Degassing-induced fractionation of multiple sulphur isotopes unveils post-Archaean recycled oceanic crust signal in hotspot lava

Abstract

Mantle source regions feeding hotspot volcanoes likely contain recycled subducted material. Anomalous sulphur (S) isotope signatures in hotspot lavas have tied ancient surface S to this deep geological cycle, but their potential modification by shallow magmatic processes has generally been overlooked. Here we present S isotope measurements in magmatic sulphides, silicate melt inclusions and matrix glasses from the recent eruption of a hotspot volcano at El Hierro, Canary Islands, which show that degassing induces strongly negative δ³⁴S fractionation in both silicate and sulphide melts. Our results reflect the complex interplay among redox conditions, S speciation and degassing. The isotopic fractionation is mass dependent (Δ³³S = 0‰), thus lacking evidence for the recycled Archaean crust signal recently identified at other hotspot volcanoes. However, the source has an enriched signature (δ³⁴S ~ + 3‰), which supports the presence of younger ³⁴S-rich recycled oceanic material in the Canary Island mantle plume.

Introduction

Chemical, physical and biological processes within the mantle, crust, atmosphere and oceans fractionate sulphur isotopes, and material exchanges among these geological reservoirs lead to characteristic sulphur isotope signatures that have varied over time¹. Sulphur isotope heterogeneity in the mantle, as sampled by sulphide inclusions^2,3,4,5, melt inclusions^6,7 or primitive volcanic rocks^8,9, thus traces secular variation in the tectonic boundary conditions that influence mantle circulation, as well as the internal processes that have established the current mantle state. On the other hand, the full scope of mantle heterogeneity is best illustrated by variations in trace element and radiogenic isotope geochemistry of ocean island basalts (OIB), which define various mantle end-members thought to feed the sources of hotspots^10,11. Therefore, it is of interest to investigate these mantle components from a sulphur isotope perspective to help set additional constraints on their origin and further our understanding of the deep sulphur cycle. This type of approach has been adopted in recent studies, which we summarise below.

Mass-independent fractionation (MIF) of sulphur isotopes is defined by non-zero Δ³³S values, where Δ³³S = δ³³S – [(1 + δ³⁴S)^0.515–1] and δ^xS_V-CDT = (^xS/³²S)_sample/(^xS/³²S)_V-CDT – 1, and is characteristic of sedimentary rocks of Archaean age (~2.5 Ga and older), reflecting the influence of photochemical processes in an atmosphere devoid of oxygen¹². Sulphur in sedimentary rocks that post-date the Great Oxidation Event at ~2.4 Ga thus has Δ³³S values around 0‰, and the associated S isotope fractionations are termed mass-dependent. However, negative, non-zero Δ³³S values have now been reported twice for young volcanic rocks from hotspot settings: in olivine-hosted sulphides from Mangaia, Cook Islands⁴, and in sulphides from the Pitcairn hotspot⁵. These anomalous signatures are thought to reflect the cycling of Archaean sulphur from Earth’s surface to the mantle by subduction, and back to the surface via mantle plumes^4,5,13. Since Mangaia is the representative end-member of the ancient ‘HIMU’ (high μ = ²³⁸U/²⁰⁴Pb) mantle component^10,11, the potential positive covariation of Δ³³S and δ³⁴S values in Mangaia sulphides was first used to suggest a specific Archaean protolith to the HIMU source characterised by negative Δ³³S and δ³⁴S values⁴. The subsequent finding of S-MIF at Pitcairn, representative of the enriched mantle I end-member^10,11 (EM-I; characterised by unradiogenic Pb isotope signatures), in association with negative δ³⁴S values, lends support to this hypothesis, potentially resolving the positively-skewed imbalance of Δ³³S values observed in Archaean surface reservoirs^5,13. These studies thus imply that a missing Archaean sulphur pool is stored in the deep mantle and occasionally resurfaces at hotspots. Other plume-related lavas from Samoa, the type locality for the third common OIB mantle isotopic end-member, EM-II (characterised by the highest ⁸⁷Sr/⁸⁶Sr ratios), show coupled variations in S and Sr isotopes that indicate recycling of younger sulphur-rich sediments into a mantle source with a near-zero Δ³³S value (mass-dependent) and a δ³⁴S value of ~3‰⁹. The distinction in S isotope signatures between different mantle reservoirs suggests a long-lived and isotopically evolving surficial input into different hotspot source regions, highlighting the importance of understanding the causes of S isotope variability in the mantle.

Magmatic processes involving sulphur, such as degassing, sulphide segregation, mixing or assimilation, may also leave an imprint on the S isotope composition of volcanic rocks^7,8,14,15, and separating these effects from the source signature can present challenges. However, melt inclusions, which represent droplets of silicate melt trapped in minerals during crystal growth, offer unique snapshots of an ascending magma at various depths¹⁶, and can help resolve this problem. Toward this end, we report the S isotope compositions of melt inclusions and matrix glasses spanning the degassing history of the volatile- and sulphur-rich magma erupted at El Hierro, Canary Islands, in 2011–2012¹⁷. We also present δ³⁴S and Δ³³S signatures of magmatic sulphide inclusions — to our knowledge, this is the first time S isotopes are measured simultaneously in sulphides, melt inclusions and matrix glass for a single eruption. The dataset provides an unusually clear picture of the mechanisms by which sulphur isotopes fractionate during degassing and sulphide saturation in natural magmas, and shows that degassing can induce large δ³⁴S fractionations of up to 10‰. In turn, this offers an exceptional opportunity to investigate in situ how the S isotope heterogeneity generated during magma evolution, ascent and eruption^7,8,18,19 can be quantitatively discriminated from that inherited from the mantle sources for hotspot volcanism. In doing so, we find that the S isotope signals at El Hierro reflect a post-Archaean origin for recycled S in the Canary Island hotspot, contrasting with the recent findings of S-MIF at other OIB localities.

Results

Geological setting and sample description

The Canary Island hotspot in the eastern Atlantic Ocean is characterised by an exotic geochemistry, producing OIB that is mostly alkaline in composition and displays isotopic affinities with the HIMU, EM and depleted MORB mantle (DMM) end-members^20,21, suggesting that the mantle source constitutes a mix of different reservoirs²¹. This context thus offers the potential to confirm or contrast S isotope signals from the Canary Islands to those observed at hotspots in the South Pacific (i.e. Mangaia/HIMU; Pitcairn/EM-I; Samoa/EM-II). Additionally, ubiquitous fluid inclusions in mantle xenoliths and occurrence of carbonatite melt^22,23 point to a mantle source enriched in volatiles. The recent submarine eruption off the south coast of El Hierro, the youngest and westernmost island of the archipelago, produced lava balloons containing olivine-hosted melt inclusions (Fig. 1a) with dissolved volatiles reaching concentrations in excess of 3000 ppm CO₂, 3 wt.% H₂O and 5000 ppm S (ref. ¹⁷). In addition, clinopyroxene and spinel (Fe–Ti oxide) phenocrysts in the same samples host abundant sulphide globules (Fig. 1b–f), revealing that the magma was saturated with an immiscible sulphide liquid for part of its history. However, sulphide inclusions are not present in olivine phenocrysts, nor do they occur as a free phase in the matrix glass.

Fig. 1

Inclusions analysed in situ by Secondary Ion Mass Spectrometry (SIMS). a Plane-polarised light photomicrograph of an olivine-hosted melt inclusions with exsolved bubbles. b Reflected light photomicrograph of homogeneous, rounded monosulphide solid solution (mss) sulphides of pyrrhotite composition hosted in a clinopyroxene phenocryst (group 1a). c Sub-angular sulphide inclusion with intergrowths of mss and intermediate solid solution (iss) (group 2), as well as small exsolutions of a Ni-rich phase at their interface (see elemental maps in Supplementary Fig. 3). Note the contact between the inclusion and silicate melt. d Irregular shaped inclusion of mss with interstitial oxide (group 2). e Group 2 zoned mss-iss sulphide surrounded by melt and a skeletal, rapidly growing spinel phenocryst. f Rounded sulphide droplet, with trellis texture and attached to the outside of an spinel phenocryst, also in contact with melt. Scale bars are 20 μm for panels a, c–e and 50 μm for panels b, f

We performed in situ sulphur isotope analyses by Secondary Ion Mass Spectrometry (SIMS) on a suite of 25 olivine- and spinel-hosted melt inclusions and 9 matrix glass chips, previously shown to have large ranges in volatile contents and sulphur speciation, with S contents linearly and positively correlated with H₂O and S⁶⁺/ΣS (ref. ¹⁷). These measurements yielded δ³⁴S values only, owing to the analytical difficulty of resolving low-abundance ³³S at low levels of S in silicate melts. We also obtained the S isotope compositions (δ³⁴S and Δ³³S) of 49 clinopyroxene- and spinel-hosted sulphide droplets, and measured their chemical composition by electron probe micro-analysis (EPMA). Table 1 summarises the isotopic composition of the various samples. Details on our analytical techniques and associated uncertainties can be found in the Methods section. The Supplementary Information includes a discussion of potential matrix effects during isotopic analysis, which appear negligible. All quoted errors are 1σ propagated analytical uncertainties (see Methods for error treatment).

Table 1 Summary of samples analysed with ranges in S content and isotope composition

Sulphur isotopes in the silicate melt

The δ³⁴S values of melt inclusions and matrix glasses (Supplementary Data 1) exhibit an exponential decline with decreasing S content (Fig. 2a), with S-rich inclusions having positive δ³⁴S values and S-poor matrix glasses (<500 ppm S) having the most negative δ³⁴S values. Spinel-hosted silicate melt inclusions have δ³⁴S ranging from +3.1 ± 0.5‰ to +0.8 ± 0.5‰ while olivine-hosted melt inclusions span a greater range from +3.2 ± 0.1‰ to −5.9 ± 0.6‰. Matrix glasses have the lowest δ³⁴S values, extending down to −8.2 ± 0.6‰ (Fig. 2a). Inclusion size (30–150 μm) show no relationship with isotopic composition.

Fig. 2

Degassing-induced sulphur isotope fractionation. a Sulphur content and isotopic composition measured in olivine-hosted (MI-ol; navy blue circles) and spinel-hosted (MI-sp; blue diamonds) melt inclusions and matrix glasses (light blue triangles). Shaded regions represent the possible ranges of isotopic fractionation due to degassing, with changing pressure and oxygen fugacity, computed with the model of Marini et al.¹⁴ with the empirical fractionation factors of Miyoshi et al.⁴² (grey fill) and Fiege et al.⁴⁵ (hatched field). Minimum and maximum starting δ³⁴S values are chosen as the δ³⁴S value of the most S-rich inclusion and the maximum δ³⁴S value, respectively. The lower and upper bounds for each field correspond to S–fO₂ covariation as shown by the grey and black solid lines in b, with varying pressure as modelled in Supplementary Fig. 5. The dashed line shows the effect of FeS segregation on melt δ³⁴S values. Error bars represent 1σ propagated analytical uncertainties (see Methods). b Melt oxygen fugacity (ΔFMQ: log fO₂ units relative to the fayalite–magnetite–quartz buffer) as a function of S content (F(S)), derived from S⁶⁺/ΣS ratios of melt inclusions and matrix glass with the equation of Jugo et al.³³. Solid lines are linear relations drawn to encompass the data, representing minimum fO₂ (grey line) and maximum fO₂ (black line) for a given S content, and are used for modelling the lower and upper-bound fractionation estimates for each model shown in a. Dashed arrows represent degassing paths calculated with D-Compress (ref. ³⁹) in the C–S–O–H–Fe system, starting at a pressure of 300 MPa, initial volatile contents as measured in the most volatile-rich inclusion, and minimum (grey arrow) and maximum (black arrow) estimated initial fO₂. The inset shows the expected trajectory of the magma, with initially high S⁶⁺/ΣS ratio (black dot) decreasing upon S degassing (black dashed arrow) until sulphide saturation is reached (yellow dot). At that point the magma follows the SCSS upon further degassing (grey dotted line). Error bars for ΔFMQ represent the analytical uncertainty on S⁶⁺/ΣS values, converted to ΔFMQ values^17,33

Sulphur isotopes and petrography of the sulphides

All magmatic sulphides analysed have Δ³³S values within error of the terrestrial mass-dependent fractionation line (i.e. 0‰), averaging 0.01 ± 0.06‰ (Fig. 3). However, they exhibit a large δ³⁴S range from +1.0 ± 0.4‰ to −9.6 ± 0.4‰, and display a variety of morphologies and compositions (Figs 1, 3–5, Table 1).

Fig. 3

Δ³³S vs δ³⁴S for clinopyroxene- and spinel-hosted sulphide inclusions from El Hierro (coloured symbols, this study), compared to other OIB (grey symbols). The majority of inclusions have mss compositions close to pyrrhotite (yellow symbols, group 1), but some also contain Cu-rich iss intergrowths (orange symbols, group 2), while others display quench textures with low EPMA totals (~95%; dark red symbols). The δ³⁴S range is comparable to that of the silicate melt, with the most negative values reached by the 2-phase and heterogeneous sulphides. Δ³³S values cluster around 0‰ (i.e. within the mass-dependent fractionation line), contrasting with the Mangaia sulphides (grey squares; ref. ⁴) and Pitcairn sulphides (grey triangles; ref. ⁵) but similar to St. Helena glass⁵². The MORB field (black square; ref. ⁵⁴) is shown for comparison. The modelled effect of degassing (Fig. 2a) follows the negative x-direction at Δ³³S = 0 (i.e. the terrestrial mass-dependent fractionation line). Error bars are 1σ propagated analytical uncertainties

The majority of sulphide inclusions analysed (n = 33) are homogeneous, smooth, circular to elliptical droplets (Fig. 1b) with compositions corresponding to monosulphide solid solution (mss) close to the pyrrhotite (Fe_1–xS) end-member (Fig. 4a, Supplementary Data 2). These sulphides only occur completely enclosed in phenocrysts, and have δ³⁴S values between +1.0 and −4.0‰ (Figs 3 and 5). We refer to them as group 1.

Fig. 4

Geochemistry of El Hierro sulphides compared to sulphides from MORB, back-arc and arc settings. a Ternary diagram showing the composition (in molar fractions) of El Hierro sulphides (circles) and sulphides from MORB (blue symbols), back-arc (purple symbols) and arc settings (pink symbols)²⁶. Different symbols are used for mss (crosses), iss (x’s) and mss–iss intergrowths or zoned sulphides where phases were not separately measured (stars). El Hierro group 1 sulphides have pyrrhotite-like (po) mss compositions with increasing Ni content towards group 2. The mss and iss phases in group 2 sulphides were analysed separately (orange and beige circles, respectively), hence the bulk compositions of those globules would fall on a straight mixing line between the two. Common sulphide mineral compositions are also shown for comparison: pyrrhotite (po), pentlandite (pn), cubanite (cub) and chalcopyrite (cpy). Bold triangle in inset shows the region of the Fe–Ni–Cu ternary used for the plot. b Ni/Cu and Fe/S ratios (wt.%/wt.%) of mss as in ref. ²⁶., showing that the early group 1 sulphides have a strong affinity with arc sulphides and plot near the upper Ni/Cu end of the arc trend. The apparent extension of the MORB trend by group 3 sulphides is likely an artefact of their lower S content due to S degassing

Fig. 5

Variations of sulphide geochemistry with isotopic composition. Covariation of sulphur (a) and nickel (b) contents with isotopic composition, showing a general decrease in δ³⁴S with decreasing S content, and different trends for Ni which separate group 2 and 3 sulphides. Symbols as in Figs 3–4. Bulk compositions of group 2 sulphides plot along the pale orange bands, which join the S (a) and Ni (b) contents of the mss and iss intergrowths

Another set of inclusions (n = 12), group 2, display heterogeneous mineralogy and more angular shapes (Fig. 1c–e). These consist of intergrowths of mss, intermediate solid solution (iss)^24,25,26 and an interstitial oxide phase. The mss in these inclusions reaches higher Ni (0.2–6.4 wt.%) and Cu (0.8–5.8 wt.%) contents than the homogeneous sulphides (0.1–2.2 wt.% Ni and 0.4–1.9 wt.% Cu; Fig. 4a and Supplementary Data 2), and sometimes contains small exsolutions of a more Ni-rich phase, likely pentlandite (Fig. 1c, Supplementary Fig. 3b). The iss has a stoichiometry corresponding to cubanite (CuFe₂S₃) (Fig. 4a). The oxide phase was too fine-grained to analyse by EPMA, but elemental maps obtained with a scanning electron microscope (SEM) confirmed that its composition is essentially iron oxide (Supplementary Fig. 4). These inclusions are hosted primarily in spinel phenocrysts, commonly occur at the edge of crystals, and are sometimes in contact with the melt or even completely outside the crystal (Fig. 1e). They are also often associated with vesicles (cf. ref. ²⁷). They have δ³⁴S ranging from −1.8‰ to −7.1‰ (Figs 3 and 5).

Four additional globules have spherical shapes resembling the homogeneous inclusions of group 1, but have trellis-textured surfaces, which are often observed for quenched sulphide melt^28,29, and also occur outside or at the edge of crystal hosts — we distinguish them as group 3 (Fig. 1f). They are characterised by the highest Fe and lowest S contents, as well as low Ni and Cu (Supplementary Data 2, Figs 4, 5). Their δ³⁴S range is similar to that of the group 2 sulphides, from −2.3‰ down to −9.6‰. However, they do not contain a discrete oxide phase as seen in group 2, and have low EPMA element totals (~95 wt.%). This may be due to incorporation of O²⁻ anions in the sulphide liquid^30,31, given that oxygen was not included in the analytical routine. One of these inclusions also has an interstitial Cu-rich phase (Supplementary Fig. 3e).

Sulphides analysed from all three groups have similar size ranges from 25 μm up to 180 μm across, and there is no systematic variation in isotopic composition or geochemistry with size. Smaller droplets (<10 μm) were also present but not analysed due to the ~15 μm size of the ion beam (see Methods).

Discussion

Decompression-induced degassing beginning at pressures of at least 300 MPa was the primary cause of CO₂, H₂O and S loss in the El Hierro magma¹⁷ (see Supplementary Fig. 5). Sulphur speciation data (S⁶⁺/ΣS) in the melt inclusions also show that the magma became more reduced during decompression and S loss to a fluid phase (Fig. 2b). However, the presence of sulphide globules in minerals also attests to S partitioning into an immiscible sulphide melt. The solubility of sulphur in magma before the separation of an immiscible sulphide liquid, also known as the sulphur concentration at sulphide saturation (SCSS)^32,33, increases steeply with increasing oxygen fugacity (fO₂) between the fayalite–magnetite–quartz (FMQ) buffer and two log units above it (FMQ + 2), corresponding to a switch from sulphide(S^2-)- to sulphate(SO₄^2-)-dominated conditions³³. The most sulphur-rich melt inclusion has a S concentration (5080 ppm) well exceeding the S solubility of a relatively reduced magma (around FMQ)^32,33, but at FMQ + 1.5 dissolved S can reach >1 wt.%³³. The S⁶⁺/ΣS ratio of that inclusion suggests a minimum fO₂ of 1.2 log units above FMQ (but likely higher; see ref. ¹⁷), hence we infer that the initially oxidised magma was not sulphide saturated. The presence of sulphide droplets therefore reflects the decrease in fO₂ until intersection of the SCSS (Fig. 2b), at which point a sulphide liquid started to separate from the silicate melt. The absence of sulphide droplets in olivine phenocrysts supports a late saturation of sulphide. Alternatively, the lack of olivine-hosted sulphide inclusions could reflect preferential late-stage nucleation near rapidly-growing Fe–Ti oxide or clinopyroxene phenocrysts¹⁷.

The crystallisation of magnetite (Fe²⁺Fe³⁺₂O^2–₄) has previously been argued to trigger sulphide saturation in hydrous, oxidised magmas from subduction zone settings^34,35,36. This process could also reasonably be invoked for the El Hierro magma, supported by the ubiquitous association of sulphides with Fe–Ti oxides. However, the El Hierro oxides, with an average composition approaching Fe²⁺_1.0Mg²⁺_0.4Fe³⁺_0.9Al³⁺_0.3Ti⁴⁺_0.4O^2–₄ (ref. ³⁷), have a ferric iron content less than half that of the magnetite end-member, and so their crystallisation may not have modified the melt Fe³⁺/ΣFe ratio sufficiently to trigger sulphide saturation³⁴. Moreover, there is no systematic variation between the FeO content of melt inclusions and their S⁶⁺/ΣS ratios¹⁷, which would be expected if spinel fractionation was the cause of magma reduction. Hence, we favour degassing-induced reduction^{17,38,39,40,41} as the cause of the intersection of the SCSS and separation of an immiscible sulphide liquid. Our observations are consistent with different degassing modelling trends obtained with the software D-Compress³⁹, which all show reduction of the magma associated with S loss when the initial magma is oxidised (Fig. 2b). It is worth noting that while sulphide droplets are abundant in some phenocrysts, overall they represent a rather insignificant fraction of the S budget of the El Hierro magma (estimated at <1% of total S from thin section observations, see Methods), and therefore sulphide separation did not appreciably decrease the S content of the magma.

We now explore the intricate isotopic interplay among gas loss, sulphide saturation, and magmatic fO₂ with quantitative calculations of S isotope fractionation during magma ascent. The positive exponential correlation of S content and δ³⁴S in melt inclusions and matrix glasses (Fig. 2a) suggests that S loss via degassing is the primary process driving the strong mass-dependent isotopic fractionation, especially after a significant amount of S has escaped. There is no need to invoke mixing of isotopically distinct melts, which might be expected to produce linear arrays in Fig. 2a, and no evidence for late-stage assimilation of seawater-influenced components, which should drive δ³⁴S to high positive values¹⁵. Isotopic fractionation of sulphur in magmatic systems depends strongly on the proportions of oxidised and reduced S species in the different phases present, since oxidised species are generally enriched in ³⁴S relative to reduced species in the order SO₄²⁻_(melt) > SO_2(gas) > H₂S_(gas) ≈ S²⁻_(melt) ≈ S²⁻_(FeS) (ref. ¹⁵). Since the proportions of oxidised and reduced S species in the melt are linked to the oxidation state of the magma, as monitored by fO₂ (ref. ³³), processes involving the exchange of sulphur between different phases (melt, gas or sulphide liquid) have the potential to induce isotopic fractionation when these exchanges involve modifications in the valence state of S. For example, the isotopic fractionation factor between a sulphide liquid and S dissolved in the melt (α_FeS–melt, where 1000lnα_FeS–melt ≅ δ³⁴S_FeS – δ³⁴S_melt) will vary strongly with the proportions of dissolved SO₄²⁻ and S²⁻ in the melt (i.e., the S⁶⁺/ΣS ratio). At oxidised conditions (e.g. ΔFMQ ≈+1.5), where significant SO₄²⁻ is present, δ³⁴S_FeS is ~3‰ lower than δ³⁴S_melt^15,42,43, such that sulphide segregation would cause the melt to become enriched in ³⁴S. At reduced conditions, S has the same valence (−2) in both sulphide and silicate melts, and the isotopic fractionation is small, with δ³⁴S_FeS about 0.4‰ higher than δ³⁴S_melt at 1150 °C (ref. ¹⁵). This fractionation and the small contribution of sulphide to the total initial magmatic S content are not sufficient to produce the strong isotopic signal we observe (Fig. 2a). Instead, the positive exponential trend between S content and δ³⁴S values in the melt implicates open-system degassing as the primary agent of isotopic change (Fig. 2a), even though gas loss is conventionally expected to favour light isotopes, a common misconception for redox-dependent isotope systems¹⁸.

The sulphur species incorporated in a fluid phase at magmatic conditions are mainly H₂S and SO₂ (refs ^15,32), in which S has a valence of −2 and +4, respectively. The proportions of H₂S and SO₂ are therefore controlled by fO₂, but they also depend on H₂O fugacity (fH₂O), which can be used as a proxy for pressure since H₂O is the major constituent (in mol%) of the fluid phase (Supplementary Fig. 6). This can be described by the equilibrium reaction:^8,14

$${\mathrm{H}}_2{\mathrm{S}}_{({\mathrm{gas}})} + 1.5{\mathrm{O}}_{2({\mathrm{gas}})} \rightleftharpoons {\mathrm{SO}}_{2({\mathrm{gas}})} + {\mathrm{H}}_2{\mathrm{O}}_{({\mathrm{gas}})}$$

(1)

Degassing-induced S isotope fractionation is therefore sensitive to both fO₂ and fH₂O (i.e. pressure)¹⁴ (Fig. 6). As a result, both of these intensive parameters must be known simultaneously to correctly model degassing-induced isotopic fractionation of sulphur during the evolution of the El Hierro magmatic system. We used the S⁶⁺/ΣS and CO₂–H₂O data of Longpré et al.¹⁷ combined with the parameterisations of Jugo et al.³³ and Iacono-Marziano et al.⁴⁴ to estimate fO₂ and total pressure, respectively, in olivine-hosted melt inclusions (Fig. 2b and Supplementary Fig. 5). We considered two sets of end-member estimates^42,45 for the fractionation factors among the relevant S species (SO₄²⁻_(melt), S²⁻_(melt), SO_2(gas), H₂S_(gas))^15,32. This allowed us to calculate a range of evolving gas–melt S isotope fractionation factors (α_gas–melt) during gradual loss of sulphur from the El Hierro melt upon ascent from 300 to 1 MPa.

Fig. 6

Effect of pressure and oxygen fugacity on S isotope fractionation. Degassing-induced S isotope fractionation modelling using the starting isotopic composition of the inferred parental El Hierro magma, showing the relative effects of pressure (P) and oxygen fucacity (fO₂). Strongly divergent paths are determined mainly by oxygen fugacity. Solid lines and dashed lines correspond to the fractionation factors of Miyoshi et al.⁴² and Fiege et al.⁴⁵, respectively

Under isothermal decompression at 1150° C¹⁷, calculations based on our estimated fO₂ and pressure ranges produce lower- and upper-bound isotopic fractionation curves encompassing our entire dataset (Fig. 2a). The most recent experimentally determined fractionation factors⁴⁵ better reproduce the strongly fractionated δ³⁴S values we observe at low S abundances (Fig. 2a); details on these calculations are provided in the Methods section, and the differences between models are outlined in the Supplementary Discussion. Supplementary Data 3 contains the calculations used to model S isotope fractionation. Equivalent calculations, assuming S loss via sulphide segregation alone, demonstrate that this mechanism did not affect the S isotope fractionation we observe (Fig. 2a). As the δ³⁴S range of sulphide inclusions mirrors that of the melt inclusions and matrix glasses, we suggest that trapped sulphides essentially track the isotopic composition of the silicate melt as sulphur is degassed¹⁵ (see below). Our calculations demonstrate that increased proportions of SO₂ (S⁴⁺) in the gas phase, which is predicted at lower pressures, favour the degassing of heavy S isotopes since S in the El Hierro melt speciates toward S²⁻ as oxygen fugacity decreases at lower pressures. While degassing of SO₂ had previously been identified as a potential trigger for magma reduction and isotopic fractionation^18,38,40,46, our data clearly demonstrate that it can exert the primary control on both in a positive feedback loop driven by decompression.

Compositional and textural trends in the magmatic sulphides can be explained as part of this framework. The smooth, homogeneous sulphides (group 1), which have the highest δ³⁴S values (Table 1, Figs 3 and 5), represent the early sulphide liquid that was entrapped as pristine mss droplets in growing crystals of clinopyroxene and spinel. Sulphides occurring at the edge of crystals and in contact with the matrix glass (groups 2 and 3) reach significantly more negative δ³⁴S values and exhibit more diverse features, such as angular shapes, heterogeneous mineralogy or quench textures. Moreover, these sulphides have a lower S content (Supplementary Data 2, Fig. 5a), suggesting that they formed or equilibrated in a S-poorer melt (i.e., at lower fS₂)^30,31, or were subject to S loss by degassing^27,47. The Ni content of group 2 sulphides also increases in both mss and iss with decreasing δ³⁴S (Fig. 5b). To first order, the isotopic composition of the sulphides should be controlled by the S isotopic composition of the dissolved sulphur species in the silicate melt from which they are segregating. The different groups thus likely represent different stages in the evolution of the sulphide melt, with the majority of droplets observed corresponding to the onset of sulphide saturation. A break in slope in the S vs FeO* trend of melt inclusions at a S concentration of ~1500 ppm may mark this onset¹⁷. This would have occurred after a significant amount of S was already degassed, but before the isotopic composition of the melt was substantially affected, consistent with the modest isotopic fractionation seen in the group 1 sulphides, with δ³⁴S between +1.0 and −2.8‰, with one exception reaching −4.0‰ (Fig. 2a, b).

The separation of a sulphide melt should also deplete the silicate melt in metals such as Ni and Cu, which have ~1000 times more affinity for sulphide liquid^25,48. Therefore, the higher concentration of Ni and Cu in the group 2 sulphides cannot be taken as evidence for later saturation, since later-formed sulphides would precipitate from a more Ni- and Cu-depleted magma. Nevertheless, their occurrence at the outer edges and outside crystals combined with their δ³⁴S range of −1.8‰ to −7.1‰ suggests that they communicated with the silicate melt at a later stage than the group 1 sulphides, and may therefore have isotopically re-equilibrated with the ³⁴S-depleted magma. The Ni and Cu partition coefficients between mss and sulphide liquid are <1 (ref. ²⁸), hence fractional crystallisation of mss should enrich the residual sulphide liquid in these elements. This scenario is consistent with the expected evolution of the sulphide melt, which should first crystallise mss, followed by iss at a later stage⁴⁹. The group 2 sulphides therefore appear to represent residual sulphide liquid that crystallised iss and higher-Ni mss at a late stage while the δ³⁴S of the magma was plummeting, after removal of Ni-poorer mss (group 1).

The group 2 sulphides also crystallised an iron oxide phase along with mss and iss^24,50, implying that the residual sulphide liquid contained more oxygen, which is controlled by the reaction:^30,31

$${\mathrm{FeS}}_{({\mathrm{sulphide}}\,{\mathrm{melt}})} + \frac{1}{2}{\mathrm{O}}_{2({\mathrm{gas}})} \rightleftharpoons {\mathrm{FeO}}_{({\mathrm{sulphide}\,{\mathrm{melt}}})} + \frac{1}{2}{\mathrm{S}}_{2({\mathrm{gas}})}$$

(2)

This is counterintuitive, considering that these later-crystallised sulphides formed at lower fO₂. However, incorporation of O in the sulphide liquid is also controlled by sulphur fugacity (fS₂)^30,31, where a lower fS₂ represents a competing effect to lower fO₂ and can result in a higher O content of the sulphide liquid. Model calculations³⁹ indicate that fS₂ decreased by several orders of magnitude more than fO₂ during degassing of the El Hierro magma (Supplementary Fig. 7), and thus exerted the dominant control on this equilibrium. The presence of O in the sulphide phase may also explain some of the lower EPMA totals observed in the group 2 and 3 sulphides (Supplementary Data 2).

The group 3 sulphides, in contrast, do not contain a discrete oxide phase and their Ni and Cu contents are low, but they also show a large δ³⁴S range towards very negative values. Their spherical shapes and fine trellis textures suggest quenching from sulphide liquid. We therefore interpret them as having formed at the latest stage from a Ni- and Cu-depleted silicate melt. Hence we propose that sulphide segregation occurred continuously once the magma reached sulphide saturation. The group 3 sulphides have the lowest S content, presumably because they did not crystallise oxide and thus retained O in the liquid. The lower S content of these sulphides should also have lowered their melting point⁵¹, allowing them to remain a liquid phase, which is consistent with the observed quench textures. The overlap between the isotopic compositions of group 2 and 3 sulphides with group 1 (Table 1, Fig. 3) supports the idea of a continuous sulphide segregation history, with groups 2 and 3 experiencing larger fractionations at a later stage, as expected from the exponential decrease in δ³⁴S of the silicate melt upon S degassing (Fig. 2a).

The chemistry of mss sulphides has been shown to vary systematically between MORB and arc magmas, due to their different crystallisation histories and redox states²⁶. The Ni content of sulphides is generally controlled by the amount of olivine fractionation that has occurred prior to sulphide saturation, since Ni is compatible in olivine. MORB sulphides have significantly higher Ni than sulphides from arc magmas (Fig. 4, Supplementary Fig. 8), which can be attributed to the lower fO₂ of MORB parental magmas and, hence, earlier sulphide saturation prior to extensive removal of olivine²⁶. In contrast, arc magmas are generally more oxidised and reach sulphide saturation at a later stage, when the magma is more depleted in Ni. The chemistry of the El Hierro sulphides is most similar to sulphides from arc and back-arc settings, with lower Ni and Ni/Cu ratios (Fig. 4, Supplementary Fig. 8), which is consistent with the relatively high oxidation state of the inferred parental magma, and thus later sulphide saturation. These results are supported by the S isotope data, and also indicate that S degassing can have an additional important effect on the composition of the sulphides and stabilisation of the sulphide melt⁴⁷.

While the HIMU mantle reservoir is characterised by extreme ²⁰⁶Pb/²⁰⁴Pb ratios that constrain its age to >1 Ga to account for sufficient uranium decay^10,11, we find no S-MIF in El Hierro sulphides (Fig. 3), which is consistent with a post-Archaean (<2.5 Ga) origin for any recycled material in the Canary Island plume. Sulphur isotope analyses on basalt glasses from St-Helena, another HIMU-type hotspot in the South Atlantic, also show no signs of MIF⁵², contrasting with the MIF-bearing Mangaia sulphides presumably derived from an older recycled source⁴. It appears that HIMU-type signatures may not originate from a single reservoir, but rather from a recurrent geological process. This is consistent with the slightly lower ²⁰⁶Pb/²⁰⁴Pb ratios of the HIMU component found in the Canary Islands and other OIB from the eastern Atlantic Ocean, such as the Cape Verde islands and Madeira⁵³, which also suggests a shorter time-integrated addition of radiogenic Pb relative to Mangaia, and thus a more recent age of formation. By analogy, the Archaean MIF signal observed at Pitcairn⁵ may not require that all EM-I OIB have Archaean plume sources. Moreover, while degassing cannot explain the negative MIF Δ³³S values observed at Mangaia⁴ and Pitcairn⁵, our results clearly show that it can impart negative δ³⁴S signatures to volcanic rocks and their S-rich inclusions, potentially overprinting the isotopic composition of the source. Degassing thus offers an alternative mechanism to produce the highly negative δ³⁴S values of the Mangaia and Pitcairn sulphides, which have been interpreted to reflect an intrinsic feature of the presumed Archaean protolith within these plumes^4,5,13. Disentangling the effects of degassing from source heterogeneity is thus critical for understanding the geodynamic implications of S isotopes at volcanic hotspots.

By removing the effect of degassing, we find that the mantle source of the El Hierro magma was significantly enriched in ³⁴S compared to chondrites and the bulk silicate Earth⁵⁴, reaching a δ³⁴S value of +3.2 ± 0.1‰ in S-rich melt inclusions. A strikingly similar value of +3.5‰ was previously reported for S-rich clinopyroxene-hosted melt inclusions from Gran Canaria where, like at El Hierro, the most S-rich inclusions also show the highest S⁶⁺/ΣS ratios⁶. The convergence of these results suggests a potential link between high oxidation state and enriched δ³⁴S in the mantle sources of volcanic hotspots. This connects two emerging views: (1) in line with other isotopic proxies^11,21,53, recycling of subducted sulphur best explains S isotope signatures observed at many OIB hotspots (Mangaia⁴, Pitcairn⁵, Samoa⁹, and now the Canary Islands), and (2) the mantle sources of OIB appear to be more oxidised than those of MORB^33,41, but less so than those of arc magmas^46,55. During subduction, altered and oxidised oceanic crust and sediments transfer redox potential to the mantle wedge underlying volcanic arcs, likely with sulphur as an oxidising agent^46,55. However, a significant fraction of S might remain in the downgoing slab since estimates of S output at arcs are an order of magnitude lower than estimates of subducted S (refs ^46,56.), which may act as a potent provider of S and oxidising power in hotspot sources. The source of subducted S can be generally constrained from the S isotope compositions inferred here. While sulphides produced by microbial sulphate reduction during alteration of the modern-day oceanic crust generally show negative δ³⁴S values^57,58, this may have been different at earlier stages in Earth’s history¹. The high δ³⁴S (+2.8‰) of recycled S at Samoa, for example, has been attributed to the positive δ³⁴S that characterizes sulphides in Proterozoic sediments⁹. On the other hand, serpentinization of oceanic peridotites may be an important sink for the S in high-δ³⁴S seawater sulphate⁵⁹, as recent evidence points toward significant fluid uptake during bending and faulting of the oceanic lithosphere before subduction^60,61. Whether the subducted slab remains more oxidising than the surrounding mantle at greater depths and over long timescales is, to our knowledge, poorly constrained, but redox heterogeneities introduced into the deep mantle via subduction^46,55,62 conceivably play an important role in global geochemical dynamics, as documented by diamond-forming processes^62,63. The survival of such heterogeneities until recycling at hotspots could be behind the documented S isotope differences between OIB and MORB, and may explain their apparent redox contrast as well.

Methods

Sample collection and preparation

Lava balloons floating on the surface of the ocean were collected during the eruptive activity at El Hierro on 27 November 2011, 6 December 2011 and 28 January 2012. These samples were crushed and sieved, and phenocrysts of clinopyroxene and spinel were picked and mounted in epoxy to expose sulphide inclusions. Silicate melt inclusions and matrix glasses were previously mounted in indium and analysed for volatile abundances and sulphur speciation¹⁷.

Secondary ion mass spectrometry

In situ sulphur isotope measurements on these same glasses (³²S, ³⁴S) and on the sulphide inclusions (³²S, ³³S, ³⁴S) were performed using a CAMECA 1270 SIMS instrument at the University of California, Los Angeles (UCLA). A Cs⁺ primary beam was focused to a 10–15 µm spot size at a mass resolving power of 5500, with ³²SH⁻ fully resolved from ³³S⁻ for the triple isotope measurements.

The sulphides were measured in multi-collection mode with three Faraday cups, with beam current set at 5–6 nA. Charge buildup on the gold-coated sample surface was compensated using an electron flood gun. A cold finger surrounding the sample connected to a Dewar flask of liquid N₂ was used to reduce hydrides in the sample chamber. A contrast aperture and field aperture were used to exclude aberrant ions. Beam centering in the Field Aperture was scanned before each analysis. Ions from an energy window of 30 V were collected. The magnetic field was set and subsequently controlled by Nuclear Magnetic Resonance. Isotope counts per second (CPS) were measured in multi-collection mode with three Faraday cups. Slits of a common size on each detector, together with entrance slits, contrast aperture and energy window, yielded a mass resolution power of ~ 5500 (mass/∆mass) with ³²SH⁻ fully resolved from ³³S⁻. Faraday cup electrometers were calibrated immediately prior to analysis.

The analytical routine included a 60 s pre-sputter to clean the sample surface, centre the beam in the field aperture and achieve sputtering equilibrium before data collection. Measurements of each isotope in counts per second were collected in 6 cycles of 10 sec each to allow time for settling of the Faraday cup electronics between cycles. Standards used to calibrate the instrumental mass fractionation (Supplementary Table 1) included Balmat (pyrite), CAR123 (pyrite), Canyon Diablo Troilite and Anderson pyrrhotite^64,65, which was also used as our in-run internal standard. The resulting calibration curve is shown in Supplementary Figure 1 and has a slope of 0.517 ± 0.002 (2σ). Triple sulphur isotope measurements (n = 55; Supplementary Table 2) (³²S, ³³S and ³⁴S) were then obtained on exposed sulphide inclusions ranging in size from 30 to 100 μm. Regularly interspersed measurements (every 5–6 measurements) of our reference material, Anderson pyrrhotite (n = 15), gave a standard deviation (1σ) on δ³⁴S of 0.4‰, compared with its known δ³⁴S of 1.4 ± 0.3‰⁶⁴.

Double isotope analyses (n = 44; Supplementary Data 1) were conducted on glasses using instrumental parameters and analytical technique identical as above except for the following differences: (1) ³⁴S was measured on an electron multiplier; (2) the primary beam current was reduced to 3.15 nA in order to balance the errors associated with high count rates on the electron multiplier (deadtime error) and low count rates on the Faraday cup (background); (3) analyses consisted of 20 cycles of 10 s each. The standard used for calibration is P1326-2, described below.

Three repeat analyses of one large sulphide inclusion (EH_4-29), done at different times during the analytical session, show good reproducibility, yielding δ³⁴S and Δ³³S values of −0.64 ± 0.06‰ (1σ) and 0.00 ± 0.06‰ (1σ), respectively. Other sulphide inclusions analysed twice (n = 5) are reproducible within error, except for the most isotopically negative inclusion (δ³⁴S = −8.9 ± 0.4‰ and −10.2 ± 0.4‰).

Matrix glasses show more heterogeneity; for example, three analyses of a single glass chip (# 9.25) give a δ³⁴S range of 2.5‰, with an average of −0.9 ± 1.3 (1σ). Other glass chips analysed twice (n = 6) show as much as 3.8‰ difference between two measurements (see Methods for discussion of uncertainty).

Description of glass standard

Our glass standard used in SIMS analyses, P1326-2, is a MORB glass from the Juan de Fuca ridge, and has previously been described for major and trace elements by refs ^66,67. It is highly homogeneous and contains very few microlites. Its volatile concentrations (CO₂, H₂O, F, S, Cl) have previously been measured by SIMS, on separate occasions by refs ^17,67, which showed good reproducibility. Ref. ¹⁷ reports 340 ± 24 ppm CO₂, 0.29 ± 0.09 wt.% H₂O, 206 ± 6 ppm F, 1296 ± 26 ppm S and 187 ± 5 ppm Cl.

Isotope ratio mass spectrometry (IRMS)

The sulphur isotopic composition of P1326-2 was independently analysed by both SO₂ continuous flow at the University of New Mexico (Thermo Delta Plus XL mass spectrometer) and SF₆ dual inlet isotope-ratio mass spectrometry at McGill University (Thermo Scientific MAT 253). The SO₂ measurements were converted to the SF₆ scale using the regression of ref. ⁶⁸. Sulphur extractions were achieved in two ways: (1) the Kiba reagent extraction method, which extracts all S species, and (2) combined chromium reduction and Thode reagent methods to separately extract S²⁻ and S⁶⁺, respectively^69,70,71. The extracted sulphur was re-precipitated as silver sulphide (Ag₂S), before fluorination to SF₆ and subsequent isotopic analysis. For the separate sulphide and sulphate extractions, we used the relative weights of Ag₂S precipitates to estimate the S⁶⁺/ΣS ratio, followed by a mass balance to obtain a bulk δ³⁴S value. Replicate extraction and analysis (n = 3) give δ³⁴S = 0.8 ± 0.1‰ and S⁶⁺/ΣS = 0.20 ± 0.03. The full results of these analyses are summarised in Supplementary Data 4.

Electron probe micro-analysis (EPMA)

The major and minor element compositions of sulphides were determined at McGill University on a JEOL 8900 electron microprobe, using an acceleration voltage of 20 kV, a beam current of 30 nA and a beam size of 3 µm. The instrument was calibrated with the following standards: AsCoNi alloy (for As, Co and Ni), pyrrhotite and chalcopyrite (for Fe, S and Cu), galena (Pb) and sphalerite (Zn). A total of 55 analyses were obtained and are presented in Supplementary Data 2, including one analysis of Anderson pyrrhotite, which agrees very well with known Anderson pyrrhotite composition (Fe_0.87S)⁶⁴.

Scanning electron microscopy (SEM)

Elemental maps were acquired for six different sulphide inclusions (Supplementary Figs 3–4) using a Hitachi SEM at Queens College, City University of New York, equipped with a Bruker Quantax 400 energy-dispersive X-ray spectroscopy (EDS) detector.

Estimation of the S budget of magmatic sulphides

Thin section observations show that sulphides, while abundant in some clinopyroxene and spinel phenocrysts, actually represent a minor component of the El Hierro magma. A conservative estimate for the mass fraction of sulphide in the bulk rock is 0.01 wt.%. Assuming an FeS composition, the sulphides would then have ~36 wt.% S. This corresponds to an upper limit for the sulphide contribution to bulk S content of 36 ppm, hence <1% of the total initial S concentration (5080 ppm).

Calculation of uncertainty

The instrumental mass bias factor α_i on the isotopic ratio ³⁴S/³²S was calculated by comparing the average of the raw measured ratios (R_avg) of an internal calibration standard with its known isotopic ratio:

$$a_{\mathrm i} = {{R}}_{\rm avg}/{{R}}_{\rm known}$$

(3)

For the sulphide inclusions the standard was Anderson pyrrhotite (R_known = ³⁴S/³²S_known = 0.0442244 or δ³⁴S_V-CDT = + 1.4 ± 0.3‰)⁶⁴. The calculation of α_i for ³³S/³²S was done similarly, assuming a δ³³S_known falling on the regression line between δ³³S and δ³⁴S shown in Supplementary Fig. 1, i.e. dependent on δ³⁴S_known. The uncertainty on α_i is given by:

$${\rm{\sigma}} _\alpha = \sqrt {\left( {\frac{{{\rm{\sigma}} _{{\rm{avg}}}}}{{{{R}}_{{\rm{known}}}}}} \right)^2 + \left( { - {\rm{\sigma}} _{{\rm{known}}}\frac{{{{R}}_{{\rm{avg}}}}}{{{{R}}_{{\rm{known}}}^2}}} \right)^2}$$

(4)

where σ_avg is the standard deviation on the average of the raw measured ratios and σ_known the uncertainty on the known isotopic ratio of the standard.

The measured ratios ^xS/³²S (R_i) were then divided by α_i to obtain the corrected, ‘true’ isotopic ratios, and then converted to δ^xS_V-CDT values:

$$\delta ^{\rm{x}}{\rm{S}}_{{\rm{V}} - {\rm{CDT}}} = \frac{{{{R}}_{\rm{i}}}}{{{{R}}_{{\rm{V}} - {\rm{CDT}}}}} - 1$$

(5)

where R_V-CDT is the isotopic ratio of the international reference standard Vienna-Canyon Diablo Troilite⁶⁴. The propagated analytical error on δ^xS values is then calculated with the following equation:

$${\rm{\sigma}} _{\delta ^{\rm{x}}{\rm{S}}} = \left( {\frac{1}{{\alpha \times {{R}}_{{\rm{V}} - {\rm{CDT}}}}}} \right) \times \sqrt {{\rm{\sigma}} _{\rm{i}}^2 + \left( {\frac{{ - {{R}}_{\rm{i}}}}{\alpha }{\rm{\sigma}} _\alpha } \right)^2}$$

(6)

where σ_i is the analytical uncertainty on the raw isotopic ratios (R_i).

Once the corrected isotopic ratios are converted to δ³⁴S_V-CDT and δ³³S_V-CDT values, Δ³³S is calculated with the following expression:

$${\mathrm{\Delta }}^{{\mathrm{33}}}{\mathrm{S = \delta }}^{{\mathrm{33}}}{\mathrm{S}}_{{\mathrm{V - CDT}}}{\mathrm{- [(\delta }}^{{\mathrm{34}}}{\mathrm{S}}_{{\mathrm{V - CDT}}}{\mathrm{ + 1)}}^k{\mathrm{- 1]}}$$

(7)

where k = 0.5166 represents the slope of the mass-dependent fractionation line in δ³³S_V-CDT - δ³⁴S_V-CDT space (Supplementary Fig. 1). The analytical uncertainty on Δ³³S then becomes:

$${\rm{\sigma}} _{{\mathrm{\Delta }}^{33}{\rm{S}}} = \sqrt {{\rm{\sigma}} _{{\mathrm{\delta }}^{33}{\rm{S}}}^2 + {\rm{\sigma}} _{{\mathrm{\delta }}^{33}{\rm{S}}}^2 \times \left[ {k \times \left( {1 + \delta ^{34}{\rm{S}}} \right)^{ - \left( {1 - k} \right)}} \right]^2 + 2{\rm{\sigma}} _{{\mathrm{\delta }}^{33}{\rm{S}}{\mathrm{\delta }}^{34}{\rm{S}}} \times \left[ {k \times \left( {1 + \delta ^{34}{\rm{S}}} \right)^{ - \left( {1 - k} \right)}} \right]}$$

(8)

Duplicates were measured on 6 sulphide inclusions, and always agreed within uncertainty (1σ), except for the most ³⁴S depleted inclusion, which had individual δ³⁴S values of −8.9 ± 0.4‰ and −10.2 ± 0.4‰ (average of −9.6 ± 1.1‰).

For the sulphide inclusions, no instrumental drift was observed between the different standard measurements during the analytical session. For the melt inclusions and matrix glasses, analyses were conducted in a 2-day span, with 31 analyses on the first day and 13 on the second day. A linear correction was applied to the first day measurements to account for instrumental drift observed on measurements of our standard, P1326-2 (n = 24) (Supplementary Fig. 2). On the second day, no instrumental drift was observed.

The sulphur isotope data on glass measurements were corrected for instrumental mass bias (α_i) using the average of measured ³⁴S/³²S ratios on P1326-2 (day 1: n = 24, α_i = R_i/R_known = 0.99223; day 2: n = 13, α_i = 0.98871) compared to our independently determined value for P1326-2 (R_known = ³⁴S/³²S_known = 0.04420 or δ³⁴S_V-CDT = 0.9 ± 0.3‰).

Errors on δ³⁴S values were computed as for the sulphide inclusions and range from 0.1 to 0.8‰.

Modelling of S isotope fractionation

The positive correlation between fO₂ (in ΔFMQ log units, derived from S⁶⁺/ΣS ratios) of melt inclusions and their S content is used to estimate fO₂ variation during degassing (Fig. 2b), but as mentioned in the main text the highest S⁶⁺/ΣS values obtained are minimum estimates, and their variability in the matrix glasses is high. To account for this uncertainty, our approach has been to place maximum and minimum bounds on ΔFMQ values based on the log fO₂ (ΔFMQ) vs. S trend shown in Fig. 2b. This results in a sensitivity analysis both minimising and maximising the extent of S-isotope fractionation.

Vapour saturation pressures based on volatile concentrations in melt inclusions were calculated using the data of Longpré et al.¹⁷ and the model of Iacono-Marziano et al.⁴⁴. The relationship between S content and the calculated pressures for olivine-hosted melt inclusions is shown in Supplementary Fig. 5. Spinel-hosted melt inclusions were excluded because their CO₂ contents are thought to be affected by disequilibrium degassing and overestimate pressures significantly⁷². Matrix glasses were also excluded because they are supersaturated with respect to CO₂ (ref. ¹⁷), giving calculated pressures significantly in excess of the hydrostatic pressure corresponding to eruption depth. A polynomial fit was applied to the data in order to obtain input pressures as a function of S content in the degassing isotopic fractionation model. Five data points from the January 2012 sample were excluded from the computation of the polynomial fit, as they exhibit a slightly different degassing behaviour, characterised by lower H₂O and S content for comparatively high CO₂.

These trends can be compared to degassing paths obtained with the software D-Compress³⁹. This model requires the initial pressure, fO₂, CO₂ and H₂O contents of the magma as inputs, which we estimate from the most volatile-rich melt inclusion. We start with an initial CO₂ content of 3420 ppm and H₂O content of 3.05 wt.%, pressures varying between 300 and 400 MPa and fO₂ conditions between ΔFMQ = 1.2 and 1.7 (ref. ¹⁷). The programme outputs the initial S content and models each of these parameters (CO₂–H₂O–S–fO₂) during decompression. While the output from D-Compress generally provides good fits to the melt inclusion data (Fig. 2b, Supplementary Fig. 5), we note two main discrepancies: (1) the output S content reaches minimum values of >1100 ppm at low pressure (~1MPa), more than twice higher than what we measure in the matrix glasses (~500 ppm); (2) computed melt S⁶⁺/ΣS ratios stay relatively constant as fO₂ decreases. This decoupling of S⁶⁺/ΣS and fO₂ in D-Compress may arise because the S⁶⁺/ΣS vs. fO₂ relationship given by Jugo et al.³³, used in our conversion of S⁶⁺/ΣS ratios to ΔFMQ values, could not be implemented in D-Compress (A. Burgisser, personal communication). Nevertheless, there is general agreement that fO₂ decreases with S and H₂O degassing. This is also supported by XANES data on Fe³⁺/ΣFe ratios in El Hierro melt inclusions (Y. Moussallam, personal communication).

Code availability

An annotated version of the spreadsheet used to construct the isotopic fractionation trends in Fig. 2b is included as Supplementary Data 3.