Amino acid standards were supplied by Merck. The ion pairing agent, heptafluorobutyric acid (HFBA) and HPLC grade acetonitrile were purchased from Sigma-Aldrich. Water was deionized on a Milli-Q system (Milli- Q reagent water system, Molsheim, France) at 18.2 Mohm. All glassware used was oven combusted at 400˚C for 4 h.

The consideration of thermodynamic [3] [7] , experimental e.g. [41] -[46] and field data e.g. [13] [14] [17] [47] [48] together with analytical issues (separation, response, retention…) guided the choice of the 10 proteinogenic amino acids that were investigated in the present study. A stock solution comprising of Glutamic acid (Glu), Alanine (Ala), Methionine (Met), Tryptophan (Trp), Proline (Pro), Glycine (Gly), Lysine (Lys), Tyrosine (Tyr), Phenylalanine (Phe) and Leucine (Leu) in Milli-Q water at a concentration of ca. 20 mg∙L⁻¹ was prepared. The quantity of each amino acid was weighted on an accurate scale and exact concentrations were determined and are given in Table 1. Diluted solutions at 1000, 200, 20 and 2 µg∙L⁻¹ (i.e. ppb) were prepared from this stock solution.

Hydrothermal fluids were sampled in titanium syringes from the Logatchev and the Ashadze hydrothermal field by the ROV Victor 6000 during the SERPENTINE cruise in winter 2007 and from the Rainbow hydrothermal field by the manned submarine Nautile during MOMARDREAM-naut cruise in summer 2007. Both cruises were conducted by the French research institute for marine sciences, Ifremer. Samples were taken at the nose of smokers to avoid seawater mixing. The accuracy of the sampling was guided by real time records of the tem- perature within the chimneys. However, mixing occurred anyway to some extent. Based on the Mg concentra- tion, the percentage of pure fluid can be calculated. Table 2 lists the samples used in this study and gives their general characteristics and the proportion of pure fluid. As soon as the fluids were recovered, 10 mL aliquots of the samples were transferred into glass tubes, frozen immediately at −80˚C and stored until analysis. Samples were not filtered to limit the risk of contamination and because of the possibility of cell lysis during pres- sure-filtration. 5 mL of each hydrothermal fluid sample were evaporated at room temperature under a gentle N₂ flux and re-extracted in 100 μL of Milli-Q water. The extracts were then transferred into V-shaped vials and let

sit until salts and precipitates would sediment. Finally, the supernatant was sampled with a glass Pasteur pipette, poured into an additional V-shaped vial and analysed. To be noted, only the DFAA will be detected with the current sample preparation approach.

Amino acids were characterised by UPLC-ESI-QTOF-MS [49] . The ACQUITY UPLC^® instrument (Waters, Milford, USA) coupled to a Micromass quadrupole time of flight (QTOF Premier) high resolution mass spectrometer (Waters, Manchester, UK). Compounds were separated on a BEH C₁₈ 1.7 µm, 2.1 × 50 mm column (Waters) at a flow rate of 250 µL/min. Mobile phase A was a solution of 10mM HFBA in (95/5) (v/v) (water/ acetonitrile) and mobile phase B was a solution of 10 mM HFBA in (5/95) (v/v) (water/acetonitrile). A gradient elution comprising of 7 steps was performed as follows: the mobile phases ratio (A:B) was (100:0) for 2 min; linear ramp to (80:20) in 0.5 min and held for 3 min; linear ramp to (0:100) in 2 min and held for 3 min; linear ramp back to (100:0) in 0.5 min and held for 5 min; which makes up the total run time to 16 min. The mass spectrometer was operated in the positive ion mode and the MS parameters were set as follows: Capillary voltage 3.0 kV, Sampling cone 25 V, extraction cone 4 V, source temperature 100˚C, desolvation temperature 350˚C and collision energy 2.0. Automated injections were performed by the ACQUITY UPLC^® Sample Manager. The protonated molecule [M+H]⁺, of which the values of the mass to charge ratio (m/z) are given in Table 1, was used as the selected ion for compound identification in a mass window of 0.05 Da. The selected amino acids were quantified using the manual integration feature of the MassLynxV4.1 software (Waters, Milford, MA, USA).

Amino acids can be positively or negatively charged which allows the use of either a positive or a negative mode for MS detection. In the present study, ion-pairing agent HFBA was used as an additive in the mobile phase to enhance the retention of amino acids. HFBA is a strong acid, with a pKa value < 2 which ensures all the amino acids to be protonated and therefore easily detected in positive mode. Also, the acidic HFBA will compete for charges with the amino acids in the ion-source and contribute to the overall ion-suppression in negative mode. In addition, amino acids are detected with higher sensitivities in the positive mode than in the negative mode [50] . The positive mode detection appeared thus most suitable. Finally, characterisation and retention of amino acids was confirmed by injection of 1000 ppb standard solutions of each amino acid separately.

Calibration was designed to investigate linearity at lowest possible concentrations as amino acids were suspected to occur at levels < 1 nM in natural samples [12] [14] -[16] . However, Gly was detectable only at rela- tively high concentration compared to the other amino acids and therefore two stronger concentrations were also injected for calibration which consists of the following injections: 2, 4 and 10 μL of the 2 ppb solution; 2, 4, 8 and 10 μL of the 20 ppb solution; 10 μL of the 200 and 1000 ppb solutions. The 1000 ppb solution was also used as an external standard and was run on a regular basis over a day of analyses for control of retention times and peak intensity. Concentrations of the standards were normalised to a 10 μL injected volume, so that linear ranges and LOD correspond to an injection of 10 μL. The linear ranges reported here have been restrained to the domain of concentrations of interest and therefore the upper limit of linearity should not be considered as the highest one. LOD were calculated and correspond to a signal to noise ratio (S/N) of 3.

Several gradients, chromatographic conditions and ion-pairing agents were investigated before the optimum compromise was reached. Under these conditions (see Materials and Methods) the 10 amino acids investigated in the present work were easily detected and retained (see retention time [Rt] in Table 1). Separation was achieved in less than 4.5 min with very nice peak shapes (Figure 1); being about 3 times faster than with HPLC which is a common feature [35] [51] -[53] . This is also faster than other work using UPLC [33] [39] [40] .

All amino acids showed linear responses within 1 to 1.5 orders of magnitude with correlation coefficients (r²) higher than > 0.9921 (Table 3). Glu, Met, Trp, Pro, Phe and Leu responses were linear from the sub-ppb to 10 - 20 ppb whereas Ala, Lys and Tyr showed a narrower range of linearity of 2 to 20 ppb (Table 3). The response for Gly was also linear although in the much higher concentration range of 19.5 - 975 ppb which can be accounted for the little retention of the molecule which in turn may be due to the heavy charge and low molecular weight of Gly. LOD was down to the sub-ppb for most of the amino acids, Leu could even be detected at a concentration as low as 25 ppt. Ala and Lys could not be evidenced below the ppb and Gly was detected only from 20 ppb (Table 3). Nevertheless, all these LOD are 1 to 4 orders of magnitude lower than the one reported to date with both similar and different analytical techniques measuring underivatized amino acids e.g. [26] [28] [29] [32] [34] [35] [39] . As a conclusion the method meets the expectations and the requirements in terms of time efficiency, linearity and low LOD.

Reproducibility was found excellent and checked by doing three consecutive injections of the 1000 ppb solu- tion. This resulted in variation of less than 2% in Rt. Over a day of analyses, we observed a slight drift in the re-

tention of the compounds that appeared less retained, however variations in Rt were still good (<13%). In addition, Rt were highly reproducible from day to day and even month to month. Indeed, twice two injection of the 1000 ppb standard were done with one month interval and the obtained variation coefficients were less than 2%.

Four high temperature (350˚C - 360˚C), ultramafic-hosted hydrothermal systems located on the Mid-Atlantic Ridge were sampled: Rainbow (36˚13N, 2300 m), Ashadze 1 (12˚58N, 4100 m) and Logatchev 1 and 2 (14˚45N, 2700 - 3000 m). Details on geological settings and fluids inorganic geochemistry can be read in Charlou et al. [[10] and references therein]. All these sites are hosted on serpentinised peridotites with intrusive gabbroids and massive sulphides deposits are present extensively. Their fluids show the characteristic signature of hot fluids interacting with ultramafic rocks, i.e. high concentration in H₂ and CH₄, low silica contents, enriched in Li compared to basalt-hosted systems and depleted in B compare to surrounding seawater. Of particular interest are the highly reducing conditions observed due to the high amounts of H₂ (Rainbow, 13 - 16 mM; Ashadze 18 - 19 mM; Logatchev 1.9 - 12.5 mM; Logatchev 2, 11 mM) [10] . However, each field exhibit a unique geochemistry and therefore every new discovered field or data on existing vents are crucial to better understand the overall hydrothermal biogeochemical processes.

The main characteristics of the hydrothermal fluids samples and the deep seawater reference sample that were analysed are given in Table 2. Deep seawater contained mainly Ala at a concentration of 3.86 nM (A on Figure 2) and Leu was detected. Ala was also reported to be a major component of seawater [54] [55] . Gly could not be detected in any of the samples which we assume is due to its higher LOD. Although Glu, Lys and Tyr had a much lower LOD, they were not detected either in hydrothermal fluids, which means that they still might be present but at concentration below 5.4, 14.3 and 2.2 nM respectively (Table 2 and Table 3). Pro possibly occurred in only one sample of the Rainbow field, however the peak area was very small, so this result will need to be confirmed by further analyses. All fluids from the Ashadze 1 vent contained Trp (C on Figure 2) as well as the purest hot fluid sample from the Rainbow field (97% pure fluid) but was below LOQ. Phe could not be quantified in the samples as it was out of the linear domain, however it was definitely detected in all hydrothermal fluids (B on Figure 2) but one, whereas it was not detected in the deep seawater sample. Finally, Leu was with no doubt detected in all samples except for the purest hot fluid of the Rainbow field and measured concentrations were in the range 114 to 413 pM (Table 2). Trp, Phe and Leu that were quite clearly evidenced in the samples showed linearity down to lower concentration than the other amino acids. Especially Leu, which had the highest response and lowest LOD, could be accurately quantified. To be noted, Isoleucine (Ile), an isomer of Leu, was not investigated in this series of experiments, nevertheless the injection at a later stage of a standard solution of Ile and a standard solution of Leu revealed that these compounds elute in two well resolved peaks separated by 10 s. Therefore could Leu be distinguish from Ile and quantified separately.

We suspect amino acids to be present in hydrothermal fluids, however that a large portion occurs as metal complexes and cannot be detected using current approach. Earlier studies have shown indeed that amino acids can form complexes with a wide variety of metals and that the side chains −OH, −COOH, −CONH₂ are often involved in the coordination [56] [57] . Notably, this phenomenon is not to be considered marginal as, for example, Cu^II transport in blood partly occur via amino acid complexation [[58] and references therein]. Furthermore, the adsorption of amino acids on clay minerals is believed to occur through metal ions [59] . Besides, there is no doubt that clays [60] , sand [61] , sediments, [62] and many minerals [63] -[65] can retain and concentrate amino acids by adsorption. Mineral surface may also facilitate condensation and oligomerization reactions [66] although recent experiments under hydrothermal conditions led to the opposite conclusion [20] . Research regarding the adsorption and reactions of amino acids on mineral surfaces has largely been reviewed elsewhere [63] [66] [67] . As a result the amino acids possibly formed in hydrothermal systems may occur either adsorbed on minerals, or as oligomers or free; the proportion of each being unknown. The oligomer portion can be recovered and analysed by hydrolysis of the samples. Consistently Fuchida and coworkers found higher amounts of total hydrolizable amino acids (THAA) than DFAA in the Mariana Through [12] . Thus the very low concentration of DFAA detected in this study may represent only a small portion of the in-situ concentration. We suggest that a desalting step prior to preconcentration might help recovering higher amount of amino acids.

The amino acid concentration of our deep seawater reference is in the same order of magnitude as total DFAA background measured by other techniques in deep seawater from various places (Personal communication, R.

Kerouel). It is also reasonably consistent with the concentrations (< 110 nM) reported by Kaiser [68] in the 2300 - 4000 m depth of the water column at a station in the Atlantic near Bermudas. Svensson et al. [17] have reported amino acids concentrations in a moderate temperature shallow hydrothermal system surrounded by sediments: Total DFAA values ranged from 330 to 78,380 nM for hydrothermal samples and 430 nM for surrounding seawater. TDHFAA values published by Horiuchi et al. [14] for a deeper (1380 m) and hot (300˚C) hydrothermal system were in the lower range of 95.9 - 1061.6 nM and 324 nM for surrounding seawater. Amino acids contents were in the 3.82 - 15.24 µM (THAA) and < 0.22 µM (DFAA) in the hottest samples (T = 160˚C - 270˚C) of a vent field located at 2800 m depth in the Mariana Through [12] . Only two studies report on the amino acid presence in ultramafic-hosted vents. Lost City (~800 m depth) that is a one of its kind vent field having low T (< 100˚C) and very high pH (11) fluids has THAA concentrations ranging from 736 - 1565 nM in the purest samples. Klevenz et al. [15] found very low total DFAA (38 nM) in their samples (fluid % > 90) from the Logatchev ultramafic-hosted vent field whereas the other basalt-hosted fields of the Mid-Atlantic ridge had concentrations varying from 52 to 377 nM in the samples containing more than 70% of fluid. All fields in their study lied by about 3000 m depth with a fluids temperature of ~350˚C. We found here even lower amounts as only picomolar concentrations were measured. All fluids reached 350˚C and were collected at depth between 2300 and 4000 m. Our results complete the current dataset on amino acids in hydrothermal fluids but still no particular pattern relating amino acid concentration or occurrence with depth, temperature, host rocks or pH seems to emerge as mentioned by some of the above cited authors. This means that more work has to be done on natural samples to unravel the complex processes amino acids are involved in and to better understand their roles in hydrothermal environments and plumes. The technique we described here should help analysing more samples, faster, with low contamination source and down to sub-nM levels.