Experimental exploration of polycyclic aromatic hydrocarbons stability in subduction zones

Abstract

Thermodynamic, experimental and field studies have suggested that organic compounds could be stable, and in some cases predominate over inorganic carbon species, within subduction zones under high pressure and high temperature (HP-HT) conditions. Beyond sedimentary organic matter of biological origin, solid organics can be inherited from hydrothermal circulation at mid-ocean ridges or abiotically formed by carbonate destabilization in the slab. To assess the fate of solid organic compounds during subduction, HP-HT experiments using piston-cylinder and multi-anvil presses have been performed at 700–1000 °C and 3–7 GPa. Different starting solids were tested, including either synthetic polycyclic aromatic hydrocarbons (PAHs) alone, with (i.e., 1-hydroxypyrene, 1-pyrenebutyric acid) or without (pyrene) oxygen-bearing functional groups, or a mixing of pyrene and antigorite. Our results show that increasing P–T conditions lead to the formation of hydrogenated (±oxygenated) graphitic carbon preserving a high level of structural disorder, far from graphite structure. We also observe the formation of aqueous fluids during experiments at 700 °C and 3 GPa with oxygen-functionalized PAHs, suggesting quick water release from solid organic compounds at HP-HT in subduction zones. Pyrene-antigorite experiments reveal various mineral assemblages depending on redox conditions. Oxidizing conditions favor the formation of magnesite-enstatite-coesite while reducing conditions promote forsterite-enstatite-graphitic carbon assemblages. Our results finally highlight the limited reactivity of solid organic compounds when exposed to aqueous fluids derived from serpentinite under reducing conditions which could facilitate the recycling of organic carbon into the deep mantle.

Introduction

The deep carbon cycling has been the subject of intense controversies over the last decade (e.g., Dasgupta and Hirschmann 2010; Kelemen and Manning 2015; Plank and Manning 2019). Among the parameters governing carbon transfer between Earth reservoirs, redox conditions (i.e., oxygen fugacity, fO₂) exert major control over carbon speciation (from C(−IV) in methane CH₄ to C(+IV) in carbon dioxide CO₂, a redox range covering inorganic and organic carbon species) and mobility in fluids at high pressure-high temperature (HP-HT; Sverjensky et al. 2014; Guild and Shock 2020; Bouilhol et al. 2022). Solid carbonates are the main carbon pool entering subduction zones (Dasgupta and Hirschmann 2010; Kelemen and Manning 2015; Plank and Manning 2019) and thus carbonate stability and dissolution in HP metamorphic fluids are relatively well studied (e.g., Gorman et al. 2006; Manning et al. 2013; Pan et al. 2013; Ague and Nicolescu 2014; Facq et al. 2014; Menzel et al. 2020; Farsang et al. 2021). Conversely, not much is conclusively known about the fate of solid organic compounds (i.e., rock-hosted compounds with C-H bonds) during slab evolution, including dehydration, at HP-HT.

Up to now, the main solid organic pool considered in deep carbon flux models (e.g., Plank and Manning 2019) is biological organic matter in oceanic sediments. Yet solid organic compounds of abiotic origin are now increasingly found in the oceanic lithosphere, resulting from magmatic degassing and/or hydrothermal circulation along the ridge system (Sforna et al. 2018; Andreani and Ménez 2019; Andreani et al. 2023) or (HP-HT) reactions during subduction (Galvez et al. 2013; Vitale Brovarone et al. 2017; Nan et al. 2020; Bouilhol et al. 2022; Debret et al. 2022; Caurant et al. 2023). The reported condensed carbonaceous matter (CCM), also often referred to as graphitic carbon, consists in a disordered macromolecular structure of high molecular weight composed of variously-functionalized and polymerized aromatic and aliphatic moieties. Under hydrothermal conditions, CCM formation is described in olivine-rich igneous rocks and serpentinites formed at (ultra)slow-spreading oceanic ridges (such as in the Northern Appenines ophiolite and at the Mid-Atlantic ridge, as reported by Andreani and Ménez 2019, Sforna et al. 2018 and Andreani et al. 2023) or near forearc trenches (such as the Yap Trench in the western Pacific Ocean as reported by Nan et al. 2020). Its formation is connected to (1) cooling of magmatic fluids leading to carbon re-speciation at mid-ocean ridges or (2) low temperature hydrothermal alteration of ferrous-iron bearing minerals, leading to hydrogen production and subsequent organic synthesis either at mid-ocean ridges and subduction zones (Milesi et al. 2016; Sforna et al. 2018; Nan et al. 2020; Andreani et al. 2023). Abiotic organic synthesis has also been shown to occur at HP-HT conditions during subduction as a consequence of carbonate destabilization and/or metamorphic reactions in an increasing number of studies (Malvoisin et al. 2012; Galvez et al. 2013; Vitale Brovarone et al. 2017; Tao et al. 2018; Bouilhol et al. 2022; Debret et al. 2022; Caurant et al. 2023). In addition to nano- to micro-diamonds, a wide variety of solid organic compounds is observed in this setting including amorphous carbon, graphitic carbon (also referred to as nano/micro-crystalline graphite, graphite-like or simply graphite) but also hydrocarbons. Recently, functional groups typical of organic acids have also been reported along with amorphous carbon, as coating the surface of diamonds hosted in fluid inclusions (Frezzotti 2019). However, our understanding of solid organic compound stability, their mechanism of formation as well as their role in the deep carbon cycle remains limited, resulting in a major knowledge gap.

A growing body of literature tackles the question of the stability of (abiotic) organic compounds in metamorphic fluids at HP-HT using thermodynamic calculations or experiments. Manning et al. (2013) first calculate the solubility of pure graphite in subduction fluids. They show that graphite could be easily dissolved, especially in CO₂-rich-fluids at HP-HT (i.e., at 0.2 GPa and 1000 °C), conversely to what was previously thought based on its refractory character. This result is confirmed experimentally by Tumiati et al. (2020), who study the solubility of graphite and amorphous carbon (glass-like carbon) in HP-HT fluids through piston-cylinder (PC) redox-buffered experiments conducted at pressures ranging from 1 to 3 GPa and temperatures of 800 °C. Pure glass-like carbon is used in their study as an analogue for disordered carbonaceous matter of biological origin. They report an enhanced dissolution of glass-like carbon compared to crystalline graphite, declining at increasing pressure. Above 3.4 GPa at 800°C, glass-like carbon could be less soluble than graphite. Therefore, they speculate that, alternatively to graphite and even diamond, disordered carbonaceous matter should be the stable carbon polymorph. This study highlights how “intermediate” solid organic compounds with a redox state comprised between those of methane and pure graphite (C(0)), should be considered in subduction zones due to their specific properties including solubility and stability, among others. Along the same lines, Sverjensky et al. (2014) thermodynamically predict the composition of fluids at equilibrium with eclogitic and peridotite minerals (i.e., for temperature between 600 and 900 °C and pressure of 5.0 GPa). They conclude for eclogitic minerals that dissolved organic acids such as acetate (CH₃COO⁻) and propionate (CH₃CH₂COO⁻) can dominate over carbonate species and methane in HP-HT fluids, depending on fO₂ and pH values. Stability of a Mg_silicate/propionate complex (Mg[OSi(OH₃)][CH₃CH₂COO]) was also predicted at 1 GPa and 800 °C from experimental data and thermodynamic calculations (Tumiati et al. 2017; Tiraboschi et al. 2018).

Similarly, Huang et al. (2017) explore the stability of acetate ions at 300 °C and 2.4–3.5 GPa in diamond anvil cell (DAC) experiments and show the formation in water of immiscible hydrocarbons coexisting with carbonate crystals. Szlachta et al. (2022) further test the stability of acetate up to 5 GPa and 600 °C, also using sodium acetate aqueous solutions. They observe the decomposition of acetate into hydrocarbons and solid carbonaceous matter beyond 4 GPa and 500 °C. Huang et al. (2023) perform DAC experiments using calcium acetate Ca(CH₃COO)₂ aqueous solutions as starting material and report the formation of hydrocarbons, carbonate CO₃²⁻ and bicarbonate HCO₃⁻ ions and CaCO₃ crystals (i.e., calcite or aragonite). Finally, Li (2016) shows the formation of hydrocarbons from sodium acetate aqueous solutions in synthetic fluid inclusions trapped in quartz crystal at 1.5–2.5 GPa and 600–700 °C, in the presence of different mineral redox buffers. Graphitic carbon (referred to as graphite in the study) formation is also reported in these experiments when buffered by nickel-nickel oxide Ni-NiO (NNO), cobalt-cobalt oxide Co-CoO and rhenium-rhenium oxide Re-ReO₂ (RRO). It is interesting to note that, although these studies focus on dissolved carbon species, they also report the formation of solid organic compounds, i.e., graphitic carbon. The later are unfortunately often poorly characterized in these studies.

While natural abiotic solid organic compounds have not yet been studied experimentally at HP-HT, the stability of solid organic compounds of biological origin was already investigated in sedimentary or subduction context. For example, kerogen stability was examined at increasing pressure and temperature in order to characterize the thermal maturation and graphitization of sedimentary organic matter (up to 1450 °C and 8 GPa; Ross et al. 1991; Bustin 1995; Beyssac et al. 2003; Zeng and Wu 2007; Nakamura et al. 2017; 2020). During these experiments, as expected from comparison with natural systems, kerogens evolve chemically and structurally from an amorphous macromolecular carbon structure with hydrogen and heteroatoms (including O, N, and S) to graphite or graphitic carbon depending on the graphitizable or non-graphitizable nature of the studied kerogen (Buseck and Beyssac 2014). This graphitizable character depend on chemical composition and the structure of the kerogen precursor (hydrogen-rich with predominantly long chain aliphatic structures vs. oxygen-rich with high content of cyclic and aromatic structures) and refers to the possibility to synthetize graphite after additional laboratory heat treatment (without pressure). Even at very high temperature, kerogens derived from oxygen-rich, strongly aromatic and cyclic precursors are non-graphitizable, leading to graphitic carbon with increasing temperature and not graphite. This contrasts with natural settings where, even if the associated kinetics are very slow, all biological solid organic compounds are converted to crystalline graphite over 700 °C thanks to hydrostatic pressure and geological timescales (Beyssac et al. 2002; Buseck and Beyssac 2014). Immature and non-graphitizable kerogens (often attributed to Type III or Type IV kerogens, rich in O and S, respectively) share some structural and chemical similarities with abiotic CCM, especially in terms of aromaticity degree, aliphaticity and heteroatom contents and could be regarded as analogues to run HP-HT experiments aiming at characterizing CCM stability. However, natural kerogens can be highly heterogeneous in chemical structure and composition depending on their biological precursors, conditions of formation and thermal maturation. As the chemistry and structure of abiotic CCM are not yet well constrained, it is accordingly difficult to find a kerogen that constitutes a relevant analogue. Moreover, the host rock of kerogen (i.e., sediments) differs from the magmatic one of CCM, which is often found spatially related to (hydrated) mantle mineral, e.g., serpentine, olivine. The host rock could have an influence on the evolution of CCM for instance via the iron content capable of interacting with other redox-sensitive elements and compounds, or via the catalytic properties of certain rock-forming minerals known to have an impact on abiotic organic synthesis. Natural observations (e.g., Debret et al. 2022; Bouilhol et al. 2022) of HP-serpentinites highlight a wide range of structural and compositional diversity of solid organic compounds, which is not observed in eclogitic meta-sediments in similar settings. The host rock could therefore impact the synthesis pathways and thermal evolution of solid organic compounds during subduction. The host rock is even reported to affect the analysis of organic matter, as highlighted by Matthewman et al. (2013) for meteorites with comparable paragenesis.

To avoid possible biases in using inappropriate kerogen references in HP-HT experiments or numerical modelling, polycyclic aromatic hydrocarbons (PAHs) could represent better analogues of abiotic solid organic compounds. PAHs (e.g., naphthalene C₁₀H₈, anthracene C₁₄H₁₀, pyrene C₁₆H₁₀) which are composed of multiple aromatic rings have already been used in this purpose in thermodynamic calculations (Milesi et al. 2016; Andreani et al. 2023). PAHs stability was also experimentally studied at HP-HT conditions by Chanyshev et al. (2015, 2017a; b) and Davydov et al. (2004). These authors conducted experiments using powdered solid PAHs from 1.3 to 8.0 GPa and up to 1300 °C in which they report PAHs turning into graphitic carbon by dehydrogenation and condensation reactions. However, the potential role of slab minerals on the reactivity of these compounds (e.g., during slab dehydration) has not been yet investigated experimentally, as is the HP-HT stability of more diverse and complex PAHs including functionalized ones. Until now, PAHs and serpentine mixing has only been studied from a planetary accretion perspective via shock pressure experiments (in the range 6–37 GPa; Mimura et al. 2005). Here, we examine the fate of CCM during subduction using HP-HT experiments along a subduction gradient (i.e., 700–1000 °C, 3–7 GPa) involving either a PC press or a multi-anvil press (MAP). We investigate different CCM analogues, including pyrene and oxygen-functionalized PAHs (i.e., 1-hydroxypyrene C₁₆H₉-OH and 1-pyrenebutyric acid C₁₆H₉–CH₂–CH₂–CH₂-COOH, which contain alcohol or carboxylic groups, respectively), to better match CCM chemical structure and composition observed at mid-ocean ridges. These solid organic compounds were used alone or, in the case of pyrene, mixed with HP serpentine (i.e., antigorite). We observe for all PAH types that the increase in pressure and temperature leads to the formation of hydrogenated (±oxygenated) graphitic carbon with a preserved structural disorder, thereby allowing the transport of carbon and volatiles into the deep Earth. The release of aqueous fluids from organics is observed with 1-pyrenebutyric acid at 3 GPa and 700 °C. Concerning the stability of pyrene during serpentinite dehydration (i.e., antigorite and chlorite breakdown), we show that various assemblages are stabilized in the used HP-HT devices. This is likely explained by the variable redox conditions imposed by different experimental assemblies, which appear as the primary factor controlling the form and mobility of carbon in subduction settings.

Material and methods

Polycyclic aromatic hydrocarbons and natural antigorite samples

Four series of experiments were conducted using different starting materials (Table 1). The first three experiment series used PAHs purchased from Sigma Aldrich, namely pyrene (sublimed grade, 99% purity), 1-hydroxypyrene (purity > 97%) and 1-pyrenebutyric acid (purity > 97%, hereafter to as PyBA). The synthetic PAHs were ground to powder using agate mortar and pestle and then loaded in platinum (Pt) capsules for HP-HT experiments.

Table 1 Experimental conditions with starting material and solid products, and Raman parameters of the D and G bands obtained for the produced graphitic carbon

The fourth series of experiments used a 50:50 mixing (by mass) of pyrene and natural antigorite powders. Antigorite came from a natural sample, i.e., Vis-9, from the Monviso meta-ophiolite (Debret 2013). To avoid weathered surface, the inner part of the field rock specimen was cut with a copper saw blade before being ultrasonically cleaned in an absolute ethanol bath and then ground using agate mortar and pestle. During grinding, as much grains as possible of opaque mineral phases (i.e., iron sulfides and magnetite) were removed by hand in order to obtain the purest possible antigorite powder.

Major element and volatile composition of the powdered antigorite was analysed at the Service d’Analyses des Roches et des Minéraux (SARM, Nancy, France; Table 2) by inductively coupled plasma optical emission spectroscopy (ICP-OES, Thermo Scientific iCAP 6500). Sample digestions for major element analysis (i.e., SiO₂, Al₂O₃, Fe₂O₃, MnO, MgO, CaO, Na₂O, K₂O, TiO₂, P₂O₅) were performed on LiBO₂ fluxed fusions following the procedure described by Carignan et al. (2001). The standard reproducibility was better than 2% for major elements based on repeated analyses of the UB-N serpentinite standard from the Centre de Recherches Pétrographiques et Géochimiques (CRPG, Nancy, France). Whole rock sulfur and carbon contents were determined on a carbon–sulfur (CS) analyser Leco SC144 DRPC by the SARM. This involved high temperature combustion of 250 mg of powdered rock sample and infrared detection of effluents. The analyses were below detection limit (<0.01%), showing the absence of detectable impurities such as sulfur and carbon phases in the powdered antigorite. Iron redox state in antigorite was determined first by X-ray absorption spectroscopy (XAS) at the iron K-edge at the LUCIA beamline of SOLEIL synchrotron (Source Optimisée de Lumière d’Energie Intermédiaire du LURE, Saint Aubin, France) on antigorite minerals in the field sample thin sections giving a Fe³⁺/ΣFe of 0.2 (with ΣFe corresponding to the total Fe²⁺ and Fe³⁺). Automatic titration at the SARM on the powdered antigorite (from which magnetite crystals were handpicked) gives an identical ratio Fe³⁺/ΣFe of 0.2 (Supplementary Methods). As both the low FeO content (Table 2) and Fe³⁺/ΣFe estimates from bulk analysis of the powdered antigorite are consistent with in situ analyses on antigorite minerals, this suggests the absence of magnetite (or other Fe-bearing accessory phases) in the powdered antigorite used for experiments. This is also consistent with petrographic observation which reveals the quasi absence of magnetite in the hand specimen (Debret 2013). The powdered antigorite (between 2 and 15 mg, depending on the experiment) was mixed with the powdered pyrene and introduced in the Pt capsules for HP-HT experiments.

Table 2 Composition (in weight %, wt%) of the powdered Monviso antigorite (sample Vis-9) used in pyrene-antigorite (50:50 by mass) experiments

High pressure experiments

The experiments were performed at Institut de physique du globe de Paris (IPGP, Paris, France) using either an end-loaded PC press or a Walker-type MAP. The PC press was favoured for low P conditions (i.e., antigorite breakdown, in chlorite stability field; Fig. 1) while the MAP was used for higher P–T conditions (i.e., chlorite breakdown, in garnet and diamond stability fields; Fig. 1). To investigate kinetic effects, various run durations were tested for the pyrene series (i.e., 1. 2. 3 and 7 days based on literature review) while the other series were run during 72 h. The pressure and temperature conditions, also reported in Fig. 1, duration of the experiments and a summary of the starting materials and run products are given in Table 1. The starting materials were encapsulated as a homogeneous powder into Pt tubes. Platinum was chosen for its non-reactivity with the starting compounds (organic matter or silicates) in the chosen P–T conditions (see Jakobsson 2012 who discusses Pt permeability for carbon). The capsules were welded on both sides with dimensions ranging from 1.8 to 7.6 mm in length and 1.6 to 2.0 mm in outer diameter, depending on the type of press and the experiment. For the PC experiments, we used a standard ½″ assemblage with talc-Pyrex pressure medium, a graphite furnace and fired magnesia MgO spacers (e.g., Blanchard et al. 2015; see Supplementary Fig. S1). The temperature was monitored using a tungsten-rhenium W–Re thermocouple (type D, W-3% Re/W-25% Re) passing through a four holes alumina Al₂O₃ tube. For the MAP experiments, a 10/5 type assemblage was used, including a MgO octahedron, a lanthanum chromite LaCrO₃ insulating sleeve, a Re furnace, MgO spacers, a zirconia disk Zr₂O for thermal insulation and an alumina disk isolating the Pt capsule from the W–Re thermocouple (e.g., Siebert et al. 2011; see Supplementary Fig. S1). The thermocouple is guided by an alumina tube. These experiments were performed with a heating rate of about 100 °C min⁻¹, and quenching was achieved by turning off the furnace power. It is important to note that the utilization of different furnaces is believed to have an impact on the redox stability of silicates and opaque phases during serpentine dehydration experiments (Maurice et al. 2020), as further discussed.

Fig. 1

Experimental pressure–temperature stability field of antigorite and chlorite and experimental conditions. Thick lines indicate key dehydration reactions, i.e., the antigorite-out (dark green, solid) and the chlorite-out (light green, dashed) from experimental studies. Only selected results are shown. It includes, for antigorite: UT95: Ulmer and Trommsdorff (1995), WS97: Wunder and Schreyer (1997), BN98: Bose and Navrotsky (1998), BP03: Bromiley and Pawley (2003), P10: Padrón-Navarta et al. (2010); for chlorite: FP04: Fumagalli (2004), FP14: Fumagalli et al. (2014). Stars and polygons represent the experimental conditions used with the PC press (in the range 3 GPa and 700–1000 °C) and the MAP (in the range 7 GPa and 700–1000 °C), respectively. Slab surface P–T paths modelled by Syracuse et al. (2010) are shown using coloured arrows, with an average path for cold and hot subduction in blue and red, respectively. Graphite-diamond and quartz–coesite transitions are represented by dashed lines. Atg: antigorite, Chl: chlorite, Grt: garnet, Ol: olivine, Opx: orthopyroxene, PhA = phase A, Ta = talc

Characterization of run products

After quench and decompression of the experiments, the capsules were embedded in KM-U resin (Presi), hand polished with silicon carbide abrasion papers (SiC; 15.3, 10 and 5 µm grain size) and absolute ethanol until the experimental material outcrops and finally cleaned in an absolute ethanol ultrasonic bath.

Raman spectroscopy

Raman analyses were performed at IPGP with a Renishaw In Via Raman spectrometer coupled to an Olympus BX61 microscope, using the 514 nm wavelength of an argon ionized laser (HÜBNER Photonics) dispersed by a holographic grating of 1800 grooves mm⁻¹. The measurements were performed with a 50× objective (numerical aperture: 0.75), a configuration allowing to obtain a planar resolution close to 1 µm. The calibration of Raman bands was performed before each measurement session using a silicon microprocessor chip with a characteristic band at 520.4 cm⁻¹. Spectra were collected in the range of 130–4000 cm⁻¹ using a Renishaw CDD camera for punctual analysis. The laser power varied between 0.25, 2.5 and 12.5 mW, depending on the fluorescence level, the nature of the probed material (minerals vs. graphitic carbon) and the associated signal to noise ratio. For graphitic carbon analyses the power was set to the minimal value in order to avoid any radiation damage. Acquisition time was set to 10 s and each spectrum corresponds to 2 cumulative 10 s cycles. Each graphitic carbon spectrum presented in figures is an average of 5 spectra collected on identical compounds. The baseline of the spectra were realized with Peakfit 4.0 software (Jandel Scientific) over the 1000–1800 cm⁻¹ region, considering a cubic function. Deconvolution of bands characteristic of graphitic carbon was performed with the Wire 5.4 software (Renishaw).

Graphitic carbon Raman spectra can be decomposed into two regions, named first and second order regions (see Henry et al. 2019 for a review). The first order region, ranging between 1000 and 1800 cm⁻¹, is composed of two main bands D and G at about 1350 and 1580 cm⁻¹, respectively. The G band (for graphite) is related to the in-plane stretching vibration of carbon atoms in an aromatic ring with E_2g symmetry; this is the only band present in pure graphite. The D band (for defect), also called D1, is the result of structural and chemical defects in graphitic carbon, precisely the breathing motion of sp² C atoms in an aromatic ring with a A_1g symmetry (Tuinstra and Koenig 1970). In graphitic carbon, there are additional minor bands, also due to disorder, namely D2 (at about 1610 cm⁻¹), D3 (at about 1500 cm⁻¹), D4 (at about 1200 cm⁻¹), D5 (at about 1260 cm⁻¹) and D6 (at about 1440 cm⁻¹). The detailed interpretation of these additional bands D2, D3, D4, D5 and D6 is reviewed by Henry et al. (2019). Briefly, the D2 band is related to the presence of disorder in an aromatic ring with E_2g symmetry. The D3 band is related to out-of-plane vibration due to chemical defects. The D4 and D5 bands are linked to C-H moieties in aliphatic hydrocarbons, as is presumably the D6 band. The most visible of these bands is D2, which nonetheless overlaps with the G band in the case of low graphitization degree but becomes more distinct as the graphitization degree increases and the G band shifts to lower wavelengths.

The second order region, between 2400 and 3400 cm⁻¹, is composed of four bands S1, S2, S3 and S4 at about 2450, 2700, 2900 and 3180 cm⁻¹, respectively. The second order region is interpreted as overtones and combinations of the bands of the first order region (Beyssac et al. 2002; Wopenka and Pasteris 1993). For instance, the S2 band at 2700 cm⁻¹ is an overtone of the D-band at 1350 cm⁻¹.

The Raman spectra of graphitic carbon evolve with increasing crystallinity, until reaching the graphite structure. Especially, in the first order region, during graphitization, the D band shifts toward higher wavelengths, sharpens and increases in intensity to exceed the G band intensity by a factor of two before finally decreasing and disappearing upon pure graphite formation. The G-band shifts toward lower wavelengths, thickens a little before sharpening and increasing until it becomes the only visible band (Buseck and Beyssac 2014). In the second order region, at low crystallinity, there is a single broad band which disassembles as graphitization proceeds in S2, S3 and S4 bands. With the increase in crystallinity, S2 band is more intense than S3 and S4 and another small band S1 appears. In graphite, S2 band is split into two separate bands G′1 (at 2690 cm⁻¹) and G′2 (at 2735 cm⁻¹).

The deconvolution of first-order region allows to derive different parameters for monitoring the level of graphitization. Here, only G, D1, D2 and D3 bands have been deconvoluted using a Gaussian Lorentzian mixed function. The parameters used in this study are the following:

• D and G band separation value in cm⁻¹ (RBS with RBS = G-band position − D1-band position). RBS value is expected to decrease as the graphitization level increases.

• G and D1 band widths (i.e., full width at half maximum, FWHM). Upon graphitization, the D1-FWHM is decreasing while the G band is gently increasing in intensity before decreasing.

• D1 and G bands height ratio (R1 with R1 = D1 intensity/G intensity). During graphitization, R1 increases until D1 band is about twice more intense than G, then it decreases.

Polishing of the sample is known to increase graphitic carbon D-band intensity and have a controversial effect on its G-band (see Henry et al. 2019 for review) in graphitic carbon. However, such effects seem to be reduced when using abrasion paper (Pasteris 1989) rather than diamond-bearing polishing solutions. To further investigate potential artefacts during polishing, Raman analysis was carried out on one sample (PC1d) unpolished and polished and gives identical results (Supplementary Fig. S2). We therefore consider polishing effects as minimal.

Scanning electron microscopy

Scanning electron microscopy (SEM) images in backscattered electron (BSE) and secondary electron (SE) modes and energy dispersive X-ray spectroscopy (EDS) analyses were collected at IPGP on carbon coated sample surface using an Auriga 40 Zeiss microscope equipped with a Bruker QUANTAX 200 (125 eV resolution). Analytical conditions were 15 kV accelerating voltage, low or high current intensity (i.e., 1.5 and 3 nA) for imaging or EDS analysis, respectively, and a 8.5 mm working distance. Elemental distributions and images were processed using the ImageJ software (Schneider et al. 2012). Olivine, pyroxene and magnesite X_Mg values with X_Mg = (Mg + Fe)/Mg_tot value were derived from SEM–EDS measurements on run products, along with mineral composition in oxide weight percent (wt%).

Time of flight-secondary ion mass spectrometry

Time-of-flight secondary ion mass spectrometry (TOF–SIMS) measurements were carried out using a TOF–SIMS IV reflectron mass spectrometer (IONTOF GmbH) located at the Laboratoire d’Archéologie Moléculaire et Structurale (Sorbonne Université, CNRS, Paris, France). The instrument is equipped with a bismuth liquid metal ion gun delivering a pulsed beam of Bi₃⁺ cluster ions. Primary ions of 25 keV struck the sample at an angle of incidence of 45° and a pulsed current of 0.45 pA, with a repetition rate of 10 kHz. The emitted secondary ions were accelerated to 2 keV (extraction at 2 kV) towards a field-free region and a single-stage reflectron (first-order compensation). Secondary ions ejected from the top few monolayers of the sample surface were post-accelerated to 10 keV before reaching the detector comprising a microchannel plate, scintillator and photomultiplier. The focusing mode of the ion column ensured a spatial resolution of 2–5 μm and a mass resolution of 5000 (FWHM) at a mass-to-charge ratio (m/z) of 500. A low-energy (~20 eV) electron gun was used between two successive pulses of primary ions for charge compensation with minimum damage to the sample surface. An optical camera in the sample vacuum chamber was used to locate the samples, which were deposited on the sample holder without any treatment or adhesive. Sputtering of the sample surface was first carried out to remove potential contaminants using a pulsed argon Ar_n⁺ cluster ion beam of 10 keV energy (n ~ 2000, 5 eV per Ar atom) over an area of 500 µm × 500 µm (ion fluence of 3.1 × 10¹⁴ ions cm⁻²). In the center of this area, images of ions in negative and positive polarities were acquired over an m/z range from 0 to 800, in a raster pattern over areas of 200 µm × 200 μm and 128 × 128 pixels (pixel size 1.56 µm × 1.56 μm). Images were recorded with a primary ion fluence of 5.2 × 10¹¹ ions cm⁻² (45 scans with a cycle time of 100 μs). Data were acquired and processed using SurfaceLab 6.6 software (IONTOF GmbH). Spectra from the total analysis area were extracted. Internal mass calibration was performed using low-mass fragment ion signals from C⁺, CH⁺, CH₂⁺, CH₃⁺, C₂⁺, C₂H₃⁺, C₃⁺, C₄⁺ for the positive ion mode and from C⁻, CH⁻, CH₂⁻, C₂⁻, C₃⁻, C₄⁻, C₄H⁻ for the negative ion mode. Ion peaks were assigned according to instrument resolution, precision and valence rule. Some surface contaminations of easily ionized elements can occur and induce very intense peaks in the spectra (e.g., Na⁺, Mg⁺, K⁺, Ca⁺ in positive polarity spectra). These peaks associated with elements absent from the starting material were not taken into account in the list of the ten most intense peaks used to compare the different experiments.

Thermodynamic calculations

We use a Gibbs free energy minimization strategy (Connolly 2005, 2009) to predict, using PerpleX software, phase equilibria of the experiments. Phase diagrams in the CO₂–SiO₂–MgO–H₂O systems were calculated using the Holland and Powell (1998)’s thermodynamic database (revised in 2002, hp02ver.dat datafile). All the activity-composition (a–X) relationships of the phases are from Holland and Powell (2003) except for antigorite which derives from Padrón-Navarta et al. (2010). Model bulk composition derives from antigorite composition in Table 2, excluding aluminium and iron, considering a 50:50 mixing with pyrene (i.e., in mass amount CO₂: 10.8, SiO₂: 37.1, MgO: 33.7, H₂O: 18.4). The fluid was modelled using the equation of state of Holland and Powell (1991, 1998) and includes CH₄, H₂ and CO₂ species. Other organic compounds were not considered in the thermodynamic modelling as the current thermodynamic database of PerpleX at HP-HT is limited to methane (CH₄) and organic acids in fluids (Sverjensky et al. 2014), while the latter could decompose into solid organic compounds (e.g., Szlachta et al. 2022), for which no thermodynamic database is available.

Results

Evolution of pyrene and oxygen-functionalized PAHs at HP-HT

In order to assess the influence of P–T conditions, three experiments using pyrene as starting material were performed during 3 days with PC press at 700 °C and 3 GPa (PC1f), 1000 °C and 3 GPa (PC1c) and with MAP at 1000 °C and 7 GPa (MAP1a) (Table 1). The pyrene yellow powder originally inserted in the Pt capsule was replaced in the three experiments by a dark grey compact layered material, as shown in SEM (Fig. 2a–d). Raman analyses of run products derived from pyrene at increasing P–T conditions after 3 days are displayed in Fig. 2e. In all the experiments, pyrene turns into graphitic carbon characterized by two marked D and G bands in the first order region (in PC1f, at 1348 and 1598 cm⁻¹), and four bands S1, S2, S3 and S4 in the second order region (in PC1f, only S2, S3 and S4 are visible at 2686, 2937 and 3201 cm⁻¹ respectively). All pyrene characteristic Raman bands have been replaced (Supplementary Fig. S3). On the SEM images, it is observed that the increase in P–T conditions impacts the crystallinity of run products, showing a more layered material in PC1c and MAP1a (1000 °C and 3–7 GPa, respectively) compared to PC1f (700 °C–3 GPa). It comes together with the Raman spectra showing a substantial change in both the first and second order regions with increasing temperatures (Fig. 2e). At a constant pressure of 3 GPa, the RBS value decreases from 251 cm⁻¹ at 700 °C to 228 cm⁻¹ at 1000 °C (Table 1). Similar decrease with increasing temperature in observed for D1 and G band FWHM values (from 56 and 54 cm⁻¹ to 30–43 and 20–22 cm⁻¹, respectively) as well as R1 ratio (from 1.60 to 0.35–0.65). S1 band appears in the second order region at 2455 cm⁻¹. Conversely, at a constant temperature of 1000 °C and increasing pressure, few changes in the Raman spectra are observed (Fig. 2e). The RBS value is the same (228 cm⁻¹) in all the experiments while the R1 ratio is only decreasing from 0.65 at 3 GPa to 0.35 at 7 GPa (Table 1). Although there are slight variations in the D1-FWHM values (30 cm⁻¹ at 3 GPa vs. 43 cm⁻¹ at 7 GPa), the G-FWHM values are comparable in the experiments carried out at constant temperature (20–22 cm⁻¹). The only noticeable change is observed in the second order region, where the intensity of the S2 band at 2707 cm⁻¹, drastically decreases between 3 and 7 GPa (Fig. 2e).

Fig. 2

SEM (secondary electrons, SE) images of experiments performed with pyrene at increasing P and T (PC press and MAP) (a–d) and associated Raman spectra (e, f). a PC1f (700 °C, 3 GPa), b PC1c (1000 °C, 3 GPa), c,d MAP1a (1000 °C, 7 GPa), e pyrene evolution with pressure and temperature for a constant duration of 3 days (i.e., 3 GPa at 700 or 1000 °C (PC press) and 7 GPa at 1000 °C (MAP)). f Pyrene evolution with time at constant P and T (i.e., 1, 2, 3 and 7 days at 700 °C and 3 GPa; PC press). Graphitic layers are visible in all experiments, showing higher definition at higher P–T conditions. All collected spectra show the so-called first order D and G bands at ~about 1350 and ~1600 cm⁻¹ and the second order S1 to S4 bands in the range 2400–3100 cm⁻¹, all characteristic of graphitic carbon (Henry et al. 2019). Raman D and G band parameters are given in Table 1 together with details on experimental conditions. au: arbitrary unit

In order to highlight possible kinetic effects, four PC experiments were performed using pyrene as starting material at 700 °C and 3 GPa during 1, 2, 3 and 7 days (PC1e, PC1d, PC1f and PC1b, respectively; Table 1). Raman analyses of run products are displayed in Fig. 2f. In all the experiments, pyrene turns into graphitic carbon characterized by two marked D and G bands in the first order region at about 1348 and 1599 cm⁻¹, respectively, and three bands S2, S3 and S4 in the second order region at about 2686, 2937 and 3201 cm⁻¹. Graphic inspection of Raman spectra shows no substantial change with experiment duration. Systematic deconvolution of the collected spectra shows little variations of G and D1 band parameters (Table 1). The RBS value is about the same in all the experiments (250–252 cm⁻¹). D1-FWHM and G-FWHM values are also comparable between experiments (D1-FWHM = 51–56 cm⁻¹, G-FWHM = 52–56 cm⁻¹). The R1 ratio slightly increases from 1.54 after 2 days to 1.88 after 7 days. However, those variations are not very significant in view of spectra deconvolution error (see also Henry et al. (2019) who reported an error between 5 and 10% for R1 ratio). This overall suggests that equilibrium was reached as early as day 1 and that all experiments performed with PAHs alone can then be considered to be in equilibrium.

To evaluate the effect of PAH functionalization during HP-HT experiments, two series of experiments were conducted for each oxygen-functionalized PAH, at 700 °C, 3 GPa using PC press (PC5 and PC4 for 1-hydroxypyrene and PyBA; respectively) and 1000 °C, 7 GPa using MAP (MAP5 and MAP4 for 1-hydroxypyrene and PyBA, respectively, Table 1; Fig. 3a). SEM observation of the run product derived from 1-hydroxypyrene at 700 °C, 3 GPa shows the same compact aspect as pyrene run product (Fig. 2a). In contrast, for the same P–T conditions, the run product derived from PyBA contains many spherical cavities less than 2 µm wide distributed throughout the material (Fig. 4a) and few larger ones (>10 µm in diameter) (Fig. 4b). The biggest cavities are fractured, relating their formation prior to decompression. TOF–SIMS imaging shows that these textural heterogeneities are also associated with chemical heterogeneities (Fig. 4c), with in particular oxygen-bearing fragment ions C₃H₃O⁺ preferentially localized in circular shapes and aliphatic fragment ions C₄H₇⁺ at their periphery in the bulk organic matrix. At 1000 °C and 7 GPa, the run products show for both oxygen-functionalized PAHs the same compact material as pyrene under the same conditions, with very well-defined layers of graphitic carbon (Fig. 2c, d). Associated Raman spectra are compared to those of pyrene in Fig. 3a under the same P–T conditions. Graphic inspection of the spectra shows no discernible differences at both P–T conditions, as confirmed by D1 and G band parameters (Table 1). Indeed, the RBS, R1 and G-FWHM values are very close for the three PAH types for the same P–T conditions (at 700 °C, 3 GPa: RBS = 250 cm⁻¹, R1 = 1.60–1.69, G-FWHM = 54–55 cm⁻¹; at 1000 °C, 7 GPa: RBS = 227–228 cm⁻¹, R1 = 0.33–0.51, G-FWHM = 43–45 cm⁻¹; Table 1), overall indicating similar crystallinity degree of the graphitic carbon obtained in the three experiment series.

Fig. 3

Raman spectra (a) and TOF–SIMS negative polarity spectra (b, c) collected on run products obtained using PAHs of a single type as starting material in PC press or MAP. a Comparison of the evolution of pyrene and two oxygen-functionalized PAHs (1-hydroxypyrene and 1-pyrene butyric acid, PyBA) at 700 °C and 3 GPa (PC) or 1000 °C and 7 GPa (MAP). Raman spectra show graphitic carbon bands that are very similar, indicating an equivalent graphitization level. See Fig. 2 for band assignment and Table 1 for Raman D and G band parameters. b,c TOF–SIMS negative polarity spectra (m/z of 0–100) of run products obtained using pyrene (b) and PyBA (c) at 700 °C and 3 GPa (PC). The first ten most intense fragment ions are indicated (see Supplementary for mass deviation). They are mainly composed of short carbon chains and oxygen-functionalized aliphatic moieties for pyrene and PyBa, respectively. au: arbitrary unit

Fig. 4

Chemical and textural heterogeneities in the run product obtained at 700 °C and 3 GPa using PyBA (PC4). a,b SEM-SE images of spherical cavities in PC4. Micrometer and tens of micrometer round holes are observable all over the polished capsule. c Two-color overlay TOF–SIMS ion images composed of the fragment ions C₃H₃O⁺ and C₄H₇⁺ and corresponding individual ion images. Fragment ions are unevenly distributed, with C₃H₃O⁺ fragment ions preferentially emitted at circular shapes and C₄H₇.⁺ fragment ions at their periphery (see Supplementary Table S1 for mass deviation)

TOF–SIMS spectra have been collected on run products obtained from pyrene and PyBA at 700 °C and 3 GPa (PC1f and PC4, respectively). Respective negative polarity spectra over the range m/z = 0–100, where most fragment ions have been detected and identified, are shown in Fig. 3b and c (see also Supplementary Table S1). PC1f TOF–SIMS spectrum is mainly composed of short carbon chains (i.e., C⁻, C₂⁻, C₃⁻, C₄⁻, C₆⁻ fragment ions) and aliphatic moieties (i.e., CH⁻, C₂H⁻, C₄H⁻ fragment ions). Conversely, PC4 TOF–SIMS spectrum shows mainly oxygen-bearing fragment ions (i.e., O⁻, OH⁻, CH₃O⁻ C₂HO⁻, C₃H₃O⁻ C₄H₅O₂⁻) in addition to short chain aliphatic moieties (i.e., CH⁻, C₂H⁻, C₄H⁻). Similar trends were seen in the positive polarity spectra (Supplementary Fig. S4a–b; Table S1).

Evolution of pyrene and antigorite mixing at HP-HT

To assess the impact of serpentinite dehydration on PAH P–T evolution, seven experiments were conducted using a 50:50 mixing (by mass) of pyrene and natural antigorite powders at temperatures ranging from 700 to 1000 °C and pressure of 3 GPa in PC press (PC2a, PC2d, and PC2b) and from 700 to 1000 °C and 7 GPa in MAP (MAP2b, MAP2a, MAP2c, MAP2e), beyond antigorite and chlorite stability fields, respectively (Table 1; Fig. 1). In all experiments, the pyrene and antigorite mixing turned into an assemblage of silicate- and graphitic carbon domains of about several hundred micrometers in size (Fig. 5).

Fig. 5

Composite SEM–EDS chemical maps (C (red), Mg (blue), Al (cyan), Si (yellow); left) and associated Raman spectra (right) of pyrene-antigorite mixing (50:50 by mass) PC press and MAP experiments performed at increasing P and T. a PC2a (700 °C, 3 GPa), b MAP2b (700 °C, 7 GPa), c PC2b (1000 °C, 3 GPa), and d MAP2a (1000 °C, 7 GPa; see Table 1 for associated experimental conditions and Raman D and G band parameters, as well as Fig. 2 for band assignment). Magnesite formation is only observed in the MAP experiments at 7 GPa and 700 or 1000 °C (see also Fig. 6). CGr: graphitic carbon, Chl: chlorite, En: enstatite, Fo: forsterite, Mgs: magnesite, Prp: pyrope. au: arbitrary unit

At 3 GPa and 700 °C (PC2a and PC2d; Fig. 5a), silicate domains are made of microcrystalline assemblages of pyroxene (i.e., enstatite with X_Mg values ranging from 0.94.to 0.95, representing around 20% of the run products visible at the surface of the polished capsule), olivine (forsterite with X_Mg values ranging from 0.91 to 0.94; around 20%) and chlorite (close to clinochlore end-member; <10%), as established by SEM–EDS (Supplementary Table S2) and Raman analysis. These assemblages are classic products of antigorite breakdown predicted by thermodynamic calculations or established experimentally (Fig. 1).

At 3 GPa and 1000 °C (PC2b; Fig. 5c), after chlorite breakdown, only the anhydrous silicate phases remain, i.e., enstatite and olivine with X_Mg values of 0.93 and 0.95, respectively. Enstatite crystals form lamellae or needles (up to 40 µm in length). No aluminium-rich phase was observed in the quenched experiment. Indeed, the bulk Al₂O₃ content is very low (Supplementary Table S2), so is the expected amount of garnet. Furthermore, the Al₂O₃ content is much higher in the enstatite obtained at 1000 °C than in that obtained at 700 °C (4.6 wt% at 1000 °C vs. 1.6–3.0 wt% at 700 °C), hence limiting garnet formation.

At 7 GPa and 700 °C (MAP2b; Fig. 5b), silicate assemblages are mainly composed of enstatite (around 25% of the run products) and olivine (<15% with X_Mg value of 0.94, comparable to the ones obtained in PC experiments at the same temperature). An aluminium-rich phase of about ~1 µm in size is observed at the center of the capsule, crystallizing between graphitic carbon layers and representing less than 2% of the capsule surface. Because of its small size, it was not possible to identify this Al-rich phase by Raman spectroscopy. Its chemical composition supports nonetheless the presence of pyrope garnet (Table S2). The Al₂O₃ content in enstatite is comparable to the one in PC experiments at 3 GPa and 700 °C (2.8 wt% at 7 GPa vs. 1.6–3.0 wt% at 3 GPa). Magnesite (X_Mg = 0.89) is also associated with both silicates and graphitic carbon sheets (Fig. 6) and accounts for about 5–10% of the run products.

Fig. 6

Magnesite formed in MAP experiments (7 GPa, 700 and 1000 °C) involving a pyrene-antigorite mixing (50:50 by mass). a Orientated magnesite formation in graphitic carbon or silicates domains (SEM-BSE). b Interplanar magnesite formation in graphitic carbon (SEM-SE). CGr: graphitic carbon, Mgs: magnesite

At 7 GPa and 1000 °C (MAP2a, MAP2c, MAP2e; Fig. 5d), the silicate minerals present notable differences in composition and relative proportions. Enstatite and garnet are always observed, representing respectively about 20–40% (depending on the experiment) and 5% of the run products. Enstatite Al₂O₃ content ranges from 0.3 to 1.5 wt%, which is less than at 700 °C at the same pressure (2.8 wt%). Conversely, X_Mg values, ranging from 0.95 to 1.00, are higher (contrarily to what is observed at 3 GPa and 1000 °C). Garnet compositions are very similar when comparing similar P–T conditions and fall between that of pyrope MgAl₂Si₃O₁₂ and almandine Fe²⁺Al₂Si₃O₁₂ (e.g., for MAP2a: 87% pyrope vs. 13% almandine based on the chemical compositions using the Droop method, Droop 1987). Olivine and magnesite relative amounts are highly variable in the run products and X_Mg values of olivine are slightly higher than at 700 °C (as for 3 GPa and 1000 °C) while X_Mg values of magnesite are extremely higher (X_Mg = 1) in the three experiment. In MAP2e, olivine (X_Mg = 0.99) represents about 25% of the run products while magnesite is found mainly in graphitic carbon layers in very small quantities (about 5%). In MAP2a, olivine (X_Mg = 0.95) represents less than 15% of the run products and crystallized in and next to the silicate assemblages, between the graphitic carbon sheets (Fig. 5d). Magnesite occurs mainly in graphitic carbon but is also observed in silicate assemblages (Fig. 6) and represents up to 5–10% of the run products. In MAP2c, olivine is absent. Magnesite represents up to 30% of the run products, and crystallized both in silicate assemblages and between the graphitic carbon sheets. Moreover, coesite is also observed in this experiment, representing up to 5–10% of the run products.

We assume that local equilibrium is reached in most experiments (except in PC2b and MAP2b where garnet is not observed), as the observed silicate assemblages are in agreement with standard dehydration assemblages observed, (Ulmer and Trommsdorff 1995; Bromiley and Pawley 2003; Pawley 2003).

In all the experiments, pyrene has been replaced by graphitic carbon, variously sheeted depending on the temperature and pressure. Raman spectra show D and G bands in the first order region and S1, S2, S3 and S4 bands in the second order one (Fig. 5). As shown in Table 1 and observed with pyrene experiments, Raman parameters associated with D1 and G bands evolve significantly with temperature (e.g., at 3 GPa, between PC2a and PC2b experiments; Fig. 5a, c.) and less with pressure (e.g., at 700 °C, between PC2a and MAP2b experiments; Fig. 5a, b). For instance, at 700 °C and 3 or 7 GPa, RBS values range between 250 and 245 cm⁻¹ while they fall between 225 and 228 cm⁻¹ at 1000 °C and 3 or 7 GPa. Compared to the experiment performed only with pyrene, at 700 °C–3 GPa and 1000–700 °C, the RBS and G-FWHM values are equivalent for the same P–T conditions (about 250 cm⁻¹ and 55 cm⁻¹ at 700 °C and 3 GPa and 228 cm⁻¹ and 21 cm⁻¹ at 1000 °C and 7 GPa) (Table 1). R1 values are slightly different. At 700 °C and 3 GPa, R1 ratio is higher (1.95 for PC2a vs. 1.60 for PC1f) while it is lower at 1000 °C and 3 GPa (0.19 for PC2b vs. 0.65 for PC1c). At 1000 °C and 7 GPa, R1 values (0.21 and 0.55) are around the one of MAP1a (0.35).

TOF–SIMS spectra have been collected on run products of pyrene and antigorite at 1000 °C and 3 GPa (PC2b; Supplementary Fig. S4e, f; Table S1) and compared to the one obtained for pyrene at 1000 °C and 7 GPa (MAP1; Fig S4c, d; Table S1). Compared to the positive polarity TOF–SIMS spectrum of MAP1 where only short carbon and aliphatic chains are observed, the positive polarity TOF–SIMS spectrum of PC2b shows oxygen-bearing fragment ions (C₂H₃O₂⁺, C₄H₅O⁺). The negative polarity TOF–SIMS spectrum of both experiments are similar, the main difference being the presence of one oxygen-bearing fragment ion (C₃H₅N₂O⁻) in PC2b.

Oxygen fugacity prevailing in antigorite-pyrene experiments

The oxygen fugacity can be constrained in the MAP experiments involving antigorite-pyrene mixing when either magnesite or coesite are observed in the obtained mineralogical assemblages.

In MAP2a, MAP2b, and MAP2e experiments where magnesite is present, fO₂ is calculated using the equilibrium EMOG between enstatite, magnesite, olivine (forsterite) and graphitic carbon C_gr (approximated to graphite):

$${\mathop {Mg_2 SiO_4 }\limits_{Forsterite}} + C_{gr} + O_2 \leftrightarrow {\mathop {MgSiO_3 }\limits_{Enstatite}} + {\mathop {MgCO_3 }\limits_{Magnesite}}$$

(1)

The equilibrium constant K can be expressed as:

$$\begin{array}{*{20}c} {ln\,K = \frac{ - \Delta_r G_1^\circ (P,T)}{{RT}} = ln\frac{aMgSiO_3 \times aMgCO_3 }{{aMg_2 SiO_4 \times aC_{gr} \times aO_2 }}} \\ \end{array}$$

(2)

with $- \Delta_r G_1^\circ$, the Gibbs free energy in J mol⁻¹, $P$, the pressure in Pa, $T$, the temperature in Kelvin, $R$, the gas constant value and $a$, the activities of components. Activity of graphite is approximated to unity while the activities of enstatite, magnesite and forsterite are calculated using a Margules model with mixing on two sites (as in Holland and Powell 1998). Accordingly, activity of a specie A in a phase χ is $a_A^\chi = X_A^{2 } \gamma_A$ with X the mole fraction of the specie A, $\gamma_A$, the activity coefficient defined from the expression $RT\,\ln \,\gamma_A = W_{A - B}^\chi X_B^2$ with and W, the Margules parameter.

Magnesite activity is only calculated at 700 °C because it is found pure at 1000 °C (its activity is then approximated to unity at this temperature). Therefore, it follows:

$$\log_{10} fO_2 = \frac{\Delta_r G_1^\circ (P,T)}{{RT\,\ln (10)}} + \log_{10} a_{MgSiO_3 } + \log_{10} a_{MgCO_3 } - \log_{10} a_{Mg_2 SiO_4 }$$

(3)

In MAP2c experiment where coesite is present, the oxygen fugacity is calculated using the equilibrium MCEG between magnesite, coesite, enstatite and graphitic carbon (approximated to graphite):

$${\mathop {MgSiO_3 }\limits_{Enstatite}} + C_{gr} + O_2 \leftrightarrow {\mathop {MgCO_3 }\limits_{Magnesite}} + {\mathop {SiO_2 }\limits_{Coesite}}$$

(4)

Yet again, activity of coesite and magnesite are approximated to unity as theses phases are found pure in composition while activity of enstatite is calculated as for EMOG equilibrium, so it follows:

$$\begin{array}{*{20}c} {\log_{10} fO_2 = \frac{\Delta_r G_2^\circ (P,T)}{{RT \ln (10)}} - \log_{10} a_{MgSiO_3 } } \\ \end{array}$$

(5)

Gibbs free energies $G_1^\circ$ and $G_2^\circ$ for reactions (1) and (4), respectively, are calculated using the thermodynamic parameters provided in Supplementary Table S3. The calculated oxygen fugacities are presented in Table 3. Oxygen fugacity values are also given as ΔQFM. The quartz-fayalite-magnetite QFM buffer is calculated at 7 GPa and 700/1000 °C using the formulation of QFM provided by Ballhaus et al. (1991) and O’Neill (1987) and the data from Holland and Powell (1998), available in Maurice et al. (2020) (see Table 3).

Table 3 Estimated redox conditions during MAP experiments with associated experimental conditions, redox buffer used for calculations, and absolute and ΔQFM fO₂ values

Discussion

P–T evolution of pyrene and two oxygen-functionalized PAHs is studied in order to assess the fate of abiotic solid organic compounds in subduction zones. These compounds were chosen since their structure and chemistry are comparable to those of abiotic CCM observed in abyssal (Andreani and Ménez 2019; Sforna et al. 2018; Andreani et al. 2023) and subduction zone (Nan et al. 2020; Debret et al. 2022) serpentinites. Low molecular weight PAHs such as pyrene and anthracene are also commonly used as simple analogues of abiotic carbonaceous matter of hydrothermal origin for thermodynamic calculations (e.g., Milesi et al. 2016; Andreani et al. 2023). Here, we first discuss the evolution of PAHs under subduction P–T conditions, and then the influence of serpentinite dehydration reactions on the metastability of PAHs.

PAHs graphitization under subduction P–T conditions

With increasing P–T conditions, pyrene Raman spectra evolve, in the first order region, from four sharp main bands at 1240, 1405, 1593 and 1626 cm⁻¹ (Supplementary Fig. S3) to two broad D and G bands at about 1350 and 1600 cm⁻¹ (Fig. 2e, f; Table 1) showing that the starting material turns into graphitic carbon. Graphitization can proceed through either oxidation, pyrolysis and dehydrogenation reactions (Zolotov and Shock 1999), all impacting the PAH carbon redox state of PAHs and hence their stability. Chanyshev et al. (2015, 2017a; b) already investigated pyrene graphitization with increasing P–T conditions (from 3 to 9 GPa, and 200 to 1000 °C) and propose that decomposition of PAHs is mainly controlled by condensation and dehydrogenation reactions of benzene rings into graphitic carbon. Although investigated P–T conditions are different here, we observe identical Raman spectra evolution with increasing P–T conditions, suggesting similar mechanisms affecting pyrene.

By comparing the Raman spectra of PC1f (700 °C, 3 GPa), PC1c (1000 °C, 3GPa) and MAP1a (1000 °C, 7 GPa) (Fig. 2e) and associated derived parameters, temperature seems to be the main parameter controlling PAHs graphitization while pressure has a reduced effect. Indeed, large differences in the Raman parameters are observed at 3 GPa between 700 and 1000 °C (RBS and G-FWHM values decrease from 251 to 228 cm⁻¹ and 54 to 22 cm⁻¹, respectively; Table 1) while there is little difference at 1000 °C between 3 and 7 GPa (RBS = 228 cm⁻¹ at both P conditions; G-FWHM values decreases from 22 to 20 cm⁻¹) regardless of the duration of the experiment (Table 1). Similar conclusions are also drawn by Chanyshev et al. (2015), who performed HP-HT experiments using various PAHs (pyrene, naphthalene, acenaphthene C₁₂H₁₀, phenanthrene C₁₄H₁₀, fluoranthene C₁₆H₁₀, benzo[a]pyrene C₂₀H₁₂ and coronene C₂₄H₁₂) at 16 GPa/25 °C and 7 GPa/500–600 °C. Overall, no pure graphite formation from PAHs is reported here or in previous experiments at similar P–T conditions (Chanyshev et al. 2015, 2017a; b). Davydov et al. (2004) report the formation of pure graphite above 1200 °C from different PAHs (i.e., naphthalene, anthracene, pentacene C₂₂H₁₄, perylene C₂₀H₁₂, and coronene), hence well above the temperatures investigated here. It is interesting to note that investigations on graphitization from natural samples predict the formation of pure graphite from kerogen at temperature close to 700 °C (Beyssac et al. 2002). The persistence of a large structural disorder in the graphitic carbon obtained in experiments involving PAHs and performed below 1200 °C can be attributed either to kinetics effects associated with the experimental settings and/or to the non-graphitizable nature of the material (Beyssac and Rumble 2014; Franklin 1951).

Possible kinetic effects were tested through a series of experiments of different durations (from 1 to 7 days) performed at 700 °C and 3 GPa (Fig. 2f; Table 1). It should be noted that kinetics experiments aim at testing the graphitic carbon meta-stability, as high activation energies and kinetic factors limit comparisons between laboratory experiments and natural graphitization (Buseck and Beyssac 2014). For all these experiments, the Raman parameters of the obtained graphitic carbon do not significantly evolve, suggesting a similar structural state (RBS, G-FWHM and R1 values all in the range 250–252 cm⁻¹, 52–56 cm⁻¹, and 1.54–1.88, respectively; Table 1). Moreover, Davydov et al. (2004) report Raman spectrum of naphthalene after 60 s at 8 GPa and 1000 °C, which is very similar to the one obtained here for pyrene after 72 h at 1000 °C and 7 GPa (RBS = 223 cm⁻¹ for naphtalene vs. 228 cm⁻¹ here for pyrene). Overall, this suggests that the obtained graphitic carbon is at meta-equilibrium. Previous studies have shown that pressure accelerates graphitization in laboratory experiments, facilitating graphite formation at temperature of about 1000 °C from graphitizable materials (Beyssac et al. 2003 and reference therein). Hence, the preservation of a high level of disorder in pyrene-derived graphitic carbon at 1000 °C suggest pyrene is a non-graphitizable material. This disorder is likely due to the persistence of H atoms in its structure, as supported by TOF–SIMS analyses (Fig. 3a; Fig. S3a, c–d).

Both oxygen-functionalized PAHs namely 1-hydroxypyrene and PyBA show Raman spectra evolution identical to that of pyrene with increasing pressure and temperature (for instance at 700 °C and 3 GPa, RBS and G-FWHM values are all in the range 250–251 cm⁻¹ and 54–55 cm⁻¹, respectively; Table 1; Fig. 3a). Hence, the presence of oxygen does not appear to influence the graphitization level of PAHs. It must however be noted that the investigated PAHs contain heteroatoms only in the form of substituted oxygen-bearing functions. These are more easily removable than the O-bearing heterocycles potentially included in natural CCM. Nevertheless, TOF–SIMS data (Figs. 3c, 4c; Supplementary Fig. S4b) show that, in the graphitic carbon structure derived from PyBA experiment at 700 °C and 3 GPa, some oxygen is preserved and is heterogeneously distributed. Therefore, the high level of disorder preserved in graphitic carbon produced by oxygen-functionalized PAHs could be due to structural oxygen together with hydrogen (as observed in pyrene). Although we have not performed kinetics experiments on oxygen-functionalized PAHs, we assume that these PAHs are also non-graphitizable materials because (1) they present Raman spectra similar to those of pyrene at 1000 °C and comparable spectra evolution with increasing P–T (Fig. 3a) and (2) oxygen-bearing functions promote the non-graphitizable nature of organic compounds during subduction (Buseck and Beyssac 2014).

Interestingly, spherical cavities are observed in PC experiments using PyBA (Fig. 4a, b), the most oxygen-rich PAH investigated. As some of these are intersected by fractures (Fig. 4b), these cavities must have formed before the quench, which induces rapid decompression. The presence of round cavities formed at HP-HT and their enrichment in O relative to the bulk matrix revealed by TOF–SIMS imaging (Fig. 4c) therefore suggest the formation of an aqueous phase (H-, and O-rich) during PyBA graphitization at HP-HT. Similar observations are reported by Nakano et al. (2020) who observed the formation of water droplets during heating at temperatures < 350 °C of an analogue to precometary-organic-matter corresponding to a mixing of functionalized aliphatic and aromatic compounds. It is therefore possible to propose here a similar scenario with some oxygen-rich moieties exsolving as water during graphitization of oxygen-functionalized PAHs.

The three PAHs explored in our study, used as analogues of abiotic CCM in serpentinite (Sforna et al. 2018; Debret et al. 2022; Andreani et al. 2023), behave as non-graphitizable materials, preserving high level of disorder in the graphitic carbon they form at increasing pressure and temperature in experiments. In subduction zones, this non-graphitizable nature could allow the transfer at depths of volatiles (e.g., O, H or N), as it was shown in previous study on non-graphitizable materials (Nakamura et al. 2020). Moreover, the comparative Raman analysis of the experiments involving oxygen-functionalized PAHs and pyrene (Fig. 3a), as well as the presence of oxygen-bearing fragment ions (Fig. 3c) and spherical cavities in the graphitic carbon obtained at 700 °C and 3 GPa in the PyBA experiment (Fig. 4a, b.), suggest a quick partial removal of oxygenated moieties from oxygen-bearing CCM with increasing temperature and pressure, certainly as aqueous water (see also Nakano et al. 2020). Overall, this suggests enhanced reactivity at shallow levels in subduction zones as well as and possible remaining-oxygen transfer to the deep mantle, as also supported by HP-HT solubility experiments carried out at 1 GPa, 550 °C with carbonaceous materials by Toffolo et al (2023).

In addition to oxygen, serpentinites are also known to play a major role in the recycling of other volatile and redox sensitive element to the deep Earth (e.g., Cannaò et al. 2020; Bouilhol et al. 2022). CCM may also preserve, if initially present, other heteroatoms such N and S and is therefore able of carrying part of the serpentinite volatile and redox budget to the deep mantle. For example, several studies report high nitrogen contents in lithologies from both abyssal (Philippot et al. 2007) or subduction contexts (>30 ppm; Pagé et al. 2018; Cannaò et al. 2020) and claim that serpentinites can therefore represent up to 60% of nitrogen fluxes to the deep Earth. Cannaò et al. (2020) also underline that HP oceanic rocks are likely to retain their nitrogen content beyond antigorite breakdown (i.e., 700 °C) suggesting that N is carried by secondary phases. It is well established that CCM has the potential to incorporate nitrogen into its structure (McCollom and Seewald 2007). Preserving heteroatoms, such as nitrogen, within solid organic compounds could potentially provide a new path for the deep nitrogen cycle.

On the P–T evolution of pyrene-antigorite mixing

During subduction, with the gradual increase in P–T conditions, serpentinites undergo successive dehydration reactions (i.e., brucite, antigorite, and chlorite breakdown) producing volatile-rich fluids (Ulmer and Trommsdorff 1995). These fluids may interact with abiotic CCM of hydrothermal origin enclosed in the serpentinites and enhance its solubility in slab-derived fluids, especially under oxidizing conditions (Debret et al. 2015; Debret and Sverjensky 2017; Duan et al. 2022; Iacovino et al. 2020; Maurice et al. 2020). Redox conditions are therefore of prime interest to constrain the fate of carbonaceous matter in subduction zones and hence, the deep carbon cycle.

As illustrated in Fig. 1, during the increase in P–T conditions, in MgO-Al₂O₃–SiO₂–H₂O (MASH) system, antigorite breaks down into forsterite, enstatite, and chlorite at 600–700 °C and 3 GPa (Ulmer and Trommsdorff 1995; Bromiley and Pawley 2003). This can be expressed as:

$$\begin{array}{*{20}c} {antigorite = forsterite + enstatite + chlorite + H_2 O } \\ \end{array}$$

(6)

A further increase in P–T conditions, over 860 °C and 3 GPa, leads to the breakdown of chlorite releasing additional aqueous fluids (Pawley 2003) following:

$$\begin{array}{*{20}c} {chlorite + enstatite = pyrope + forsterite + H_2 O} \\ \end{array}$$

(7)

In PC experiments involving a mixing of antigorite and pyrene at 700 °C and 3 GPa (PC2a, PC2d), we observe the formation of classical silicate assemblages following the breakdown of antigorite (Eq. 6), including forsterite, enstatite and chlorite (Fig. 5a; Table 1) and the breakdown of chlorite. PC experiment at 1000 °C and 3 GPa (PC2b) leads also unsurprisingly to forsterite and enstatite after chlorite breakdown (Fig. 5c; Table 1). Hence, the presence of pyrene did not influence the silicate paragenesis at HP-HT. Similarly, pyrene graphitization is not modified by dehydration reactions, at least when considering the structural evolution of the graphitic carbon formed. Raman spectra show the same graphitization level as in the pyrene experimental series (RBS and G-FWHM values are all in the range 250–251 cm⁻¹ and 54–55 cm⁻¹, respectively; Table 1). However, TOF–SIMS spectra of PC2b (Supplementary Fig. S4e–f) show the presence of oxygen-bearing fragment ions which were absent in those obtained for the pyrene experiments, indicating that oxygen is incorporated into the graphitic carbon following antigorite dehydration, overall supporting potential reactivity. Nevertheless, graphitic carbon and silicates are localised in distinct areas and no graphitic carbon is observed within the silicate domains (Fig. 5a c). This suggests that pyrene is poorly solubilized in the fluid phase produced by antigorite and chlorite breakdown in PC experiments.

Conversely, in MAP experiments, in addition to classical dehydration products formed after antigorite and chlorite breakdowns, i.e., enstatite, pyrope ± forsterite (Eqs. 6 and 7; Fig. 1), the formation of magnesite is observed in all the experiments, at both 700 °C and 1000 °C. Magnesite forms in variable relative proportions both at the expense of graphitic carbon and as an interstitial phase in silicate domains (Figs. 5b, d, 6). These observations support ongoing reactions between silicate and pyrene domains leading to the oxidation of pyrene-derived organic phases into carbonates. The residual graphitic carbon shows similar graphitization levels relative to the experimental series carried out with pyrene alone (identical RBS values of 228 cm⁻¹ and G-FWHM values in the range 20–27 cm⁻¹, for the experiments performed at 1000 °C and 7 GPa; Table 1). Magnesite formation can be explained by the consumption of forsterite formed by antigorite dehydration and the oxidation of graphitic carbon, expressed in the CO₂–MgO–SiO₂–H₂O system as:

$${\mathop {Mg_3 Si_2 O_5 (OH)_4 }\limits_{Antigorite}} + C_{gr} + O_2 = {\mathop {MgCO_3 }\limits_{Magnesite}} + {\mathop { 2MgSiO_3 }\limits_{Enstatite}} + 2H_2 O$$

(8)

The nature of the fluid, here expressed as H₂O will depend on the redox conditions, which are not controlled here.

Coesite is also observed in experiment MAP2c in the silicate paragenesis without forsterite suggesting competition between the two phases under these P–T conditions.

Overall, two major features emerge from the HP-HT experiments involving antigorite-pyrene mixing. Firstly, the formation of magnesite is evident in MAP experiments, whereas this phase is absent in PC experiments. Secondly, there is a variability in the silicate assemblages (i.e., presence of forsterite vs. coesite) formed at 1000 °C and 7 GPa using MAP. Both features could be explained by variable redox conditions either among MAP experiments or in between PC and MAP experiments. The influence of fO₂ on silicate and carbonate assemblages was approached through PerpleX pseudosections in the simplified CO₂–MgO–SiO₂–H₂O system at 3 and 7 GPa (Fig. 7). At low oxygen fugacity, organic carbon (represented by CH₄, as detailed in the “Materials and methods” section) remains stable in the presence of enstatite and forsterite. In contrast, a higher oxygen fugacity favours the formation of magnesite, either in equilibrium with olivine or coesite (in addition to enstatite), as oxygen fugacity increases. This suggests that there are variable fO₂ conditions either among MAP experiment and between PC and MAP experiments. PC experiments appear to be equilibrated at low oxygen fugacity in the CH₄-Fo-En field, whereas MAP experiments seem to be equilibrated at higher oxygen fugacity, either in the Mgs-Fo-En or the Mgs-En-Coe fields. These pseudosections show that oxygen fugacity is a critical parameter affecting the stability of carbon phases in the experiments.

Fig. 7

Phase diagram pseudosections in the CO₂–SiO₂–MgO–H₂O systems for an antigorite-pyrene mixing (50:50 by mass) at a 3 GPa and b 7 GPa plotting temperature vs. oxygen fugacity (fO₂). In both pseudosections, at low oxygen fugacity (i.e., below about −32 and −20 at 700 °C and 1000 °C, respectively), magnesite formation is not achievable. As oxygen fugacity increases, magnesite appears at the expense of forsterite, and coesite stabilises. The composition of the fluid change, from CH₄-dominated at low oxygen fugacity to H₂O-dominated at high oxygen fugacity. Polygons represent the oxygen fugacity of MAP experiments calculated based on the equilibrium EMOG buffer (red polygons for experiments MAP2a, MAP2b and MAP2e) or MCEG buffer (green polygon for experiment MAP2c) (Table 3). White arrow represents the fO₂ imposed by the PC graphite furnace based on Hermann and Spandler (2008). Cobalt-cobalt oxide (Co-CoO), quartz-fayalite-magnetite (QFM), and magnetite-hematite (MH) buffers are given for better readability. There were calculated and ploted thanks to oxygen buffer calculator Copyright © 2021 Michael Anenburg, available at: https://fo2.rses.anu.edu.au/fo2app/, consulted the 28/07/24. RRO buffer should impose fO₂ within a range of ΔQFM −1 and ΔQFM +3, between NNO and MH buffers. Light blue fields are magnesite-bearing fields. Coe: coesite, En: enstatite, Fo: forsterite, Mgs: magnesite, Qz: quartz, Tlc: talc

This raises the question of what buffers the oxygen fugacity in both PC and MAP experiments, with fO₂ being either internally regulated by the newly-formed assemblages or externally controlled by the experimental device.

During experiments, antigorite dehydration products or graphitizing pyrene can internally buffer fO₂. The redox state and redox capacity of fluids produced by antigorite dehydration have been the subject of recent controversies (Debret and Sverjensky 2017; Merkulova et al. 2017; Bretscher et al. 2018; Piccoli et al. 2019; Iacovino et al. 2020; Maurice et al. 2020; Evans and Frost 2021; Duan et al. 2022). The presence of hematite or hemo-ilmenite in HP-HT oceanic rock samples implies that the breakdown of antigorite can produce fluids that equilibrate nearby ΔQFM +3 (Debret et al. 2015). This suggests that such fluids have the potential to oxidize graphitizing pyrene into magnesite, as observed in Fig. 6. The formation of oxidizing fluids during antigorite breakdown is attributed to the reduction of iron in serpentinites (Debret et al. 2014; 2015).

$$Fe_2 O_3 (s) + H_2 (aq) = 2FeO(s) + 2H_2 O(aq)$$

(9)

This reduction has to be coupled with the oxidation of a reduced component, which can be express in our experiment as:

$$C_{gr} (s) + H_2 O(aq) = CO_2 (s) + H_2 (aq)$$

(10)

Such a redox process can therefore be expressed as:

$$2Fe_2 O_{3\, antigorite } + C_{graphitizing\,pyrene} = 4FeO_{antigorite} + CO_2$$

(11)

During the experiment, the oxidation of pyrene is therefore controlled by the Fe³⁺/ΣFe ratio in antigorite (as magnetite has been manually removed, see “Materials and methods” section). The initial iron redox state of the antigorite used is 0.2, which corresponds to approximately 4.0 10⁻⁵ mol of antigorite capable of oxidizing 4.7⋅10⁻³ mol of carbon (see Supplementary Methods for detailed calculations). This suggests that iron reduction can only account for the oxidation of less than 1% of the total carbon content in our experiments. Consequently, this process seems to be unrealistic in explaining the large amount of magnesite observed (Fig. 6). Moreover, antigorite iron reduction cannot explain the absence of magnesite in PC experiments.

The other internal buffer of fO₂ is the graphitic carbon. In natural settings, an increasing body of literature shows occurrences of solid organic compounds under reduced conditions in subduction settings. As an example, Vitale Brovarone et al. (2017) report the formation of graphitic carbon in meta-ophicarbonates from the Lanzo Massif (Western Alps) linked to the circulation of highly reducing H₂-bearing fluids. Recently, abiotic disordered carbonaceous matter was found in meta-ophicarbonates from the Monviso meta-ophiolite (Western Alps) that recorded low fO₂ conditions (ΔQFM −4) (Caurant et al. 2023). These disordered carbonaceous compounds (Vitale Brovarone et al. 2017; Caurant et al. 2023) can lower down the fO₂ thanks to their reactivity. Indeed, due to the small size of its coherent aromatic domains and its hydrogenated character, graphitic carbon can be easily functionalized through H–O substitutions during redox changes. As heating, and hence graphitization progress, the size of the coherent domains increases and the structure dehydrogenates. It is therefore plausible that the graphitic carbon buffer capacity decreases with increasing temperature, leading to paragenesis equilibrated at increased fO₂ (i.e., as in MAP experiments relative to PC experiments). This hypothesis was tested through experiment PC2b (1000 °C, 3 GPa), where the chlorite breakdown is reached. The absence of magnesite in this experiment rules out the role of graphitic carbon in controlling the fO₂ in our experiments, therefore calling for alternative scenarios.

An external control of oxygen fugacity might be attributed to the assembly used in the HP device, which includes the pressure medium, the thermal insulator sleeve, the furnace, the spacers and the capsule (Supplementary Fig. S1). As the present experiments are “unbuffered”, i.e., the redox conditions are not imposed by a defined buffer added to the experiment (for instance using a double capsule technique e.g. Eugster and Skippen 1967), all the parts of the assembly exert an intrinsic fO₂ on the experiment. This fO₂ can be imposed to the sample by the available H₂ content in the assembly, which highly depends on the preparation of the parts (that in particular should be dried to avoid moisture) (Luth 1989; Matjuschkin et al. 2015). As a result, the intrinsic fO₂ applied to an experimental sample is very variable between studies and experiments depending on the assembly used and its preparation.

During our experiments, the MgO spacers could have buffered the fO₂ at more oxidized conditions compared to other materials (Kägi et al. 2005). However, MgO is used in both assemblies and therefore cannot explain the different mineral assemblages observed in our experiments. Similarly, capsules are made of Pt in both devices, ruling out a difference in H permeability to explain the variation in fO₂. The parts in assemblies that differ between PC press and MAP experiments are the pressure medium (talc-Pyrex in PC press vs. MgO in MAP; see Supplementary Fig. S1), the thermal insulator sleeve (again talc-Pyrex in PC press vs. LaCrO₃ in MAP) and the furnace (graphite in PC press vs. Re in MAP). The talc-Pyrex part (present in PC) is critical because the water released from talc dehydration increases H₂ availability and therefore the fO₂ equilibration with a buffer (which can be another part of the assembly). Pyrex can prevent H₂ diffusion during a few hours (Matjuschkin et al. 2015), sofaster equilibration of fO₂ with the different parts of the assembly is expected in PC press experiments compared to the MAP ones. LaCrO₃ (used in MAP) was used by Maurice et al. (2020) and was reported as to be “inert with respect to fO₂”, without significant reducing or oxidizing power.

The remaining possible control on fO₂ that could explain the different mineral assemblages between the two experimental devices is the furnace, PC and MAP experiments employing graphite and rhenium furnaces, respectively. Maurice et al. (2020) performed antigorite dehydration experiments in MAP (at 3 GPa and 600–900 °C) using either a graphite or a LaCrO₃ furnace. They observe the absence of hematite in graphite furnace experiments, while hematite formed in LaCrO₃ furnace experiments. The presence of hematite indicates fO₂ conditions beyond the magnetite-hematite (MH) redox buffer, i.e., at ΔQFM +3. In contrast, Merkulova et al. (2017), who also used a graphite furnace in their PC experiment (at 2 GPa and 600–900 °C) do not report the presence of hematite. This supports the buffering capacity of the graphite furnace, which imposes fO₂ conditions within the range ΔQFM −2 and QFM (value from Truckenbrodt et al. 1997), thereby preventing the formation of hematite. This ΔQFM range is equivalent to an absolute log₁₀fO₂ ranging from −16.13 to −14.13 at 700 °C and 3 GPa. Hermann and Spandler (2008) report a similar value, close to the NNO buffer, i.e., log₁₀fO₂ = −14.83 for a comparable assembly at 700 °C and 3 GPa in PC press (i.e., Teflon foil, salt outer sleeve, Pyrex glass, graphite furnace and MgO inserts) coupled with a gold capsule. It should be noted that the employing of a Pt capsule in our experiments rather than a gold one should accelerate the equilibration of fO₂ at the value imposed by the furnace as platinum has an higher permeability to H₂ than gold (Jakobsson 2012; Chou 1986).

In our experiments, the buffering capacity of the graphite furnace in PC press can therefore prevent magnesite formation. Conversely, the rhenium furnace used in MAP experiments does not have the same buffering capacity, potentially allowing an increase in oxygen fugacity during serpentine dehydration. Indeed, the rhenium furnace, as well as rhenium oxides (i.e. RRO buffer) should impose fO₂ within the range ΔQFM −1 and ΔQFM +3, between NNO and MH buffers (nonetheless extrapolated from high temperature low pressure conditions in Pownceby and O’Neill 1994). This could explain the formation of magnesite and the alternating presence of coesite or forsterite in MAP experiments. By comparing PerpleX calculations with fO₂ estimates for PC experiments (using values from Truckenbrodt et al. 1997) (Fig. 7), it must be noted that the estimated range of fO₂ values (from −16.13 to −14.13) should theoretically allow magnesite appearance. Such a disagreement might be attributed to simplifications with thermodynamic modelling or neglecting organic compounds other than graphite and methane. It is also possible that the relative amount of magnesite predicted in thermodynamic modelling calculations is too low (less than 1%) to be readily observed.

In MAP experiments, the log₁₀fO₂ values calculated after EMOG (−13.54 at 700 °C or ΔQFM −3.03; −8.45 at 1000 °C or ΔQFM −2.38) and MCEG (−7.31 or ΔQFM −1.24) buffers are in the magnesite stability fields (Fig. 7; Table 3). However, based on the PerpleX predictions at 7 GPa, these values are also in the coesite stability field while only one experiment presents the correct equilibrated mineral assemblage (i.e., MAP2c; Table 1). This could be attributed to slight variations in the relative proportions of antigorite vs. pyrene in the starting materials (see “Materials and methods” section). A high carbon concentration will favour the stabilization of coesite relative to forsterite (see Supplementary Fig. S5). The formation of magnesite in these experiments occurs below ΔQFM −2, suggesting that oxygen fugacity is even lower in PC experiments where magnesite is not observed.

In natural settings, several studies have highlighted large fO₂ variations within subducting slabs, ranging from ΔQFM −4 to ΔQFM +4 (Debret and Sverjensky 2017; Merkulova et al. 2017; Bretscher et al. 2018; Piccoli et al. 2019; Iacovino et al. 2020; Maurice et al. 2020; Evans and Frost 2021; Duan et al. 2022). This variability can be attributed to the protolith, notably the redox state and budget of abyssal serpentinites (Bretscher et al. 2018; Evans and Frost 2021), as well as other slab components, such as sediments (Caurant et al. 2023; Padrón-Navarta et al. 2023). In particular, the reaction between carbonated lithologies and serpentinite-derived fluids is likely to produce reducing conditions (Galvez et al. 2013; Vitale Brovarone et al. 2017; Caurant et al. 2023; Maffeis et al. 2023; Eberhard et al. 2023), while the redox impact of the organic carbon counterpart during serpentinite dehydration is unknown. While thermodynamic modelling allow to predict the stability of organic compounds in fluids under reducing conditions (Sverjensky et al. 2014), we show here with our experimental approach that under such reducing conditions, solid organic carbon is unlikely to react with serpentine-derived fluids and tends to remain metastable within the dehydrated silicate assemblages. As graphitization progresses, the size of its coherent domains increases and its structure dehydrogenates, hence leading to decreased reactivity with increasing temperature for the graphitic carbon. This limited reactivity likely restrains carbon solubility and mobility in fluids. This ligns with Szlachta et al. (2022) who observed the decomposition of aqueous acetate ions into graphitic carbon at 5 GPa and 600 °C, calling for carbon metastability as a solid rather than in the fluid phase. Overall, this shows that, under reducing conditions, solid organic compounds are retained in HP-HT parageneses, thus limiting carbon transfer to the mantle wedge and promoting their recycling into the deep mantle. In contrast, serpentinite dehydration under oxidizing conditions seems to oxidize graphitic carbon, resulting in the formation of oxidized carbonate ions in fluids, as evidenced by the presence of magnesite within silicate domains in our experiments (Figs. 5, 6). Hence, significant transfer of carbon from the slab to the mantle could be limited by an increase of oxygen fugacity within the slab during prograde metamorphism. This decoupling behaviour of carbon was recently reported for natural rocks by Bouilhol et al. (2022) who highlight the presence of carbonates in HP metamorphic veins and the presence of various types of graphitic carbon in the residual dehydrated serpentinites. Therefore, beyond the encountered P–T conditions, it is the redox reaction pathway of metamorphic rocks that becomes of primary interest in better understanding the fate of carbon during subduction.

Conclusion

HP-HT experiments have been performed using three types of solid PAHs, with some containing oxygen-bearing functional groups. These PAHs served as analogues for hydrothermal abiotic CCM increasingly reported in oceanic rocks. The experiments were conducted along a subduction gradient ranging from 700 to 1000 °C and 3 to 7 GPa. Despite different P–T conditions, we find, as previously reported, similar structural evolution of PAHs with increasing pressure and temperature, leading to the formation of graphitic carbon showing nonetheless various chemistry affecting its O and H contents. It therefore preserves a high structural disorder far from pure graphite structure, which makes it potentially reactive as subduction progresses. Moreover, aqueous fluid seems to be released from one oxygen-functionalized PAH type at low P–T conditions (700 °C, 3 GPa).

Experiments with mixing of pyrene and natural antigorite powders have reproduced the prograde dehydration reactions occurring in CCM-hosting serpentinites, i.e., antigorite and chlorite breakdown (700–1000 °C, 3–7 GPa). Depending on the experimental device used, various mineral assemblages were stabilized. In PC experiments, a conventional silicate assemblage following antigorite breakdown is observed, consisting of forsterite, enstatite, chlorite preserving graphitic carbon derived from PAH graphitization. In contrast, MAP experiments yielded an unconventional assemblage, consisting of enstatite, garnet, coesite, and magnesite. Thermodynamic calculations show that these differences can be explained by the redox conditions, likely induced by the experimental device as previously observed (Maurice et al. 2020). Under reduced conditions, CCM would be poorly reactive during serpentinite dehydration, leading to its recycling into the deep mantle. In contrast, more oxidizing conditions allow carbonate formation, potentially facilitating the transport of carbon to the mantle wedge. Thus, it appears that redox conditions are the main parameter controlling the metastability of PAHs and CCM in subduction context with potentially a large impact on the deep carbon cycle.