Element concentration maps of clinopyroxene from sample DS2217 show that the core of the host grain between the monomineralic lamellae of albite is depleted in Al, Fe, and Na compared to the grain periphery (Fig. b–d). There, these elements occur spatially concentrated, showing thin straight lines parallel to one of the cleavage plane directions of the host grain, which coincides with one of the albite lamellae directions and the orientation of very thin lamellar inclusions between grain rims and grain cores (Fig. a; for higher magnification photos, the reader is referred to the previous study by Fig. 5e and f). These tiny inclusions between the grain rims and cores of the clinopyroxene appear to occur only in sample DS2217. They were not observed by optical methods in any other sample in the series and were not described in the previous study.

Clinopyroxene, which hosts bimineralic lamellae of quartz and pargasite, has element concentrations around these inclusions that are depleted in Al content and enriched in Ca content; an example of this is sample M65 (Fig. S2).

As outlined in Sect. , we calculated the hydroxyl content of the NAMs with two independent calibrations: and . The hydroxyl contents determined with the calibration of are 45 ± 2 % and 42 ± 4 % lower for clinopyroxene and garnet, respectively, than those of and 30 ± 5 % higher for orthopyroxene. The results of both calibrations are reported in Table , but, in order to avoid confusion, in the following, we refer only to the values determined using the absorption coefficient of .

The approach of treating absorption bands (6) and (7) in garnet separately for the quantification of hydroxyl shows that the difference between minimum and maximum estimates, by using bands (1–5) and (1–7), respectively, for a given calibration does not exceed 17 µgg-1 and is on average 6 ± 4 µgg-1 (Table ). This rather small difference particularly affects the garnet grains with the lowest hydroxyl content but is unlikely to be relevant for the interpretation of the data set. For reasons of simplicity, the following discussion will refer to the minimum estimates for hydroxyl in garnet based on the integral absorbances of bands (1–5).

Individual infrared spectra show that the structural hydroxyl content in grain rims and cores varies for a particular sample and mineral phase (Fig. ). This variation includes differences in absorbance resulting from different orientations of anisotropic grains in non-polarised light. Therefore, the average values obtained from 2–5 grains separately for rims and cores (Table ) likely have an uncertainty of 25 % or less , which causes many of these average values to overlap within uncertainty. However, sample-specific rim / core ratios of these average hydroxyl contents are systematically below unity for orthopyroxene (Fig. ), suggesting that orthopyroxene rims experienced late hydroxyl loss. The ratios for clinopyroxene range from unity upwards, which is consistent with the uptake of hydroxyl at the grain rims in some samples. The intragranular distribution of hydroxyl in individual grains of orthopyroxene and clinopyroxene demonstrates that the difference between core and rim contents is beyond analytical uncertainty (Figs. S3 and S4). The ratios for garnet cluster around unity, indicating indifferent behaviour. The late hydroxyl loss in orthopyroxene should be considered when interpreting the water content of this mineral from highly retrograde samples.

The hydroxyl contents determined with the calibrations of and differ significantly. Both calibrations are based on different approaches. While the former uses wavenumber-dependent molar absorption coefficients, the latter calibration relies on mineral-specific counterparts. In the case of clinopyroxene, the absorption band at approximately 3350 cm-1 observed in most of the WGR samples (Table ) is absent in the augitic clinopyroxene used in the calibration of . Besides, the quantification after has been shown to be in good agreement with the εi of , whose structural H2O content was determined by secondary ion mass spectrometry, and to be more generally applicable to clinopyroxene . These aspects suggest that the hydroxyl content in the analysed clinopyroxene is more likely in the range of 31–384 µgg-1. Similar arguments could be made for the hydroxyl contents quantified for orthopyroxene and garnet. For example, the analysed orthopyroxene shows no absorption at the wavenumbers 3060 and 3300 cm-1, which were used in the calibration of , which in turn shows no absorption at about 3625 cm-1. All garnet absorption wavenumbers used in the calibration (3512 and 3571 cm-1) barely agree in number and position with those observed in this study. For this reason, more weight may be given to hydroxyl contents quantified using the spectrum-specific approach, and their mineral-specific counterparts may be regarded as maximum estimates. However, since the two estimates differ by less than 0.5 orders of magnitude and therefore are not expected to have any influence on the interpretation of the data set, and for comparison purposes, the values from were used in the diagrams.

Evidence for UHP metamorphism was obtained earlier for all studied eclogite bodies, either by the current mineral chemistry applied to geothermobarometers, index mineral inclusions, reconstructed mineral chemistries from mineral microstructures, multi-equilibrium thermodynamic calculations, or a combination of the above methods . There are two lines of evidence for differences in hydroxyl content during this UHP metamorphism. One relates to sample DS1438, whose garnet hydroxyl content is 2 orders of magnitude higher than that of garnet from any other sample (Fig. c). Since this is the only sample from an outcrop studied whose peak metamorphic mineral assemblage contains hydrous minerals, zoisite, and phengite , the high structural hydroxyl content in its garnet is most likely related to the more hydrous conditions for the whole rock during peak metamorphism. Possible hydroxyl sources may include an inherited (pre-UHP) hydroxyl-rich precursor whole-rock chemistry or local fluids during UHP metamorphism. The hydroxyl concentration ratio between clinopyroxene and garnet in this sample is close to unity (1.2), lower than in any of the studied samples (3.5–81.8) but similar to that of zoisite-bearing mantle eclogite (1.5–2.0; ; Fig. ).

The hydroxyl content of clinopyroxene provides the other indication. All eclogite samples with evidence for equilibration at UHP by orthopyroxene barometry, i.e. by the current mineral chemistry in addition to mineral microstructures (DS0326, 2-4A, M65, DS1409), have cores of clinopyroxene with relatively uniform hydroxyl content (114–146 µgg-1; Table ). All other samples that have evidence for UHP metamorphism only by mineral microstructures but not the current mineral chemistry show large variation across all such samples (40–539 µgg-1). In addition, the highest hydroxyl content does not occur in clinopyroxene of the zoisite-bearing sample (Fig. a). This suggests that the post-peak metamorphic evolution of eclogite modified the hydroxyl content of the minerals.

Importantly, the eclogite with hydrous garnet (sample DS1438) and the orthopyroxene-bearing eclogites with anhydrous garnet (2-4A, DS0326, DS1409, M65) share oriented mineral inclusions of pargasite + quartz in clinopyroxene. This indicates that the oriented mineral microstructure was formed after UHP metamorphism regardless of the presence or absence of a hydrous mineral (i.e. differences in hydrous conditions) during UHP metamorphism. The fact that the remaining bimineralic eclogite samples (DS1405, DS2204, DS2217, UL-96-2) have oriented inclusions without pargasite in clinopyroxene cores, but have clinopyroxene with the lowest and highest hydroxyl contents in the entire sample suite (Fig. ), suggests that the processes that formed the extreme mineral hydroxyl contents and the pargasite lamellae are different (i.e. decoupled from each other). It should be noted that the major mineral element chemistry of garnet from all samples (Table S3) obtained from shows little to no correlation with the hydroxyl content of clinopyroxene (Fig. S5). Likewise, the hydroxyl content of orthopyroxene is either the lowest of the three major minerals or similar to that of garnet (Fig. ). Apparently, the presence or absence of orthopyroxene in the mineral assemblage also had no significant influence on the hydroxyl content or the hydroxyl distribution between the other two major minerals.

The Al content in orthopyroxene, which shares a mineral paragenesis with garnet, is known to sensitively record metamorphic pressures and is therefore often used as a geobarometer. The combination of such pressure estimates with the total average structural hydroxyl content of clinopyroxene, i.e. the major host for water in eclogite in the absence of hydrous minerals, shows an inverse correlation with individual values that largely overlap in analytical uncertainty (Fig. ). The rather large uncertainty results from the low statistics on anisotropic pyroxene. However, when the rim average hydroxyl content of clinopyroxene is used, the inverse correlation is more pronounced and the increase in hydroxyl content exceeds possible orientation effects (Fig. ). This correlation indicates that decompression led to the incorporation of hydroxyl into clinopyroxene. The increase in structural hydroxyl content observed within individual grains from the clinoyproxene core to the rim regions, in addition to the overall averages, indicates (i) a reliable intracrystalline difference in hydroxyl content beyond analytical uncertainty, which (ii) affects pre-existing grains after the formation of a unique mineral inclusion microstructure (Fig. S4).

Since all samples are decompressed (i.e. collected from the surface), the cause of the correlation shown in Fig.  is likely not the pressure change itself but rather the mineral–chemical response to decompression (i.e. re-equilibration) that allowed clinopyroxene to incorporate additional hydroxyl. An inverse correlation between the jadeite component and the total structural hydroxyl content in clinopyroxene supports this assumption (Fig. S5b). Orthopyroxene tends to show a similar inverse relationship in the metamorphic pressure range of 4.0–5.5 GPa, which is consistent with the experimentally demonstrated dependence of the amount of structural hydroxyl on the Al content . An exception is the strongly retrogressed sample DS2216, which shows a lower orthopyroxene hydroxyl content than expected. This low content could be related to late hydrogen loss, which is particularly plausible given the fractured nature of the crystals, as this allows diffusion to also efficiently affect the crystal core regions (cf. Sect. 5.1 and Fig. 8c in the preceding study).

Although the orthopyroxene-free eclogites have clinopyroxene with a large variation in hydroxyl content (Fig. a), the samples (with and without orthopyroxene) with the highest abundances in clinopyroxene (DS2216 and DS2217, respectively) are only a few hundred metres apart (Fig. ). This suggests that the structural hydroxyl variation in orthopyroxene-free eclogite is in part also related to different degrees of retrogression.

The inverse correlation between structural hydroxyl in clinopyroxene and quantified pressure (Fig. ) indicates that the retrogression after UHP was related to the availability of water or fluids. The same conclusion can be drawn independently from the hydroxyl content in individual crystals of clinopyroxene, which increases from the grain cores to the rims; this gradient overprints a pre-existing decompression mineral microstructure (Fig. S4). The close proximity of chemically distinct eclogites with the highest structural hydroxyl content in clinopyroxene suggests that the main pathways for the fluids may have been the foliation planes. A rough structural extrapolation of eclogite exposure with the highest structural hydroxyl content in clinopyroxene (or degree of retrogression; samples DS1405, DS2216, DS2217) could indicate two corridors between the coast and the hinterland in which efficient retrogression occurred (Fig. ). More samples would have to be analysed to test this hypothesis. Nevertheless, the presumed corridors are situated in between two formerly separated UHP areas and may explain why evidence for UHP metamorphic conditions in eclogite exposed in this area was difficult to detect for a long time.

The source of hydroxyl structurally bound to pargasite occurring as oriented lamellae together with quartz in clinopyroxene (Figs. b, S1, S2, and S4) could either be external (e.g. from an infiltrating fluid) or internal (from the water content of a precursor clinopyroxene). Several arguments are against the first variant and in favour of the second variant. First of all, an infiltrating fluid that caused the formation of pargasite in clinopyroxene within the amphibole stability field would be expected to have simultaneously infiltrated associated orthopyroxene crystals. If this is true, then orthopyroxene is expected to record metamorphic conditions in the stability field of amphibole but not of diamond (Fig. ). Furthermore, the occurrence of bimineralic inclusions in clinopyroxene is independent of the structural hydroxyl content of the host mineral, which increases with retrogression and is thus a function of retrogression (Fig. ). Since the bimineralic inclusions occur regardless of the degree of retrogressive overprint (i.e. in peak metamorphic and retrogressed samples), their origin is unlikely to be related to the availability of external fluids. In addition, the compositional halos around the pargasite and quartz inclusions of more diopsidic, less aluminous clinopyroxene indicate that the pargasite formation resulted from the local decomposition of the jadeite component rather than an external fluid (Fig. S2). Finally, the K2O content of the lamellar pargasite (Table S1) is close to the detection limit of the electron microprobe but should, arguably, be significantly higher if the lamellae were in contact with an infiltrating fluid due to the fluid mobility of K, as is assumed for comparable samples from the Alps .

With three exceptions (DS1438 as part of a zoisite–eclogite and DS1405 and DS2217, which contain abundant secondary plagioclase), the clinopyroxene host has up to several hundred micrograms per gram (µgg-1) lower structural hydroxyl content than is typically found in comparable samples, e.g. coesite- and quartz-eclogite from the Erzgebirge and the Kokchetav massif and eclogite xenoliths (both metasomatised and pristine) reported from the Siberian, Slave, Kaapvaal, and West African cratons (; Fig. ). This deficit in structural hydroxyl, when considered with the general deficit in molecular water, suggests that the monomineralic quartz lamellae (Fig. c and d) formed under comparatively dry conditions, as was apparently the case for the bimineralic lamellae of quartz + pargasite (Fig. a and b). Reintegration of hydroxyl currently bound in lamellar pargasite into the clinopyroxene precursor chemistry would increase its structural hydroxyl by about 200 µgg-1 per percent lamellar volume. The volume fraction of lamellar pargasite was not quantified, but backscattered electron images suggest only a few percent (Figure S1b). If combined with the low hydroxyl content of the host (119–364 µgg-1 or less; Table ), then up to several volume percent of pargasite could have theoretically exsolved from a precursor clinopyroxene, since clinopyroxene is capable of accommodating up to more than 2000 µgg-1 of structural hydroxyl . Fluid-mediated metasomatism, on the other hand, has been shown to form amphibole lamellae in clinopyroxene and orthopyroxene in the absence of quartz lamellae . Such lamellae have not been previously reported from the samples studied but appear to occur as parallel submicrometer-sized inclusions in clinopyroxene in sample DS2217, as indicated by a strong absorption peak at 3622 cm-1 (Fig. a; ) and the spatial concentration of Na and Al typical for amphibole and layered silicates (Fig. ). The size (thickness), mineralogy (close spatial association with quartz lamellae), and occurrence (presence in clinopyroxene but absence in orthopyroxene) of the pargasite lamellae in samples of this study differ from the microstructure of the amphibole lamellae in the samples thought to have formed by metasomatism reported by and provide little support for the assumption that the bimineralic lamellae formed by a similar process.

UHP eclogites of the Kokchetav massif show that hydroxyl absorption in clinopyroxene increases with the vacancy concentration in the pyroxene structure (i.e. the Ca-Eskola component; ). This relationship has been confirmed experimentally , suggests that samples rich in exsolved quartz lamellae may also be rich in pargasite lamellae, and is consistent with a cogenetic origin of the observed bimineralic lamellar inclusions.

Finally, the distribution of structural hydroxyl between clinopyroxene and garnet, DCpx/Grt = cH2OCpx/cH2OGrt, is greater than 40 for the most overprinted samples (i.e. those with the highest hydroxyl content in clinopyroxene: DS1405, DS2217) and significantly less than 40 for all others (Fig. ). This systematic is consistent with experimental data showing an opposite pressure dependence of the hydrogen content in both mineral phases and references therein. By implication, the oriented mono- and bimineralic lamellae in the eclogitic clinopyroxene formed at high pressure and before the retrogressive overprinting that caused an increase in structural hydroxyl in clinopyroxene (Fig. ). The source of the hydrous fluid is unconstrained, but the infiltration of fluids can explain the crystallisation of hydrous phases such as the matrix biotite, which is a source of rheological weakening and thus strain partitioning. Such areas of increased deformation of gneiss have recently been shown to be more effective in retrogressive overprinting of enclosed UHP eclogite .