Analysis of short-term changes in the Arabidopsis thaliana glycerolipidome in response to temperature and light




Although the influence of temperature, particularly cold, on lipid metabolism is well established, previous studies have focused on long-term responses and have largely ignored the influence of other interacting environmental factors. Here, we present a time-resolved analysis of the early responses of the glycerolipidome of Arabidopsis thaliana plants exposed to various temperatures (4, 21 and 32°C) and light intensities (darkness, 75, 150 and 400 μmol m−2 s−1), including selected combinations. Using a UPLC/MS-based lipidomic platform, we reproducibly measured most glycerolipid species reported for Arabidopsis leaves, including the classes phosphatidylcholine (PC), phosphatidylethanolamine (PE), phosphatidylserine (PS), phosphatidylinositol (PI) phosphatidylglycerol (PG), monogalactosyldiacylglycerol (MGDG), digalactosyldiacylglycerol (DGDG) and sulfoquinovosyldiacylglycerol (SQDG). In addition to known lipids, we have identified previously unobserved compounds, such as 36-C PGs and eukaryotic phospholipids containing 16:3 acyl chains. Occurrence of these lipid species implies the action of new biochemical mechanisms. Exposition of Arabidopsis plants to various light and temperature regimes results in two major effects. The first is the dependence of the saturation level of PC and MGDG pools on light intensity, likely arising from light regulation of de novo fatty acid synthesis. The second concerns an immediate decrease in unsaturated species of PG at high-temperature conditions (32°C), which could mark the first stages of adaptation to heat-stress conditions. Observed changes are discussed in the context of current knowledge, and new hypotheses have been formulated concerning the early stages of the plant response to changing light and temperature conditions.


Changes in environmental conditions directly affect membrane properties. Probably the best studied case is the influence of temperature, which has been shown to exert a major effect on various physical properties of biological membranes. Most notably, it influences membrane fluidity, which is reduced with decreasing temperature (Alonso et al., 1997) and increased at high temperatures (Quinn, 1988). Temperature-related changes involve membrane permeability to water, solutes and protons (Lande et al., 1995), and also have a major impact on the activity of membrane-localized proteins (Whiting et al., 2000). Plants balance these changes by regulating the level of saturation of membrane glycerolipids, as the presence of unsaturated bonds decreases the phase-transition temperature (Nishida and Murata, 1996). Williams et al. (1988) and Tasseva et al. (2004) found that levels of desaturated phosphatidylcholine (PC) and monogalactosyldiacylglycerol (MGDG) were increased in cold-grown Brassica napus leaves compared to control plants. In Arabidopsis, Welti et al. (2002) observed an increase in desaturation in all measured phospholipids after 3 days of cold acclimation. Similarly, mutants deficient in acyl desaturation are more susceptible to cold stress (reviewed by Upchurch, 2008), while cold-tolerant species (Sakamoto et al., 2004; De Palma et al., 2008) or varieties (Horvath et al., 1983) have more desaturated glycerolipids. In addition to decreasing the saturation level, freezing tolerance has also been shown to be achieved by changing the ratio of bilayer- to non-bilayer-forming membrane lipids in the chloroplast outer membrane (Moellering et al., 2010). Heat stress, on the other hand, has the opposite effect. In the grass Agrostis stolonifera, the levels of saturated lipids increase after heat stress, while tolerant varieties have greater amounts of saturated lipids at normal temperature (Larkindale and Huang, 2004). This increase in saturation has been also observed in other plant species (Horvath et al., 1998; Nishiyama et al., 1999) and in Synechocistis (Gombos et al., 1994). In this cyanobacterium, the change was identified as essential for stabilizing photosynthetic complexes at high temperatures.

Most studies on changes in lipid composition in response to changing environments use a targeted approach (Iba, 2002; Williams et al., 1995), with the exception of a few comprehensive studies, such as that of Welti et al., 2002. In addition, most studies have concentrated on analyzing the influence of low temperatures on lipid metabolism, with other environmental parameters such as light intensity being largely neglected. Finally, most studies are endpoint studies, with the analysis performed after days or even weeks (Williams et al., 1988; Welti et al., 2002; Larkindale and Huang, 2004).

We are interested in the response of higher plants towards changing environments, particularly changing temperature and light intensity, and how the metabolism is reconfigured over time. Here we investigate the response of the lipidome of A. thaliana when exposed to changes in temperature (from 21 to 32°C and from 21 to 4°C) or light (transfer into darkness, low light or high light), and combinations of these treatments. In contrast to most previous studies, we concentrated on the early response (from 20 min up to 6 h) after exposing plants to the new condition. The main aims of this study were to address three main questions: (i) which lipid species respond to changing light intensity and temperature, (ii) when changing both light and temperature, are synergistic, additive or antagonistic effects observed, and (iii) can the observed changes be explained by existing knowledge or do they require the creation of new hypotheses?

The results described below show that there is a fast lipid species-specific response to changing light intensity, most notably in the dark. Changing temperature alone does not have a major effect on lipid composition; however, changing temperature in parallel with changing light conditions has either a synergistic effect (for 21 and 32°C) or a clear antagonistic effect (for 4°C). A shift in the distribution of specific lipid species with respect to their degree of saturation in the dark was also observed, which is most likely explained via differential sensitivity of de novo acyl synthesis compared to the activity of desaturases under dark conditions. Furthermore, the data obtained suggest that the early changes observed have a more pronounced influence on the structure of the plasma membrane compared to the thylakoid.

Finally, in addition to the original goal of this study, our comprehensive analysis allowed us to perform a detailed analysis of the distribution of the various lipid species with regard to chain length and degree of saturation. This allowed us to investigate whether existing biochemical knowledge explains the distribution of molecular species sufficiently. This is the case for the majority of observations, but some observations require stating new hypotheses about substrate specificity of enzymes involved in the lipid metabolism.

Results and Discussion

Experimental design

We obtained lipid profiles for plants exposed to eight environmental conditions, comprising combinations of three temperature (4, 21 and 32°C) and four light regimes [darkness, low light (75 μE), normal light (150 μE) and high light (400 μE)] (Figure 1, see Experimental procedures). In case of the 4°C/normal light conditions, we used 85 μE instead of 150 μE in order to avoid cold-induced photooxidative stress (Wise, 1995) and for consistency with previous studies on cold acclimation (Hannah et al., 2005). For convenience, in the following sections, the conditions are indicated by letter codes 4-L (cold/normal light), 4-D (cold/darkness), 21-HL (control temperature/high light), 21-L (control temperature/normal light; control conditions), 21-LL (control temperature/low light), 21-D (control temperature/darkness), 32-L (heat/normal light) and 32-D (heat/darkness). For each condition, we analyzed samples from 19 time points, covering the first 360 min of the plant response at 20 min resolution, where time point 0 corresponds to the start of the treatments. Three independent biological replicates were analyzed at each time point for each treatment, resulting in a total of 435 samples (18 time points in three replicates for eight conditions plus three replicates of time point 0 common for all treatments). Our UPLC/MS platform allowed quantification of the abundance of 253 lipid species. From this pool, 92 glycerolipids are used in this analysis based on peak intensity and measurement reproducibility. Our analysis method allows direct comparison of the abundance of lipid species belonging to the same lipid class, but not of those belonging to different lipid classes due to different ionization efficiencies, mostly determined by the head group (see Experimental procedures).

Figure 1.

 Design of the experiment: light and temperature regimes used in the experiment.
Each box represents one experiment. Its placement on the plot indicates the light intensity and the temperature applied to the plants.

Molecular species abundance supports and extends knowledge about biosynthetic pathways

In the first analysis, we wished to determine to what extent the distribution of lipid species within each class can be explained by known biochemical routes. Figure 2(a) summarizes the various lipids detected reproducibly, and compares the peak intensities of each lipid species within each class. All main classes of glycerolipids in plants were measured, with phospholipids being represented by PC (14 species), phosphatidylethanolamine (PE) (19 species), phosphatidylserine (PS) (six species), phosphatidylinositol (PI) (eight species) and phosphatidylglycerol (PG) (14 species), and galactolipids being represented by MGDG (ten species), digalactosyldiacylglycerol (DGDG) (11 species) and sulfoquinovosyldiacylglycerol (SQDG) (10 species). Table S1 provides information on the acyl chain composition as derived from fragmentation data. The coverage is comparable to that of a previous lipidomic study (Devaiah et al., 2006). The pool of measured compounds is extended for PG and SQDG, although 34-C, 36-C and 44-C PS are absent. On the other hand, due to co-elution problems, the very common species 34:2 PI and 34:1 PI, although detected qualitatively, could be not reliably quantified.

Figure 2.

 The relative abundances of lipid molecular species are mostly explained by known activities of biosynthetic enzyme and desaturases.
(a) Relative abundance of lipid species. The bars show raw intensity values for all measured samples, colored in shades of gray according to the level of desaturation of the compounds: white and the darkest gray indicate 0 and 6 double bonds, respectively. Error bars indicate one standard deviation.
(b) Paths of fatty acid desaturation in plant glycerolipids. Reactions are based on available published data. The fatty acids shown in each class were either determined from our fragmentation data or by Devaiah et al. (2006). The substrate specificities of FATA and FATB were reported by Salas and Ohlrogge (2002). A detailed description of other reactions and corresponding references are given in the main text. Reactions shown in black are the most common, and explain the intensity values for the most abundant compounds. Reactions shown in gray explain compounds that are present in small amounts and may arise from residual activities of biosynthetic enzymes for substrates other than the main ones.

Distribution of fatty acids among lipid classes according to chain length.  The largest proportion of lipid species analyzed in this study contain 36 and 34 acyl carbons, indicating two 18-C fatty acid chains or a combination of 16-C and 18-C chains. Less abundant were species with 38 acyl carbons or more, containing one chain of 20–26 carbons (very long chain fatty acids, VLCFAs) plus one 18-C chain (Devaiah et al., 2006), or species containing 32 acyl carbons (two 16-C chains). Compounds containing 34 and 36 acyl carbons are generally present in all glycerolipid classes. However, 36-C PI is not as abundant as 36-C in other classes, and 36-C PG has not been reported previously in Arabidopsis (Welti et al., 2002; Devaiah et al., 2006). This study shows that 36-C PG is present in Arabidopsis, albeit at low signal intensity. Compounds containing 38 or more acyl carbons are found only in the phospholipid classes PC, PE and PS, although 38:6 DGDG was also found, but at very low signal intensity. On the other hand, PG and PI were the only lipid classes containing 32-C species.

This uneven distribution of the fatty acids is in accordance with previously reported synthesis mechanisms, and can be explained by substrate preferences of lipid synthesis enzymes and the compartmentation of fatty acid substrates in chloroplast and ER membranes. For instance, the dominance of 34-C and 36-C species in most glycerolipid classes is due to the fact that palmitic acid (16:0) and oleic acid (18:1) are the main products of fatty acid synthesis (Ohlrogge and Browse, 1995), and that glycerol-3-phosphate-1-acyltransferase (GPAT) and lysophosphatidyl acyltransferase (LPAT) acyltransferases, either chloroplastic or in the ER, show a strong preference for incorporating these two fatty acids (Murata and Tasaka, 1997; Kim et al., 2005). Glycerolipids containing VLCFAs are almost exclusively extra-chloroplastic, as they are synthesized via fatty acid elongase (FAE) complexes in the ER starting from 18-C chains (Mietkiewska et al., 2007). Study of an interesting Arabidopsis line synthesizing massive amounts of VLCFAs (35S–FAE1; Millar et al., 1998) showed that substrate availability is the major factor limiting VLCFA presence in chloroplast membranes, as high levels of VLCFAs were incorporated into MGDG and DGDG in the case of this mutant. In contrast, PS enrichment for VLCFAs may be explained by substrate preference, because the base exchange phosphatidylserine synthase (BE-PSS), suspected to be the only enzyme responsible for PS synthesis in Arabidopsis (Mizoi et al., 2006), is strongly biased towards substrates containing VLCFAs (Vincent et al., 2001). The distribution of 32-C species also seems to be influenced by substrate preference. The PG synthesis enzymes phosphatidylglycerolphosphate synthases 1 and 2 (PGPS1 and PGPS2) both exhibit high preference for 32-C CDP-diacylglycerol (CDP-DAG; Muller and Frentzen, 2001). In addition, VLCFA accumulation in PG in 35S–FAE1 is the lowest among all classes measured (Millar et al., 1998), so there also appears to be discrimination against VLCFAs in PG synthesis. For PI, there is evidence for activity of PI synthases for 32-C CDP-DAG (Lofke et al., 2008). Although it was not the preferred substrate, it is possible that this affinity is enough to yield the low amount of 32-C PI observed.

The appearance of 36-C PG in this study is worth noting, given that PG has been regarded as containing exclusively prokaryotic DAG (Fritz et al., 2007; Benning, 2009). The reason why eukaryotic PG was not detected previously is currently unclear. Some evidence suggests that eukaryotic phosphatidic acid (PA) may be channeled directly to PA phosphatase (Awai et al., 2006; Fritz et al., 2007), preventing eukaryotic PG from reaching detectable limits. In addition, high amounts of 36-PG in the chloroplast achieved by genetic engineering have been shown to be detrimental to the plant (Millar et al., 1998). In plants other than Arabidopsis, 36-PG has been detected in the plasma membrane and mitochondria (Lynch and Steponkus, 1987; Dorne and Heinz, 1989). In addition, Uemura et al. (1995) detected PG in Arabidopsis plasma membrane preparations. This finding implies the existence of eukaryotic PG and suggested already the presence of 36-C PG.

Degree of desaturation is dependent on class and chain length.  The lipid species analyzed in our study have between 0 and 7 double bonds, with the number of double bonds varying for acyl chains of different lengths and classes. The 16-C acyl chains show only a low level of desaturation, with the notable exception of MGDG and DGDG. In MGDG, 16:3 is the most abundant 16-C chain, while in DGDG, 16:0 and 16:3 are almost equally abundant. In PG, 16:0 and 16:1 are present at almost equal amounts. Although present in small quantities, we found 16:2 and 16:3 in PC, PE, PI, PG and SQDG.

In contrast, 18-C chains are usually mostly desaturated. The most abundant C18 acyl residues are 18:2 in phospholipids and 18:3 in galactolipids. VLCFAs usually have one unsaturation, although saturated chains are also found. It is worth mentioning that, unlike the other VLCFAs, 20-C acyl chains are usually highly desaturated, with a pattern more similar to that of 18-C acyl chains.

As a consequence of the various levels of fatty acid desaturation and the preference of the glycerolipid classes for various combinations of acyl chain lengths, the distribution of unsaturated acyl chains between various glycerolipids can be deduced. 32-PG contains either no or one double bond and 32-PI usually contains three unsaturated bonds. Phospholipids containing 34 acyl carbons usually have two or three double bonds (mostly due to the 18-C chain). The most abundant 34-C galactolipids contain six double bonds for MGDG, three or six in the case of DGDG, and three for SQDG. For phospholipids containing 36 acyl carbons, the most abundant species contain between four and six double bonds, whereas the most abundant species for galactolipids contain six double bonds. We described below how these patterns can be related to known activities of lipid biosynthetic enzymes.

From the above results, it is evident that desaturation has a preference for 18-C chains. However, the presence of monoenoic and dienoic membrane desaturases in both the ER (FAD2 and FAD3) and the chloroplast (FAD6, FAD7 and FAD8) cannot explain this preference, as they show high activity towards both 16-C and 18-C chains (Nishiuchi et al., 1995; Covello and Reed, 1996). Thus, we would expect all 16-C and 18-C chains with one or two double bonds to be fully desaturated to three. The key point seems to be the first unsaturation introduced into the glycerolipids, which is not catalyzed in the ER membrane but by a soluble enzyme in the chloroplast lumen (SSI2, formerly known as FAB2; Mckeon and Stumpf, 1982; Lightner et al., 1994; Kachroo et al., 2007) or FAD5, a chloroplast membrane-localized enzyme (Kunst et al., 1989; Heilmann et al., 2004a). SSI2 has a strong preference for 18-C chains (Kachroo et al., 2007). Therefore, it provides the initial double bond required by monoenoic desaturases, but mainly for 18-C chains. For this reason, 36-C glycerolipids, containing two 18-C chains, can be desaturated either at the chloroplast or the ER membrane. On the other hand, in 34-C glycerolipids, containing one 16-C and one 18-C acyl chains, only one acyl chain (the 18-C chain) is highly desaturated. At the chloroplast, FAD5 can introduce an initial double bond for 16-C acyl chains, but the enzyme is specific for MGDG (Heilmann et al., 2004b). If DGDG is synthesized from MGDG, only MGDG and DGDG will have double bonds introduced by FAD5 and thus serve as substrates for further desaturation via FAD6, FAD7 and FAD8. We are left with the question of how to explain the relatively high abundance of 34:4 PG, which has one more double bond than we would expect according to the previous hypothesis, as PG is mainly chloroplastic. This extra double bond is probably a trans double bond catalyzed by FAD4, a desaturase that is highly active for 16-C chains of PG (Gao et al., 2009). As opposed to the cis double bond introduced by FAD5, the FAD4-catalyzed trans double bond prevents the further desaturation of 16-C chains, and therefore 34-C PGs compounds with more than four double bonds are either very low in abundance or absent (Devaiah et al., 2006; Welti et al., 2002; this study). In addition, FAD4 is probably responsible for the single double bond present in 32:1 PG, which would explain why it is not further desaturated to 32:3 PG.

Importantly, as we mentioned before, we found also low-abundant phospholipids containing 16:1, 16:2 and 16:3 acyl chains. It is known that SSI2 desaturase has some activity for 16-C chains, producing 16:1 fatty acids. Presence of 16:1 fatty acids in the ER is due to the residual activity of Acyl-ACP thioesterase A (FATA) for 16:1 fatty acids, which can ultimately be desaturated to 16:3 by the consecutive action of FAD2 and FAD3, once it is part of a glycerolipid. In addition, we also found low-abundant compounds containing 18:0 fatty acid, which can be explained by a small pool of this compound that was not desaturated by SSI2. Its presence at the ER is also due to the residual activity of Acyl-ACP thioesterase B (FATB) for 18:0 fatty acid. Another species that occurs rarely is 36:7 PG, which has only been detected previously in transgenic tobacco with increased levels of 36-C PG (Fritz et al., 2007). The authors found that the extra double bond is trans, suggesting that FAD4 can also use 18-C chains as a substrate.

As mentioned in the previous section, VLCFA-containing glycerolipids are all extra-chloroplastic, with the exception of the relatively low-abundance 38:6 DGDG. In all cases, one VLCFA is present in combination with a 18-C chain (Devaiah et al., 2006). For compounds containing 40 or 42 acyl carbons, we did not observe more than four double bonds, as while the 18-C chain can have up to three double bonds as described, at most one double bond was detected in the 22-C or 24-C chain. This is consistent with the findings of Millar et al. (1998), who observed a low level of polyunsaturated VLCFAs in the Arabidopsis high-VLCFA 35S–FAE1 line. Therefore, taking into account that the double bonds in 22-C or 24-C are probably introduced by SSI2 into the 18-C precursor chain in chloroplasts, we hypothesize that FAD2 monoenoic desaturase does not use them as a substrate for further desaturation. Remarkably, the 35S–FAE1 line had a high content of polyunsaturated VLCFAs in chloroplastic lipids, suggesting that prokaryotic desaturases do not exclude VLCFAs, in contrast to eukaryotic desaturases. However, our results suggest that there is a low preference of eukaryotic FAD2 and FAD3 desaturases for 20-C chains, as lipids with 38 acyl carbons show up to six double bonds (18:3/20:3). Finally, the presence of 38-C DGDG was interesting, given that this class is chloroplastic. This can be explained by the fact that there is a flux of eukaryotic DAG into the chloroplast, and that MGDG containing eukaryotic DAG is converted mostly to DGDG (Browse et al., 1986).

In summary, we conclude that a large part of the complexity observed for the lipids analyzed here can be explained by current knowledge on plant lipid metabolism. Fatty acid chain lengths and the desaturation level in glycerolipids are determined by the specificity of lipid metabolic enzymes and substrate availability, influenced by cellular compartmentalization of the metabolic pathways. A very good example is the very low content of 16:3 in eukaryotic lipids. This is not because FAD2 and FAD3 desaturases are not able to desaturate 16-C chains, but because the availability of 16:1, as opposed to 18:1, is very limited. Likewise, substrate availability is also a limiting factor for synthesis of galactolipids containing VLCFA. Here, low abundance of VLCFA in chloroplasts makes them inaccessible for chloroplastic GPAT and LPAAT acyltransferases (Millar et al., 1998). On the other hand, substrate specificity of the enzymes appears to be more important in case of VLCFA in PS and 32-C in PG. An interesting question emerging from the data is why galactolipids exhibit more complete desaturation than phospholipids. While one or two predominant species are present in galactolipids for a given acyl carbon number, with phospholipids there are usually several highly abundant species (Figure 2a). Williams et al. (2000) already pointed this out, and suggested that reshuffling of acyl chains in PC is probably the cause for its incomplete desaturation. We are not aware that acyl exchange operates in galactolipids, so this may well be the reason for this difference in desaturation of phospholipids.

Changes in lipid species profiles caused by environmental conditions

As stated above, the main aim of this study was to investigate the early changes in the glycerolipids of A. thaliana leaves upon changing environments, specifically light and temperature. The changes in all lipids analyzed as a function of treatment and time suggest a fairly complex picture (Figure 3 and Table S1). To identify treatment effects and their inter-relationships, we used anova and PCA.

Figure 3.

 Profiles of the measured glycerolipids.
Values were normalized row-wise to the median value for all conditions. The scale is logarithmic. Values lower than the median are shown in shades of red; values higher than the median are shown in shades of blue. Asterisks indicate the species changing significantly in at least one of the conditions with respect to normal conditions (anovaP-value = 1e–06).

The anova analysis revealed that 34% of all analyzed molecular species exhibit a significant response to at least one applied condition (anovaP value 1e−06; FDR value 3.29e−06; pairwise Tukey test 1e–06; see Table S1). The ‘responsive’ groups are chemically heterogeneous, including species of PCs, PEs, PSs, PIs, MGDGs and SQDGs. The PCA shows that, although the individual experimental conditions are not clearly separated, component 1 distributes the samples according to the light intensity (Figure 4). This is in agreement with the results of the anova analysis, in which the light conditions were identified as the most determinant factor related to changing intensities of specific lipid species, with the highest significance for darkness treatments. The anova analysis shows that the darkness effect is further enhanced by high temperature and diminished by low temperature. More importantly, the PCA clearly shows that the effects of temperature on dark conditions are reflected by the same principal component as light conditions. The loadings of principal component 1 identify 34:6 PE, 34:6 PC, 32:3 PI, 34:5 PI and several species of MGDGs as being most important for the separation, whereas component 2 and all higher-order components do not separate any biologically meaningful groups. In order to reduce the data complexity, we separated temperature and light gradient, producing two datasets: the first, containing three different temperature conditions under normal light and the second, containing four different light conditions under normal temperature. The various light conditions are again separated on the PCA plot, but the temperature gradient gives no clear separation (Figure 4c,d). Supervised analysis using anova models gave similar results for the light gradient, but also identified a small group of temperature-responsive compounds, the most significant ones being 36:4 PG, 36:5 PG and 36:6 PE.

Figure 4.

 Principal component analysis of temperature- and light-dependent lipid changes.
Conditions are color-coded; time points are represented by increasing size of the symbols.
(a) A plot of first two principal components; light conditions.
(b) A plot of first two principal components; temperature conditions.
(c) PCA of the data subset including only samples collected at the control temperature (21°C); first two principal components.
(d) PCA of the data subset including only samples collected under control light conditions (150 μE); first two principal components.

In summary, analysis of the changes observed in lipids after 6 h of eight treatment combinations demonstrates that light conditions have the major influence on the lipid profiles, with temperature modulating the scale of these changes. We discuss the likely sources of this light-dependent variation below. The modulation of the changes by temperature is probably an example of the influence of temperature on metabolic rate (Gillooly et al., 2001), given that cold diminishes darkness effects while heat exacerbates them. The most significant changes mainly concern MGDG and PC species. This may be related to the high flux of these two classes during lipid metabolism. The changes induced specifically by temperature are not sufficient to result in a separation in the PCA. However, the changes found at high temperature were significant, and we hypothesize that they are the first signs of adaptation to heat stress of Arabidopsis membranes. Low temperature, on the other hand, did not cause any significant changes compared with control conditions. This strongly supports the notion that the well-documented response in lipid composition leading to adjustment of membrane fluidity is a late effect (Williams et al., 1988).

Treatment-induced changes: darkness increases desaturated species and triggers 16:3 accumulation in PC, PE, PI and SQDG.  We identified above several classes of lipids that show significant changes in steady-state concentration during the first 6 h of treatment. As the next step in our analysis, we took advantage of the high kinetic resolution of this analysis, allowing a closer look at identified responses and describing the major groups of responsive compounds.

Hierarchical clustering analysis (HCA) analysis identifies two major groups of compounds (Figure 5) that show characteristic behaviors under the various conditions. The first group consists of mostly low-abundant, highly desaturated species belonging to the PC, PE, PI and SQDG classes. All the species belonging to this group have either been reported to be scarce or have not been previously found in plants. In terms of their response to environmental conditions, their concentration increases under two of the three dark conditions, i.e. 21-D and 32-D, with accumulation being faster and more intense (approximately 1.4-fold) at the higher temperature (32-D). Under darkness with the temperature kept at 4°C (4-D), this effect is essentially absent, demonstrating the antagonistic effect of light and low temperatures, as also observed for other molecular responses (C. Caldana et al., Max Planck Institute of Molecular Plant Physiology, Potsdam-Golm, Germany, unpublished results).

Figure 5.

 Heat map of profiles for the lipid species identified as treatment-induced by anova analysis.
The color code described in Figure 3 was used for the heatmap. The three distinct trends in treatment-induced changes are indicated by labels of different colors. These clusters were identified manually, based on the specificity of their response to particular environmental conditions. The manual grouping is in close agreement with unsupervised clustering, as shown by the HCA tree.

The second main group comprises most of the responsive lipid species. Within this group, most species had a clear response towards light intensity that was modulated by temperature. Thus a strong decrease in their relative concentration was observed for the two dark conditions at higher temperature (21-D and 32-D). However, a decrease was also clearly apparent in darkness when kept at 4°C, albeit to a lesser extent, demonstrating a less dramatic attenuating effect than for the first group, for which low temperature essentially eliminated the dark response. Another light-related effect on the second group is the strong increase in response to high light conditions (21-HL), and the weak but still significant decrease under low light conditions (21-LL). Almost half of the lipid species belonging to this group are MGDGs, followed by PCs, PEs and one PG. A common feature for all classes is that the most abundant species (i.e. MGDG 34/36:6 or PC 36:5) are missing, whereas less desaturated species are represented.

In the simplest case, the influence of temperature on the magnitude of dark-induced changes could be due to the influence of temperature on metabolic rates. Within the physiological range, enzyme reaction rates increase with temperature, because temperature increases the number of substrate molecules with sufficient or more energy for activation of the reaction (Mahan et al., 2004). A 10°C increase in temperature will usually increase the reaction rate by around twofold (Atkin and Day, 1990). In addition, lateral diffusion, a parameter that influences enzymatic rates in membrane-based reactions, slows down exponentially when temperature decreases (Matos et al., 2007). Increasing the lateral diffusion would mean a higher probability of fatty acids encountering desaturases, thus high temperatures would facilitate desaturation reactions, while cold temperature could affect them negatively, strengthening or weakening the darkness effects observed in this study, respectively.

As Figure 5 shows, we can also identify a group of five lipids responsive to 32°C treatment independently of the light intensity. Two of them are present at higher levels at this temperature than under control conditions (32:0 PG and 36:5 SQDG), while the other three (36:6 PE, 36:4 PG and 36:5 PG) decrease in response to heat (32-D and 32-L). These two responses may have a compensatory effect to counter, via modulating desaturation, increased membrane fluidity at high temperatures. As we can observe, the three decreasing species in 32-L are highly desaturated. As well, 32:0 PG, increasing in this condition, is saturated. 36:5 SQDG also increases in heat stress, and is highly desaturated but it is more saturated than the most abundant species for their acyl carbon number. Thus, despite the high desaturation of 36:5 SQDG, an increase in its proportion with respect to 36:6 SQDG may be relevant for membrane fluidity. In summary, there is a decrease in highly unsaturated species, and an increase in species with higher saturation compared to the most abundant species for the class/acyl carbon number. Interestingly, the three heat-responsive glycerolipid classes are anionic lipids. Given the importance of anionic lipids in stabilization of membrane proteins (Yu and Benning, 2003), these changes could be important for protecting proteins from heat stress. We consider that this set of changes constitutes the primary response of Arabidopsis membranes to heat stress. Particularly remarkable is the decrease in the level of 36:5 PG within 1 h after imposition of heat stress.

Another notable observation concerns a significant difference in the variance between low- and high-abundance compounds. The most abundant species of PGs, PIs, PEs, MGDGs and SQDGs show the lowest changes, and belong to the ten lipid species with the smallest variance coefficients (Figure S1). In contrast, the lipid species that exhibit the highest treatment-specific responses are all of low abundance. This observation indicates that, within the time scale of the experiment, the treatments applied do not lead to major re-organization of the membrane lipid composition, but rather trigger specific changes in particular low-abundance compounds. Given their low abundance, it is tempting to speculate that some of the changes may be involved in signaling cascades, as expected as part of an early response. However, this hypothesis obviously requires further targeted experimentation.

The relationship described between abundance and variance overlaps with the saturation level of the acyl chains. The PC and MGDG species that decrease under dark conditions (21-D and 32-D) are always less desaturated than the most abundant compound of the same class and carbon number. A likely explanation is that fatty acid synthesis is strongly repressed in leaves in the dark (Ohlrogge and Jaworski, 1997), while desaturation is not light-responsive (Browse et al., 1981). If new (saturated) fatty acids are not produced, desaturation of the acyl chains in glycerolipids increases with respect to the total pool size. Thus the decrease observed for the less desaturated species is essentially due to their shifting into the large pool of highly desaturated and more abundant lipid species (i.e. 36:5 PC or 34:6 MGDG). Our experimental results did not identify a statistically significant change in highly abundant lipid species of most classes, probably due to the large pool size; however, minor changes were seen for 36:5 PC, 34:3 PI and 36:6 SQDG. In the case of 36:5 PC, the increase was very probably channeled to 36:6 PC.

Likewise, accumulation of newly detected species at 21-D and 32-D is explained as follows (see also Figure 2b). Most of these species contain 16:3, a fatty acid arising from desaturation of 16:1, which may be produced by residual activity of SSI2 and since it is a substrate of FATA and FATB (Salas and Ohlrogge, 2002) it can also be transported to the ER. Lipid species containing 16:1 always co-elute with very abundant species containing 16:0 and the same mass. Therefore, we cannot detect them under normal light and temperature conditions. When no new 16:1 is synthesized, as under dark conditions, all species containing 16:1 are desaturated to 16:3 species, hence the tremendous rise in their concentration.

In agreement with our results, diurnal studies have shown that the amount of 18:1 acyl chains increases during the day and decreases overnight (Browse et al., 1981; Ekman et al., 2007), while 18:2 and 18:3 accumulate during the night and decrease over the day (Ekman et al., 2007). In their analysis, Ekman et al. (2007) reported that MGDG exhibits very small diurnal variations in the fatty acid profile; we found that all except the most abundant molecular species of MGDG exhibited a significant decrease under dark conditions. Ekman et al. (2007) analyzed the total amount of fatty acids per class rather than quantifying individual molecular species; thus, the high content of 18:3 in the unchanging 34:6 and 36:6 species could mask a change in the content of this fatty acid in other species.

On the other hand, we observed an increase in low-desaturated compounds of PC and MGDG under 21-HL, suggesting that lipid metabolism is not only sensitive to darkness but also to higher than usual light intensities. A surplus of NADPH at high light intensities (Stitt, 1986) can possibly explain this effect, stimulating fatty acid synthesis. This increase in less desaturated species may suggest that the activity of the desaturases is not enough to convert any surplus into more highly desaturated species, thus creating a situation opposite to that under dark conditions. The data suggest one further conclusion. Several of the PC species displaying this feature are highly abundant, whereas similarly behaving MGDG species are present in minor amounts, suggesting that the overall saturation status of the PC pool is more affected than that of the MGDG pool. Due to the preferred localization of PCs and MGDGs in the plasma membrane and thylakoids, respectively, the changes observed would have a more pronounced influence on the structure of the cell membrane compared to the thylakoids.

The fact that the main lipid classes responding to light conditions are PCs and MGDGs fits very well with the fact that PCs and MGDGs are the primary sites for de novo fatty acid allocation in Arabidopsis (Browse et al., 1986). As the flux through these classes is greater than through other classes, it is more likely that they are more sensitive to changes in precursor supply. Hence, when the flux of new acyl chains (with one or no double bonds) decreases in response to the dark treatments as described above, there is a drop in the abundance of species with a low degree of desaturation.

General correlation patterns between measured lipids

Most compounds exhibited fluctuations over time for each condition. Remarkably, a few patterns of fluctuation were shared by a large number of species. Concerted variation of metabolites has been observed in a number of metabolomic studies, and is thought to be a result of both the metabolic pathway structure and environmental fluctuations (Weckwerth and Morgenthal, 2005; Morgenthal et al., 2006). Thus, the correlation patterns can potentially provide information on the activity of the underlying pathway. High correlations were observed principally for compounds within the same class, although not to the same degree for all classes (Figure S2). For instance, most correlation values for DGDGs are above 0.6; whilst, in contrast, high correlations among species with different numbers of acyl carbons within PCs and PGs were comparatively few. For 36-C PC, high correlations were observed between consecutive species in the desaturation pathway but decreased even after two desaturations. An intermediate case is that of PE, where very high correlations arose within VLCFA-containing species, with moderately low correlations between highly desaturated 34-C PE and the rest of the PE species. According to Camacho et al. (2005), high correlations may arise from an enzyme controlling a pair of metabolites. In view of this, we expect most of the DGDG profiles to be controlled by DGDG synthases, while FAE complexes would have a strong influence on PE species with 38–42 acyl carbons. In the case of 36-C PC, variations may perhaps be explained by the activity of FAD2 or FAD3. With regard to correlations among different classes, we observed high scores within two groups: one formed by classes PC, PE, MGDG and DGDG, and the second formed by PI, PS, PG and SQDG. Remarkably, high correlations were found within classes localized in different compartments and in branches of the pathway that are apparently unlinked. The fact that all the compounds in the first group are important sinks of de novo synthesized fatty acids (Browse et al., 1986; Hocquellet et al., 2005) suggests there may be a link between correlations and fluxes. The flux of fatty acids towards the second group, comprising anionic glycerolipids, is substantially lower (Browse et al., 1986).


The untargeted lipidomic approach, combined with sampling under a variety of environmental conditions, provides a comprehensive dataset containing not only most reported membrane lipids, but also previously unobserved compounds, such as eukaryotic lipid species containing 16:3 acyl chains or 36-C PG. The patterns of lipid abundance have been discussed in the context of compartmentalization of biosynthetic pathways and enzyme specificity. The distribution of lipid species abundance can be explained by current knowledge about plant lipid metabolism. The appearance of new compounds, on the other hand, allowed formulation of new hypotheses about biochemical mechanisms leading to their formation, such as the putative activity for FAD2 and FAD3 desaturases for 20-C chains. The analysis of the short-term response of membrane components to various temperature and light conditions showed that the light regime has a major effect on the levels of a range of lipids, with opposite effects induced by darkness compared to high light conditions with regard to the responses of glycerolipids. Heat treatment was shown to have a considerable impact, with 36:5 PG being a sensitive indicator of high temperature. Conversely, cold treatment was shown to have no specific effect on lipid profiles during the first 6 h of treatment. Overlapping conditions of cold/dark and heat/dark indicate that the temperature regime has a significant modulating effect on the lipidome of Arabidopsis plants exposed to dark conditions. Analysis of the changes at the level of head group chemistry, acyl chain length and saturation level indicate that the observed changes in membrane lipid composition probably originate from light/dark control of the lipid biosynthesis machinery, with ongoing desaturation and acyl editing processes, with temperature probably affecting the rate of the enzymatic reactions.

Experimental Procedures

Plant material

Arabidopsis thaliana Col-0 plants were grown for 4 weeks under short-day conditions (8 h light/16 h night), and then for 2 weeks under long-day conditions (16 h light/8 h night). Control plants were kept at 21°C and a light intensity of 150 μmol m−2 s−1 (subsequently abbreviated as μE). Treatments were applied 4 h after commencement of the light period. There were four light regimes (high light, 400 μE; normal light, 150 μE; low light, 75 μE; and darkness) at control temperature of 21°C. Cold-treated plants were exposed to 4°C and two light regimes: 85 μE and darkness. Heat-treated plants were exposed to 32°C and two light regimes: 150 μE and darkness. For the eight environmental conditions, three rosettes were harvested every 20 min during the first 6 h of each treatment. The plant material was immediately frozen in liquid nitrogen and powdered using a cryogenic grinding robot (Labman Automation,; Stitt et al., 2007).

Lipid extraction

Aliquots of 100 mg were prepared in 1.5 ml Eppendorf tubes (, under constant freezing. The lipids were extracted from each aliquot in 1 ml of a mixture of chloroform/methanol/water (1:2.5:1) at −20°C. Extractions were performed while shaking for 30 min at 4°C (Thermo Stat Plus, Eppendorf). Following this extraction step, the tubes were spun down for 5 min at 4°C in a table-top centrifuge (Eppendorf). The organic phase was dried down in a SpeedVac (Heto-Holten, and resuspended in 100 μl isopropanol/hexane/water (55:20:25, upper phase). Between sample collection and measurements, samples were stored dry at −80°C.

UPLC/MS measurement

UPLC separation was performed using a Waters Acquity UPLC system (Waters, using a C8 reverse-phase column (length 150 mm, internal diameter 2.1 mm, particle size 1.8 μm; Waters) at 60°C. The mobile phases consisted of 1% 1 m NH4OAc and 0.1% acetic acid in water (A) and methanol/isopropanol (5:2, 1% 1 m NH4OAc, 0.1% acetic acid) (B). The injection volume was 5 μl. The following gradient profile was applied: a linear 17 min gradient starting from 65% B to 100% B, followed by a 2.5 min isocratic period at 100% B, before a decrease to 65% B over 0.5 min. Finally, the column was re-equilibrated for 5 min, giving to a total run time of 25 min. The flow rate of the mobile phase was 350 μl min−1.

The mass spectra were acquired using a Waters SYNAPTTM mass spectrometer equipped with an ESI interface. All spectra were recorded as centroid data in negative ionization mode at a capillary voltage of 3.0 kV. The sample cone was set to 30, and the extraction cone was set to 4.7. The source temperature was held at 120°C, with a desolvation temperature of 400°C. A cone gas flow of 25 l h−1 and a desolvation gas flow of 800 l h−1 were used in the ionization source. All spectra were recorded with the extended dynamic range mode on, covering a mass range from m/z 50–1600. The scan time was set to 0.08 sec, and the inter-scan delay time was set to 0.02 sec, leading to a scan rate of 10 scans per second. All spectra were measured in MSE mode, where low-collision energy MS spectra and high-collision energy MS/MS spectra are recorded alternating. In the high-energy MS/MS mode, the same settings as for the low-energy MS mode were used, except the collision energy was set to a ramp between 15 and 30 eV. All spectra were calibrated on the fly, with the separately injected m/z of 554.2615 [M-H+] (leucine enkephalin) employing the lock mass function of the two-spray ionization source.

Peak identification and quantification

Processing of chromatograms, peak detection and integration were performed using QuanLynx within the MassLynx software (version 4.1, Waters). In a targeted approach, a set of Arabidopsis lipid species previously described by Devaiah et al. (2006) were searched for in our samples based on their exact m/z (including an error tolerance of 10 ppm) and their expected retention time (RT). These m/z and RT parameters were then assembled into an automated method within QuanLynx for systematic extraction and detection of possible lipid peaks. The detected peaks were checked manually, and peak areas were integrated in MassLynx. The detected peaks were then exported as a matrix containing the RT, the m/z and the intensity of all lipid species detected in each sample.

In addition to the automated MS peak detection routine described above, we also performed a manual search for possible MS/MS fragments that could be indicative of the fatty acid composition of the detected peaks. These searches were performed on selected, representative spectra.

Data analysis

Data normalization, visualization, HCA, anova and PCA analysis were performed using R software (Ihaka and Gentleman, 1996; Warnes, 2009). Relative chromatogram intensities were normalized to the total ion count in the sample. Outliers were removed if they exceeded three standard deviations from the mean in a window of five adjacent time points. HCA was performed using Euclidean distance and average linkage. Significantly changing profiles were defined as those with a P value <1e−06 (FDR value 3.29e−06) by anova. Correlation analysis based on Spearman correlation coefficients was performed using R software (Ihaka and Gentleman, 1996).


We would like to thank A. Eckardt, A. Skirycz, A. Bolze, B. Luo, C. Scherling, D. Riewe, D. Sanchez, D. Hincha, F. Lippold, G. Wolter, J. Lisec, J. Witkowicz, L. Sieburth, L. Hong, M. Korn, M. Kraemer, M.Hundertmark, P. Do, S. Bem, S. Jozefczuk and Z. Bieniawska for help in harvesting plant material. We are grateful to Äenne Eckardt, Gudrun Wolter and Antje Bolze for technical assistance with sample handling for UPLC/MS measurements. The study was funded by the Max Planck Society.