Thank you for your comments on our manuscript, we really appreciate them.

The sample with the highest hydroxyl content in garnet (DS1438) comes from an outcrop previously studied by Terry et al. (2000; 10.2138/am-2000-11-1207), which provides metamorphic estimates of 34-39 kbar and 820 °C (sample 1066b). These estimates are based on the garnet-clinopyroxene thermometer of Krogh Ravna (2000; 10.1046/j.1525-1314.2000.00247.x) in combination with multi-equilibrium thermobarometry using the TWQ program version 2.02 of Berman (1991; Can. Min. 29:833). Both samples contain polycrystalline inclusions in clinopyroxene of quartz, which are interpreted to be after coesite and thus can be considered as independent evidence of earlier UHP metamorphism. The enstatite-in-clinopyroxene thermometer of Nimis and Taylor (2000; 10.1007/s004100000156) applied to sample DS1438 suggests 788+/-30 °C (for a choosen pressure of 40 kbar, Spengler et al., 2023), which overlaps in error with the temperature derived from sample 1066b. Therefore, the metamorphic temperatures and (minimum) pressures applicable to sample DS1438 are fairly well constrained. However, a low H2O activity of 0.00036 is included in the estimates of Terry et al. (2000), who argue that "This low activity of H2O may be consistent with the proposed very low or proposed absence of H2O content in fluid inclusions in garnet in the [nearby] microdiamond-bearing kyanite gneiss." This may suggest that the reason for the high hydroxyl content in the garnet of sample DS1438 is less an indication of the availability of external fluids during UHP metamorphism than of (the presence or type of nominally hydrous minerals in) the prograde mineral composition (whole rock chemistry), which differs markedly between the eclogite samples studied and the nearby microdiamond-bearing kyanite gneiss. Since sample DS1438 is the only one in the set of 10 eclogite samples that contains a nominally hydrous mineral during UHP metamorphism, we consider this sample less suitable to compare with systematics in the other samples. Unfortunately, we are not aware of any experimental study on the influence of water saturation on the hydroxyl content of nominally anhydrous minerals during UHP metamorphim.

All clinopyroxene grains analysed contain oriented inclusions of quartz (some together with pargasite), which may indicate a Ca-Eskola component in the precursor clinopyroxene, which in turn, as experimentally determined, dramatically increases the incorporation (saturation level) of hydroxyl in clinopyroxene (Bromiley and Keppler, 2004; 10.1007/s00410-003-0551-1). If one considers the reverse process, i.e. the reduction of the Ca-Eskola component by exsolution of quartz during early decompression, the hydroxyl content can be above the HP saturation level, so that a nominally hydrous phase is formed at the same time. Such pargasite (in close association with quartz) was found, with the exception of sample DS1438, only in clinopyroxene of Opx-bearing eclogite. This is consistent with the view that the hydroxyl in clinopyroxene of DS1438 was close to saturation (or at least comparably high) during UHP conditions, as in all Opx-bearing eclogites, but different from the remaining Opx-free eclogites (which have oriented quartz but no pargasite in clinopyroxene). Once pargasite has formed by exsolution in clinopyroxene, the hydroxyl content in the clinopyroxene host grain is expected to have decreased. However, further decompression allows for an increase in the hydroxyl content in clinopyroxene, as the (new) saturation level (of Ca-Eskola-free clinopyroxene) increases with decreasing pressure (Bromiley and Keppler, 2004). In a natural environment, such an increase may require the availability of water, which we believe may be related to the degree of retrogression in the samples (our Fig. 9). For clarity, it is perhaps better to note that the hydroxyl content in clinopyroxene measured and shown in Fig. 9 is not the hydroxyl content that was present at UHP conditions (because at UHP, the oriented inclusions of quartz + pargasite were still dissolved and therefore the OH content in clinopyroxene was higher). All OH amounts refer to post-exsolution. Plotting theses against pressure estimates of orthopyroxene (which is very sensitive to retrogression) illustrate how these OH contents vary with the accumulated retrogression in individual samples (i.e. the progression of Al diffusion in orthopyroxene).

Since the hydroxyl content in orthopyroxene is lower than that in garnet and clinopyroxene, the presence or absence of orthopyroxene in the peak metamorphic mineral assemblage is expected to have little affect on the hydroxyl distribution. However, an in-situ origin of pargasite in clinopyroxene in a chemically closed system implies that the Opx-bearing eclogite samples (all af which contain such pargasite) and most Opx-free eclogites (which do not contain such pargasite) had different hydrogen contents in the bulk rock during UHP metamorphism. The reason for this difference is unkown, but is probably related to the origin of the two groups of eclogite. The Opx-free eclogites were considered to be part of the Baltica crust, while the Opx-bearing eclogites show a garnet Ca-Cr-systematics known for mantle rocks (Spengler et al., 2023).

We agree that a small number of anisotropic grains with different orientations analysed using unpolarised light has a significant impact on the uncertainty of the hydroxyl content determined for each sample. The recent study by Qiu et al. (2018; 10.2138/am-2018-6620) suggests that the uncertainty of the hydroxyl content is +/-25% in most cases when averaging 2 grains. Since we used averages of 3-6 grains of clinopyroxene for the Opx-bearing eclogites shown in Fig. 9, the uncertainty per sample is expected to be less than 25%. The average clinopyroxene hydroxyl content in the samples with the highest metamorphic estimates (4.7-5.2 GPa) shown in Fig. 9 is 132 µg/g, while that with the lowest metamorphic estimate (2.2 GPa) is 174 µg/g, about 30% higher. Therefore, the inferred increase in structural hydroxyl content with decreasing metamorphic pressure may outweigh possible orientation effects.

We will be pleased to add information on the determination of metamorphic pressures to facilitate the interpretation of Fig. 9.

We would like to thank reviewer 2 for the detailed comments on our manuscript. The critical view constructively scrutinises the argumentation presented and leads to further improvements. We would like to respond to all comments below.

The mineral trace elements were not analysed in the samples examined and are not available. However, the previous study Spengler et al. (2023; 10.5194/ejm-35-1125-2023) includes EMPA data, which show that the K2O content of Cpx does not exceed 0.02 wt% (analysed in spot mode, Table 4) and that of re-integrated Cpx containing Qz+Amp inclusions has a range of 0.01-0.03 wt% (analysed with a widened beam, Table 5). In addition, we had also analysed in sample DS1438, but not reported, the composition of two types of Amp that occur as inclusions in Cpx together with Qz and as a retrograde phase between Cpx and Grt. The former Amp type contains 0.01-0.03 wt% K2O and the latter 0.07-0.17 wt%. These data suggest that the oriented inclusion-bearing Cpx, the Amp inclusions and the re-integrated Cpx all contain K as a trace element component with quantities lower than those reported form the Alpine samples CM31/03 and SKP31 (Konzett et al., 2008; 10.1016/j.lithos.2008.09.002).

The cited study of Bromiley & Keppler (2004; 10.1007/s00410-003-0551-1) makes two major conclusions, i.e. the H2O storage capacity in Jd increases with decreasing P and a Ca-Eskola bearing Jd-Di solid solution dramatically increases the H2O storage capacity in Cpx. We refer to both in our reply to RC1. In short, the initial decompression of WGR eclogite (from the stability fields of Coe and Ca-Eskola to that of Qz) is expected to decrease the storage capacity for H2O in Cpx, from which follows that the release of Qz may be associated with that of H2O. This is consistent with the joint exsolution of Qz+Amp in WGR Cpx, before further decompression caused an increase in the now Ca-Eskola-free Cpx (as shown in Figure 9). The finding from Bromiley & Keppler (2004) is consistent with the other study of Katayama & Nakashima (2003; 10.2138/am-2003-0126) showing that the hydroxyl absorbance in Cpx correlates positively (clearly increases) with the Ca-Eskola component. Conversely, this means that the decomposition of Ca-Eskola (by decompression) is accompanied by a release of H2O (provided there was enough H2O stored in Cpx prior to decompression). The study of Koch-Müller et al. (2004; 10.2138/am-2004-0701) shows that eclogite xenoliths from different depth levels of the Siberian lithospheric mantle differ in H2O content. Samples from the deepest levels contain less H2O in Cpx than those from shallower levels. However, these measurements illustrate a systematic of the regional SCLM, which does not necessarily reflect a systematic for the H2O storage capacity in Cpx, because the samples studied do neither share their evolution nor environment (the deep and shallow SCLM levels may chemically differ). Nevertheless, high H2O contents in Cpx from the shallow Siberian SCLM are consistent with our finding that H2O increases in WGR Cpx during decompression. The study of Gose & Schmädicke (2022; 10.1111/jmg.12642) on eclogite from the Erzgebirge shows indeed no major difference in H2O content between HP and UHP samples, but the difference in P of their samples account also only to about 10 kbar and clusters at the Qz/Coe phase transition. This is very different from the WGR samples that document >30 kbar decompression, from the stability field of diamond to that of Qz. This difference in P seems to be large enough to show (i) the release of H2O (by formation of oriented Qz+Amp during Ca-Eskola breakdown) and (ii) the subsequent uptake of H2O (as structural hydroxyl) in Cpx during further retrogression along the exhumation path. In summary, we cannot recognise the criticism raised by the reviewer when looking at the details of the studies mentioned.

The major element mineral chemistry of the samples studied is described in the previous study by Spengler et al. (2023). In fact, sample DS1438 has Grt with the highest CaO content in the sample set, 12.5 wt.%. The remaining Opx-free eclogites contain Grt, whose CaO content ranges between 7.8 and 10.9 wt%. Since the difference in H2O content between the former and the latter Grt is about an order of magnitude, the rather small difference in CaO content is probably not the cause of the difference in H2O content. For reasons of clarity for the reader, we are happy to add a corresponding note in the revised version of the manuscript.

To the specific comments:

lines 10-14: Correctly we said "Apart from extreme values ...". Figure 8 shows that, apart from three samples (Zo-bearing, retrogressed), the H2O content in WGR eclogitic Cpx is <200 ppm, while that of pristine eclogite xenoliths is dominantly above, consistent with the studies mentioned by the reviewer: Bizimis & Peslier (2015; 10.1016/j.chemgeo.2015.01.008) 260-576 ppm in Cpx of Hawaiian garnet-pyroxenite, Huang et al. (2014; 10.1007/s00410-014-1092-5) 211-1496 ppm in Cpx of Roberts Victor eclogite (the range covers Type I = metasomatised and Type II = non-metasomatised). To make this point even clearer to the reader, we would also like to include the data from these other references.

line 18-20: We will add some explanatory sentences to the revised version regarding the Cpx mineral chemistry described above.

line 29: The reviewer rightly stresses that the decomposition of (pure) Jd cannot lead to the formation of symplectite. However, what we meant, and what should be made clearer in the revised version, is that in practice the isochemical decomposition of the Jd component in Omp (which changes the composition of the Cpx from omphacitic to diopsidic) often occurs in the form of symplectites consisting of diopsidic Cpx and Pl (Anderson & Moecher, 2007; 10.1007/s00410-007-0192-x). This is a fairly typical mineral texture for eclogite, which, unlike a kimberlite eruption, was exhumed by a tectonic process.

line 95-96: The sentence preceding the quotation refers to "... hydrous minerals as part of the peak UHP metamorphic mineral assemblage ... could not be identified in the thin sections prepared." Therefore, we assume that the attentive reader understands without further words that due to the small outcrop size, chemical equilibration of the sample studied (DS1438) with hydrous peak metamorphic minerals can reasonably be assumed.

line 100: The reviewer is correct in the conclusion that Cpx in sample DS2217 should be Na-poor. Table 4 of Spengler et al. (2023) shows that this Cpx contains only 6-7 mol% Na-pyroxene.

pages 6-7: The reviewer raises an important point. The number of grains from which IR spectra were collected is given for each sample, mineral phase and grain domain in Table 1. The 1 sigma uncertainties of the quantified hydroxyl contents shown in Figure 8 were obtained from the calibration of Bell et al. (1995; 10.2138/am-1995-5-608), who reported uncertainties for the integral molar absorption coefficients of 10% for Grt, 4.4% for Cpx and 4.0% for Opx. However, these may be too low to be applied to our analytical results. The uncertainty in determining the thickness of the relatively thick grains (rock slabs) is estimated to be 1% or less. Major sources of error are the baseline correction of the spectra and the absorption variations resulting from the limited number of different orientations of anisotropic grains in unpolarised light (see also AC1 to RC1). Both uncertainties are reflected in the variation of the sample-specific hydroxyl contents determined per mineral domain, which are shown in Figure 6 and estimated to be 10% (1 sigma). The uncertainty in the accuracy of the values ​​due to the calibration of the absorption coefficients is 10% or less (Bell et al., 1995; Libowitzky & Rossman, 1997; 10.2138/am-1997-11-1208). This gives a total uncertainty of 20%. Therefore, the error bars in Figure 9 should be updated.

line 168: The reason for including the mineral chemical variation in Cpx of sample DS2217 as Figure 5 was to highlight presumably hydrous phases of sub-micron width occurring in peripheral grain domains. These phases appear to be responsible for the absorption around 3622 1/cm, which is absent in most spectra of the other samples (Figure 3a). However, we can comply with the reviewer's request and include an example of an elemental distribution map of Cpx with Qz+Amp inclusions as Supplementary Information, as the information content is more likely to serve healthy curiosity than the subject of the manuscript.

lines 183-190: Sample 1066b from Fjørtoftvika, reported by Terry et al. (2000; 10.2138/am-2000-11-1207), contains phengite and comes from the same outcrop as sample DS1438. However, we could not detect phengite in any of our samples, but secondary biotite (in the matrix).

line 209: The reviewer is correct: looking at the entire data set, the ranges of the measured hydroxyl contents in the grain cores and grain rims are similar (as described in this sentence) and overlap in error. Looking at individual samples, the ranges between the two differ outside the error (as shown in Figure 6b and noted in the following sentence).

lines 212-213: As the reviewer correctly recognised, the number of bands in each sample is not always 7, as can also be seen in Figure 4, and varies between 4 and 7 depending on the sample and analysed grain domain (area), as shown in Table 1. As also shown in Table 1, the presence of specific bands in Grt does not depend on the presence of Opx in the mineral assemblage, which subdivides the dataset and reflects two distinct mineral chemistry trends of Grt (Figure 6a,b in Spengler et al., 2023). Therefore, we see no evidence of a clear correlation between bands and mineral assemblage or Grt chemistry. However, looking at the total number of absent bands in the entire dataset, the Opx-bearing eclogite has fewer than the Opx-free eclogite. Whether this is random or significant is beyond the scope of the study.

lines 251-254: We appreciate the reviewer's critical view and would like to note that even if the uncertainty of the individual measurement of hydroxyl is 20%, the Opx rims always show the lowest individual hydroxyl content and the Opx cores always the highest, with the exception of the strongly retrogressed sample DS2216 (Figure 6b, for a photo of Opx from this sample see Figure 8c in Spengler et al., 2023). This difference leads to different average values in Opx rims and cores per sample (Table 1), whose ratio OH_rim/OH_core is less than 1 (Figure 7). The two average values per sample do indeed have a large overlap in terms of uncertainty (although the uncertainty of the average values is slightly lower than that of the individual measurements). However, this overlap does not call into question the systematic nature of (i) the lower average values of the rims compared to the cores and (ii) the lower individual minimum and maximum values in the rims compared to the cores. Both indicate that the Opx rims have suffered a loss of hydroxyl - within the limits of analytical uncertainty, unless the obverved systematic would be by chance.

We would like to thank the reviewer for the other point raised, which relates to the different hydroxyl contents in rims and cores of Cpx. Actually, we refer in lines 254-255 to: "Ratios for clinopyroxene cluster from unity upwards, are consistent with uptake of hydroxyl at grain rims in some samples." Opx-bearing eclogite has Cpx with the highest difference in hydroxyl content between rim and core precisely in those two samples that have the lowest pressure estimates, DS0326 and DS2216. This ratio between rim and core can thus be considered an independent argument for hydroxyl uptake in Cpx during decompression-assisted retrogression, a relationship that we overlooked when analysing the data.

lines 273-274: Obviously, the use of the term "inherited" leads to confusion with a presumed prograde history (for which we actually provide no evidence at all). To increase clarity, we would like to replace the title of the subsection with "Differences in hydroxyl content during peak metamorphism" and reword this subsection slightly.

lines 275-278: Our answer is given in the same lines: "... the high structural hydroxyl content in its garnet is most likely related to the more hydrous conditions for the whole-rock during peak metamorphism." In principle, two possibilities are conceivable: Either the composition of the whole-rock (outcrop) of sample DS1438 differs in that it was inherently more hydroxyl-rich during UHP metamorphism, or the whole-rock was exposed to local fluids during UHP metamorphism, unlike the other samples. However, the subject of this study is not the cause of the exceptionally high hydroxyl content in the Grt of sample DS1438. More important, this sample shares oriented mineral inclusions in Cpx with the other samples, suggesting that the observed mineral microstructure appears to be independent of differences in hydrous conditions (either "inherited" from whole-rock chemistry during crystallisation at the peak of metamorphism or caused by local fluids during crystallisation at the peak of metamorphism). The main conclusions are drawn from the other samples. The data does not seem to be able to provide a deeper answer, although it would be interesting to explore this, we fully agree. A few additional explanatory sentences should make this clearer.

Fig. 8: As mentioned in our answer above (lines 10-14), the information on the xenolith data can be expanded.

line 282: We will use the term "Opx-barometry" in a rephrased sentence for clarity.

lines 283-284: The reviewer draws our attention to a point, which is described in our AC1 and which should be explained in more detail (in this and the next subsections). Following the logic, "All other samples" contain either retrogressed Opx (i.e. there is evidence of regression) or no Opx (they differ in overall rock chemistry from the Opx-bearing eclogites). The point is that, prior to exsolution of Qz, Opx-bearing eclogite appears to have been more "wet" compared to Opx-free eclogite (with the exception of DS1438). Thus, the former could, during initial decompression from the stability field of Ca-Eskola & Coe to that of Qz, exsolve Amp (together with Qz) while the latter did not (Qz only). Subsequent decompression increased the hydroxyl content of Cpx in those samples that accumulated retrogression. We will reword this part of the manuscript in the revised version to increase clarity.

lines 284-285: A reference to the previous study by Spengler et al. (2023) will allow the reader to obtain details of the desired mineral chemistry.

lines 302-308: The quoted sentence is indeed long. It will be reworded in the revised version.

line 316: Yes. The Cpx that now hosts the oriented Amp inclusions used to be without them, as Amp is not stable under UHP conditions. The same logic applies to the orientated Qz inclusions, as Qz is also not stable at UHP and the Opx-bearing eclogites do not contain grains of free silica (i.e. SiO2 was not part of the metamorphic mineral assemblage, making "overgrowth" of SiO2 during Cpx growth unlikely). Thus, Qz+Amp in Cpx are of secondary origin, for which two main scenarios can be considered: either the source of the chemical components that formed the orientated inclusions is external or internal. In the latter case, the (solid solution) precursor Cpx is expected to be more hydrous than the current (exsolved) host Cpx.

lines 317-319: The reviewer puts forward an interesting alternative hypothesis, which is challenged by the following remarks, which in turn could be added to the revised version:

- Amp, which occurs as lamellae together with Qz in Cpx grain cores, has a lower K content than Amp grains in the matrix (see first comment in this reply).

- Oriented inclusions of Qt+Amp in Cpx were never observed exclusively at the rims of the Cpx grains, as would be expected if they originated from an external fluid source. Instead, they occur either homogeneously in whole Cpx crystals (e.g. sample DS1438 in Figure 4c in Spengler et al., 2023) or with variable density in the grain cores, while they disappear towards the grain rims (e.g. sample M65 in Figure 2c,e in Spengler et al., 2023).

- All oriented Qz inclusions in Cpx that occur together with Amp have a crystallographic orientation relationship with the Cpx host (recognisable by a common angle of extinction under crossed-polarised light).

- The hydroxyl loss in Opx rims (Figure 7) seems to be inconsistent with an infiltration by an external fluid.

- It is highly unlikely that a penetrating fluid that partially alters the core of a Cpx crystal would leave the core of a neighbouring Opx crystal untouched, so that the latter retains a very low alumina content, indicating metamorphic conditions in the stability field of diamond, i.e. where Amp is not stable.

line 333: The requested chemically integrated analyses of the microstructure were presented in the previous study (Spengler et al., 2023).

line 342: The reviewer's question can be answered by repeating the reference from line 337.

lines 353-355: The reviewer is correct, and we fully agree, that partial melting is likely to result in partitioning of H2O into the melt. In response, we replace the passage on melting with a sentence suggesting that 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.

The compositional data of the three minerals can easily be inserted into the manuscript and briefly included in the discussion, but require 5 additional columns to be presented. Since Table 1 is already page-filling, a new table for the compositional data is probably advisable.

We fully agree that the K content in Cpx is generally too low to be reliably analysed with EPMA. To clarify for the less informed reader, we can note that quantified values close to the detection limit are rather indicative.