Hydrocarbon phenotyping of algal species using pyrolysis-gas chromatography mass spectrometry
© Barupal et al. 2010
Received: 23 October 2009
Accepted: 21 May 2010
Published: 21 May 2010
Skip to main content
© Barupal et al. 2010
Received: 23 October 2009
Accepted: 21 May 2010
Published: 21 May 2010
Biofuels derived from algae biomass and algae lipids might reduce dependence on fossil fuels. Existing analytical techniques need to facilitate rapid characterization of algal species by phenotyping hydrocarbon-related constituents.
In this study, we compared the hydrocarbon rich algae Botryococcus braunii against the photoautotrophic model algae Chlamydomonas reinhardtii using pyrolysis-gas chromatography quadrupole mass spectrometry (pyGC-MS). Sequences of up to 48 dried samples can be analyzed using pyGC-MS in an automated manner without any sample preparation. Chromatograms of 30-min run times are sufficient to profile pyrolysis products from C8 to C40 carbon chain length. The freely available software tools AMDIS and SpectConnect enables straightforward data processing. In Botryococcus samples, we identified fatty acids, vitamins, sterols and fatty acid esters and several long chain hydrocarbons. The algae species C. reinhardtii, B. braunii race A and B. braunii race B were readily discriminated using their hydrocarbon phenotypes. Substructure annotation and spectral clustering yielded network graphs of similar components for visual overviews of abundant and minor constituents.
Pyrolysis-GC-MS facilitates large scale screening of hydrocarbon phenotypes for comparisons of strain differences in algae or impact of altered growth and nutrient conditions.
The world requires a sustainable source of energy for the future. Autotrophic organisms have been proposed to reduce the energy dependence of world economy on the fossil oil . Specifically, biofuel derived from microalgae  have been under active investigation. Hypothetical yield per hectare, land requirements, eco-friendly production, simple life structure and available scientific technologies are the major advantages to use the microalgae for large-scale biofuel production . The hydrocarbon content of algae, specifically fatty acids, isoprenoids and triacylglycerides , have the potential to compensate for future decline of crude oil production  if algae growth and harvest can be sustained under economically and energetically feasible parameters. Genetic and environmental factors affect the lipid constituents of microalgae [2, 4] as well as algae biomass growth. Depending on the species-strains and conditions, lipids can constituent up to 80% of algal dry mass . For biotechnological applications of algal biofuels , existing analytical and computational tools are required to rapidly screen and characterize the strains and environmental conditions. Classic methods such as analysis of total triglycerides use a transmethylation procedure to shave off fatty acids  but are not be able to screen for alkanes or uncommon hydrocarbons found in Botryococcus. Similarly, current metabolomics methods analyze free fatty acids by trimethylsilylation and GC-TOF mass spectrometry  which is not amenable to total fats and pose difficulties to extend towards volatile aliphatics, long chain hydrocarbon and complex lipids in a single analytical method. We here aim at providing a rapid and automated way to assess total hydrocarbon and lipid contents in algae for fast strain discriminations that still can be visualized for the underlying changes in chemical complexity.
Amid the algal species relevant for biofuel research are different strains of Scenedesmus obliquus, Dunaliella salina, Botryococcus braunii and Chlamydomonas reinhardtii. One requirement for the use of special strains in biofuel production is that the microalgae produce lipids under normal and stress conditions. Although the triglyceride content of C. reinhardtii is very low it is used as a common model organism to study metabolism and metabolic networks under different nutrient and light conditions . Hydrocarbon contents of algal species have been characterized using a variety of analytical tools and procedures. Among the traditional techniques for lipid analysis from algae are gravimetric and spectrophotometric techniques (Nile Red staining) . Gradient centrifugation experiments can be used for rapid analysis and quick comparison of the lipid content of algal species producing high hydrocarbon . More selective and sensitive technologies such as gas chromatography (GC) and liquid chromatography (LC) coupled to mass spectrometric detectors  provide quantitative and extensive qualitative data of biofuel constituents of microalgae . Electrospray ionization and atmospheric pressure chemical ionization mass spectrometry  and comprehensive two-dimensional liquid chromatography  (LCxLC) were used for the analysis of triacylglycerides. Analysis of free fatty acids, sterols and waxes is usually performed with gas chromatographic approaches using flame ionization and mass selective detectors . Plant specific phospholipids and galactolipids can be quantitatively profiled using electrospray ionization tandem mass spectrometry . In addition to these compound-specific techniques, pyrolysis-gas chromatography (pyGC-MS) has been used in combination with pattern recognition for fingerprinting of soils, bacteria, lignin and cellulosic analysis as well as analysis of complex organic matter, but much less applied for lipid profiling. High molecular weight lipids in B. braunii were previously characterized using pyGC-MS  with the focus on understanding of pyrolysis fragmentations and pitfalls in pyGC-MS, referring to structures that were identified using classical analytical methods such as nuclear magnetic resonance  and thin layer chromatography . However, no comprehensive characterization of the overall hydrocarbon patterns was performed using substructure annotations. In addition, these methods had never been shown to be highly useful for rapid and quantitative comparison of multiple species.
We here present a comprehensive approach using existing computational tools and pyrolysis coupled to gas chromatography/mass spectrometry to phenotype the lipid rich B. braunii and the model organism C. reinhardtii based on their hydrocarbon content. To this end we used mass spectral deconvolution algorithms (AMDIS)  which we combined with a result filtering algorithm (SpectConnect)  for comparing multiple data sets. In order to get a deeper insight into the chromatogram content we first queried chemical and natural product databases to obtain already known algal compounds. Additionally we performed a substructure analysis  of deconvoluted mass spectra together with mass spectral library search. An elution order analysis of several substance classes was performed based on their distinct mass spectral patterns. In order to discriminate between hydrocarbon rich species and hydrocarbon low abundant species we applied unsupervised multivariate statistics and furthermore visualized hydrocarbon abundances with the Cytoscape visualization software. Results can be directly interpreted with respect to the difference in hydrocarbon type and abundance between different algae species that may be important to evaluate strains and hydrocarbon outputs in biofuel reactors.
GC-MS analyses yield complex three-dimensional raw data sets (time x mass x intensity) which need to be deconvoluted as many fragment ions may be shared between two chromatographically co-eluting compounds. To extract and purify unique mass spectra from complex mixtures, noise analyses and baseline drift corrections are necessary for each ion trace, with subsequent peak picking and mathematical deconvolution using the most unique model ion elution curves . All pyrograms were processed using the freely available AMDIS software  using optimized parameters as given in the method section, yielding up to 512 deconvoluted components in B. braunii. However, AMDIS is known to evolve a high number of false positive or false negative peak detections, depending on the parameter settings. Therefore, different numbers and identities of components may be detected for biological replicates of the same species. Moreover, when different species or environmental conditions are compared, huge differences in peak numbers and compound identities can be expected. Dot-product similarity alignment algorithms [28–31] will fail in such cases. Instead, we have directly uploaded the *.ELU files resulting from AMDIS processing to the SpectConnect  web-tool to find the complement of peaks that are consistently detected in multiple chromatograms of one of the study design classes (e.g. per species). Based on the spectral similarity and retention-time shift corrections, SpectConnect filtered out all peaks that were not detected or below a selected threshold signal, generating three matrices (for relative amount, amount and integrated signals) which were later used for multivariate statistics.
Annotation of pyGC-MS components using AMDIS substructure classifiers.
B. braunii race A
B. braunii race B
AMDIS deconvoluted spectra were subsequently used for compound annotation by searching against the NIST05 mass spectral library at a 70% similarity threshold. Biofuel-related pyrolysis products such as fatty acids and their methyl esters, isoprenoids, sterols, vitamins, aromatics and branched and unbranched alkane-type compounds were identified as intact molecules up to 600°C thermal pyrolysis temperatures (Additional file 2, table S1). Hence, it is likely that components detected by pyGC/MS consist of intact compounds as well as fragments of thermally degraded larger biomolecules. Increasing lengths of side chains were distinguished by an increase of 14 mass units reflecting CH2-units. Further peaks were identified using the apparent molecular weights and querying the Dictionary of Natural Products database (DNP). DNP queries retrieved 85 entries for Botryococcus DNP of which 17 compounds were matched to pyrograms by molecular weight and mass fragmentation pattern (Additional file 3, table S2). Botryococcus is well known for its high content of long chain hydrocarbon contents. Up to 70% of the dry weight are long chain hydrocarbons . Botryococcenes are important hydrocarbons in this alga for which we identified lycopadiene (C40) and botryococcene (C32) in pyGC-MS data based on the molecular ions, mass spectral patterns and retention times (Additional file 3, table S2). The biomarker compound lycopadiene  is specifically only found in B. braunii race A. DNP comprised only five Chlamydomonas metabolites which did not yield additional pyrogram compound annotations.
We have established that pyGC-MS offers a fast track to phenotype algae strains in a cost effective manner, providing both qualitatively and quantitatively important information. Preparation for analysis was minimal and could potentially further robotized, unlike methods that are classically used in lipid analysis or metabolomics. Specifically, constraining AMDIS deconvolution results was important to yield clean and consistent data sets that could be used to compare and contrast hydrocarbon contents. For the first time we have used substructure annotation algorithms, mass spectral library matching and chemical database search to automatically assess hundreds of pyrolysis peaks in algae research. Visualization of result data sets in network graphs offers an improved tool to highlight differences between algae cultures that can much easier be interpreted than commonly used table representations or box-whisker graphs. We suggest that the integrated approach presented here is an efficient strategy for hydrocarbon phenotyping of microalgae in a rapid and automated manner. This method can be applied to small scale as well as large scale research projects considering screening and physiological studies on microalgae for biofuel applications.
Algal samples were harvested and quenched similar to two previously published reports [9, 36]. Three different algal strains, B. braunii UTEX LB-572, B. braunii UTEX-572 and C. reinhardtii strain CC125 were used in this study. Botryococcus and Chlamydomonas strains were cultivated in modified CHU-13 [37, 38] and Tris acetate phosphate (TAP) medium respectively at 23°C under constant illumination with cool-white fluorescent bulbs at a fluence rate of 70 μmol m-2 s-1 and with continuous shaking. At the incubation site, 1 mL cell suspensions were injected into 1 mL of -70°C cold quenching solution composed of 70% methanol in water using a thermo block above dry ice. Centrifuge tubes containing the solution during harvest were cooled in a pre-chilled cooling box to keep sample temperature below -20°C. Cells were collected after centrifugation with a rotational speed of 13,200 rpm for 2 min with the centrifuge and rotor cooled at -20°C. Supernatant was decanted and residual liquid carefully removed. The pellet was flash frozen in liquid nitrogen and lyophilized at -50°C in a 2 mL round bottom Eppendorf tube.
Lyophilized samples of B. braunii and Chlamydomonas were immersed in 2 ml Eppendorf tube with 20 μl of isopropanol. 15 μl of sample were dispensed in pyrolyzer cup and placed on the pyrolysis auto sampler. Pyrolysis was performed with a single-shot pyrolyzer (Model 2020, Frontier Laboratories) directly connected to an Agilent 6890 GC/MS system equipped with a fused silica capillary column (DB5-HT 30 m length, 250 μm inner diameter, 0.25 μm film thickness). An Agilent 5973 quadrupole mass spectrometer was used as detector (EI + at 70 eV). The analysis was performed at a pyrolysis temperature of 600°C for 10 seconds. The pyrolyzer-GC injector interface temperature was set to 320°C. The GC-MS conditions were as follows: oven temperature was held at 50°C for 1 min and then increased up to 325°C at 10°C min-1 and hold at 325°C for 2 min. The GC-MS interface temperature was set to 250°C. Helium was used as a carrier gas with a constant flow of 1.2 ml min-1. Split ration was 1:40. The selected mass range was 35-600 Da and the selected scan speed was 2 scans per second. Select pyrograms can be downloaded from the supplement section.
The data from GC-MS analysis were deconvoluted in batch mode using the freely available Automated Mass Spectral Deconvolution and Identification System (AMDIS) spectral deconvolution software package (v2.65, NIST Gaithersburg) . AMDIS deconvolution settings were as follow: resolution was medium, sensitivity was low, shape requirement was medium and component width was kept at 10. The SpectConnect online service  was used to cross-reference multiple chromatograms for filtering out inconsistent signals . The AMDIS substructure algorithm from the post-processing analysis task was utilized to annotate deconvoluted components with putative substructures. Selected ion chromatograms (SIC) were extracted for 55 m/z using AMDIS. Each component of a given Botryococcus sample was searched against the NIST electron impact mass spectral library (NIST05) for compound annotations. AMDIS generated *.FIN files were parsed using Textpad to extract the NIST MS library hits, substructure classifiers, amount, relative abundances and retention time. The following AMDIS substructure classifiers were summarized as hydrocarbon-related compounds: n-C4H9-; n-C4H9-O; n-C4H9-O; C4 H9; C5 H11; C6 H13; C6 H13 n; C7 H15; C8 H17; C9 H19; C10 H21; C11; C11 H23; C11 H23; (CH3)3-C; (CH2); (CH2)6; (CH2)6-; (CH2)6-C; (CH2)6-CO; (CH2)6-CO; (CH3)2 > C = C; (CH3)3-C.
The relationships between components and substructure classifiers were converted into Cytoscape SIF networks (see supplement data). The network was visualized in organic layout using Cytoscape version 2.6. For visualization the node size was adjusted according to the relative abundance of components. Dictionary of Natural Products (DNP 17.1 Copyright 2008 Taylor & Francis Group)  was queried for Botryococcus and the 85 retrieved compounds were searched in the pyrolysis GC-MS data using the molecular weight as particular m/z to extract the SIC. SpectConnect generated matrices were normalized to the total signal and utilized for principal components analysis in Statistica 8.0 (StatSoft, Tulsa, USA).
DKB was funded through NSF grant MCB-0820823. Substructure annotation and compound identification was funded for TK and OF through NIH grant R01 ES013932.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.