Detection of phosphates originating from Enceladus’s ocean – Nature

CDA data analysis

Short description of the CDA chemical analyser subsystem

The chemical analyser is the subsystem of the CDA¹⁰ that provides compositional information about an impacting dust particle. Depending on the trajectory of the particle, it either hits the central rhodium target (chemical analyser target (CAT), diameter 0.17 m), the surrounding gold target (diameter 0.41 m) or the inner wall of the instrument. This work deals only with impacts on the CAT. If a dust particle impacts the CAT with sufficient energy, it is totally vaporized and partly ionized, forming an impact plasma of target and particle ions together with electrons, neutral molecules and atoms. The instrument separates this plasma, the positive component of which is linearly accelerated towards a multiplier about 19 cm away, used to generate a ToF spectrum. The spectrometer is sensitive to positive ions only. The mass resolution (m/Δm), derived from laboratory experiments with the instrument, depends on the atomic masses of the ions. At 1 u, m/Δm is 10, increasing to 30 m/Δm at 100 u and up to 50 m/Δm at 190 u, although these values vary markedly with impact conditions.

Acquisition of a spectrum can be triggered by charge thresholds being exceeded on either the target or the multiplier ion detector (when a certain abundant ion species arrives there, typically H⁺, H₃O⁺ or Na⁺). In the case of a multiplier-trigger, the spectra show only ions with masses higher than the triggering ion species; the triggering species and those with lower masses do not appear in the spectrum. For this work, all P-rich spectra were triggered by impact. The spectra are logarithmically amplified, digitized at eight-bit resolution and sampled at 100 MHz for a period of 6.4 μs after the trigger. The recording period of the high-rate sampling mode allows the detection of ions with mass of up to 185–200 u, assuming that instrument recording is triggered by the impact itself.

Because ToF is proportional to the square root of the mass:charge ratio of ions, its spectrum in an ideal case also represents a mass spectrum for identical ion charges. The ions created by impact ionization in the impact speed regime considered here (roughly 10 km s^–1) are almost exclusively singly charged. Unfortunately, ToF is also influenced by the broad distribution of initial ion velocities, slightly varying flight paths and plasma-shielding effects⁴⁰. For that reason, species of identical mass are distributed over a range of sampling points and the mass resolution drops below integer values, usually at 20–30 u and above. Therefore, we prefer to show original CDA spectra with an x axis showing ToF rather than mass; the latter could mislead the reader to intuitively assume an unrealistically high accuracy.

For a detailed description of the instrument see ref. ¹⁰; the calibration routine is described in refs. ^6,20. For the dataset in this work the mass calibration was generally done using sodium (Na⁺) and sodium hydroxide ((NaOH)Na⁺) mass lines as reference and a stretch factor of a = 473 ns.

CDA dataset

All data used for this work are listed in Extended Data Table 1 and, like all CDA data, are archived on the Small Bodies Node of the Planetary Data System (PDS–SBN), at https://sbn.psi.edu/pds/resource/cocda.html. We used 15 time periods between 2004 and 2008 covering times of CDA pointing that were favourable for the detection of E-ring dust. Times this early in the Cassini tour were chosen because CDA contamination from salts deposited on the CDA impact target during deep Enceladus plume crossings was then negligible³⁹. Within these periods a total of 7,353 spectra from E-ring grains were recorded. A Lee filter was applied to the spectra to improve signal to noise. Of these spectra, 962 have been categorized as Type 3 (Extended Data Table 1).

From this set of Type 3 spectra, nine stand out as belonging to a previously unknown subtype with a unique pattern of three peaks of m/z above 120 u; these are the P-rich spectra discussed above. Individual events are listed in Extended Data Table 2. Remarkably, all these spectra were triggered on the instrument by the impact charge of the particle whereas only 36% of all Type 3 spectra in the dataset were triggered in this way (Extended Data Table 1). The remaining 64% of Type 3 spectra were triggered by a particular ion species reaching the multiplier (Methods). By applying binomial statistics, we get a probability that all P-rich spectra were impact charge triggered by chance of only 0.01%. Therefore, we assume a systematic effect in spectral recognition for P-rich grains. Indeed, it is plausible that the spectrum of a P-rich grain is not identified as Type 3 if it is triggered by an ion species at the multiplier. Type 3 spectra are in most cases triggered by Na⁺, which is then missing in the spectrum. However, without the Na⁺ line and in the absence of other typical Type 3 features such as (NaCl)_nNa⁺ or (Na₂CO₃)Na⁺, P-rich spectra are not identifiable as being Type 3 following the criteria normally used, which focus on spectra dominated by chloride, carbonate and hydroxide peaks^3,11.

We therefore used only those 345 spectra of the dataset that—like the new phosphate-rich spectra—have been triggered by impact (Extended Data Table 1) as a reference to calculate the frequency of the new phosphate-bearing subtype within the family of Type 3 grains and, with that, the average phosphate concentration in salt-rich Type 3 grains. It is noteworthy though, that even in the unlikely case that there was no such systematic selection effect and we use the entire dataset for the inference of average phosphate concentration, the lower limit will still be ΣPO₄^3– above 0.3 mM, more than 100-fold the mean concentration in Earth’s oceans, and the conclusions of this work are unaffected.

Laboratory methods simulating CDA mass spectra

Identification of sodium phosphates

The largest peaks in all nine CDA spectra always appear at 23 and 63 u, indicative of Na⁺ and (NaOH)Na⁺ cations that are the defining features in all salt-rich spectra from grains emitted by Enceladus (Type 3; ref. ³), thus these spectra contain sodium salts in an alkaline matrix. We used the other three characteristic peaks, at 125, 165 and 187 u (allowing a divergence of ±1 u that reflects the mass uncertainty of CDA^7,10), to perform a rigorous database search—in both our own database base⁴¹ as well as several external databases (for example, NIST Chemistry WebBook, MassBank North America and MassBank North Europe (https://massbank.eu/MassBank)). We found phosphates to be the only possible class of Na-rich compounds that produce a pattern that is in even remote agreement with the observed cationic peak pattern. After that we widened the accepted mass range to ±2 u away from the observed masses and relaxed the restriction to Na-rich compounds to determine whether there are other species (without sodium) forming a characteristic mass line pattern similar to the three mass lines (125, 165 and 187 u). The only match remotely close was methyl-2,2-dimethyl-3-(2,2-dichloroethenyl) cyclopropanoate, showing prominent peaks at m/z 61(−2), 127(+2), 163(−2) and 187. The mass spectrum can be found at ref. ⁴².

It is questionable whether a sufficient concentration of this type of chlorinated organic ester would be stable in Enceladus’s alkaline ocean. Even if we assume that this unlikely substance could be mixed with sodium salts in an ice grain, the spectrum has additional features not matching the CDA spectrum—most prominently a peak at m/z 91, which is absent in all nine CDA spectra. Similarly, any combination of exotic organic compounds we found that remotely reproduces the five prominent CDA peaks does have a number of additional peaks that are absent in CDA spectra.

In conclusion, sodium phosphates are the simplest and—to the best of our knowledge—the only explanation for all the peaks observed in the CDA spectra (Figs. 1 and 2). These salts have already been predicted to be relatively abundant in Enceladus’s ocean based on recent geochemical modelling¹⁷, and the observed peak pattern can be readily reproduced in mass spectra from our analogue experiment, as described in the following section.

LILBID experiments simulating CDA spectra

We used the laser-induced liquid beam ion desorption (LILBID) facility at Freie Universität Berlin to simulate CDA spectra. The experimental setup is described in detail in ref. ²² and here we provide only a brief overview. A micrometre-sized water beam is irradiated by a pulsed infrared laser (20 Hz, 7 ns pulse length) at a wavelength of 2,840 nm and variable laser energy. On absorbing the laser energy, the water beam explosively disperses into neutral and charged macroscopic, molecular and elemental fragments. Cations or anions, depending on the instrument polarity, are accelerated and analysed in a ToF mass spectrometer. Detected signals are amplified, digitized and recorded with a LabVIEW-controlled computer. Typically 300–500 individual spectra are co-added and averaged during one effective measurement. The laboratory setup is calibrated before every measurement using a 10⁻⁶ M NaCl solution²². The recorded mass spectra showed a mass resolution of 600–800 m/Δm (full-width at half-maximum).

As discussed in Methods, we found phosphates to be the only Na-rich compounds that are in even remote agreement with the observed cationic peak pattern. Because the appearance of certain peaks and their amplitudes are very sensitive to the phosphate species used and their concentrations, we individually tested a variety of combinations of aqueous solutions of Na₃PO₄, Na₂HPO₄, NaH₂PO₄ and Na₂HPO₃, as well as mixtures of these compounds at different concentrations with the goal of reproducing the CDA peak pattern. Concentrations were varied between 0.01 M and about 5 M for NaH₂PO₄, 0.5 M for Na₂HPO₄ and 1.5 M for Na₃PO₄.

We found that a mixture of Na₃PO₄ and Na₂HPO₄ most closely reproduced the CDA peak pattern of P-rich spectra (Fig. 1), whereas spectra of the pure substances always exhibited differences (Extended Data Figs. 1 and 2).

We also tested for the presence of phosphites (using Na₂HPO₃, in which P has an oxidation state of +III) and hypophosphite (using NaH₂PO₂, in which P has an oxidation state of +I). Both substances created mass lines not in agreement with CDA observations: (Na₂HPO₃)Na⁺ (149 u) for phosphite and (NaH₂PO₂)Na⁺ (111 u) for hypophosphite. For phosphite we inferred an upper limit concentration of 5 mM to be in agreement with the absence of a mass line at 149 u in CDA spectra (Extended Data Fig. 3).

The phosphate salts used in our experiments were:

• Na₃PO₄: Sigma-Aldrich, 96% purity

• Na₂HPO₄: Roth, over 99% purity

• NaH₂PO₄•2H₂O: AppliChem, over 98% purity

• NaH₂PO₃•5H₂O: Sigma-Aldrich, over 98% purity

• NaH₂PO₂•H₂O: Sigma-Aldrich, over 99% purity

Theoretical considerations regarding the observed characteristic pattern of phosphate ions

We observed the peak sequence at masses 187, 165 and 125 u as a characteristic pattern indicating the presence of phosphorus (P) salts in Enceladus’s ice grains, present in all nine of the individual CDA ice grain spectra. Phosphate-containing salts from the solvation and extraction of minerals at a basic pH value produced a number of anion species in solution that are coupled through a multistep acid–base thermodynamic equilibrium and proton exchange—for example, PO₄^3– ↔ HPO₄^2– ↔ H₂PO₄^–. For the given high salt concentrations, abundance of counter cations are close to anion abundances in solution (Na_nPO₄^(3–n)–). After ice grain impact, ionic aggregates showed additional clustering with abundant cations such as Na⁺ resulting in characteristic mass peaks due to species including (Na₂HPO₄)Na⁺ and (Na₃PO₄)Na⁺. Owing to the basic pH of the solution forming the ice grains (or the liquid desorption matrix of the laboratory experiment), these two species are expected.

However, we did not detect a (NaH₂PO₄)Na⁺ adduct in CDA spectra but rather a (NaPO₃)Na⁺ peak at 125 u, a charged molecular aggregate of a metaphosphate with phosphorus in the same oxidation state (+V) as the other orthophosphates. We attribute the absence of (NaH₂PO₄)Na⁺ and the appearance of (NaPO₃)Na⁺ instead to an elimination reaction during desorption (LILBID) or particle impact (CDA):

NaH₂PO₄ + heat/energy → NaPO₃ + H₂O, followed by the addition of Na⁺.

The fact that the (NaPO₃)Na⁺ peak at 125 u is formed from hydrogen phosphates (linked via an acid–base equilibrium) in our analogue experiments, but not from Na₃PO₄ alone, supports this picture. However, it should be noted that we could produce a (NaH₂PO₄)Na⁺ peak (143 u) if high concentrations (over around 0.05 M) of NaH₂PO₄ or Na₂HPO₄ (Extended Data Fig. 2) were used in the laboratory experiments or if the pH value of the solution was forced below about 9.0. We then find a clear correlation of the amplitudes of dihydrogen phosphate (143 u) with metaphosphate (125 u). In the CDA data the 143 u peak appears to be below the noise level of the detector. In our laboratory spectra the 125 u metaphosphate peak is generally three- to fivefold more intense and it is therefore likely that small amounts of (NaH₂PO₄)Na⁺ (143 u) were buried in the noise of the CDA spectrum. Such a signature was indeed tentatively observed in one of the nine spectra (event 3 in Extended Data Table 1).

To test these ideas we calculated the thermochemical properties of the water elimination of NaH₂PO₄ + heat/energy → NaPO₃ + H₂O with the Jaguar quantum chemistry programme package at the PBE0-D3/6-311+G(d,p)/PBE level of theory. Two variants have been calculated: (1) with Na⁺ as a counter ion and (2) without Na⁺ (involving a H₂PO₄^– anion). The reaction is slightly endergonic in the gas phase and even more favoured in/with water. The barrier of the concerted elimination reaction appears to be low if water molecules are involved. One may even consider it to be a water-assisted reaction, in particular for the case of Na⁺ as counter ion (Extended Data Fig. 4).

Hydrothermal experiments and calculations of chemical equilibria

Experimental methods and procedures

Hydrothermal experiments were performed with a Dickson-type hydrothermal apparatus (Toyo Koatsu Co. Ltd) used in a previous study⁴³. In this apparatus, water–rock reactions occur in a flexible reaction cell that comprises a gold bag with a titanium head. The flexible reaction cell was connected to a stainless steel sampling tube with a gold-coated inner wall, to avoid catalytic reactions. The reaction cell was heated at 500 °C for 3 h in air before each experiment to remove potential contamination by organic matter. The reaction cell was placed in a stainless steel autoclave, which was held at a pressure and temperature of 30 MPa and 150 °C, respectively, during the experiments. The reaction cell was pressurized by the addition of surrounding water into the autoclave using a hand pump. The autoclave was heated with an external heater. Due to the flexibility of the gold bag, the reaction cell deforms plastically without fracturing in response to the addition of surrounding water. This allowed us to undertake online sampling of fluid without significant changes in pressure.

The starting rock material was prepared by powdering a few pieces of the commercially available Jbilet Winselwan carbonaceous chondrite (CM-2 type)²⁵ in an agate mortar. The starting solutions were prepared by dissolving NaHCO₃ powder and NH₃ solution in ultrapure water such that the initial ΣCO₂ was 0.2 M for Run no. 1 and 1.0 M for Run no. 2, and ΣNH₃ was 0.5 M. One way to achieve high ΣCO₂ in an Enceladus experiment is to introduce gaseous CO₂ in the reaction cell and dissolve it by pressurization. However, this is technically challenging in our experimental system and we cannot control the [ΣCO₂] and pH of the starting fluids. Therefore, we achieved high ΣCO₂ by the addition of NaHCO₃ powder. We added 1.0 M Na in Run no. 2, higher than Na concentrations in Enceladus’s ocean³. If Na phosphate (for example, Na struvite) had precipitated in such Na-rich fluids of Run no. 2, this would have affected the buffer system that controls ΣPO₄ in fluids. Although the temperature and pressure dependences of the thermodynamic constants of Na struvite have not been experimentally determined, the saturation index (log([Na⁺][Mg²⁺][HPO₄^2–]/K_Na-struvite), where K_Na-struvite denotes the equilibrium constant from ref. ¹⁷ of Na struvite at 1 bar and 25 °C, is calculated to be below −1.6 using the measured Na, Mg and ΣPO₄ concentrations and in situ pH. Because the fluid Mg concentrations of Run no. 2 were below the detection limit (Extended Data Fig. 8b), Na struvite was likely to have been undersaturated in the experiment. The presence of Ca sulfate in the starting rock materials (CM chondrite) also does not affect our conclusion regarding the hydroxyapatite-calcite buffer system. If the hydroxyapatite-gypsum (or hydroxyapatite-anhydrite) buffer controls ΣPO₄^3–, [ΣPO₄^3–] should dramatically increase with [SO₄^2–]; however, this did not occur (Extended Data Table 3). The initial amounts of starting rock powder and solution were around 7 g and 10 ml, respectively. In each experiment we collected several aliquots (about 1.2 ml) of fluid for analysis. As such, water:rock mass ratios in the reaction cell changed from around 1.4–0.9 due to the removal of multiple fluid samples.

The collected fluid samples were analysed by inductively coupled plasma atomic emission spectroscopy (ICP–AES: SPS5510, Hitachi High-Tech) for dissolved Na, K, Mg, Ca, Si and Al content; by ICP mass spectrometry (ICP–MS: Agilent 8000) for dissolved Fe content; by ion chromatography (ICS-1600, DIONEX) for dissolved Cl and SO₄ content; and by gas chromatography (GC-2010, Shimadzu) for ΣCO₂ content (Extended Data Table 3). ΣPO₄^3– content was analysed by both tandem ICP–MS (ICP–MS/MS: Agilent 8000, Agilent) and ion chromatography. The former methodology quantified dissolved total P in fluids at high sensitivity, including phosphate and potentially phosphite, whereas the latter directly measured ΣPO₄^3– with moderate sensitivity. The results of the two analyses showed that, basically, all dissolved total P in the fluids was phosphate (Extended Data Table 3). As such, we used ICP–MS/MS results to quantify ΣPO₄^3– in the fluid samples. In situ pH values at 150 °C during the experiments were calculated using the geochemical code PHREEQC v.3 (ref. ⁴⁴), based on the measured pH at 25 °C and dissolved elements, molecules (phosphate and sulfate) and ΣCO₂ in the diluted fluid samples. To calculate the in situ pH of the alkaline, Na-carbonate-rich fluids, we constrained the charge balance from the measured pH at 25 °C in two ways. First, Na was used to compensate for charge imbalance due to analytical uncertainties. Second, ΣCO₂ was used to compensate for charge imbalance. We adopted the median value between the two values as the in situ pH, and the difference between the median and upper/lower values as the error. We also conducted mineralogical and chemical analyses of solid residues after the experiments using an X-ray diffraction spectrometer (XRD: MiniFlex600, Rigaku) and a scanning electron microscope with an electron probe microanalyser (EPMA: JXA-8530F, JEOL).

The ΣPO₄^3– concentrations controlled by the hydroxyapatite-calcite and whitlockite-calcite buffer systems were calculated with the equilibrium constants obtained from the SUPCRT92 programme⁴⁵. In the calculations, the thermodynamic properties of whitlockite (β-Ca₃(PO₄)₂) and hydroxyapatite were taken from Robie et al.⁴⁶.

Elemental composition of carbonate and phosphate

Extended Data Figs. 5 and 6 show typical secondary electron microprobe images and elemental mapping for thick sections of solid residues after the experiments (Run nos. 1 and 2). In the solid residues we found large grains of precipitated Ca carbonate (CaCO₃, size over 100 μm; Extended Data Fig. 5 and Extended Data Table 4). Because the typical size of Ca carbonate in the sample before the experiments was 10 μm or less (Extended Data Fig. 7), the large Ca carbonate grains found after the experiments are highly likely to have been precipitated during the experiments through ΣCO₂ sequestration. This view is supported by decreases in ΣCO₂ in fluids (Extended Data Fig. 8). Our XRD analysis shows that the precipitated Ca carbonates are most likely calcite in both of the experiments (Run nos. 1 and 2; Extended Data Fig. 10).

To investigate the composition of the phyllosilicate phase before and after the experiments, we performed EPMA analysis for the phyllosilicate matrix around Ca phosphate grains. Extended Data Fig. 9 shows a ternary diagram with the results of EPMA analysis for phyllosilicate, Ca phosphate and Fe oxides in the samples before and after the experiments. Phyllosilicates in the starting materials were mostly serpentine; however, mixtures of serpentine and saponite appeared in the samples after the experiments. Extended Data Fig. 9 indicates a linear trend of a mixing line of the compositions between Ca phosphate and serpentine (and saponite mixtures).

Extended Data Table 4 presents the results of spot analysis for Ca phosphate and carbonate found in the solid residues after the experiments of Run nos. 1 and 2. At the micrometre scale, Ca phosphate minerals usually coexist with other secondary minerals such as serpentine and saponite (Extended Data Fig. 6), and thus the results of the spot analysis might not be explained by the simple chemical composition of an alteration mineral. As reported in Extended Data Table 4, Grain C with the highest P content had high total mass fractions of around 95–100 wt%, suggesting that the contribution of calcite to Ca:P atomic ratios is small (less than 2–3% of Ca mole abundance), whereas serpentine and saponite would coexist with Ca phosphate of Grain A, Vein B and Grain D. These hydrated secondary minerals can reduce the total mass fractions of spot analyses by 5–10% owing to the presence of H₂O and OH. Because the total mass fractions for Grain A, Vein B and Grain D are also high in general (over 87 wt%; Extended Data Table 4), calcite is presumed to be rare in Ca phosphate grains, suggesting that the contribution of calcite to Ca:P ratios is small.

Grain C of Run no. 2 has very high P:Si ratios (Extended Data Table 4c). The measured Ca:P mole ratios are 1.32–1.37. Grain D of Run no. 2 has relatively high P:Si ratios, above 1.0 (Extended Data Table 4d). The Ca:P mole ratios of Grain D range from 1.29 to 1.34. The Ca:P ratios of hydroxyapatite, whitlockite [β-Ca₃(PO₄)₂] and Mg-whitlockite [Ca₉MgH(PO₄)₇] are 1.67, 1.50 and 1.29, respectively. Ca phosphates collected after Run no. 2 (Grains C and D) are inferred to consist of mixtures of multiple Ca phosphates. Due to the trace amounts of Ca phosphates in the solid residues, these could not be identified in the XRD spectra (Extended Data Fig. 10)

In Ca phosphate grains of Run no. 1 (Grain A and Vein B), P:Si mole ratios are generally low (roughly 0.1–1.0) (Extended Data Table 4a,b), suggesting that Ca phosphates coexist with serpentine and saponite. There is a relatively wide variation in the Ca:P mole ratios of Grain A and Vein B, from approximately 1.1 to 1.5 (Extended Data Table 4a,b). A high Ca:P ratio of around 1.5 requires the presence of hydroxyapatite and/or whitlockite. To explain the low Ca:P ratios (below 1.29) of the analysed spots on Grain A and Vein B, the presence of Ca-poor phosphate, such as NH₄MgPO₄, would be needed.

Our results showing high C:P ratios (1.32–1.37) for Run no. 2 and low Ca:P ratios (1.25–1.27) for Run no. 1 are consistent with our interpretation of the buffer system that controls ΣPO₄³⁻ concentrations in the fluid samples. The Ca phosphate grains (C and D) of Run no. 2 contain high fractions of phosphate with high Ca:P, such as hydroxyapatite. This observation is consistent with the control of ΣPO₄³⁻ concentrations in fluid samples of Run no. 2 by the hydroxyapatite-calcite buffer (Fig. 3). In contrast, the observation of Ca phosphate grains with low Ca:P for Run no. 1 (for fluid with ΣCO₂ = 0.01 M and reaction time 1,000 h) is consistent with the control of ΣPO₄³⁻ concentrations by Ca phosphate with lower Ca:P ratios, such as whitlockite, rather than control by hydroxyapatite (Fig. 3). This fluid sample was collected in the final sampling of Run no. 1 (Extended Data Fig. 8a), and the solid residue was collected after this final sampling. We consider that Ca phosphate with relatively low Ca:P, such as whitlockite [β-Ca₃(PO₄)₂], Mg-whitlockite and NaMgPO₄, would have been precipitated before the final sampling in response to an increase in Mg concentration in the fluid (Extended Data Fig. 8), and that the whitlockite-calcite buffer would have controlled the P concentration of the fluid at the end of experimental Run no. 1.

Chemical equilibrium considerations regarding phosphate buffering systems

Extended Data Fig. 11 shows the calculated results of [ΣPO₄³⁻] in fluids as a function of temperature for different pH and [ΣCO₂] values, assuming the whitlockite-calcite buffer system (see above for the equation). Although our experimental data suggest that the whitlockite-calcite buffer could have controlled ΣPO₄³⁻ concentrations in the fluid at the end of Run no. 1, we consider that this could have been an experimental artefact reflecting transient states of supersaturation of Mg-bearing phases. In the experiments, the dissolution of anhydrous primordial minerals (for example, olivine and pyroxene) would have provided Mg and Ca for the fluids. These Mg and Ca ions would have been removed by reactions with CO₃²⁻, forming carbonate minerals (Extended Data Fig. 5). However, at the end of Run no. 1, ΣCO₂ in the fluids was depleted (Extended Data Fig. 8) owing to CO₂ sequestration via carbonate mineral formation; therefore, dissolved Mg and Ca concentrations increased in the fluids (Extended Data Fig. 8). In longer-period reactions, excess Mg and Ca ions would have been consumed through the formation of serpentine and saponite. Thermodynamic equilibrium calculations show that Mg and Ca concentrations in fluids that interact with carbonaceous chondritic rocks are typically low (0.001–0.010 mM each) in alkaline fluids (pH over 9.0) in the presence of serpentine and saponite^36,47. In Enceladus, water–rock reactions would be driven closer to acid–base equilibrium because the reactions can occur over much longer time scales than apply in our experiments. If the ocean of Enceladus were consistent with the expectation of low Mg and Ca concentrations (for example, under 0.1 mM each), then ΣPO₄³⁻ in fluids in contact with chondritic rocks would be controlled by the hydroxyapatite-calcite buffer system.

In this study, we considered hydroxyapatite and whitlockite [β-Ca₃(PO₄)₂] as phosphate minerals of buffer systems that control [ΣPO₄³⁻] in fluids. However, previous work has suggested the possibility of the occurrence of other phosphate minerals in water–rock reactions in Enceladus¹⁷, including Mg-whitlockite and merrillite. Using the equilibrium constants of dissolution of Mg-whitlockite and merrillite at 1 bar and 25 °C (ref. ⁴⁸), and assuming [Mg²⁺] as roughly 10⁻⁵ M in the fluid samples (Extended Data Table 3), ΣPO₄³⁻ values controlled by the Mg-whitlockite-calcite and merrillite-calcite buffer systems (that is, 6 × 10⁻⁵ and 7 × 10⁻⁶ M, respectively, for ΣCO₂ = 0.1 M) are about 1/30 to 1/250 that of the hydroxyapatite-calcite buffer (that is, 2 × 10⁻³ M for ΣCO₂ = 0.1 M). Although the temperature and pressure dependences of the thermodynamic equilibrium constants of both Mg-whitlockite and merrillite are unknown, the observed ΣPO₄³⁻ values in our experiments are unlikely to be explained by these buffer systems.

We also assumed that calcite is the carbonate mineral phase for the buffer system, but dolomite is usually more abundant than calcite in CI chondrites⁴⁹. However, ΣPO₄^3– values controlled by the hydroxyapatite-dolomite buffer are orders of magnitude higher than those controlled by the hydroxyapatite-calcite buffer for ΣCO₂ = 0.1 M and the typical Mg concentration in our experiments ([Mg²⁺] = 10 M^–5). Therefore, we conclude also that the hydroxyapatite-dolomite buffer may not explain the observed ΣPO₄³⁻ in our experiments.