High H2O Content in Pyroxenes of Residual Mantle Peridotites at a Mid Atlantic Ridge Segment

Global correlations of mid-ocean-ridges basalt chemistry, axial depth and crustal thickness have been ascribed to mantle temperature variations affecting degree of melting. However, mantle H2O content and elemental composition may also play a role. How H2O is distributed in the oceanic upper mantle remains poorly constrained. We tackled this problem by determining the H2O content of orthopyroxenes (opx) and clinopyroxenes (cpx) of peridotites from a continuous lithospheric section created during 26 Ma at a 11°N Mid-Atlantic Ridge segment, and exposed along the Vema Transform. The H2O content of opx ranges from 119 ppm to 383 ppm; that of cpx from 407 ppm to 1072 ppm. We found anomalous H2O-enriched peridotites with their H2O content not correlating inversely with their degree of melting, although H2O is assumed to be incompatible during melting. Inverse correlation of H2O with Ce, another highly incompatible component, suggests post-melting H2O enrichment. We attribute a major role to post-melting temperature-dependent diffusion of hydrogen occurring above the melting region, where water-rich melt flows faster than residual peridotites through dunitic conduits cross-cutting the uprising mantle. Accordingly, estimates of the H2O content of the MORB mantle source based on H2O in abyssal peridotites can be affected by strong uncertainties.


Results
We used infrared spectroscopy to measure the concentration of water or, more precisely, of hydroxyl (OH) defects, in both cpx and opx (see Methods). All measured pyroxene grains display several obvious absorption bands between 2800 and 3700 cm −1 (Supplementary Fig. 1) characteristic of the OH-stretching vibration regions in cpx and opx reported in earlier studies [16][17][18] . The H 2 O content of VLS peridotites ranges from 407 to 1072 ppm in cpx and 119 to 383 ppm in opx (Table 1) and shows negligible core-rim variability ( Supplementary Fig. 2). The H 2 O content of both cpx and opx in our samples displays weak positive or no correlation with the spinel chromium number (Cr# = 100Cr/(Al + Cr), in mole fraction) (Fig. 1a), a robust index reflecting degrees of melting of the host peridotites 19 . Similar results were obtained also between H 2 O content and other melting indices, such as Cr in pyroxene (Fig. 1b). These results are surprising because it is experimentally established and naturally observed that H 2 O behaves incompatibly during mantle melting, with a partition coefficient similar to that of Ce 19,20 ; therefore, the H 2 O content in melting residues is expected to decrease with degree of melting, contrary to our results. In fact, the concentration of Ce (Table 1) in the VLS peridotite pyroxenes anticorrelates with their degree of melting 9 , in line with Ce being incompatible but in contrast with the behaviour of H 2 O (Fig. 2).

Discussion
In order to explain these results, we explored two alternative possibilities. One, the correlation H 2 O content-degree of melting in our peridotites is due to processes occurring in the sub-ridge mantle during melting. Two, it is due to processes taking place after melting. The first alternative implies a number of assumptions. One is that the pyroxene H 2 O content is not modified during post melting uplift of the mantle peridotites. This assumption has been shown to be valid in ODP Leg 153 peridotites drilled from the Mid Atlantic Ridge near 23°N, where opx contain H 2 O in the 159-270 ppm range 21 . Also favouring this assumption is our observation that H 2 O is distributed homogeneously within individual opx grains; core-rim profiles show no obvious H 2 O content variations, suggesting no H diffusive loss or addition ( Supplementary Fig. 2). Moreover, cpx and opx H 2 O contents are positively correlated with a partition coefficient of 3.0 (Fig. 3), a value close to an average value of 2.6 obtained for pyroxenes in peridotites from oceanic ridges and xenoliths 22 . However, if the pyroxene H 2 O content of the Vema mantle-derived peridotites were due to sub-ridge melting processes, we would expect elements as incompatible as H 2 O, i.e., Ce, Nd, Yb, etc., to behave like H 2 O during melting. Their concentration in pyroxenes would then correlate positively with the concentration of H 2 O. This is clearly not the case, as shown for instance by a plot of Nd versus H 2 O in the Vema cpx ( Supplementary Fig. 3). Ce should also follow H 2 O and be higher in the depleted pyroxenes (Fig. 2). However, Ce in contrast to H 2 O, is lower in the depleted pyroxenes 9 .
The behaviour of H 2 O, different from that of elements with similar partition coefficients, is hard to reconcile with the distribution of H 2 O in the Vema mantle peridotites being due solely to partial melting processes. Given the degrees of melting of the Vema peridotites estimated from Cr# of spinel and pyroxenes, and given the of 0.008. These additional arguments suggest that the observed water enrichment reflects post-melting hydration in the mantle.
Among post-melting processes we consider first serpentinization and contact metamorphism. The influence of serpentinization is unlikely because hydrogen diffuses sluggishly into pyroxene grains at low temperature (<300 °C) and pressure [26][27][28] . Serpentinization of the Vema peridotites at the depth of <4 km, occurred mostly near ridge axis 14    Low-pressure contact metamorphism can occur in the lower oceanic crust where discontinuous magma chambers are surrounded by mantle rocks. Degassing from a magma chamber can potentially refill depleted rocks by diffusion in the aureola zone. Aggregated melts differentiating in a magma chamber are characterized by high δ 18 O values; such a process would therefore produce a positive correlation between water content and δ 18 O. This is clearly not the case in the Vema peridotites, as shown in Supplementary Fig. 5.
Experimental studies of the influence of water on melting and phase assemblages in the upper mantle have shown a water content of ~200 ppm in residual nominally anhydrous minerals after incipient melting of lherzolite at the vapour-saturated solidus with pressure ranging from 2.5 to 4 GPa 30-32 . This raises the possibility that some of the investigated upper mantle peridotites may represent parcels of the upper mantle that did not go through significant melting; in fact, the maximum water contents in both pyroxenes of Vema mantle-derived peridotites resemble closely those found in these studies (~300 and 900 ppm for opx and cpx, respectively). However, mantle rocks exposed along the VLS formed originally towards the northern edge of the 80-km-long eastern MAR segment from the sub-axial mantle column, which inevitably went through the melting region. Given the divergent upwelling flow of the solid mantle beneath a ridge segment, it is unlikely that off-axis mantle rocks after incipient melting may converge towards ridge axis.
One possible H 2 O-rich source for post-melting enrichment of the subridge rising mantle might be H 2 O-rich, low degree melts originating by off-axis incipient melting. The production of these low degree melts at the edges of the sub-ridge melting region has been hinted at by several studies 21,22,27,28,[30][31][32][33] . These small quantities of H 2 O-rich incipient melt may be channelled toward the axis along the sloping base of the thermal lithosphere. Distal melts trapped in the lithosphere-asthenosphere boundary may carry amounts of H 2 O comparable to those we estimated in equilibrium with Vema residual pyroxenes: incipient melts from depleted upper mantle with ~200 ppm H 2 O would contain about 2-3 wt% H 2 O 24,34 .
Oceanic peridotites, representing fragments of the uppermost zone of the subaxial melting column, show often vein lithologies suggesting interaction with melts [35][36][37] . We explore the hypothesis that the H 2 O rich melts recorded by our Vema residual peridotites are fractions of distal low-degree melts that migrated along lithosphere-asthenosphere boundary channels towards ridge axis (Fig. 4). These low density/low viscosity melts may tend to accumulate toward the top of the sub-ridge melting column and to react with residual peridotites before dispersing within the low-H 2 O aggregate melts in sub-axial dunite channels.
The H 2 O content of peridotite pyroxenes may help estimate the H 2 O content of the interacting melts, thanks to the experimentally established hydrogen partition coefficients between olivine and pyroxenes = . ± .  38,39 . In fact, the amount of water in olivine may be determined from the water content of coexisting pyroxenes using mineral-mineral partition coefficients and the bulk water concentration for peridotite may be then estimated using mineral modes 22 (see Methods). Accordingly, the melt H 2 O content retrieved from pyroxenes H 2 O contents may range from 1.0 to 2.5 wt% (average 1.6 ± 0.4 wt%) when olivine H 2 O content is estimated from opx, and from 1.0 to 3.0 wt% (average 1.8 ± 0.6 wt%) when estimated from cpx ( Table 2). The differences between melt water contents estimated from opx and from cpx ( Table 2)    O melts are unlikely to form by "normal" subridge high-degree partial melting; they may be generated only by very low degrees of melting ranging from ~0.1% to ~0.9%, assuming equilibrium hydrous melting 42 and 200 ppm H 2 O in the mantle source. These low amounts of incipient melts reflect a low interconnectivity between solid matrix grains and thus, a low attitude to melt migration. A simple compilation of global data indicates that, although never reaching the extreme Vema enrichments, excess water contents are common in "residual" mantle rocks worldwide (Fig. 5 and  Supplementary Table 2). Global water contents estimated for equilibrium melts statistically assembled by ridge segments (Fig. 5) do not match any MOR basalts nor mineral-hosted melt inclusions 5,25 . In general, melt H 2 O contents increase with increasing degree of fractionation (decreasing MgO) 43 ; an alternative explanation could be that conductive heat flux from the surface causes cooling along the flanks of the melting region. As a consequence, most of the low degree melt produced in the distal parts of the H 2 O rich melting region migrating towards the lithosphere-asthenosphere boundary (LAB) crystalizes and fractionates, increasing the melt H 2 O content and decreasing the melt freezing point 43,44 .
Next we will investigate the role of post-melting metasomatism and re-equilibration of water in the nominally anhydrous mantle rocks minerals, assuming that H 2 O-rich incipient melts may be extracted and focused toward ridge axis. Refractory peridotite residues may be more susceptible to shallow mantle metasomatism than fertile lherzolites 45 . Therefore, the slight positive correlation between H 2 O content and degree of melting may simply reflect post melting metasomatic enrichment of incompatible elements, including H 2 O, in the more refractory peridotite. Shallow mantle metasomatism should mobilize not only H 2 O, but also LREE. However, the VLS peridotites are characterized by cpx with LREE-depleted patterns 9 and spinel with less than 0.1 wt% TiO 2 , indicating weak or no mantle metasomatism 9 . Modelling based on REE and Ti-Zr in residual cpx also indicates weak re-fertilization of the residual source by small (~0.2%) amounts of partially aggregated melt 9 . If re-fertilization was the main cause, we would expect the peridotite H 2 O content to correlate positively with chemical indices reflecting the extent of metasomatic/melt-rock reactions such as Ce/Yb, and Na 2 O in cpx (Supplementary Fig. 3). Absence of these correlations suggests refertilization is not the main cause of H 2 O enrichment in the VLS peridotites ( Supplementary Fig. 6).  25 . Isotherms are indicated by thin red lines. White thick dashed line marks the region of anhydrous melting, i.e., the sub-region where water is completely exhausted from peridotite nominally anhydrous minerals 25 . Solid thick blue line marks the upper boundary of the region that contributes to melt production (full rainbow scale), i.e. where production rate is positive. The lighter rainbow scale area marks the mantle region where a parcel of melt with a given degree of melting freezes if not extracted from the melting region, i.e., where production rate is negative. The 1100 °C isotherm is assumed to approximate the lithosphereasthenosphere boundary layer (LAB) 65 . Water-rich low degree melts, produced at the edge of the sub-ridge melting region, percolate at the base of the lithosphere (blue ellipses) where they migrate towards ridge axis. www.nature.com/scientificreports www.nature.com/scientificreports/ Hydrogen diffusion into pyroxenes at mantle depths after cessation of sub ridge partial melting might have occurred in East Pacific Rise 27 (Hess Deep) and Mid Atlantic Ridge 21,46,47 peridotites. H diffusion in and out of pyroxenes is favoured by high temperature. The Vema mantle peridotites have risen slowly from mantle depths after melting has stopped roughly 20 km below the seafloor (Fig. 4). The rate of ascent is similar to the half spreading rate 8 , i.e., ~15 mm/yr: the ascent within the mantle up to the base of the sub ridge lithosphere (~5 km below seafloor) lasts roughly 1 Ma. A good interval of this upward motion will take place at temperatures ranging from ~1250 °C to ~900 °C. During ascent of the peridotite, melts may flow in dunitic conduits and veins cross cutting the uprising mantle. The spacing of these conduits could be quite narrow 48 for a nearly continuous melt extraction as we observe at the VLS. Given its high diffusion coefficient, H could diffuse from H-rich to H-depleted zones in the surrounding mantle 49 . The diffusion coefficient D of H in pyroxene within the predicted temperature range will vary from ~1 × 10 −10 to ~1 × 10 −12 m 2 /s, while REE diffusion coefficients are of the order of 1 × 10 −21 m 2 /s, that is, orders of magnitude lower. Thus, in 1 Ma hydrogen could travel as far as ~11 ÷ 112 m (diffusion length scale = 2*√D t), while a REE would be able to travel only ~0.35 mm. This process would explain why H is decoupled from other incompatible elements in our peridotites ( Supplementary Figs. 3 and 7). Higher hydrogen diffusion coefficients in cpx than that in opx would also explain why hydrogen has been mostly taken   www.nature.com/scientificreports www.nature.com/scientificreports/ up by cpx rather than by opx, since mineral-mineral (cpx-opx) water partition coefficients recorded by our data are about twice those determined experimentally (Fig. 3). However, the hydrogen diffusion coefficients in natural opx reported in literature 50 are of the same order of magnitude as hydrogen diffusion coefficients for diopside 51 , although few experiments are available for natural orthopyroxenes, with total absence of hydrogen diffusion data for pure enstatite 52 .
Diffusion of hydrogen was already suggested to explain low variability of water content relative to Ce in nominally anhydrous mantle minerals from Colorado Plateau peridotite xenoliths 49 . However, hydrogen diffusion requires a prolonged post-melting residence time of mantle rocks at mantle conditions. The negative correlation between opx water contents and mantle equilibration temperatures observed comparing peridotites from ODP Legs 153 and 149 supports this idea 46,47 , i.e., lower equilibration temperatures imply longer H diffusion times during mantle rise and consequently higher water contents. Oxygen isotopes tell here the same story. In fact, δ 18 O and water content for opx and cpx define a slight negative trend ( Supplementary Fig. 5) suggesting that at nominally zero W/R ratio, high temperature H 2 O diffusion may have produced a slight depletion in oxygen isotope composition. High temperature H 2 O incorporation occurs upon cooling according to the oxygen isotopic equilibrium fractionation in opx and cpx, that in turn depends on equilibration temperatures. This explains why opx and cpx oxygen isotope contents do not correlate positively with their H 2 O contents.
Our post-melting models of H 2 O enrichment of sub-ridge mantle peridotites do not exclude that different parcels of pre-melting mantle along the VLS may have contained different amounts of H 2 O, and that these different H 2 O contents, in addition to temperature, may have caused temporal variations of degree of melting. However, post-melting H diffusion and H 2 O redistribution make it difficult to reconstruct pre-melting H 2 O contents. If our models are correct, it would follow that, due to post melting mobility of H in the suboceanic upper mantle, estimates of the H 2 O content of the pre-melting mantle source of MORB based on the H 2 O content of abyssal peridotite pyroxenes, may be affected by strong uncertainty.

Methods ftiR measurements of H 2 O content.
Water content determinations followed methods of ref. 53 . Infrared measurements were carried out in 10 to 15 relatively large and clean grains of each mineral picked from the peridotites. The selected grains were mounted in a self-supporting epoxy matrix, and double polished to a thickness of ~0.2 mm. We obtained infrared unpolarized spectra at wavelengths ranging from 1250 to 4000 cm −1 on a Nicolet 5700 FTIR spectrometer, coupled with a Continuum microscope at USTC, using a KBr beam-splitter and a liquid-nitrogen cooled MCT-A detector. A total of 128 scans were accumulated for each spectrum at a 4 cm −1 resolution. The aperture size was set from 30 × 30 μm to 100 × 100 μm, depending on the size and quality of the mineral grains to be analyzed. Accurate determination of OH concentrations in optically anisotropic minerals obtained by non-polarized light on unoriented grains was proven to be a reliable method both theoretically 54 and practically 55 . The mineral water content was calculated by the transformed Beer-Lambert law: where C is the water content of minerals in ppm, A is the non-polarised integral absorbance, I is the absorption coefficient (7.09 ppm −1 cm −2 for cpx and 14.84 −1 cm −2 for opx, ref. 56 ) and t is the thickness in cm, measured by a digimatic indicator for each grains. The average value was used to obtain water content. Uncertainties in water contents calculated from Eq. (1) derive from: (i) using non-polarized infrared beams on non-oriented minerals (<10%); (ii) baseline correction (<5%); (iii) variable sample thickness (<3%); and (iv) differences between the absorption coefficients (<10%) of our samples and those of the samples used by ref. 57 due to differences in composition. The total uncertainty is estimated to be less than 20-30%.
Water content of melt in equilibrium with residual peridotite. The water contents of the percolating melts in equilibrium with the residual peridotites were estimated calculating bulk water concentration of peridotites using mineral modes and determining bulk H 2 O partition coefficient between peridotite and melt using experimentally determined partition coefficients between minerals and melt 58 : are the partition coefficients of water between bulk peridotite and melt, and between olivine-pyroxenes and melt; and X j are the mineral abundances of olivine, cpx and opx, respectively.
The partition coefficients of water between minerals and melt are given by definition: Partition coefficients of water between peridotite mineral assemblages and melt determined experimentally may depend from mineral chemistry (i.e., Al 2 O 3 ), oxygen fugacity and P-T conditions [22][23][24]38,39 . Here we assume water partition coefficients determined experimentally at upper mantle pressure (1-3 GPa) and temperature (1230-1380 °C) of ref. 38  ). Assuming that the mineral phases are in equilibrium with the melt and that the adopted partition coefficients reflect exactly the www.nature.com/scientificreports www.nature.com/scientificreports/ observed equilibrium, then Eqs. (2) and (3) are equivalent. Taking into account uncertainties in the mineral water content determinations and in the experimental determined partition coefficients, Eq. (3) provides different values of melt water content for each mineral phase of the residual peridotite:  Our samples contain no olivine relicts. Although we were able to estimate mineral abundances in our samples following methods of refs. 9,59 (Tab. 2), we were not able to measure any olivine water contents. Thus, olivine water contents have to be inferred from water contents of pyroxenes adopting mineral-mineral partition coefficients, i.e.: are mineral-mineral partition coefficients. Here, we adopted those of ref. 38   for each mineral phase abundances (X j ). These results suggest that melt water contents estimated from opx and cpx represent the lower and upper limits of water contents of melts in equilibrium with residual peridotites regardless of their mineral abundances.
Oxygen isotopic ratios. Following method of ref. 60 , oxygen isotopes were measured at the Consiglio Nazionale delle Ricerche-Istituto di Geoscienze e Georisorse of PISA by laser fluorination 61 , reacting 1 to 1.5 mg opx and cpx fragments in F2 gas 62 . We irradiated the samples with a 25 W CO 2 laser operating at a wavelength of 10.6 μm (ref. 63 ). Three pre-fluorination steps were made before running new sets of analyses in order to remove the moisture in the holder and in the line. O 2 produced during laser fluorination together with excess fluorine were passed through potassium chloride salt; excess fluorine was converted into a potassium-fluoride salt and chlorine gas. A cryogenic trap cooled at liquid nitrogen temperature was then used to freeze chlorine. After purification, O 2 was trapped over a cold finger filled with 5 A zeolites 62 , and transferred to a Finnigan Delta Plus Mass Spectrometer for oxygen isotopic analysis. The international quartz standard NBS 30 and in-house laboratory standard Quartz Merck Standard (QMS) were measured at the beginning of each analytical session. Mineral sequences were started after the standards reached the accepted values: five to six standards were measured during each set of analyses. The average δ 18 O value of NBS30 and QMS is 14.05 ± 0.17‰ (1σ), and 5.24 ± 0.15‰ (1σ), respectively. All δ 18 O values are relative to SMOW (standard mean ocean water, 18 O/ 16 O = 2005.2 × 10 −6 ). At least two fragments were analyzed for each mineral, and the variations within the same sample are less than the precision of standards.