Metabolic ¹⁵N-labeling introduces the stable isotope in each nitrogen atom in the labeled organism. Therefore, every observable peptide from mixtures of unlabeled (¹⁴N) and ¹⁵N-labeled samples should occur in pairs and thus be suitable for quantification. In a first step, labeled protein extracts were analyzed for full incorporation of ¹⁵N into all proteins after 14 days of continuous growth in presence of K¹⁵NO₃. Comparison of protein identifications in database searches with and without ¹⁵N as a parameter for fixed modification (Tab. 1) showed that 98.3% of all proteins of the labeled cells were identified only in their labeled form, while for 2% of the proteins, a residual minor peak was observed also for the ¹⁴N form. This indicates that ¹⁵N incorporation into the proteome of plant cell cultures is as complete as it can be achieved with 98 atom% enriched KNO₃. Most importantly, we observed no significant protein to protein variation for the incorporation of the label.

In the following experiments, protein extracts of unlabeled (¹⁴N) cell cultures were mixed with extracts of ¹⁵N-labeled cell cultures at known mixing ratios of 1:1, 1:4 and 4:1. Quantitative analysis of the intensity ratios of labeled and unlabeled peptide pairs indeed corresponded to the mixing ratios (Fig. 1). Taking the doubly-charged peptide ALGVDTVPVLVGPVSYLLLSK as an example, the quantitation revealed that in unlabeled cell extracts no pairing peak was found at the expected position of the labeled peptide (Fig. 1A). In contrast, the ¹⁵N-labeled extract contained no visible peak at the expected position of the ¹⁴N-form of the respective peptide (Fig. 1B). In the 1:1 mixture of the two protein extracts, labeled and unlabeled form of the peptide were clearly identified and quantified with a ratio of 0.92 (Fig. 1C), in the 1:4 mixture the intensity ratio of labeled and unlabeled form was 2.5 (Fig. 1D), while in the 4:1 mixture the ratio was found to be 0.18 (Fig. 1E). In average in all of the above protein extract mixtures, 85% of all identified proteins were quantified, and the average intensity ratios of all quantified proteins were not significantly different (χ²-test) from the expected values (Tab. 2). However, it has to be kept in mind, that ratio quantitation can become limited by the signal-to-noise ratio of the mass spectrometer 26 especially if there are large differences between the two samples. In those cases, it will not be possible to accurately reproduce the precise mixing ratios in the measured ratios of the peak pairs.

There are two main reasons why proteomic quantification using full ¹⁵N-labeling is not as straight forward for automated high-throughput analysis. Firstly, the pairs of tryptic peptides are separated by a mass difference that depends on the number of ¹⁵N atoms in the respective peptide sequence. Therefore, the peptide sequence has to be determined by a database search before the mass difference to the labeled or unlabeled partner can be calculated, and this mass difference is different for each unique peptide (Tab. 3). This is actually the most critical disadvantage of the ¹⁵N-labeling method as it makes automated peak pair recognition impossible during data-dependent acquisition of mass spectra or in automated analysis of the raw files. Secondly, the isotopic distribution of the mass spectrum of a ¹⁵N-labeled peptide is slightly altered. This is so because the labeling substance, in this case KNO₃, was used only in an isotopic enrichment of 98 atom% ¹⁵N. As the number of nitrogen atoms as a proportion of the total number of atoms in a peptide is rather high, the isotopic distribution of the peptide ion is influenced by the degree of enrichment of ¹⁵N. Similarly to previous work using full ¹⁵N- labeling as a method for quantification in proteomics 2444, we do not account for this effect. However, other research groups have developed algorithms that correct for isotope enrichment effects 46 or make use of it for quantitative analysis 47. Despite some drawbacks in the automated analysis, full ¹⁵N-labeling is a cheaper alternative to metabolic labeling with SILAC (Tab. 5). With the concentrations of amino acids used for SILAC labeling of plant cell cultures 35 and the concentrations of nitrate used in the labeling medium in this study, the K¹⁵NO₃ labeling method is up to 10-times less expensive than SILAC labeling.

For metabolites, quantitation was based on the ion intensities of characteristic fragment ions 39. In total, among the nitrogen-containing compounds, 14 amino acids (ala, asp, cys, glu, gly, ile, lys, met, phe, pro, thr, trp, tyr, val) and two polyamines (spermidine, putrescine) were analyzed. We estimated an average of 91% incorporation of ¹⁵N-label into metabolites, ranging from 70% for alanine to the maximally achievable 98% for tryptophan and putrescine.

Mass isotopomer analysis of the saturated ¹⁵N-labeled cells indicated the expected portion, namely approximately 2% of residual ¹⁴N in all observed ¹⁵N-containing metabolite preparations (Fig. 2A–D, ¹⁵N samples). The observed mixing ratios of the unlabeled and labeled metabolite extracts show the expected tendencies. For example, ¹⁵N/¹⁴N ratios for glycine were found to be 1.0, 4.3 and 0.3 in 1:1, 1:4, and 4:1 mixtures (Fig. 2A). However, all 4:1 ratios were over-estimated without correction for the contribution of natural abundances of ¹³C (Fig. 2A–D, higher isotope peak in ¹⁴N-extracts). Natural abundance of nitrogen and carbon isotopes as well as discrimination effects during metabolism may account for these observations and the differences between metabolites 4849. In contrast to peptide analysis, we therefore conclude that when analyzing mixing ratios of compounds with one or few N-atoms the natural isotope abundance of ¹³C (1.10%) needs to be considered, whereas the contribution of natural abundance of ¹⁵N (0.37%) may be neglected (Fig. 2 A–D). In addition two further sources of errors were tested: (1) The two original cell cultures, labeled and unlabeled may have different metabolite pools and (2) equal combination of labeled and non-labeled samples might be inaccurate. Exemplary metabolites which do not contain nitrogen, demonstrate slight non-significant changes in pool size and negligible errors in the 1:1, 1:4, and 4:1 mixtures (Fig. 2 E–F). While glycine, inositol or fumarate levels were unchanged in the two original cell cultures (Fig. 2.A, E–F), threonine (Fig. 2B), glutamate (Fig. 2C), and to a minor extent spermidine (Fig. 2D) had accumulated in the ¹⁵N-labeled cells. Although the two cell cultures were grown in the same medium composition and under the exact same growth conditions, changes in metabolite contents may have resulted from slight differences in handling during harvesting (e.g. intensity of washing, speed of freezing). As a consequence, we propose that those pool-size changes in the originating cells need to be experimentally considered for the purpose profiling experiments of the N-containing metabolic complement.

Using the unmixed extracts (¹⁴N-extract and ¹⁵N-extract, Fig. 2) to estimate the contribution of natural abundance of ¹³C to higher isotope fragmentation peaks, we calculated the expected ratios of ¹⁵N/¹⁴N for all metabolites in the 1:1 mixture (Tab. 4). The calculation also takes into account that pool sizes (ion intensities) for certain metabolites were significantly different in the unmixed original cells. It becomes apparent that in most cases the measured ratios in the 1:1 mixture are rather accurate (deviation < 10%), and only for those metabolites, which display large differences in concentration in the original unmixed cells, measured ratios were over- or underestimated. Alanine was found in higher concentrations in unlabeled cells and therefore the expected mixing ratio was strongly underestimated due to signal-to-noise limitation of the ¹⁵N/¹⁴N ratios. The contrary holds true for methionine and putrescine which show higher concentrations in the ¹⁵N-labeled cells leading to a strong over-estimation of expected ratios. These examples underline the importance of including the unmixed extracts in the analysis. It has been described also for quantitative protein analysis that large ratios lead to larger relative errors, basically because the ratio is then defined by the signal to noise ratio of the mass spectrometer 26.

The examples of different metabolite pool sizes (ala, met, glu, putrescine) in the ¹⁴N-cells compared to the ¹⁵N-cells actually reflect the true experimental situation, in which two cell extracts of unknown concentrations would be mixed in a 1:1 ratio, and those compound differing significantly from a 1:1 ratio of the ion intensities are concluded to be of different pool size in the original cells. The same experimental strategy generally also applies to proteins, and has already been successfully been used to detect proteins specifically interacting with a bait protein or peptide from background interaction partners or to describe temporal changes in protein abundance 293032. However, in the experiments described here, no variation between proteins was observed, since protein extracts were mixed based on total protein content, while for metabolite analysis, frozen cell material was mixed on a fresh weight basis (see Methods for details). Thus, the labeling and mixing strategy described here is well suitable for accurate detection of small changes and also ideal for screening for large deviations from an expected ratio, e.g. in presence/absence studies.

A drawback of the ¹⁵N-labeling strategy for metabolomic analyses can be seen in the fact that quantitative information can only be obtained for nitrogen-containing metabolites. Thus, for quantitative information of a full metabolite profile, metabolic labeling with ¹³C would be a more promising strategy, as has already been proposed in an experiment using yeast cells 39. However, for protein analysis full ¹³C-labeling is not suitable: The high proportion of carbon atoms in tryptic peptides would lead to very large mass differences between labeled and unlabeled peptides making unambiguous detection of labeled and unlabeled ion pairs very difficult.

Arabidopsis cell cultures derived from leaves 50 were grown in JPL medium 51 with 19 mM potassium nitrate as the sole nitrogen source. Potassium nitrate was supplied either in normal form or in ¹⁵N-enriched form (98 atom% K¹⁵NO₃, Sigma-Aldrich). Cells were subcultured every 7 days using 10% of the culture volume. The remaining cells were harvested by suction over a filter plate and frozen at -80°C.

Frozen cells were ground and extracted in 50 mM TRIS-HCl pH 7.5, 1% NP-40, and a protease inhibitor mixture (Complete Tabs, Roche). Cell debris was pelleted by centrifugation and 200 μg of Protein was precipitated with TCA. Protein was resuspended in a small volume (approx. 20μL) of 6 M Urea/2 M thiourea pH8 and in-solution digested with trypsin after reduction in DTT, alkylation with iodoacetamide and dilution of the sample with 4 volumes of 50 mM NH₄HCO₃ was performed as described 30.

Tryptic peptide mixtures were then desalted on STAGE tips 52 before LC-MS/MS analysis using nanoflow HPLC (Proxeon Biosystems, Odense, Denmark) and a linear ion trap instrument (LTQ, Thermo Electron, San Jose, CA, USA) as mass analyzer. Peptides were eluted from the analytical column (Reprosil C18, Dr. Maisch GmbH, Tübingen, Germany) by a linear gradient running from 10 % to 50 % acetonitrile in 110 minutes and sprayed directly into the LTQ mass spectrometer. Proteins were identified by tandem mass spectrometry (MS/MS) by information-dependent acquisition of fragmentation spectra of multiple-charged peptides. Additional data-dependent fragmentation (MS3) was used to confirm identity of ambiguous proteins 53. Fragment spectra were then searched against a non-redundant version of the Arabidopsis protein Database (TIGR6) using the Mascot algorithm (Matrix Science, UK, http://www.matrixscience.com). The following search parameters were applied: Peptide mass tolerance 800 ppm, MS/MS tolerance 0.8 Da, methionine oxidation and carbamidomethylation of cysteine were set as variable modifications. For each sample, two database searches were carried out, one without fixed modifications ('14N-search') and one with ¹⁵N-amino acids as a fixed modification ('15N-search').

Quantitative Analysis was done using the open-source software MSQuant (http://msquant.sourceforge.net) which provides a validation and quantitation platform for protein mass spectrometry. The MSQuant software has previously been successfully applied to quantitative analysis involving SILAC labeling 3132. For quantitation involving full ¹⁵N-labeling, ratios between the centroids in the ion chromatograms of the eluting 'heavy' (i.e. ¹⁵N-labeled) and 'light' (i.e. unlabeled) tryptic peptide peaks were calculated and averaged over the duration of the respective peaks in the total ion chromatogram 26. Labeled and unlabeled forms of peptides were found to co-elute (Fig. 3) and the quantification was based on the average of independently determined ratios for each peptide. Ratios obtained from different peptides identifying the same protein were averaged. A final standard deviation for protein ratios was calculated from the ratios of the individual peptides or, when the identification was based on a single peptide, from the ratios obtained from the different mass spectra of this peptide pair.

The labeled or unlabeled counterparts to peptides identified in the database searches were detected automatically in MSQuant based on their amino acid sequence and mass to charge ratio (Fig. 4). For example, if a peptide was identified in the ¹⁴N-search, the mass to charge ratio of the corresponding peak in the ¹⁵N-labeled form was calculated from the expected mass shift based on the amino acid sequence of the ¹⁴N-isoform of the identified peptide and its charge state.

Metabolites were extracted from 120 mg cell material, which was shock frozen in liquid nitrogen and ground to a fine powder. Alternately different ratios of labeled and non labeled powder were combined to a final amount of 120 mg and subsequently treated as a single sample. Care was taken to keep samples frozen during manipulation. Frozen material was extracted with 360 μL methanol. Extraction temperature was 70°C. 200 μL chloroform and 400 μL water were added for partitioning polar metabolites as described 54. Polar extracts were dried in a vacuum concentrator over night. Derivatization was conducted with 10 μL 20 mg mL^-1 methoxyamine hydrochloride/pyridine, and 17.5 μL 20 mg mL^-1 N-methyl-N-(trimethylsilyl)-trifluoroacetamide. A mixture of alkanes was added for calculation of retention time indices 55.

GC-MS based metabolite profiling performed according to the procedure described previously 56. Briefly, 1μL of sample was injected at 230°C into a GC 6890 (Agilent Technologies, Palo Alto, CA, USA) gas chromatograph mounted with a Rtx-5Sil MS capillary column, 30 m × 0.25 mm inner diameter with 0.25 μm film thickness, and a 10 m integrated guard column (Restek GmbH, Bad Homburg, Germany). The temperature program comprised 1 min at 70°C, a 9°C min^-1 ramp to 350°C, and 5 min at 350°C. The transfer line to the mass-spectrometer was set to 250°C. Carrier gas was helium at a flow rate of 0.6 mL min^-1 and operated at constant flow. A time-of-flight mass-spectrometer was used, Pegasus III TOF-MS system (Leco, St. Joseph, MI, USA) with the electron impact ionization source set to -70 eV and 250°C. Data acquisition was set to 20 spectra s^-1 and the mass range restricted to m/z = 70–600. All other settings of the mass spectrometer were according to manufacturer's instructions. Chromatograms were acquired, deconvoluted and processed with ChromaTOF™ software (LECO, St. Joseph, MI, USA). Metabolites were identified using the GMD library (Golm Metabolome Database, http://csbdb.mpimp-golm.mpg.de/csbdb/gmd/gmd.html; 5758). Mass spectral and RI comparison was performed by NIST02 (National Institute of Standards and Technology, Gaithersburg, MD, USA) mass spectral search and comparison software accepting mass spectral matches >650 on a scale of 1000 and retention time index matches ± 5.0 RI units. Mass spectra and fragmentation pattern of amino acid derivatives were interpreted as described 59 in order to identify ¹⁵N-labeled forms of the respective amino acids.

Absolute ion currents and ratios in ion intensities of metabolites were directly derived from mass spectral fragmentation pattern exported from the ChromaTOF™ software. If available, for each metabolite, more than one fragment ion was used for quantitation 39. For each sample type (original cell cultures, and the three mixtures), five independent samples were analyzed and the ion intensities for each fragment ion were averaged across the five replica samples.