Two Isomeric C16 Oxo-Fatty Acids from the Diatom Chaetoceros karianus Show Dual Agonist Activity towards Human Peroxisome Proliferator-Activated Receptors (PPARs) α/γ

The peroxisome proliferator-activated receptors (PPARs) function as ligand-activated transcription factors that convert signals in the form of lipids to physiological responses through the activation of metabolic target genes. Due to their key roles in lipid and carbohydrate metabolism, the PPARs are important drug targets. However, for several of the PPAR drugs currently in use, adverse side effects have been reported. In an effort to identify compounds from marine organisms that may serve as molecular scaffolds for the development of novel and safer PPAR-targeting drugs, we performed a bioassay-guided screening of organic extracts made from organisms supplied by the Norwegian Biobank of Arctic Marine Organisms (Marbank). Among several interesting hits, we identified two poorly described isomeric oxo-fatty acids from the microalgae Chaetoceros karianus for which we provide the first evidence that they might display dual specificity towards human PPARα and PPARγ. Principal component analysis showed that C. karianus stood out from other Chaetoceros species, both with respect to the metabolic profile and the PPAR activity. The isolation of these compounds holds the potential of uncovering a PPAR pharmacophore with tunable activity and specificity.


Introduction
The peroxisome proliferator-activated receptors (PPARs) are operating as lipid sensing receptors at the crossroads between carbohydrate and lipid homeostasis, and metabolic disorders. Functioning as ligand-activated transcription factors, they integrate signals in the form of lipids (fatty acids, phospholipids, eicosanoids, and other oxygenated fatty acids) to physiological responses through the activation of metabolic target genes [1,2]. A first step in this process is the binding of lipid ligands to the PPAR ligand-binding domain (LBD) [3,4]. In humans, the PPAR isoforms PPARα and -γ are mainly expressed in liver and adipose tissue, respectively, while PPARδ is more ubiquitously storage [6][7][8]. Moreover, the activation of PPARγ maintains normal insulin sensitivity through upregulation and secretion of adipokines such as adiponectin and leptin from adipose tissue [9]. In parallel, PPARα increases lipid uptake and energy expenditure in the liver by upregulating targets involved in fatty acid transport, activation, and oxidation [10,11]. Altogether this protects the organism against high levels of circulating triacylglycerols and harmful free fatty acids [12] and ensures normal insulin function in peripheral tissues [9]. However, when adipose tissue increases past critical mass, it attracts immune cells like NK cells and macrophages, and starts to secrete pro-inflammatory cytokines like tumor necrosis factor alpha (TNF-α), interleukin-1 beta (IL-1β), and resistin, which results in suppressed insulin action and increased low density lipoprotein (LDL)-cholesterol [13][14][15][16]. Liver steatosis is another detrimental condition, fueled by PPARγ during a high-fat state [17]. This leads to reduced fatty acid oxidation and ultimately liver failure, which together with the adverse adipokine secretion exacerbates obesity.
With their key roles in lipid and sugar metabolism, the PPARs have for many years been top priority drug targets [18]. Consequently, several PPAR ligands with well-recognized biological effects are in clinical use, like the hypolipidemic fibrates, acting as PPARα activators, and the anti-diabetic thiazolidinediones (TZDs), targeting PPARγ [19,20]. However, adverse side effects have been reported, especially for the TZDs [21,22]. One way to avoid these problems has been to find or design partial [23,24] or dual [25,26] agonists which in general seems to come with fewer side effects. Thus, the aim of the present work was to identify compounds from marine organisms that regulate PPAR activity, and that may serve as molecular scaffolds for the development of novel, safer PPAR-targeting drugs.
The diatom genus Chaetoceros contributes heavily to the overall primary production in the northern and Arctic seas [27,28]. Northern diatoms are characterized by potentially high chemical diversities and bioactivity levels [29,30]. Several species from this genus, kept in stock cultures by the Norwegian Biobank of Arctic Marine Organisms (Marbank, www.imr.no/marbank/en), were used as a library in the bioassay-guided screening for PPAR ligands reported herein. In the present study we identified two poorly described, closely related oxo-fatty acids ( Figure 1) from the diatom Chaetoceros karianus with an apparent dual specificity towards human PPARα and PPARγ. The isolation of these isomeric compounds with dual PPAR α/γ agonist activity holds the potential of uncovering a pharmacophore with tunable activity and specificity.  (1) and (10E)-9-oxohexadec-10-enoic acid (2).

Results and Discussion
The main goal of this study was to identify novel PPAR agonists or modulators from arctic marine organisms. Any such hit compound might later form the basis of a structure-activity-relationship (SAR) effort, where the aim would be to develop drugs for the treatment of diabetes and metabolic syndrome. To this end, we performed a primary screening where we transiently expressed Gal4-LDB constructs comprising the ligand binding domain (LBD) Figure 1. Chemical structures of the two isolated oxo-fatty acids, (7E)-9-oxohexadec-7-enoic acid (1) and (10E)-9-oxohexadec-10-enoic acid (2).

Results and Discussion
The main goal of this study was to identify novel PPAR agonists or modulators from arctic marine organisms. Any such hit compound might later form the basis of a structure-activity-relationship (SAR) effort, where the aim would be to develop drugs for the treatment of diabetes and metabolic syndrome. To this end, we performed a primary screening where we transiently expressed Gal4-LDB constructs comprising the ligand binding domain (LBD) of human PPARα, -γ, and -δ fused to the yeast Gal4 DNA-binding domain in COS-1 cells, a fibroblast-like cell line derived from monkey (Cercopithecus aethiops) kidney. Testing of hundreds of extracts (40 mg/mL) from Marbank in the cell assay system allowed us to identify and confirm the PPAR modulatory activity from a diverse set of organisms (Table S1). The diatom species Chaetoceros karianus (M11031) interestingly showed a consistent activation of PPARα and -γ. Importantly, the same extracts were not able to activate human PPARδ, or human Liver X receptor (LXR) α or -β (not shown).
To verify the findings from C. karianus, Flash chromatography fractions of fresh materials from the Marbank were prepared as described in the experimental section by reconstituting organic extracts in ACN/H 2 O, defatting with hexane, reconstituting in ACN/H 2 O, and eluting from a reverse-phase C18 solid phase extraction (SPE) column. The new eluates (100 mg/mL) were used to stimulate COS-1 cells, this time expressing full-length human PPARα or -γ together with retinoic acid X receptor alpha (RXRα), the natural PPAR heterodimerization partner. Gene activation was assessed on Luciferase reporters driven by natural promoters; human CPTA1 for PPARα and human PLIN1 for PPARγ. As demonstrated (Figure 2A,B), the C. karianus extract was able to activate both promoters. When using reporters with mutated PPAR recognition elements (PPREs), the extract's ability to activate was significantly reduced ( Figure 2C,D), corroborating PPAR-dependent agonism. of human PPARα, -γ, and -δ fused to the yeast Gal4 DNA-binding domain in COS-1 cells, a fibroblast-like cell line derived from monkey (Cercopithecus aethiops) kidney. Testing of hundreds of extracts (40 mg/mL) from Marbank in the cell assay system allowed us to identify and confirm the PPAR modulatory activity from a diverse set of organisms (Table S1). The diatom species Chaetoceros karianus (M11031) interestingly showed a consistent activation of PPARα and -γ. Importantly, the same extracts were not able to activate human PPARδ, or human Liver X receptor (LXR) α or -β (not shown).
To verify the findings from C. karianus, Flash chromatography fractions of fresh materials from the Marbank were prepared as described in the experimental section by reconstituting organic extracts in ACN/H2O, defatting with hexane, reconstituting in ACN/H2O, and eluting from a reverse-phase C18 solid phase extraction (SPE) column. The new eluates (100 mg/mL) were used to stimulate COS-1 cells, this time expressing full-length human PPARα or -γ together with retinoic acid X receptor alpha (RXRα), the natural PPAR heterodimerization partner. Gene activation was assessed on Luciferase reporters driven by natural promoters; human CPTA1 for PPARα and human PLIN1 for PPARγ. As demonstrated (Figure 2A,B), the C. karianus extract was able to activate both promoters. When using reporters with mutated PPAR recognition elements (PPREs), the extract's ability to activate was significantly reduced ( Figure 2C,D), corroborating PPAR-dependent agonism. Monkey kidney fibroblasts, COS-1 cells, were transfected with plasmids expressing human, full-length PPARα, retinoic acid X receptor alpha (RXRα), and a CPT1A promoter-driven luciferase reporter (A) or with a CPT1A reporter with PPAR response elements (PPREs) mutated (C). Correspondingly, cells were transfected with plasmids expressing human, full-length PPARγ and RXRα and a PLIN1 promoter-driven luciferase reporter (B) or with a PLIN1 reporter with PPREs mutated (D). Cells were treated with a dilution series of SPE eluates (100% ACN). 1:1 dilution corresponds to ~100 μg dry weight/mL. The results represent mean relative light units (RLU) ± s.d. of two independent assays performed in duplicate. The activities of the positive controls (Rosiglitazone or Bezafibrate) were set to 100%. D: DMSO, B: 50 μM Bezafibrate, R: 1.0 μM Rosiglitazone.
In order to track the activity found in the extract, we developed a HPLC-based activity profiling protocol (see Experimental section). The chromatogram (210-600 nm) of a semi-preparative HPLC separation (1.25 g extract) and the corresponding activity profile of the time-based fractionation (40 fractions of 60 s each) is shown in Figure 3A. The activity was found in the fractions collected Monkey kidney fibroblasts, COS-1 cells, were transfected with plasmids expressing human, full-length PPARα, retinoic acid X receptor alpha (RXRα), and a CPT1A promoter-driven luciferase reporter (A) or with a CPT1A reporter with PPAR response elements (PPREs) mutated (C). Correspondingly, cells were transfected with plasmids expressing human, full-length PPARγ and RXRα and a PLIN1 promoter-driven luciferase reporter (B) or with a PLIN1 reporter with PPREs mutated (D). Cells were treated with a dilution series of SPE eluates (100% ACN). 1:1 dilution corresponds to~100 µg dry weight/mL. The results represent mean relative light units (RLU) ± s.d. of two independent assays performed in duplicate. The activities of the positive controls (Rosiglitazone or Bezafibrate) were set to 100%. D: DMSO, B: 50 µM Bezafibrate, R: 1.0 µM Rosiglitazone.
In order to track the activity found in the extract, we developed a HPLC-based activity profiling protocol (see Experimental section). The chromatogram (210-600 nm) of a semi-preparative HPLC separation (1.25 g extract) and the corresponding activity profile of the time-based fractionation (40 fractions of 60 s each) is shown in Figure 3A. The activity was found in the fractions collected between 14 and 22 min ( Figure 3A). The pooled fractions of several injections were concentrated and a second semi-preparative HPLC separation based on peak signal absorption at 227 nm was applied (Figure 3, insert). The major peak of activity was found in fractions 10, 11, and 12 ( Figure 3B,C). A dual, yet weakly isoform-specific, agonist activity profile was observed, where fractions 10 and 11 increased PPARα activation over the solvent controls by 224 and 167%, respectively, and fractions 10, 11, and 12 increased PPARγ activation by 175, 388, and 211%, respectively. All the remaining fractions showed minimal activity and were not considered further.  Figure 3A). The pooled fractions of several injections were concentrated and a second semi-preparative HPLC separation based on peak signal absorption at 227 nm was applied (Figure 3, insert). The major peak of activity was found in fractions 10, 11, and 12 ( Figure 3B,C). A dual, yet weakly isoform-specific, agonist activity profile was observed, where fractions 10 and 11 increased PPARα activation over the solvent controls by 224 and 167%, respectively, and fractions 10, 11, and 12 increased PPARγ activation by 175, 388, and 211%, respectively. All the remaining fractions showed minimal activity and were not considered further. Isolation of the two active compounds, 1 and 2, was achieved by reverse-phase SPE and subsequent purification with semi-preparative HPLC. The compounds were tracked with the aid of ultra-performance liquid chromatography-electrospray tandem mass spectrometry (UPLC-ESIMS) and photodiode array detection (UPLC-PDA). In Figure 4A,B, typical base peak chromatograms of an SPE eluate, acquired in both positive and negative mode, are shown. The defatting step with hexane removed apolar lipids that might interfere with our screening system, while the SPE step was designed to retain polar molecules and peptides, which were not of interest in our nuclear receptor screen, as they translocate poorly to the cell nucleus and rarely have been reported to activate PPARs. As shown in Figure 4A,B, the isolated compounds ionize differently in positive and negative mode and appeared to not be the main components of the fraction mixture based on MS signal intensities. Isolation of the two active compounds, 1 and 2, was achieved by reverse-phase SPE and subsequent purification with semi-preparative HPLC. The compounds were tracked with the aid of ultra-performance liquid chromatography-electrospray tandem mass spectrometry (UPLC-ESIMS) and photodiode array detection (UPLC-PDA). In Figure 4A,B, typical base peak chromatograms of an SPE eluate, acquired in both positive and negative mode, are shown. The defatting step with hexane removed apolar lipids that might interfere with our screening system, while the SPE step was designed to retain polar molecules and peptides, which were not of interest in our nuclear receptor screen, as they translocate poorly to the cell nucleus and rarely have been reported to activate PPARs. As shown in Figure 4A,B, the isolated compounds ionize differently in positive and negative mode and appeared to not be the main components of the fraction mixture based on MS signal intensities. This allowed us to determine the molecular formula of the two compounds, which turned out to be C16H28O3 for both molecules, showing that they were structural isomers with a number of ring double bond equivalents of three.
The compounds were isolated as colorless oils in quantities of 60 µg of compound 1 (approx. 95% purity) and 53 µg of compound 2 (approx. 75% purity). The UV spectra of both compounds were similar with a single absorption maximum (λ max ) at 227 nm (1) and 226 nm (2)   Detailed analyses of one-and two-dimensional NMR data, including 1 H, correlation spectroscopy (COSY), total correlation spectroscopy (TOCSY), selective TOCSY (SELTOCSY), gradient-selected heteronuclear single quantum coherence spectroscopy (g-HSQC), gradient-selected heteronuclear multiple bond correlation (g-HMBC), HSQC-TOCSY, and nuclear overhauser effect spectroscopy (NOESY) spectra determined in CD3CN as well as the comparison with published spectral data [31,32], afforded complete 1 H and 13 C assignments for compounds 1 and 2 ( Table 1). The data showed the presence of a trans-coupled ( 3 JH-H = 15.9 Hz) pair of olefinic protons at 6.08/6.84 ppm in 1 and 6.08/6.85 ppm in 2. In both isomers the lower-field olefinic proton (i.e., 6.84 ppm and 6.85 ppm) was additionally coupled ( 3 JH-H = 7.0 Hz) to a methylene proton resonance at 2.22 ppm, while the higher-field olefinic proton (i.e., 6.08 ppm) only showed weak 4 JH-H coupling (1.5 Hz) to the same methylene protons, but no 3 JH-H coupling in addition to its olefinic neighbor. At the same time, the g-HMBC spectra showed the presence of a carboxyl-carbon at 175.0 ppm (1) and 175.2 ppm (2), as well as a carbonyl-carbon at 201.4 ppm (identical shift for both compounds).  Detailed analyses of one-and two-dimensional NMR data, including 1 H, correlation spectroscopy (COSY), total correlation spectroscopy (TOCSY), selective TOCSY (SELTOCSY), gradient-selected heteronuclear single quantum coherence spectroscopy (g-HSQC), gradient-selected heteronuclear multiple bond correlation (g-HMBC), HSQC-TOCSY, and nuclear overhauser effect spectroscopy (NOESY) spectra determined in CD 3 CN as well as the comparison with published spectral data [31,32], afforded complete 1 H and 13 C assignments for compounds 1 and 2 ( Table 1). The data showed the presence of a trans-coupled ( 3 J H-H = 15.9 Hz) pair of olefinic protons at 6.08/6.84 ppm in 1 and 6.08/6.85 ppm in 2. In both isomers the lower-field olefinic proton (i.e., 6.84 ppm and 6.85 ppm) was additionally coupled ( 3 J H-H = 7.0 Hz) to a methylene proton resonance at 2.22 ppm, while the higher-field olefinic proton (i.e., 6.08 ppm) only showed weak 4 J H-H coupling (1.5 Hz) to the same methylene protons, but no 3 J H-H coupling in addition to its olefinic neighbor. At the same time, the g-HMBC spectra showed the presence of a carboxyl-carbon at 175.0 ppm (1) and 175.2 ppm (2), as well as a carbonyl-carbon at 201.4 ppm (identical shift for both compounds). The quantity of the isolated material was too low to obtain one-dimensional direct 13 C-NMR data, and the DEPT spectra only showed few 13 C-resonances of low signal/noise. The carbonyl-carbon showed 2 J C-H and 3 J C-H correlations in the g-HMBC spectra to a 3 J H-H coupled methylene proton resonance at 2.54 ppm (identical shift for both compounds) and the lower-field olefinic proton, respectively. From these data it became clear that the carbonyl-function was interspersed between the olefinic double bond and an aliphatic chain, i.e., the compounds contained an α,β-unsaturated ketone. For further structure determination of compound 1, the assessment of the g-HSQC, g-HMBC, and high-resolution COSY data clearly showed the presence of five methylene groups between the carboxyl-carbon and the olefinic moiety, i.e., the molecule contained a 7,8-double bond. The exact assignment of the methylene groups between the C-9 carbonyl and the terminal methyl group was ambitious, as the protons of five of the six methylene groups appeared in a narrow window between 1.29 ppm and 1.32 ppm. Using high-resolution COSY data, it was possible to identify four of the six methylene groups. Assignment of the remaining two methylene groups was only possible using the HSQC-TOCSY experiment. This is a hybrid experiment that gives through-bond correlations between a 13 C-attached proton to all other coupled protons. The coupled protons can be seen along a line at the same 13 C chemical shift from the carbon atom attached to the primary proton (i.e., the HSQC crosspeak). The assignment of the remaining methylene groups was then possible because individual methylene groups were well resolved in the 13 C dimension even though they overlapped in the 1 H dimension. NMR assignment of compound 2 started at the terminal methyl group. Long-range TOCSY data (mixing time parameter d9 = 160 ms) showed a correlation between the methyl protons and the olefinic protons, and hence these protons belonged to a common 1 H spin system. Such a correlation was absent in a TOCSY spectrum acquired using a mixing time of 80 ms. The assignment of the methylene groups between the terminal methyl and the olefinic moiety was in this case ambitious due to the fact that the NMR sample of compound 2 contained about 25% of compound 1. However, careful examination of the g-HMBC, g-HSQC, SELTOCSY, and COSY data starting from the terminal methyl group showed that compound 2 contained a 10,11-double bond, while the carbonyl group was at the same position as in 1.
Mass spectrometry-aided structural elucidation of 1 and 2 was performed by interpreting HRMS and HRMS/MS analyses of the isolated compounds ( Figure 5). Compound 1 showed a fragmentation pathway leading to an intense [M − 140] − fragment ion apparent at m/z 127.1339 ( Figure 5A). This ion is consistent with α-cleavage of the ketone across C8-C9, i.e., the bond joining the olefinic moiety in  Figure 5B). The prominent m/z 123.1038 product ion in the product ion spectrum of 2 could be the result of water loss from m/z 141.1503 ( Figure 5B).
As hypothesized, based on the dereplication described above, the isolated compounds turned out to be 9-oxohexadec-7-enoic acid (oxo-FA 1) (Figures 1 and 5A) and 9-oxohexadec-10-enoic acid (oxo-FA 2) (Figures 1 and 5B). After establishing the structures of the two molecules 1 and 2, comprehensive literature searches revealed that the (E)-isomers of these molecules had recently been described from other marine organisms [31,32]. In 2005 a new C 16 fatty acid-based oxylipin pathway was demonstrated in axenic cultures of the diatom Thalassiosira rotula, and compound 1 was isolated along with other novel oxo-and hydroxy unsaturated fatty acids, but no biological activity was reported [31]. Compound 2 was isolated in a later study aimed at identifying natural products with anti-inflammatory activity from the red algae Gracilaria verrucosa [32]. Compound 2 significantly inhibited the production of the pro-inflammatory mediators NO and IL-6 in lipopolysaccharide-activated murine macrophage cells [32].
Having refined the oxo-fatty acids 1 and 2 to approximately 95% and 75% purity, respectively, we wanted to reevaluate the weak PPAR isoform specificity observed during the bioassay-guided isolation. To this end we treated COS-1 cells expressing Gal4-LBD chimeras with 100 µM of 1 and 2. As can be seen in Figure 6, the dual PPARα/γ specificity profile was retained, while the potency was somewhat reduced. Still, oxo-FA 1 seems to activate PPARγ slightly better than oxo-FA 2 ( Figure 6B), while the opposite is true for PPARα ( Figure 6A). The low amount of material isolated (60 µg and 53 µg of oxo-FA1 and oxo-FA2, respectively) prevented further pharmacological characterization of the two compounds. In a follow-up study, we intend to characterize the dose-response curves of these oxo-fatty acids with respect to PPAR α/γ binding and confirm the functional activation of these receptors by studying their PPAR-dependent regulation of relevant, endogenous target genes.
With their dual PPARα/γ specificity, these oxo-fatty acids might make a good start for a structure-activity-relationship (SAR) study to identify PPAR-targeting anti-diabetic drug leads. This could be achieved by comparing our oxo-fatty acids (1 and 2) with other cis-and trans-hexadecenoic acids that do not possess the carbonyl group in their structure. Moreover, the activity profile of the C. karianus oxo-fatty acids, as well as derivatives thereof, should be compared with the profile of other promising dual PPARα/γ agonists, like Saroglitazar [35]. Saroglitazar is marketed in Asia and displays significant improvement in both glycemic as well as dyslipidemic parameters with no evidence of conventional side effects (as reviewed in [25]). From a structural perspective, a recent study by Yore et al. [36] who made use of quantitative MS to identify metabolic changes in transgenic Glut4 mice is highly interesting. Despite being obese, the studied animals showed an increased glucose tolerance. Lipidomic analysis of their adipose tissue revealed the existence of branched fatty acid esters of hydroxy fatty acids, so called FAHFAs. The most abundant form both in mice and human white and brown adipose tissue is 9-PAHSA (palmitic acid-hydroxy stearic acid), having the hydroxyl/ester group placed on the same carbon as our 9-oxo-FAs. Even though the PAHSAs signal through GPR120 to enhance insulin-stimulated glucose uptake, it could be possible that the corresponding unbranded fatty acids (with the hydroxy-/keto group on C9) might have equally beneficial effects with regard to whole body insulin sensitivity when functioning as PPAR agonists.
Having refined the oxo-fatty acids 1 and 2 to approximately 95% and 75% purity, respectively, we wanted to reevaluate the weak PPAR isoform specificity observed during the bioassay-guided isolation. To this end we treated COS-1 cells expressing Gal4-LBD chimeras with 100 μM of 1 and 2. As can be seen in Figure 6, the dual PPARα/γ specificity profile was retained, while the potency was somewhat reduced. Still, oxo-FA 1 seems to activate PPARγ slightly better than oxo-FA 2 ( Figure  6B), while the opposite is true for PPARα ( Figure 6A). The low amount of material isolated (60 μg and 53 μg of oxo-FA1 and oxo-FA2, respectively) prevented further pharmacological characterization of the two compounds. In a follow-up study, we intend to characterize the dose-response curves of these oxo-fatty acids with respect to PPAR α/γ binding and confirm the functional activation of these receptors by studying their PPAR-dependent regulation of relevant, endogenous target genes.  Finally, to address whether the identified metabolite/activity profile is unique to the planktonic diatom Chaetoceros karianus, we employed a metabolomics approach to compare it with different, but related microalgae. Four additional strains of species belonging to the genus Chaetoceros were obtained from the Marbank collection. Extracts and SPE eluates were generated from Cheatoceros decipiens (M09048), Chaetoceros diadema (M09047), Chaetoceros furcellatus (M08023), Chaetoceros karianus (M11031), and Chaetoceros socialis (M08009) as described in the Experimental section. To assess the analogies and differences in the composition of the metabolites of the five species, a non-targeted metabolite fingerprinting of SPE eluates was conducted. ACN/H 2 O-soluble constituents from these species were analyzed by UPLC-HRESIMS as described in the Experimental Section. Results from the UPLC-HRESIMS were further processed by peak picking and alignment for further Principle Components Analyses (PCA). PCA demonstrated that striking differences exist between the metabolic profiles of the five Chaetoceros species (Figure 7). Interestingly, the most distinct profiles were those obtained from C. karianus and C. diadema.
When data obtained by UPLC-HRESIMS in positive ionization mode were subjected to principal component analysis (PCA) (Figure 7A), it appeared that C. furcellatus and C. decipiens are similar with regard to metabolite production. In positive ionization mode, the replicates were perfectly clustered, indicating an appropriate method repeatability. Negative ionization mode data showed similar clustering ( Figure 7B), but in this case C. furcellatus and C. socialis were almost overlapping. Both in positive and negative mode, it appeared that the metabolic profiles of C. furcellatus, C. decipiens, and C. socialis were closely related. On the other hand C. karianus and C. diadema were less related to the other three species and to each other with regard to metabolite production. In negative ionization mode, the replicates were not as well clustered as in positive mode and showed some deviation ( Figure 7B). This somehow lower reproducibility could be attributed to instrumental drift and injection problems. Interestingly, and in line with the different metabolic profiles, the different Chaetoceros extracts displayed different PPAR α/γ agonist activity ( Figure 7C). Even though the relative position of the species was shifted, the C. karianus PPAR α/γ activity was found to be high and at a distance from both C. furcellatus and C. socialis, and C. diadema and C. decipiens which seemed to be more closely related, respectively. Whether this is due to differences in oxo-fatty acid levels or different levels of related or unrelated metabolites is not known and should be addressed in future studies.

Biological Material
For primary screening we used organic Flash fractions from the Marbank collection [37]. Inocula for the mass cultivation of Chaetoceros karianus were from our in-house culture collection. The microalga was isolated from north Norwegian spring bloom stocks in 2009. Mass cultivation was performed at 4.5 °C in a temperature and irradiation-controlled room, where the microalgae were grown in 0.

Biological Material
For primary screening we used organic Flash fractions from the Marbank collection [37]. Inocula for the mass cultivation of Chaetoceros karianus were from our in-house culture collection. The microalga was isolated from north Norwegian spring bloom stocks in 2009. Mass cultivation was performed at 4.5 • C in a temperature and irradiation-controlled room, where the microalgae were grown in 0.3 m 3 transparent plexiglas cylinders with external illumination. At the start of the culture period, mean scalar irradiance across the cultivation cylinder was approximately 125 µmol m −2 ·s −1 and at the time of harvesting it was approximately 35 µmol m −2 ·s −1 due to self-shadowing. The photoperiod was 14:10 (L:D). Illumination was supplied by Osram daylight fluorescent tubes.
The culture medium was Millipore filtered (0.1 µm) 33.5 ppt. seawater growth media with a concentration of inorganic nutrients equivalent to f/10 (Sigma-Aldrich, St. Louis, MO, USA) and a surplus of silicon (50 µmol). The growth rate of the cultures were monitored twice weekly by measuring the in vivo fluorescence and by microscopic inspection of the cells. The biomass concentrations at the start of the experiments were approximately 2 µg·Chla·L −1 and by time of harvested, approximately 350 µg·Chla·L −1 . To monitor biomass development as Chla, a Turner fluorometer (TD-700) was applied. The cells were harvested by sieving the culture though a 10 µm mesh size plankton net, thus letting any (small) amounts of bacteria to be flushed out. After collection, the cells were scraped off the plankton net and stored in 25 mL plastic tubes at −80 • C.

General Extract Preparation
The samples were freeze-dried, ground, and stored at −20 • C before extraction with Milli-Q water (2×, 24 h and 30 min, respectively) at 5 • C (20 mg/mL). The extracted slurries were then centrifuged and the supernatant (the H 2 O extract) was removed. The pellet was then freeze-dried and ground again.
To produce the organic extract, the pellet was extracted with MeOH/CH 2 Cl 2 (1:1) (2× at 5 • C, 24 h and 30 min, respectively) making a 20 mg/mL organic extract. The extract was then filtered through Whatman filters (nr. 3, Ø 125 mm) and the filtrate was evaporated in vacuo and stored at −20 • C.

Specific Extract Preparation for Bioassay-Guided Fractionation
Organic extracts (0.5-2 g) from the library prepared as described above (Section 3.2.1) were reconstituted in ACN/H 2 O (9:1) using an ultrasound bath (30 min) to make a 20 mg/mL suspension. The suspension was then filtered, recovered in a separatory funnel, and defatted with hexane (1:1, 2×). The polar organic extracts were evaporated to dryness and reconstituted in ACN/H 2 O (1:1), before being subjected to vacuum-aided solid phase extraction (SPE) on Strata C18-E cartridges (Phenomenex, Torrance, CA, USA). The eluates were recovered by a 5-fold concentration (100 mg/mL).

Semi-Preparative High-Performance Liquid Chromatography (Semiprep HPLC-PDA)
The SPE-eluates were subjected to semi-preparative reverse-phase chromatography on a Luna C18 column (250 × 10 mm, 5 µm) (Phenomenex, Torrance, CA, USA). A 6 mL/min flow of a mobile phase consisting of 0.1% formic acid (A) and 0.1% formic acid in acetonitrile (B) was employed with gradient elution starting with 5% B, and rising to 100% B within 30 min. Then isocratic elution with 100% B was performed before switching back to the initial conditions. Time-based (60 s) fractions were collected in a Waters 2767 sample manager after detection with a Waters 2998 PDA detector (Waters, Milford, MA, USA). Fractions collected were then tested in our bioassay platform, and those showing significant activity were pooled and subjected to a second round of semiprep HPLC-PDA. In this second step, we applied a gradient elution program starting with 70% A (10 mM ammonium formate and 0.1% formic acid) and 30% B (10 mM ammonium formate and 0.1% formic acid in acetonitrile), rising to 100% B within 30 min. Then isocratic elution with 100% B was performed before switching back to the initial conditions. Compounds 1 and 2 were collected in separate fractions by following the PDA signal trace.

Plasmids
The pSG5-Gal4-hPPARα-LBD, pSG5-Gal4-hPPARδ-LBD, and pSG5-Gal4-hPPARγ-LBD encoding Gal4 DNA-binding domain (DBD; aa 1-147) fused in frame with the SV40 nuclear localization signal 1, and the ligand binding domain (LBD) of human PPARα (aa 168-468), PPARδ (aa 140-441), and PPARγ (aa 205-505), respectively, were received as generous gifts from Dr. Krister Bamberg (AstraZeneca, Mölndal, Sweden). The same was true for the pGL3-5×UAS-SV40 luciferase reporter consisting of five Upstream Activation Sequences (UAS; 25 bp phasing) and a SV40 early promoter in front of the Firefly Luciferase gene. The plasmid encoding full-length human PPARγ, pcDNA3.1-hPPARg2-FLAG, was made by PCR amplification of human PPARG cDNA, using custom-made primers (harboring kozak and FLAG-tag), and subcloning it into pcDNA3.1(+) (Life Technologies, Carlsbad, CA, USA) between NheI and NotI. The corresponding PPARα plasmid, pcDNA3.1-hPPARa-FLAG, was made by PCR amplification of human PPARA cDNA and subcloning it into pcDNA3.1-hPPARg2-FLAG between NheI and BamHI, exchanging PPARG with PPARA, and keeping the FLAG tag. The plasmid encoding full-length human RXRα, pcDNA3-hRXRα, has been described earlier [38]. The human PLIN1-driven reporters, pGL3-hPLIN1-3 del and pGL3-hPLIN1-3 del-PPREmut, have been described before [7]. The human CPT1A-driven reporters, pGL3-hCPT1AInt and pGL3-hCPT1AInt-PPREmut, were received as a gift from Prof. Diego Haro Bautista and have been described previously [39]. The vector pRL-CMV (Promega, Madison, WI, USA), constitutively expressing Renilla Luciferase, was used as a control of transfection efficiency. All cloned plasmids have been sequenced. Cloning primer sequences are available upon request.  High resolution mass spectra (UPLC-QToFMS) were recorded in positive and negative mode, making use of both electrospray (ESI) and atmospheric pressure chemical ionization (APCI) interfaces on a Xevo G2 QToF mass spectrometer (Waters, Milford, MA, USA). The following settings were applied in both ESI negative and ESI positive MS: Capillary voltage 0.5 kV, source and desolvation temperatures of 130 • C and 400 • C, respectively, cone and desolvation gas at 20 L/h and 800 L/h, respectively. The instrument was run in MS e resolution mode with a scan time of 0.1 s, centroid mode. Low energy (no collision energy) and high energy (ramp collision energy from 30 to 50 V) spectra were acquired in the mass range from m/z 100-1200. The instrument was regularly calibrated with sodium formate with leucine enkephaline applied as a lock mass.
Chromatographic analysis was performed using an Aquity 50 × 2.1 mm (i.d.) 1.7 µm UPLC BEH C18 column (Waters, Milford, MA, USA) with a mobile phase consisting of 0.1% formic acid (A) and 0.1% formic acid in acetonitrile (B). Different linear gradient programs were applied for scouting and compound-directed chromatography. The longer scouting program applied in the initial studies on complex mixtures was as follows: 90% A was kept for the first 2 min, then elution from 10 to 100% B over 25 min with a 10 min hold at 100% B before the eluent was switched back to the initial conditions. The shorter program was as follows: 98% A kept constant for 2 min, then elution from 2 to 100% B over 3 min and hold at 100% B for 2 min before switching back to the initial conditions. A mobile phase flow rate of 0.6 mL/min was used, and 2-10 µL samples were injected.

Ultra-Performance Liquid Chromatography Coupled to Photodiode Array Detection (UPLC-PDA)
The chromatographic features applied in the UPLC-PDA analysis were similar to those described above (Section 3.3.1), including column type, mobile phase, and elution properties. The UV spectra were recorded with an Acquity PDA detector (Waters, Milford, MA, USA) between 210 and 600 nm. The wavelength resolution was 1.2 nm.

Metabolomics
Diluted SPE-eluates (1 µg/mL) of the species C. decipiens, C. diadema, C. furcellatus, C. karianus, and C. socialis were analyzed following the protocol described above (Section 3.3.1) in UPLC-MS e in both ESI positive and negative mode. The chromatograms obtained from the UPLC-MS e analyses were processed by the MarkerLynx 1.3 application manager (Waters, Milford, MA, USA) for mass signal extraction and alignment. MarkerLynx makes use of ApexTrack for peak detection and integration, listing the detected peaks as their m/z values and retention times along with their associated intensities. When the data are collected, they are automatically delivered to the Extended Statistics software integrated in MarkerLynx and multivariate statistics are applied for unsupervised principal components analyses (PCA). The parameters applied in the MarkerLynx application manager were as follows: XIC windows, 0.02 Da; marker intensity, 1200; mass windows, 0.02 Da; retention window, 0.1; the Apex Track peak parameters were automatically calculated by MarkerLynx. The isotopes were not included in the peak detection process.

Conclusions
We have identified two isomeric oxo-fatty acids from Chaetoceros karianus with apparent dual PPARα/γ agonist activity. These oxo-fatty acids also show a limited isoform-specificity. (7E)-9-oxohexadec-7-enoic acid mainly activates PPARγ, while (10E)-9-oxohexadec-10-enoic acid primarily activates PPARα. Principal component analyses indicate that C. karianus differs from other Chaetoceros species, both with respect to the metabolic profile and the PPAR activity. This finding holds the potential of defining a PPAR pharmacophore with tunable activity and specificity.