanding of LD biology in plants. However, due to the predominant focus of these efforts on seeds, leaves and pollen tubes, sparingly little is known about LDs in roots.

That is, one of the earliest reports of root LDs is their observation in cress (Lepidium sativum L.), where LDs are abundantly found in early differentiating statocytes and then decrease in number during differentiation (Hensel, 1986). In leek (Allium porrum L.) seedlings, LD number and TAG content were observed to increase in root tissues when their sterol biosynthetic pathway was chemically blocked with fenpropimorph (Hartmann et al., ²⁰⁰²). In cotton (Gossypium sp.), LDs of root tissue have also been analyzed in seedlings by direct organelle mass spectrometry (MS) and showed a distinct TAG composition enriched in cyclic fatty acids (Horn et al., ²⁰¹¹). Among the potential root LD proteins identified, a root‐expressed hydroperoxide lyase of Medicago truncatula was confirmed to localize to LDs (De Domenico et al., ²⁰⁰⁷). While these studies suggest that root LDs have distinct functions, much of the root LDs' function in general is still unclear.

In this study, we show that Arabidopsis root LDs act as a sink for acyl chains during heat stress‐induced membrane remodeling, and, simultaneously, that LD‐stored SEs strongly decrease to serve as a potential source for sterols in membranes, with the highest relative increase being observed in the acylated steryl glycosides (ASGs). Furthermore, we isolated LD‐enriched protein fractions for subsequent proteomic analyses from the Arabidopsis double mutant trigalactosyldiacylglycerol 1‐1 sugar dependent 1‐4 (tgd1‐1 sdp 1‐4), which has strongly elevated TAG levels (Fan et al., ²⁰¹⁴). In our proteomic analysis of the LDs from this mutant, 34 previously described LD proteins were detected. Furthermore, we selected 24 highly enriched proteins in the LD fraction that have not been reported to be LD‐localized for further analysis. For 14 of these candidate LD‐associated proteins, we observed that LD localization in transiently transformed Nicotiana tabacum pollen tubes and/or Nicotiana benthamiana leaf cells. Among these, LD proteins were two 2,3‐oxidosqualene cyclases, suggesting that LDs in roots are hubs for triterpenoid synthesis. Finally, we observed that the isolated root LDs were also enriched in certain triterpenes and triterpene esters, indicating that root LDs can act as a storage site for non‐TAG lipids as well.

Materials and Methods

Plant lines and growth conditions

Lipidomic and proteomic experiments were carried out with Arabidopsis thaliana (L.) Heynh. Columbia‐0 (Col‐0) and the tgd1‐1 sdp1‐4 double‐mutant line (Fan et al., ²⁰¹⁴).

Seeds of Arabidopsis lines were surface‐sterilized with 6% (w/v) sodium hypochlorite and 0.1% (v/v) Triton X‐100 and germinated on ½‐strength Murashige & Skoog medium (½MS; Duchefa Biochemie, Haarlem, the Netherlands) medium (Murashige & Skoog, 1962) containing 0.8% (w/v) agar with or without 1% (w/v) sucrose as indicated. Plants were grown at a light intensity of 100 μmol s⁻¹ m⁻² for microscopy and 150 μmol s⁻¹ m⁻² for lipidomics, and a temperature of 23°C. The temperature was shifted to 37°C for heat stress treatment.

For LD enrichment and subsequent proteomic or lipid analysis, high‐yield axenic root cultures of tgd1‐1 sdp1‐4 were cultivated by adapting an established protocol (Hétu et al., ²⁰⁰⁵). In short, seeds of the oil‐rich mutant tgd1‐1 sdp1‐4 were surface‐sterilized, placed on sterile steel grids on top of solid ½MS + 1% sucrose medium, stratified for 72 h, and subsequently grown for 7 d. One‐week‐old seedlings were transferred to 100‐ml Erlenmeyer flasks on the steel grid and supplemented with 10 ml liquid ½MS + 1% sucrose medium. The culture was agitated at 85 rpm for 11 d with regular exchange of the medium every 3 d. Then, the medium was changed to 15 ml ½MS + 3% sucrose medium and seedlings were grown an additional 11 d before harvest, exchanging the medium every third day.

Lipidomic sample preparation and measurements

Lipids were extracted from cutoff roots corresponding to c. 5 mg dry weight. To inactivate phospholipase activity, samples were initially incubated in boiling water for 5 min. Lipids were sequentially extracted with 1 ml of chloroform : methanol (1 : 2, v/v), 1 ml of chloroform : methanol (2 : 1, v/v), and 1 ml of chloroform. For each extraction step, samples were vortexed thoroughly, centrifuged for 10 min at 1500  g and the supernatants collected in a new tube. To the combined supernatant, 0.75 ml of 300 mM ammonium acetate was added, samples were vortexed thoroughly, centrifuged for 5 min at 1500  g , and the lower phase transferred to a new tube. Extracts were evaporated to dryness and dissolved in chloroform : methanol : 300 mM ammonium acetate in H₂O (300 : 665 : 35, v/v/v). The dry weight of the remaining residue was determined, and the amount of the internal standard adjusted accordingly. Samples were analyzed via direct infusion nanospray MS on an 6530 Accurate‐Mass Q‐TOF liquid chromatography mass spectrometry (LC‐MS) instrument (Agilent, Santa Clara, CA, USA) equipped with a ChipCube interface as previously described (Welti et al., ²⁰⁰²; Gasulla et al., ²⁰¹³; Gutbrod et al., ²⁰²¹).

Sterols and other triterpenes were extracted from roots corresponding to c. 5 mg dry weight. The material was ground with a glass rod after the addition of 1000 μl of MTBE : MeOH (3 : 1) incubated for 1 h at 4°C with shaking. Then, 5 μg of cholestanol standard and 500 μl of ddH₂O were added, followed by vortexing and centrifugation for 10 min at 4°C. The upper organic phase was recovered into a new tube for measurement, and the dry weight of the remaining residue was determined for normalization. To measure free sterols and triterpenes, 50 μl were transferred to a GC‐vial and evaporated to dryness. Then, the lipids were dissolved in 50 μl hexane and measured by GC‐FID, or they were taken up in 15 μl of anhydrous pyridine, derivatized with 15 μl of MSTFA, and measured by gas chromatography mass spectrometry (GC‐MS). For total sterol and triterpene measurements, 50 μl of each sample was first evaporated and saponified with 1 ml of 6% methanolic potassium hydroxide at 80°C for 2 h. Lipids were then recovered by adding 500 μl of ddH₂O and 1 ml hexane three times. The hexane extracts were combined and evaporated under a N₂ stream, before being treated as above for GC‐MS measurements.

GC‐MS measurements were performed on an 8890 GC system equipped with a HP‐5MS UI 30 m column coupled to an 7250 GC/Q‐TOF (Agilent). Helium was used as the carrier gas at a flow rate of 1.1 ml min⁻¹. The inlet temperature was set to 250°C. The oven gradient started at 120°C, held for 1 min, then increased at 15°C min⁻¹ until 300°C, and held for 10 min. The ion source and the transfer line temperature were set to 200°C and 280°C, respectively, and 70 eV was used as the electron energy. The mass‐to‐charge ratio range was recorded from 50 to 500. GC‐FID measurements were performed via an 8860 GC system equipped with a HP‐5MS UI 30 m column (Agilent). Helium was used as the carrier gas at a flow rate of 6.5 ml min⁻¹; the inlet temperature was set to 250°C and the oven gradient started at 120°C with an increment of 5°C min⁻¹ up to 300°C.

Isolation of root LD‐enriched fractions and preparations for proteomic analysis

Harvested root material was separated from other plant tissues and drained from the remaining medium by drying between paper towels and applying gentle pressure. The resulting root pads of two separate cultures were pooled into one biological replicate.

For the protein extraction, all processes and materials were kept on ice. Grinding buffer (50 mM Tris–HCl pH 7.5, 10 mM KCl, 0.4 M sucrose, 200 μM protease inhibitor PMSF; Carl Roth, Karlsruhe, Germany) and sea sand were added to the root pads, which were subsequently ground. To remove cellular debris and sea sand, the homogeneous suspension was centrifuged for 1 min at 100  g . An aliquot was taken from the supernatant and precipitated in 96% ethanol at −20°C, representing the total protein fraction. The remaining supernatant was overlaid with washing buffer (50 mM Tris–HCl pH 7.5, 10 mM KCl, 0.2 M sucrose, 200 μM protease inhibitor PMSF; Carl Roth, Karlsruhe, Germany) and centrifuged for 35 min at 100 000  g and 4°C in a swing‐out rotor. The floating fat pad was mechanically collected, emulsified in a small volume of washing buffer, and centrifuged in a fixed angle rotor for 35 min at 100 000  g . The floating fat pad was collected again, taken up in 96% ethanol to remove fat, and stored at −20°C to precipitate proteins. This fraction was considered the LD fraction.

Precipitated root protein pellets were subjected to defatting by undergoing two washes with 80% ethanol, followed by drying and an additional wash with 96% ethanol. The obtained proteins were dissolved in 6 M urea, 5% sodium dodecyl sulfate (w/v) solution, and their concentrations were determined using the Pierce BCA protein assay kit (Thermo Fisher Scientific, Waltham, MA, USA). Twenty micrograms were subjected to in‐gel digest, as described previously (Shevchenko et al., ²⁰⁰⁶; Rappsilber et al., ²⁰⁰⁷).

Proteomic LC‐MS/MS measurements

An EASY‐nLC 1200 system (Thermo Fisher Scientific) in conjunction with an Exploris 480 mass spectrometer (Thermo Fisher Scientific) was employed for the LC‐MS/MS analysis of root‐derived peptides. For this purpose, peptides were separated on 20 cm frit‐less silica emitters (CoAnn Technologies, Richland, WA, USA) with a 0.75‐μm inner diameter, packed in‐house with ReproSil‐Pur C₁₈ AQ 1.9 μm resin (Dr Maisch, Ammerbuch‐Entringen, Germany). The column was maintained at a constant temperature of 50°C. Elution of peptides was carried out over 115 min using a segmented linear gradient from 0 to 98% Solvent B (Solvent A: 0% ACN, 0.1% formic acid; Solvent B: 80% ACN, 0.1% formic acid) at a flow rate of 300 nl min⁻¹. The data‐dependent acquisition mode was utilized to acquire mass spectra. For full proteome samples, MS¹ scans were obtained at an Orbitrap resolution of 120 000, covering a scan range of 380–1500 (m/z). The maximum injection time was set to 100 ms, and a normalized AGC target of 300% was utilized. Precursors with charge States 2–6 were selectively chosen for fragmentation, and up to 20 dependent scans were acquired. Dynamic exclusion was enabled with an exclusion duration of 40 s and a mass tolerance of ±10 ppm. A 1.6 (m/z) isolation window with no offset was established, accompanied by the application of a normalized collision energy of 30. Acquisition of MS² scans was performed at an Orbitrap resolution of 15 000, while maintaining a fixed first mass (m/z) of 120. The maximum injection time was 22 ms, and the normalized AGC target was set to 50%.

Computational processing of proteomic MS/MS data

MS/MS raw data were processed in the maxquant software (v.1.6.2.17) for feature detection, peptide identification, and protein group assembly (Cox & Mann, 2008). Mostly, default settings were used with additional settings as specified in Supporting Information Tables S1 and S2. The TAIR10 protein database (Lamesch et al., ²⁰¹²) was used as a reference for identification. Label‐free quantification was performed to obtain iBAQ (intensity‐based absolute quantification and label‐free quantification) values for protein abundances that indicate the relative molarity of the proteins (Cox & Mann, 2008; Schwanhausser et al., ²⁰¹¹; Cox et al., ²⁰¹⁴). Further data analysis was done in perseus v.1.6.2.2 (Tyanova et al., ²⁰¹⁶). Proteomic raw data can be found in the PRIDE database (Vizcaíno et al., ²⁰¹⁴) under the identifier PXD051152 and PXD068568 (https://www.ebi.ac.uk/pride/). They were uploaded through the jPOST environment (Okuda et al., ²⁰²⁵) All metadata can be found in Tables S1 and S2.

Protein localization was annotated based on the Plant Proteome Database (PPDB; Sun et al., ²⁰⁰⁹) as of 14 June 2022. LD localization was assigned based on previous studies (Kretzschmar et al., ^{2018, 2020}; Fernández‐Santos et al., ²⁰²⁰; Scholz et al., ²⁰²⁴). riBAQ values of proteins were calculated by dividing all individual iBAQ values in one sample by the sum of all iBAQ values in this sample and multiplying by 1000. Only proteins that were detected in all five replicates of one of the cellular subfractions were considered for further analysis.

Molecular cloning and microscopy of N. tabacum pollen tubes and Arabidopsis roots

Open reading frames of selected candidate genes were amplified from Arabidopsis floral, root or leaf cDNA and cloned into pLatMCC‐GW (Müller et al., ²⁰¹⁷) via Gateway cloning as described in Müller et al. (2017). All primers used in this study are listed in Table S3. Nicotiana tabacum L. growth, pollen transformation, and pollen tube growth were performed as described previously (Müller et al., ²⁰¹⁷). Images were taken using either a ZEISS LSM780 confocal microscope or a ZEISS LSM 980 with Airyscan 2 confocal microscope (Carl Zeiss Microscopy Germany GmbH, Oberkochen, Germany). This second microscope was also used for monitoring root LDs via the Airyscan 2 function. Detailed settings for all micrographs are described in Table S4. For microscopy of root LDs, roots were fixed in 4% formaldehyde prepared in 25 mM PIPES buffer (pH 7.0) for 15 min with agitation. After three washes with 50 mM PIPES buffer, the roots were stained with 4 μg ml⁻¹ BODIPY 493/503 in 50 mM PIPES (pH 7.0) for 30 min before microscopy.

LD size and number were directly quantified in fiji imagej (v.1.54; Schindelin et al., ²⁰¹²): In brief, an region of interest (ROI) of up to 100 μm × 100 μm was selected in each root micrograph, the LDs within the selected area were recognized using the Threshold (MaxEntropy) function, and the resulting particles after thresholding were quantified with the particle analysis tool. Due to the increased number of LDs in roots grown on medium with sucrose, quantification of the respective images was carried out combining image segmentation in ilastik v.1.4.1post1 (Berg et al., ²⁰¹⁹) and quantification in fiji imagej (v.1.54; Schindelin et al., ²⁰¹²). Segmented images were exported from ilastik and converted into masks in fiji imagej using the Threshold (Default settings) function. Subsequently, ROIs across the root section (c. 150 μm height along the root axis) were selected, and segmented LDs within the ROI were quantified with the particle analysis tool.

Molecular cloning and candidate localization studies in N. benthamiana leaves

Open reading frames of selected candidate genes were amplified from cDNA prepared with Maxima Reverse Transcriptase (Thermo Fisher Scientific) according to the manufacturer's instructions using leaf RNA that had been extracted using the Spectrum Plant Total RNA Kit (Merck KGaA, Darmstadt, Germany). Constructs were amplified with the Phusion High‐Fidelity DNA Polymerase (Thermo Fisher Scientific) following the manufacturer's instructions. Gateway cloning into the mCherry‐encoding plant binary vectors pMDC32‐ChC and pMDC32‐ChN was carried out by traditional or fast Gateway® cloning as described in Müller et al. (2017). Vector construction of pMDC32‐ChC and pMDC32‐ChN has been described previously in Kretzschmar et al. (2020) and Doner et al. (2021), respectively. The construct of MmDGAT2 in pMDC32, which was used for co‐expression experiments, has been described in Cai et al. (2019). Localization of candidates was analyzed in 28 d‐old leaves of Nicotiana benthamiana Domin that were transiently transformed by infiltration with Agrobacterium tumefaciens (strain LBA4404) harboring candidate expression vectors. N. benthamiana plant growth, leaf infiltration, and BODIPY 493/503 staining were performed as previously described (Gidda et al., ²⁰¹⁶; Kretzschmar et al., ²⁰²⁰). Leaf micrographs 3 d‐postinfiltration were captured as single (0.4 μm) optical z‐sections and saved as 512 × 512 pixel images using a Leica SP5 confocal laser scanning microscope (CLSM) (Leica Microsystems, Wetzlar, Germany). Excitations and emission signals for fluorescent proteins and BODIPY were collected sequentially as single optical sections in double‐labeling experiments like those described in Gidda et al. (2016), with no detectable crossover observed at the settings used for data collection.

Binary plasmids for putative N‐glycan biosynthetic enzymes (At1g78800, At1g16570, At2g40190, At2g47760, and At5g38460) were constructed using Gateway Technology (Invitrogen) with the destination vectors pGWB405m (Nakagawa et al., ²⁰⁰⁷; Segami et al., ²⁰¹⁴) as described previously (Yamaguchi et al., ²⁰²⁵). Each full‐length genomic coding region without the stop codon was amplified from Arabidopsis Col‐0 genomic DNA. Each entry clone was produced by ligating each PCR product into the pENTR plasmid using an In‐Fusion Cloning Kit (Takara, Shiga, Japan), and recombined into the destination vector pGWB405m via LR Clonase II (Invitrogen), creating binary vectors for expressing GFP‐fused proteins. The binary vectors were transformed independently into A. tumefaciens strain GV3101. The transformed cells harboring each plasmid were infiltrated into the leaves of 3‐wk‐old N. benthamiana plants. The transformed cells expressing α‐DOX1‐RFP were used as an LD localization marker. Two days postinfiltration, the whole plants were subjected to heat stress (50°C, 30 min) for LD induction, and epidermal leaf cells were subsequently observed under a fluorescence microscope (BZ‐X800; Keyence, Osaka, Japan). The fluorescent signal of GFP was examined using a GFP filter (excitation, 450–490 nm; emission, 500–550 nm) and that of RFP was examined using a TRITC filter.

Extraction, derivatization and measurement of triterpenes from isolated LDs

LD‐enriched fractions (derived from c. 5 to 8 g plant material) and total cellular extracts (derived from the same material but only 1% of the volume of the sample were taken) were obtained as described above for proteomic samples and extracted two times with 1 ml of methanol and two times with 1 ml ethyl acetate. The methanol and ethyl acetate extracts from each fraction were combined and evaporated under N₂ stream. These dried extracts were then redissolved in hexane. A total of 0.2 μg cholestanol was added to the total extract fractions and 1 μg to the LD fractions as an internal standard. Triterpenes and free sterols in the samples were then extracted, derivatized and measured by GC‐MS while membrane glycerolipids and TAGs were determined by direct infusion nanospray MS as described for lipidomic samples above.

Statistical analysis of data

Statistical analysis of LD numbers (Figs 1, S1, S4) and metabolite data (Figs 2, 9, S5–S7) was carried out with the R Statistical Software (v.4.2.2; https://www.R‐project.org/), while proteomic data (Figs 3, S8–S11) were analyzed with the Perseus software platform (Tyanova et al., ²⁰¹⁶). Data cleaning, analysis and plotting in R was performed using the following libraries: dplyr (Wickham et al., ²⁰²³), stringr (Wickham, 2022), tidyr (Wickham et al., ²⁰²⁵), purrr (Wickham & Henry, 2023), car (Fox & Weisberg, 2019), stats (R Core Team, ²⁰²²), multcomp (Hothorn et al., ²⁰⁰⁸), ggplot2 (Wickham, 2016), ggpubr (Kassambara, 2023), rcolorbrewer (Neuwirth, 2022), grid (R Core Team, ²⁰²²), gridextra (Auguie, 2017), gtable (Wickham & Pedersen, 2024). For statistical analysis, pairwise comparisons were carried out using Welch's t‐test. Adjustments for multiple comparisons were made with the ‘p.adjust’ function of the stats package in R, applying the Benjamini–Hochberg correction. ANOVA calculations and post hoc Tukey tests were done with the ‘aov’, ‘lm’, and ‘glht’ functions of the ‘stats’ and ‘multcomp’ libraries in R. Data for principal component analysis (PCA) calculations of metabolites in the LD‐enriched fraction were normalized so that each compound's average across all samples equaled 1. Subsequent calculations of the PCA plot were done using the ‘prcomp’ function of the stats package, setting the parameters ‘center’ and ‘scale.’ both to ‘TRUE’. Volcano plots of proteomics data were calculated from log₂‐transformed and imputed riBAQ values using the ‘Volcano plot’ visualization tool of Perseus. False discovery rate was set to 0.01; otherwise default settings (two‐sided t‐test, number of randomizations: 250, no grouping in randomizations, S0: 0.1) were used. The resulting values were exported and visualized in python (v.3.10.6; Python Software Foundation, http://www.python.org) using the libraries matplotlib.pyplot (Hunter, 2007), pandas (McKinney, 2010), and numpy (Harris et al., ²⁰²⁰).

Fig. 1.

Lipid droplets (LDs) occur in increased numbers in parts of the root elongation zone and accumulate under heat stress. The roots of 7‐d‐old Columbia‐0 (Col‐0) Arabidopsis seedlings vertically grown on plates (½‐strength Murashige & Skoog medium (without sucrose) were fixed and stained with BODIPY 493/503. Plants were grown at 23°C (a) and, in the case of heat stress, moved for 24 h to 37°C before fixation and analysis (a, b). Median planes of different root zones (see Supporting Information Fig. S1) are imaged by confocal laser scanning microscope (CLSM) (a, b). Images display overviews (first and third columns) and magnifications (second and fourth column). Bars, 50 and 10 μm, respectively. For quantitative image analysis, areas up to 100 μm × 100 μm of each root micrograph were selected, and the LDs within the selected areas were quantified in number (c) and size (d) using fiji imagej. The root cap was omitted from the analysis. Data were analyzed from n ≥ 5 individual roots; however, in the meristematic zone of control‐treated roots, only two roots showed LDs in the micrograph and could be analyzed for LD size in (d). Statistical analysis was done using Welch's t‐test with Benjamini–Hochberg correction for multiple comparisons. Asterisks mark significant differences for the following P‐values: *, P < 0.05; **, P < 0.01; ***, P < 0.001; ns, not significant. Plots display mean ± SD.

Fig. 2.

Composition and abundance of lipids in control and heat‐stressed roots. Lipids were extracted from roots of 12‐d‐old Columbia‐0 (Col‐0) Arabidopsis seedlings grown vertically on plates (½‐strength Murashige & Skoog medium (½MS) without sucrose). Plants were cultivated at 23°C and, in the case of heat stress, moved for 24 h to 37°C before lipid extraction. Lipids were analyzed by electrospray ionization‐MS/MS. (a) Absolute values for the molar amounts of individual lipid species were determined, and their relative proportion in the respective lipid class was calculated in mol %. (b) Total amounts of lipid classes were determined as sums of all individual lipid species of the respective lipids class. Values are from n = 5–10 biological replicates, and are shown as mean ± SD. Statistical differences were calculated by Welch's t‐test using Benjamini–Hochberg correction for multiple comparisons and are represented as follows: ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001. ASG, acylated steryl glycosides; DGDG, digalactosyldiacylglycerol; FS, free sterols; IR, insoluble residue; MGDG, monogalactosyldiacylglycerol; ns, not significant; PA, phosphatidic acid; PC, phosphatidylcholine, PE, phosphatidylethanolamine; PG, phosphatidylglycerol; PI, phosphatidylinositol; PS, phosphatidylserine; SE, sterol esters; SG, steryl glycosides; SQDG, sulfoquinovosyldiacylglycerol; TAG, triacylglycerol; MS, mass spectrometry.

Fig. 9.

Overview of terpene metabolism. Enzymes and compounds found associated with lipid droplets (LDs) are highlighted in green. AT, acyl transferase; CAS1, CYCLOARTENOL SYNTHASE 1, MRN1, MARNERAL SYNTHASE 1; OSC, oxidosqualene cyclase; SQE, squalene epoxidase; SQS squalene synthase; THAS1, THALIANOL SYNTHASE 1; specific proteins in capital letters. R represents in all cases an acetyl or acyl group.

Fig. 3.

Analysis of protein enrichment in the lipid droplet (LD) fraction of Arabidopsis roots from the mutant tgd1‐1 sdp1‐4 grown in axenic root culture. The riBAQ dataset was imputed and the values were log₂ transformed. Then, the difference between the LD‐enriched and total protein fractions was calculated for each protein (Supporting Information Dataset S14). Additionally, the corresponding P‐values (−log₁₀) were determined. (a) A volcano plot was generated based on these values and the upper right corner was enlarged in (b). A false discovery rate (FDR) of 0.01 was used to distinguish between significant and nonsignificant differences (red lines). Known LD proteins (Table 1) and selected candidate LD proteins (Table 2) are highlighted. FEY3, FOREVER YOUNG 3; GPAT5/9, GLYCEROL‐3‐PHOSPHATE ACYLTRANSFERASE5/9; LEW1, LEAF WILTING 1; LPEAT1, LYSOPHOSPHATIDYLETHANOLAMINE ACYLTRANSFERASE1; MRN1, MARNERAL SYNTHASE 1; PCME, PRENYLCYSTEINE METHYLESTERASE; THAS1, THALIANOL SYNTHASE 1; TE, total extract.

Results

Heat stress in Arabidopsis roots leads to LD proliferation, membrane remodeling, and the accumulation of TAG

Plant LDs have been primarily studied in seeds and also more recently in leaves (Guzha et al., ²⁰²³). By contrast, LDs in roots are not well explored. We therefore aimed to get an overview of LD abundance in the different developmental zones of the root in Arabidopsis (Fig. S1). We imaged BODIPY 493/503‐stained LDs of 7‐d‐old seedlings that were vertically grown on ½MS without sucrose. In most zones of the root, low numbers of LDs were observed; however, in the early elongation zone (adjacent to the meristem), we observed a substantial increase in the number of LDs, reaching an average count of 3.27 LDs mm⁻² within the single‐plane micrograph of the root zone. To investigate potential connections between the number of observed LDs and membrane lipid adaptation and turnover in roots, we analyzed LD abundance in heat‐stressed roots after incubating the seedlings for 24 h at 37°C (Fig. 1a). Following the heat stress treatment, the number and size of LDs greatly increased throughout the whole root and especially in the meristematic zone (Fig. 1b,c). Similar trends were observed for 12‐d‐old Col‐0 roots grown on ½MS, including sugar, albeit with a higher number of LDs in LD‐poor root zones under control conditions (Fig. S2). For later proteomic analysis, we used the tgd1‐1 sdp1‐4 mutant which is known to accumulate TAG (Fan et al., ²⁰¹⁴). This mutant has a much higher number of LDs than the wild‐type (WT), but their number does not significantly change under heat stress (Figs S3, S4). As this mutant grows a bit slower than the WT, we imaged 14‐d‐old roots that showed a similar length and development to 12‐d‐old Col‐0 roots.

As these results implicated LDs as a possible sink during heat‐induced lipid remodeling, we analyzed the root lipidome of similarly heat‐stressed or control seedlings in a targeted manner by ESI‐MS/MS and GC‐MS (Datasets S1–S6). While the lipids were measured quantitatively, we first assessed the relative compositions of lipid species within the individual lipid classes, since adaptations here are important for the regulation of membrane fluidity during stress (Fig. 2a). The results indicate that 18:3 containing molecular lipid species strongly decreased, while 16:0‐ and 18:0‐containing species relatively increased after heat stress. These changes are especially pronounced in phosphatidylcholine (PC), where the relative abundance of the 36:5 and 36:6 species dropped from 25% to 11% and 12% to 3%, respectively. At the same time, for example, 34:1 and 36:2 species increased from 2% each to 11% and 6%, respectively. In TAG, this trend was not observed. Instead, species containing more saturated acyl chains decreased, as the relative proportion of 54:3 and 54:2 dropped from 9% to 1% and 2% to 0.3%, respectively. For most lipid classes, the average number of double bonds decreased in all membrane lipids. For instance, in PC from 3.8 to 3.0 or in monogalactosyldiacylglycerol from 5.4 to 4.3 (Fig. S5). In TAG, the opposite trend occurred, as the average number of double bonds increased from 5.7 to 6.4. In addition, the absolute amounts of most membrane lipid classes decreased while TAG levels increased 7‐fold (Fig. 2b). Regarding sterol metabolism, the trend was reversed. Here, the storage form, SEs, was almost completely depleted, while the membrane‐localized sterol forms increased, foremost the ASGs (fourfold). Regarding the sterol moiety of the SEs, sitosterol was the most abundant sterol; however, it was also most strongly depleted after heat stress (Dataset S6).

Again, similar trends in lipid remodeling were observed for Col‐0 roots grown on ½MS, including sugar (Fig. S6; Datasets S7, S8). However, the decrease for example in 36:6 MGDG and DGDG and 36:5 in PC going along with an increase of more saturated species was not as strong as in the roots growing on medium without sugar (Fig. 2a). Furthermore, the increase in TAG and ASG was also lower, and the strong decrease in total membrane lipids and SEs was not observed (Figs 2b, S7; Datasets S9–S12). These weaker changes could be due to growth on medium containing sucrose.

The tgd1‐1 mutant is known to have reduced relative levels of 18:3‐containing lipids (Xu et al., ²⁰⁰³), and we could observe this in all lipid classes investigated (Fig. S6). Similar to the WT, these levels further decreased under heat stress, while 16:0‐containing species relatively increased. We also found strongly increased TAG levels, similarly as reported for leaves of the tgd1‐1 sdp1‐4 mutant (Fan et al., ²⁰¹⁴) as the levels were increased more than 50‐fold in roots in comparison with the WT (Fig. S7). While there was a trend that TAG levels were elevated under heat stress also in the mutant, the changes were not significant.

Root LD fractions contain numerous strongly enriched proteins

The remarkable shifts in lipid composition coupled with enhanced LD abundance following heat stress in Arabidopsis roots indicate that LDs can act as both a sink and a source for acyl chains and sterols, respectively. Next, we sought to identify additional functions of root LDs by studying their proteome. For LD isolations, high yields of biomass are required. Furthermore, the relatively low number of LDs in the root tissue (Fig. 1) needed to be elevated to obtain a floating LD‐enriched phase suitable for LD protein isolation. In order to achieve this, we utilized an axenic root culture in high sucrose media (Hétu et al., ²⁰⁰⁵) of the oil‐rich mutant line tgd1‐1 sdp1‐4 (Fan et al., ²⁰¹⁴). The LD‐enriched and total cellular fractions were collected from five biological replicates for proteomic analysis. Of the resulting MS data, relative iBAQ values were calculated as ‰ of the summed iBAQ values of all proteins. In total, 4734 different protein groups were identified (Dataset S13), 4318 in the LD fraction and 4134 in the total cellular fraction. In the LD fraction, we detected 34 protein groups that have been previously reported as LD‐associated (Table 1). These annotated LD proteins were equivalent to 22.4% of the total protein abundance (rIBAQ values) in the LD fraction but only 0.33% in the total fraction, implying effective enrichment of LDs by a factor of c. 70. In order to unravel whether any other organelles were systematically co‐enriched, we binned the abundance of all proteins with the same assigned subcellular localization (according to the PPDB; http://ppdb.tc.cornell.edu/; Sun et al., ²⁰⁰⁹). This analysis indicated that besides LD proteins only plastoglobular proteins were enriched at more than an average factor of 4 (Fig. S8). All other organelles were either only slightly enriched or depleted.

Table 1.

Previously described lipid droplet (LD) proteins identified in Arabidopsis tgd1‐1 sdp1‐4 axenic root cultures.

Protein name	AGI	riBAQ in ‰	LD enrichment	N replicates detected in
α‐DOX1	AT3G01420	45.8	46.3	5
CAS1	AT2G07050	0.3	421.5	5
CB5‐E	AT5G53560	1.0	3.5	5
CLO3	AT2G33380	0.2	n.d. in TE	5
CLO4	AT1G70670	2.1	23.4	5
ERD7	AT2G17840	3.2	51.2	5
GPAT4	AT1G01610	0.9	325.1	5
GPAT8	AT4G00400	0.02	n.d. in TE	4
HSD1	AT5G50700	0.10	n.d. in TE	4
HSD3	AT3G47360	0.09	n.d. in TE	4
HSD4/7	AT5G50690	0.5	n.d. in TE	5
LDAH1*	AT1G10740	0.004	n.d. in TE	1
LDAH2	AT1G23330	0.05	n.d. in TE	5
LDAP1	AT1G67360	12.5	51.0	5
LDAP2	AT2G47780	0.02	n.d. in TE	4
LDAP3	AT3G05500	135.8	120.8	5
LDIP	AT5G16550	4.8	54.2	5
LDNP	AT5G04830	6.8	77.5	5
LDS1	AT1G43890	0.9	6.6	5
LIDL1	AT1G18460	0.08	n.d. in TE	5
LIDL2	AT1G73920	0.3	n.d. in TE	5
LIME1/FUFM1	AT4G33110	0.02	n.d. in TE	4
LIME2	AT4G33120	3.7	207.4	5
MAGL8	AT2G39420	0.3	15.7	5
MYOB14	AT4G13160	0.3	111.8	5
OBL1	AT3G14360	0.2	n.d. in TE	5
OBL3	AT1G45201	0.5	11.3	5
OBL4	AT1G56630	0.4	n.d. in TE	5
OBL5	AT5G42930	0.04	n.d. in TE	5
OLE8*	AT3G18570	0.05	n.d. in TE	3
PUX10	AT4G10790	0.2	10.3	5
SMT1	AT5G13710	1.9	29.9	5
UFAO1	AT1G30130	0.5	159.0	5
UFAM1/UFAD1	AT3G23510	0.12	n.d. in TE	5

^*Protein identification is based only on one peptide. The proteins were derived from LD‐enriched (LD) and total (TE) protein fractions. The values provided represent the average riBAQ (n = 5) of the LD‐enriched fractions. The enrichment of LD proteins was determined by dividing the average LD riBAQ by the average TE riBAQ. n = 5. n.d., not detected.

The most abundant, known LD protein in the LD fraction was LDAP3, with LDAP1 and the LDAP interaction partner LDIP also being detected in comparatively high abundance (Table 1). Interestingly, the dioxygenase α‐DOX1 was the second most abundant LD protein, even though in seedlings and leaves it was described as being of much lower abundance (Kretzschmar et al., ²⁰²⁰; Scholz et al., ²⁰²⁴). By contrast, there was no high‐abundant protein from the caleosin or oleosin protein families, whose members dominate the LD proteome of seeds, seedlings and leaves (Brocard et al., ²⁰¹⁷; Kretzschmar et al., ²⁰²⁰; Doner et al., ²⁰²¹; Scholz et al., ²⁰²⁴).

Considering our proteomic results, we reasoned that other proteins enriched in the LD fraction could potentially also localize to LDs. To ensure data reliability, we implemented a filter that only considered proteins meeting two criteria: First, they must have been identified with at least two unique peptides; and second, they must have been detected in all five replicates of either the LD or the total cellular extract. In order to identify potential LD proteins, differences of log₂‐transformed riBAQ values between LD and total cellular fractions were calculated. Furthermore, the statistical significance of this difference was calculated through Student's t‐test for each protein (Dataset S14). Ultimately, significance and differences between samples for each protein were depicted as a volcano plot (Fig. 3). Twenty‐seven of the proteins (Table 2) that had statistical significance and high enrichment factors were chosen for transient expression experiments in tobacco pollen tubes and/or N. benthamiana leaves to assess their subcellular localization by established protocols (Müller et al., ²⁰¹⁷; Scholz et al., ²⁰²⁴). In both transient expression systems, the coding sequence of the protein of interest was appended to a fluorescent tag (mCherry) and potential co‐localization to LDs was determined using the neutral‐lipid‐specific dye BODIPY 493/503. In N. benthamiana leaves, the DIACYLGLYCEROL ACYLTRANSFERASE 2 (DGAT2) gene from Mus musculus was co‐expressed to induce the proliferation of LDs (Cai et al., ²⁰¹⁹).

Table 2.

Candidate Arabidopsis proteins investigated for LD localization.

AGI	Acronym	Name	riBAQ LD	Enrichment
Detected on LDs through microscopy
AT5G60620	GPAT9	GLYCEROL‐3‐PHOSPHATE ACYLTRANSFERASE 9	0.54	77
AT3G11430	GPAT5	GLYCEROL‐3‐PHOSPHATE sn‐2‐ACYLTRANSFERASE 5	0.32	n.d. in TE
AT1G01610	GPAT4	GLYCEROL‐3‐PHOSPHATE sn‐2‐ACYLTRANSFERASE 4	0.91	325
AT1G80950	LPEAT1	LYSOPHOSPHATIDYLETHANOLAMINE ACYLTRANSFERASE1	0.17	n.d. in TE
AT5G10050		Putative short‐chain dehydrogenase	1.23	1974
AT5G04070		Putative short‐chain dehydrogenase	0.19	n.d. in TE
AT1G44170		Putative aldehyde dehydrogenase	6.55	91
AT1G78800		Glycosyl transferase family	1.00	n.d. in TE
AT5G15860	PCME	Prenylcysteine methylesterase	1.08	180
AT4G27760	FEY3	FOREVER YOUNG	0.91	132
AT4G33180		Putative hydrolase	0.19	n.d. in TE
AT4G13160	MYOB14	MYOSIN BINDING PROTEIN 14	0.30	n.d. in TE
AT1G30130	UFAO1	UNSATURATED FATTY ACID OXIDASE 1	0.51	n.d. in TE
AT5G59960		Protein of unknown function	1.13	283
AT1G11755	LEW1	LEAF WILTING 1	0.26	153
AT5G48010	THAS1	THALIANOL SYNTHASE 1	4.90	255
AT5G42600	MRN1	Marneral synthase 1	0.31	n.d. in TE
Not detected on LDs through microscopy
AT4G38540	MO2	Monooxygenase 2	2.97	97
AT1G17430		Putative hydrolase	0.43	n.d. in TE
AT5G01220	SQD2	SULFOQUINOVOSYLDIACYLGLYCEROL 2	3.71	102
AT4G23430		Putative short‐chain dehydrogenase	5.01	40
AT5G15910		Putative short‐chain dehydrogenase	3.83	42
AT1G72175		Putative zinc finger protein	0.15	n.d. in TE
AT1G25520	PML4	PHOTOSYNTHESIS‐AFFECTED MUTANT 71 LIKE 4	0.20	n.d. in TE
AT3G23175		Lesion inducing protein‐related	0.25	n.d. in TE
AT5G01750		Protein of unknown function	2.53	29
AT4G33360	FLDH	Farnesol dehydrogenase	3.51	75

The proteins were derived from LD‐enriched (LD) and total (TE) protein fractions. The values provided represent the average riBAQ (n = 5) of the LD‐enriched fractions. The enrichment of LD proteins was determined by dividing the average LD riBAQ by the average TE riBAQ. n = 5. n.d., not detected.

Several LD‐enriched proteins localized to LDs and/or the ER

Several putative or characterized enzymes were among the candidate root LD proteins tested for subcellular localization, highlighting that LDs could play an active role in root metabolism. First, we analyzed three candidates with acyl transferase activity that are involved in lipid metabolism (Fig. 4). In addition to the GPAT enzymes, GPAT4 and GPAT8, which were previously described to be LD‐localized in plants (Fernández‐Santos et al., ²⁰²⁰), GPAT5 and GPAT9 were also highly enriched in the LD fraction. In contrast to the other GPATs in Arabidopsis, GPAT9 is a member of a distinct family of plant GPATs (Waschburger et al., ²⁰¹⁸) that is described as the only GPAT enzyme with a preference to add acyl chains to the sn1‐ not the sn2‐position of glycerol 3‐phosphate (Shockey et al., ²⁰¹⁶; Singer et al., ²⁰¹⁶). As shown in Figs 4, S4, GPAT9 localized almost exclusively to LDs in pollen tubes, while in leaves, the localization was only partially on the LD surface. GPAT5 and the previously described GPAT4 (Fernández‐Santos et al., ²⁰²⁰) displayed a localization to both LDs and the ER in pollen tubes (Figs 4, S9) but GPAT5 showed a ring‐like pattern surrounding LDs upon transient expression in N. benthamiana. Furthermore, the acyltransferase LYSOPHOSPHATIDYLETHANOLAMINE ACYLTRANSFERASE 1 (LPEAT1) was assayed. LPEAT1 transfers acyl chains to lysophosphatidylethanolamine and to some extent also to lysophosphatidylcholine (Stalberg et al., ²⁰⁰⁹; Jasieniecka‐Gazarkiewicz et al., ²⁰¹⁷). Like GPAT5, LPEAT1 localized not only to LDs but also to the ER in pollen tubes; however, it targeted clearly to LDs in leaves (Fig. 4). Interestingly, AlphaFold‐based structure predictions revealed that each of the respective acyltransferases has a broad hydrophobic surface on one side of the protein that could be involved in the interaction with the LD monolayer.

Fig. 4.

Acyltransferases enriched in the lipid droplet (LD) fraction of Arabidopsis roots localize to LDs in pollen tubes and leaf cells. mCherry‐tagged root LD proteins annotated as acyltransferases were expressed in either Nicotiana tabacum pollen tubes or Nicotiana benthamiana leaves. In leaves, the formation of LDs was induced by co‐expression with the DIACYLGYLCEROL ACYLTRANSFERASE2 of Mus musculus (MmDGAT2). LDs in leaves were subsequently stained with BODIPY 493/503 or the endoplasmic reticulum (ER) was visualized using a co‐expressed ER marker (ERD2‐CFP). Images are single planes obtained by confocal laser scanning microscope (CLSM). GLYCEROL‐3‐PHOSPHATE ACYLTRANSFERASE (GPAT9) clearly colocalized with BODIPY‐stained LDs in pollen tubes and partially in leaves. GPAT5 and LYSOPHOSPHATIDYLETHANOLAMINE ACYLTRANSFERASE 1 (LPEAT1) localized to LDs and the ER in pollen tubes and to LDs in leaves. Each image is representative of at least six pollen tubes or four leaf areas. Bars, 10 μm. As shown on the right, all three corresponding protein structures, as predicted by AlphaFold2, show a hydrophobic surface on one side of the protein; hydrophobicity indicated by red color.

Further protein candidates that showed localization to LDs at least in N. benthamiana, included ALDEHYDE DEHYDROGENASE 4 (ALDH4, At1g44170) and two other proteins annotated by TAIR to contain the NAD(P)‐binding Rossmann‐fold and, thus, may have putative dehydrogenase functions (Fig. 5). The latter two proteins, encoded by At5g10050 and At5g04070, respectively, showed no specific LD localization when transiently expressed in tobacco pollen tubes, but conversely, displayed ring‐like localization around LDs in N. benthamiana leaves. ALDH4 showed a similar pattern, although its annular localization around the leaf LDs is more diffuse. AlphaFold predicted protein structures encoded by At5g10050 and At5g04070 also both possess a hydrophobic surface not unlike the previously mentioned GPAT5, GPAT9 and LPEAT1 (Fig. 4). The predicted structure of ALDH4 is less compact than the other two aforementioned proteins, as one structural domain stretches away from the protein's center, which contains a hydrophobic region at its distal end.

Fig. 5.

Putative dehydrogenases enriched in the lipid droplet (LD) fraction of Arabidopsis roots localize to LDs in leaves. mCherry‐tagged proteins were expressed in either Nicotiana tabacum pollen tubes or Nicotiana benthamiana leaves. In leaves, the formation of LDs was induced by co‐expression with MmDGAT2. LDs were stained with BODIPY 493/503 or the endoplasmic reticulum (ER) was visualized using co‐expressed ERD2‐CFP. Images are single planes obtained by confocal laser scanning microscope (CLSM). The two putative short‐chain dehydrogenases, At5g10050 and At5g04070, and the putative aldehyde dehydrogenase At1g44170 localize to the ER and in the case of At5g10050 and At1g44170 also to the plasma membrane. In leaves, all proteins were localized at LDs. Each image is representative of at least seven pollen tubes or four leaf areas. Bars, 10 μm. As shown on the right, all three corresponding protein structures, as predicted by AlphaFold2, show a hydrophobic surface on one side of the protein, with this surface being at the end of an arm for At1g44170; hydrophobicity indicated by red color.

Additional putative and characterized enzymes that localized to LDs included a glycosyltransferase family protein (At1g78800), the oxidoreductase FOREVER YOUNG (FEY3, At4g27760; Callos et al., ¹⁹⁹⁴), a putative hydrolase (At4g33180), and PRENYLCYSTEINE METHYLESTERASE (PCME, At5g15860) that can demethylate isoprenylcysteine methylesters of proteins that have a prenylation as lipid modification at their C‐terminal cysteine residues (Deem et al., ²⁰⁰⁶). At1g78800 and PCME localized to LDs, and in the case of At1g78800 also to the plasma membrane of pollen tubes (Fig. 6). In addition, when co‐expressed with an ER marker, At1g78800 and PCME also displayed ER co‐localization (Fig. 6). When expressed in N. benthamiana, the fluorescent signal of each individual protein formed clear rings surrounding LDs in leaves. The glycosyltransferase family protein (At1g78800) is an ortholog of yeast ALPHA‐1,3/1,6‐MANNOSYLTRANSFERASE 2 (Alg2) which is involved in protein N‐glycosylation (Gomord et al., ²⁰¹⁰). We also tested other putative N‐glycan biosynthetic enzymes (At1g16570, At2g40190, At2g47760, and At5g38460; Gomord et al., ²⁰¹⁰). The ortholog of yeast Alg1 (i.e. At1g16570) also displayed localization to LDs in N. benthamiana leaves (Fig. S10), while other enzymes (At2g40190, At2g47760, and At5g38460) did not show any localization to LDs (Fig. S10).

Fig. 6.

Various enzymes enriched in the lipid droplet (LD) fraction of Arabidopsis roots are localized to LDs. mCherry‐tagged proteins were expressed in either Nicotiana tabacum pollen tubes or Nicotiana benthamiana leaves. In leaves, the formation of LDs was induced by co‐expression with MmDGAT2. LDs were stained with BODIPY 493/503 or the endoplasmic reticulum (ER) was visualized using co‐expressed ERD2‐CFP. Images are single planes obtained by confocal laser scanning microscope (CLSM). In pollen tubes, the putative glycosyltransferase At1g78800 localized to LDs, the ER, and the plasma membrane, while the oxidoreductase FOREVER YOUNG3 (FEY3) only localized to the ER and the plasma membrane. The putative hydrolase At4g33180 did not show any co‐localization with LDs or the ER but was found in more cloud‐like structures and the PRENYLCYSTEINE METHYLESTERASE (PCME) targeted both to LDs and the ER. In leaves, all proteins were localized at LDs. Each image is representative of at least six pollen tubes or four leaf areas. Bars, 10 μm. As shown on the right, the protein structures of At1g78800 and FEY3, as predicted by AlphaFold2, show a hydrophobic surface on one side of the protein; hydrophobicity indicated by red color.

Localization of FEY3 and a protein of unknown function, At4g33180, expressed in leaves was more diffuse but both accumulated as rings around LDs. However, neither of the two proteins showed an obvious LD localization in pollen tubes (Fig. 6). One candidate protein without predicted enzymatic function was MYOSIN BINDING PROTEIN 14 (MYOB14), which has only recently been identified to associate with LDs (Omata et al., ²⁰²⁴). Omata et al., also described LD localization for UNSATURATED FATTY ACID OXIDASE 1 (UFAO1), which we also investigated. We could confirm the clear LD association between MYOB14 and UFAO1 in both pollen tubes and leaves (Fig. 7). MYOB14 has been proposed as a linker between myosins and LDs, whereby its predicted protein structure suggests that a hydrophobic helix on one side of the protein might interact with LDs, while a putative myosin binding domain could reside in a long adjacent helix. UFAO1 was hypothesized to work together with LIPID DROPLET METHYLTRANSFERASE 1 (LIME1) in furan‐containing fatty acid biosynthesis based on homology to a set of proteins studied in the photosynthetic bacterium Cereibacter sphaeroides (Lemke et al., ²⁰²⁰; Omata et al., ²⁰²⁴). Another protein of unclear function, encoded by At5g59960, also targeted to LDs in both transient expression systems (Fig. 7).

Fig. 7.

Several proteins of unknown function that are enriched in the lipid droplet (LD) fraction of Arabidopsis roots are localized to LDs. mCherry‐tagged proteins were expressed in either Nicotiana tabacum pollen tubes or Nicotiana benthamiana leaves. In leaves, the formation of LDs was induced by co‐expression with MmDGAT2. LDs were stained with BODIPY 493/503 or the endoplasmic reticulum (ER) was visualized using co‐expressed ERD2‐CFP. Images are single planes obtained by confocal laser scanning microscope (CLSM). In pollen tubes, MYOSIN BINDING PROTEIN14 (MYOB14) localized to LDs, while the protein of unknown function At5g59960 and UNSATURATED FATTY ACID OXIDASE1 (UFAO1) localized to LDs and the ER. In leaves, all proteins were localized at LDs. Each image is representative of at least nine pollen tubes or four leaf areas. Bars, 10 μm. As shown on the right, the protein structures of At5g59960 and UFAO1, as predicted by AlphaFold2, show a hydrophobic surface on one side of the protein, while MYOB14 harbors a hydrophobic α‐helix; hydrophobicity indicated by red color.

Furthermore, we tested 10 additional candidates, including several with described or predicted enzymatic function, that did not localize to LDs (Figs S11–S14).

Finally, we investigated three proteins involved in terpenoid metabolism (Fig. 8). One of these, the cis‐prenyltransferase LEAF WILTING 1 (LEW1), which is involved in dolichol biosynthesis (Zhang et al., ²⁰⁰⁸; Kwon et al., ²⁰¹⁶), localized to the ER in pollen tubes but displayed localization surrounding LDs in N. benthamiana leaves. The other two proteins are members of the larger protein family of oxidosqualene cyclases (OSCs), THALIANOL SYNTHASE 1 (THAS1) and MARNERAL SYNTHASE 1 (MRN1). The protein family of OSCs also contains CAS, a previously reported LD‐localized protein (Kretzschmar et al., ²⁰¹⁸) that catalyzes the first committed step in phytosterol synthesis. All OSCs share the common substrate (3S)‐2,3‐oxidosqualene, a hydrophobic triterpenoid, but synthesize a strong variety of cyclic triterpenes with one to five rings (Hoshino, 2017; Fig. 9). THAS1 (Xiang et al., ²⁰⁰⁶) and MRN1 (Xiong et al., ²⁰⁰⁶) have been well explored in regard to their enzymatic function and are both part of gene clusters that harbor genes coding for enzymes that further modify their products (Field & Osbourn, 2008; Huang et al., ²⁰¹⁹). THAS1 and MRN1 targeted LDs in both tissues. Interestingly, both THAS1 and MRN1, based on their structural prediction, have a hydrophobic flat face with a hole in the middle that might give access to their substrate. To test whether enzymes acting downstream of thalianol synthesis (Huang et al., ²⁰¹⁹) are associated with LDs, we assayed both THALIANDIOL DESATURASE (THAD) and THALIANOL HYDROXYLASE (THAH). However, both enzymes targeted the ER in pollen tubes and reticular structures in leaves (Fig. S15).

Fig. 8.

Various enzymes enriched in the lipid droplet (LD) fraction of Arabidopsis roots that are involved in isoprenoid metabolism are localized to LDs. mCherry‐tagged proteins were expressed in either Nicotiana tabacum pollen tubes or Nicotiana benthamiana leaves. LDs were stained with BODIPY 493/503 or the endoplasmic reticulum (ER) was visualized by co‐expressing ERD2‐CFP). Images are single planes obtained by confocal laser scanning microscope (CLSM). THALIANOL SYNTHASE 1 (THAS1) and MARNERAL SYNTHASE1 clearly colocalized with LDs in pollen tubes and in leaves, while the undecaprenyl pyrophosphate synthetase LEAF WILTING 1 (LEW1) targeted the ER in pollen tubes and LDs in leaves. Each image is representative for at least 10 pollen tubes or 4 leaf areas. Bars, 10 μm. As shown on the right, the protein structures of THAS1 and MRN1, as predicted by AlphaFold2, show a hydrophobic surface on one side of the protein (hydrophobicity indicated by red color) with a whole in the middle that might facilitate the uptake of the substrate (black arrow). The structure of LEW1 displays an amphipathic α‐helix.

Heat stress impacts the LD proteome

Lipidomic and microscopy data indicated that heat stress affects LD number and core LD compounds like TAG and SEs (Figs 1, 2, S1). To test whether this would be reflected in the protein coat of the LDs, we repeated the LD isolation from axenic root cultures tgd1‐1 sdp1‐4, this time including an additional heat stress treatment of 24 h at 37°C (Datasets S15 and S16). Again, total cellular extracts and LDs were isolated. In the total cellular fraction, the total abundance of LD proteins was similar. Furthermore, for the control treatment, known LD proteins were again highly enriched (Fig. S16). However, enrichment factors in the heat treatment were much lower (Fig. S17), maybe because misfolded proteins or whole organelles were sticking stronger to the LD surface. As of this, no further candidates for testing LD localization were picked based on this experiment. Nevertheless, potential changes were assessed in regard to the LD protein composition.

As the riBAQ values in the LD fraction have a bias depending on the enrichment, we compared the data on a basis of the composition of all LD proteins in %. From this analysis, several changes were observed (Fig. S18). For example, LDAP levels were reduced and the levels of CBE5, LDS and OBL5 increased.

Precursors and products of OSCs are enriched in LDs

The localization of THAS1 and MRN1 to root LDs and their reported enzymatic function brought up the question of whether products and intermediates of the terpenoid metabolism are stored in LDs. To tackle this question, we isolated root LDs again from the tgd1‐1 sdp1‐4 mutant but this time analyzed them with respect to their metabolite content using a combination of ESI‐MS/MS, and GC‐MS (Datasets S17–S20). For both membrane and neutral lipids as well as terpenoids, we calculated recovery scores, that is how much of the metabolites were recovered in the LD‐enriched fractions in comparison with the total fraction in % (Fig. 10a). During the isolation process, hydrophobic compounds could change their localization and move into the monolayer or the hydrophobic matrix of LDs. Furthermore, lipids from plastoglobules could get co‐enriched, even though the enrichment factor in regard to proteins was much lower. Nevertheless, this approach can indicate whether certain lipids are associated with LDs. The free forms of the major phytosterols, β‐sitosterol, stigmasterol, and campesterol each had a recovery score of less than 3%. These values support the notion that the main membrane sterols are present throughout the whole endomembrane system, and do not accumulate in LDs. However, phytosterol recovery is still noticeably higher than that of membrane phospholipids, which mostly were recovered to < 1% in the LD‐enriched fraction. Remarkably, there were strong differences among the membrane lipids: for example, the recovery of PC was 10 times higher than that of PE, indicating a distinct glycerophospholipid composition of root LDs. The higher recovery of SQDG could be caused by a co‐enrichment of plastoglobules, even though other major plastidial lipids like MGDG and DGDG were recovered only at very low levels.

Fig. 10.

Terpenoids are enriched in lipid droplets (LDs) isolated from roots of tgd1‐1 sdp1‐4 mutant plants. LDs were enriched from roots of the Arabidopsis tgd1‐1 sdp1‐4 mutant grown under axenic root cultivation conditions. Aliquots were taken from an initial total cell extract and the final LD‐enriched fractions and subsequently analyzed for abundance of various metabolites. Detected compounds were grouped into membrane glycerolipids, TAGs, free sterols, other nonesterified compounds, and esterified compounds. The enrichment scores of individual compounds in the LD‐enriched fraction compared with the initial cell extract was calculated (a). The detected quantities of the individual compounds in all samples (normalized to the average of each compound across all samples) were used for a principal component analysis. This analysis supported different enrichment properties of the five compound groups (b). Finally, the three compounds cycloartenol, β‐amyrin, and thalianol and their esterified forms could also be detected in root lipidomics samples obtained from wild‐type seedlings after control and heat stress treatment (c; same biological samples as initially presented for free sterols in Fig. 2). Values are from n = 4 biological replicates, and are shown as mean ± SD. Statistical differences were calculated by Welch's t‐test using Benjamini–Hochberg correction for multiple comparisons and are represented as follows: ns, P > 0.05; *, P < 0.05; **, P < 0.01. DGDG, digalactosyldiacylglycerol; FS, free sterols; MGDG, monogalactosyldiacylglycerol; ML, membrane lipids; ns, not significant; PA, phosphatidic acid; PC, phosphatidylcholine; PE, phosphatidylethanolamine; PI, phosphatidylinositol; PS, phosphatidylserine; SQDG, sulfoquinovosyldiacylglycerol; TAG, triacylglycerol.

Recovery values were drastically higher for esterified sterols, ranging from 33 ± 15% in campesteryl esters to 89 ± 33% for cycloartenyl esters. This suggests a clear distinction between free sterols, which are components of cellular membranes throughout the endomembrane system, and SEs that are stored in LDs. While higher than membrane lipids, the recovery score of TAG (11 ± 4%) was notably lower than the SEs. Interestingly, certain TAG species were more prevalent in root LDs (Dataset S11), indicating that select TAG pools are more strongly enriched in the LD fraction than others that might reside in other cellular compartments, such as the plastids, as lenses in the ER, or as small LDs that did not detach from the ER or were too small to be collected.

Precursors of sterols and other triterpenes also appeared to be enriched in LDs, as squalene reached a recovery factor of 44 ± 24%. Squalene epoxide was also detected in the LD fraction but levels were too low to be reliably quantified in the total extract. Interestingly, we also found esters of farnesol (at a score of 12.4 ± 7%) but not free farnesol.

Finally, we were able to identify the two triterpenes β‐amyrin and thalianol and their esters (Figs 9, 10a). The recovery score of free thalianol was higher than that of most sterols, reaching on average 14 ± 5%, while β‐amyrin was recovered at 8 ± 5%. The only sterol that reached similar levels to thalianol was cycloartenol (17 ± 8%), which is also synthesized by an LD‐localized enzyme (Kretzschmar et al., ²⁰¹⁸). In addition, several putative thalianol derivatives had recovery scores ranging from 8 to 18%. Unlike SEs, esters of thalianol (14 ± 5%) were recovered from LDs with similar efficiency as free thalianol, implicating that both molecules are likely similarly associated with LDs. Recovery of β‐amyrin esters (51 ± 20%) was remarkably greater than free β‐amyrin. A PCA of the values across samples also showed that the different metabolite classes exhibit similar patterns of enrichment in LDs (Fig. 10b).

As the LDs analyzed for their composition were derived from axenic roots grown under high sucrose levels, we aimed to additionally identify triterpenes in Arabidopsis roots grown under more standard conditions, that is on agar plates. For this, we reanalyzed the GC‐MS data on lipid extracts previously obtained for free sterol analysis (Fig. 2; Dataset S5). Indeed, we were able to identify both the esterified and free forms of thalianol and amyrin (Figs 10c, S19; Datasets S5, S11), albeit at much lower levels than for phytosterols (Fig. 2).

In conclusion, these proteomic and metabolomic data support the notion that OSCs, their precursors and direct products (and their esters) are enriched on and in LDs (Fig. 9).

Discussion

LDs act simultaneously as a sink and source during membrane remodeling

Our investigation of root LD composition identified compounds known to accumulate within LDs of other tissues, such as TAGs and SEs, but also distinct proteins and metabolites that hint at them having additional, previously unknown functions.

Based on our initial observations, the early elongation zone of Arabidopsis roots grown under control conditions accumulates LDs to a much higher extent than other root zones (Fig. S1). The reason for this local maximum of LDs is unclear; however, it is tempting to speculate that they accumulate here before rapid cell elongation, which might demand resources for membrane lipid synthesis. Similarly, it has been suggested that LDs in pollen tubes help deliver precursors of membrane phospholipids to the apical membrane which is constantly extended (Ischebeck, 2016). LD numbers in all zones of the Arabidopsis root increase after plants have been subjected to 1 d of heat stress (Fig. 1). Concomitantly, we observed an increase in the total amount of TAGs in roots (Fig. 2), comparable to previous reports from Arabidopsis seedlings and leaves (Mueller et al., ²⁰¹⁵; Shiva et al., ²⁰²⁰; Scholz et al., ²⁰²⁴), and tobacco pollen tubes (Krawczyk et al., ^2022a). As such, processes in roots are probably similar to the ones proposed in leaves, albeit with fewer contributions of plastidial lipids: LDs and their core component TAG serve as sinks to sequester acyl chains from degraded membrane lipids, as the membrane is remodeled to adapt to the change in temperature. Supporting this hypothesis, the average number of double bonds in root TAGs increased after heat stress, while the average number of double bonds in membrane lipids decreased, as polyunsaturated acyl chains are channeled from membrane lipids into TAGs to decrease membrane fluidity at higher temperatures (Fig. S5).

Interestingly, SEs that occur at similar levels as TAG in unstressed roots decrease during heat stress (Fig. 2). This suggests that SEs act as a source for the increase in acylated sterol glycosides (Dataset S6). While the cellular function of acylated sterol glycosides is unclear, one could speculate that they stabilize membranes or membrane domains under stress. In conclusion, LDs might simultaneously act as a sink and source to sustain membrane lipid homeostasis under stress.

Root LDs are distinct from LDs of other organs

As the use of the mutant in an axenic root culture is an artificial system, it should be assessed whether the LD proteins found in roots in this study are expressed in WT roots grown under more regular conditions (Klepikova et al., ²⁰¹⁶). This comparison, taking into account all known LD proteins, showed strong similarities in regard to the abundance of specific members of protein families (Dataset S21). For example, LDAP3 and LIME2 are the most abundant isoforms of the families both on the protein and transcript level, while LDAP2 and LIME1 are hardly found. Similarly, CALEOSIN4 is by far the most abundant of all eight isoforms. Also the high abundance of α‐DOX in roots is reflected in the transcriptome, as its level there is more than 100‐fold higher in roots (without apex) than in mature leaves or seeds. Overall, most other detected LD proteins are either most strongly expressed in roots on the transcript level (like THAS1 and MRN1) or show a ubiquitous expression pattern.

When comparing now the root LD proteome (Table 1) to LD proteomes of other tissues, (Brocard et al., ²⁰¹⁷; Kretzschmar et al., ^{2018, 2020}; Scholz et al., ²⁰²⁴), it appears most similar to leaves, primarily because no oleosins are found and LDAPs (Gidda et al., ²⁰¹⁶) are the main LD proteins. Nevertheless, there are some striking differences also between LDs of Arabidopsis leaves and roots. First, in leaf LDs obtained from the same tgd1‐1 sdp 1‐4 mutant line, CALEOSIN3 is the most abundant LD protein (Scholz et al., ²⁰²⁴), while in roots, the two detected proteins of the caleosin protein family comprised only minor amounts of the LD protein fraction (Table 1). Conversely, α‐DOX1 is the second most abundant protein in root LDs, while it is only present in small amounts in leaves (Scholz et al., ²⁰²⁴). There, α‐DOX1 and CLO3 have been described to act in concert in the oxidation of α‐linolenic acid to 2‐hydroxy‐octadecatrienoic acid (Shimada et al., ^{2014, 2015}). Notably, in tomato, the expression of an α‐DOX1 gene was reported to be responsive to salt stress and wounding (Tirajoh et al., ²⁰⁰⁵); thus, it cannot be ruled out that the conditions of our root cultures similarly induced α‐DOX1 expression even though in a previous global determination of transcript levels, α‐DOX1 was much higher in roots than in leaves (Klepikova et al., ²⁰¹⁶). In any case, even if the hyper accumulation of α‐DOX1 in root LDs is stress‐mediated, the independence of its accumulation from high protein amounts of caleosins is quite striking and in contrast to leaf LDs, where both α‐DOX1 and CLO3 are upregulated after different stresses (Scholz et al., ²⁰²⁴). Hence, putative reaction products of α‐DOX1 might be processed differently at root LDs compared with leaf LDs.

Further differences in metabolic reactions at LDs are implied by a number of root LD proteins that were not detected in leaves. These include steroleosins, for which several isoforms were detected. HSD4 and/or 7 were most abundant. These two proteins are identical on the protein level and therefore not distinguishable in proteomic datasets. Regarding HSD4/7, it is interesting to note that these particular steroleosins have not been reported in other plant tissues so far, even though other members of the protein family have been reported in seeds and seedlings (Baud et al., ²⁰⁰⁹; Kretzschmar et al., ²⁰²⁰). For the Arabidopsis steroleosin HSD1, hydroxysteroid dehydrogenase activity on mammalian sterols has been reported (d'Andrea et al., ²⁰⁰⁷), which led to speculation that steroleosins could have a role in the conversion of different brassinosteroids thereby regulating their biological activity (Chapman et al., ²⁰¹²). However, no endogenous substrates of any steroleosin have been identified so far; thus, a potential role of HSD4/7 in the regulation of root brassinosteroid activity is highly speculative.

Other proteins like GPAT9 and LPEAT1 were identified as LD‐associated in roots in this study but have clearly important functions in all plant cells (Shockey et al., ²⁰¹⁶; Jasieniecka‐Gazarkiewicz et al., ²⁰¹⁷). These proteins and others localize not only to LDs but also to the ER (Figs 4, 5, 6, 7, 8), raising the question of why this dual targeting is observed. One reason could be that due to overexpression effects, the binding sites on the LDs get saturated leading to ER targeting instead. Furthermore, proteins might bind to the ER first and then move over to the LDs as proposed for oleosin (Beaudoin & Napier, 2002) and various mammalian LD proteins (Kory et al., ²⁰¹⁶; Song et al., ²⁰²²), and are imaged while en route. Another possibility is that proteins shuttle between LDs and the ER as a mechanism to regulate their activity. For example, GPAT9 might reside inactively at LDs before moving to the ER, where it more likely finds its substrate acyl‐CoA (Bates, 2016).

Even though the TAG levels strongly increased and the lipidome was remodeled in roots under heat stress (Figs 2, S6, S7), the LD proteome was not affected in the same manner. This was similar to the leaves of the same mutant under heat stress, where the proteome did not show any striking changes (Scholz et al., ²⁰²⁴).

Overall, it is poorly understood how lipid metabolism is regulated to sustain lipid homeostasis and remodeling under stress. A study in tobacco pollen tubes indicated that there is no strong regulation at the transcript level, as transcripts coding for lipid‐related genes were not changing strongly during heat‐induced lipid remodeling (Krawczyk et al., ^2022a,b). Also in roots, many proteins that are key in lipid remodeling or have been implicated in heat stress, such as acyl transferases and lipases (e.g. see Dataset S22) are not strongly affected at the transcript level (Kilian et al., ²⁰⁰⁷). Exceptions are FATTY ACID DESATURASE3 and LYSOPHOSPHATIDYLCHOLINE ACYLTRANSFERASE 2, which are > 2‐fold decreased under 3 h of heat stress. HEAT INDUCIBLE LIPASE1, on the contrary, a plastidial enzyme that degrades MGDG and is heat‐induced in leaves (Higashi et al., ²⁰¹⁸), is also 15‐fold upregulated in roots. Important genes involved in triterpene synthesis, the pacemaker HMGR and cycloartenol synthase are not affected (Kilian et al., ²⁰⁰⁷), while MRN1 is upregulated and THAS1 is decreased (Dataset S22).

Arabidopsis root LDs are hubs for triterpene synthesis and storage

Apart from enzymes involved in glycerolipid synthesis, we found enzymes of terpenoid metabolism at root LD, including the 2,3‐oxidosqualene cyclases MRN1 and THAS1 (Table 2; Fig. 8). These proteins share strong sequence similarities with the 2,3‐oxidosqualene cyclase cycloartenol synthase which catalyzes the first committed step in phytosterol biosynthesis and is also localized at LDs (Table 1; Kretzschmar et al., ²⁰¹⁸; Kretzschmar et al., ²⁰²⁰; Scholz et al., ²⁰²⁴).

Both MRN1 and THAS1 are part of gene clusters, wherein adjacent genes encode enzymes that modify the initial reaction product of MRN1 or THAS1 (Field & Osbourn, 2008; Huang et al., ²⁰¹⁹). THAS1 expression in particular appears to be root‐specific and the metabolites derived from thalianol have been reported to modify the Arabidopsis root microbiome (Huang et al., ²⁰¹⁹). The substrate of both enzymes, 2,3‐oxidosqualene and its precursor squalene, is highly hydrophobic and appears to be enriched in the core of root LDs (Fig. 10). The discrete hydrophobic surfaces of these individual enzymes may enable LD binding and as a result, allow these enzymes to access metabolites stored within the LD cores through a hole in this surface (Fig. 8). Interestingly, their products, which are amphipathic due to a hydroxy group (Fig. 9), are also enriched in LD fractions (Fig. 10). This was also found for cycloartenol indicating that the products of 2,3‐oxidosqualene cyclases reside in the LDs for some time before moving to the ER where the downstream enzymes are localized (Fig. S15). Alternatively, the products of 2,3‐oxidosqualene cyclases could get esterified directly at the LDs, as esters of triterpenes were also enriched in LDs. In conclusion, root LDs appear as prime hubs for triterpene synthesis and storage in Arabidopsis. In other plants, they might have this major function also in other tissues, as for example rosemary leaves, birch bark and olive fruit store high amounts of triterpenes (Jäger et al., ²⁰⁰⁹).

Competing interests

None declared.

Author contributions

PS, PWN, TLS, IF, PD, RTM and TI designed the research. PS, JD, ACV, ACC, PWN, MSSL, SS, LH, MN, FD, KFB, LMP, MB, EL, YI, KG and YW performed the research. PS, PWN, ACV, MSSL, MB, TLS, JE, IF, KG, PD, RTM and TI analyzed the data. PS, TLS and TI wrote the paper with the help of all authors. PS, JD, ACV, and ACC contributed equally to this work.

Disclaimer

The New Phytologist Foundation remains neutral with regard to jurisdictional claims in maps and in any institutional affiliations.

Supporting information

Dataset S1 Membrane lipids of wild‐type roots – absolute levels.

Dataset S2 Membrane lipids of wild‐type roots – relative composition of species per class.

Dataset S3 Triacylglycerols of wild‐type roots – absolute levels.

Dataset S4 Triacylglycerols of wild‐type roots – relative composition of species.

Dataset S5 Free sterols, and triterpenoids and their esters of wild‐type roots – absolute levels and relative composition.

Dataset S6 Sterol derivatives of wild‐type roots.

Dataset S7 Membrane lipids of wild‐type and tgd1‐1 sdp1‐4 roots – absolute levels.

Dataset S8 Membrane lipids of wild‐type and tgd1‐1 sdp1‐4 roots – relative composition of species per class.

Dataset S9 Triacylglycerols of wild‐type and tgd1‐1 sdp1‐4 roots – absolute levels.

Dataset S10 Triacylglycerols of wild‐type and tgd1‐1 sdp1‐4 roots – relative composition of species.

Dataset S11 Free sterols, and triterpenoids and their esters of wild‐type and tgd1‐1 sdp1‐4 roots – absolute levels and relative composition.

Dataset S12 Sterol derivatives of wild‐type and tgd1‐1 sdp1‐4 roots.

Dataset S13 Proteins found in Arabidopsis roots of the tgd1‐1 sdp1‐4 mutant – normalized riBAQ and rLFQ values.

Dataset S14 Comparison of proteins in LD‐enriched fractions to total protein fractions isolated from Arabidopsis tgd1‐1 sdp1‐4 roots. Log₂ transformed and imputed data.

Dataset S15 Proteins found in Arabidopsis roots of the tgd1‐1 sdp1‐4 mutant under control and heat treatment – normalized riBAQ and rLFQ values.

Dataset S16 Comparison of proteins in LD‐enriched fractions to total protein fractions isolated from Arabidopsis tgd1‐1 sdp1‐4 roots control and heat treatment. Log₂ transformed and imputed data.

Dataset S17 Terpene and sterol enrichment in LDs – raw values.

Dataset S18 Terpene and sterol enrichment in LDs – processed data.

Dataset S19 Triacylglycerol (TAG) enrichment in LDs – processed data.

Dataset S20 Membrane lipid enrichment in LDs – processed data.

Dataset S21 Expression levels of LD protein‐coding genes in different organs of Arabidopsis thaliana.

Dataset S22 Expression levels of genes associated with lipid remodeling and triterpene biosynthesis.

Fig. S1 Lipid droplets are enriched in parts of the elongation zone in Arabidopsis wild‐type seedlings.

Fig. S2 Comparison of root LDs from Col‐0 and the tgd1‐1 sdp1‐4 mutant under control conditions.

Fig. S3 Comparison of root LDs from Col‐0 and the tgd1‐1 sdp1‐4 mutant under heat stress.

Fig. S4 Root LDs are depleted in the early meristematic zone and accumulate under heat stress.

Fig. S5 The average number of double bonds decreases in most lipid classes in Arabidopsis seedlings subjected to heat stress.

Fig. S6 Composition of glycerolipids in control and heat‐stressed roots from Col‐0 and the mutant tgd1‐1 sdp1‐4.

Fig. S7 Abundance of lipids in control and heat‐stressed roots from Col‐0 and the mutant tgd1‐1 sdp1‐4.

Fig. S8 Enrichment of different organellar proteomes in the LD‐enriched fraction.

Fig. S9 Subcellular localization of GLYCEROL‐3‐PHOSPHATE ACYLTRANSFERASE 4 (GPAT4) in Nicotiana tabacum pollen tubes.

Fig. S10 Subcellular localization of Arabidopsis N‐glycan biosynthetic enzymes in Nicotiana benthamiana leaves.

Fig. S11 Subcellular localization of selected Arabidopsis candidate root LD proteins in Nicotiana tabacum pollen tubes.

Fig. S12 Subcellular localization of Arabidopsis putative dehydrogenases in Nicotiana tabacum pollen tubes.

Fig. S13 Subcellular localization of selected candidate Arabidopsis root LDs proteins with unknown function in Nicotiana tabacum pollen tubes and Nicotiana benthamiana leaf cells.

Fig. S14 Subcellular localization of candidate Arabidopsis root LD proteins with unknown function in Nicotiana tabacum pollen tubes.

Fig. S15 Subcellular localization of various Arabidopsis enzymes acting downstream of thalianol synthase.

Fig. S16 Analysis of protein enrichment in the LD fraction of Arabidopsis roots of the mutant tgd1‐1 sdp1‐4 grown in axenic root culture.

Fig. S17 Analysis of protein enrichment in the LD fraction of heat‐stressed Arabidopsis roots of the mutant tgd1‐1 sdp1‐4 grown in axenic root culture.

Fig. S18 Analysis of protein enrichment in the LD fraction of heat‐stressed Arabidopsis roots of the mutant tgd1‐1 sdp1‐4 grown in axenic root culture.

Fig. S19 Triterpenoids are found in roots of Col‐0 and the mutant tdg1‐1 sdp1‐4.

Table S1 Metadata for proteomic analysis – unstressed roots.

Table S2 Metadata for proteomic analysis – heat‐stressed and unstressed roots.

Table S3 List of oligonucleotides used in this study.

Table S4 List of microscopy settings for pollen tubes and root LDs.

Please note: Wiley is not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office.

Data availability

Data are available in the article Supporting Information (Datasets [Link], [Link], [Link]) and the proteomic raw data under the identifiers PXD051152 and PXD068568 at https://proteomecentral.proteomexchange.org/ui.