1. Statistical support for the two-group GMM classification should be made more transparent.

The two-group division of barite δ¹⁸O–δ³⁴S data is central to the manuscript. However, the Methods state that the Bayesian Information Criterion indicated three components as the optimal GMM solution, whereas the authors selected two components because the three-component model produced negative regression slopes and because two components are consistent with the expected AOM–OSR end-member framework. This choice may be reasonable, but it requires clearer justification. Please report the BIC values for one-, two-, three- and preferably four-component models. The rejected three-component solution should be shown in the Supplement, together with a short explanation of why negative slopes are considered geochemically implausible rather than reflecting mixed growth generations, local fluid mixing, non-coeval sulfate reservoirs, or sampling heterogeneity. I also recommend a sensitivity test showing whether the two groups remain robust when Forsmark and Laxemar are treated separately, or when individual crystals/samples rather than individual SIMS spots are used as the statistical unit. Because multiple SIMS spots may come from the same crystal or fracture, they are not fully independent observations; this should be considered when evaluating the robustness of the clustering and regression results.

• Reply: We thank the reviewer for the opportunity to clarify our reasoning behind the clustering approach. It is important to note that two barite groups and their slopes are purely statistical entities, since they should reflect multiple fractionation processes occuring separately in space and time, rather than continuous fractionation trends, as in porewater profiles published earlier (e.g., Antler et al., 2013, 2014, 2015). The moderate statistical support for a three-component solution (ΔBIC = 5) and previous evidence for AOM and OSR at Laxemar and Forsmark (Drake et al., 2017, 2018) suggest it is geochemically more plausible to assume that barite in our samples mostly reflects sulfate reservoirs affected by these two endmember proccesses. Our underlying assumption is that compiling datapoints affected by these two endmember processes should result in clusters reflecting the average trend of the respective processes. We test this assumption by comparing the results of the clustering with the δ¹³C values of associated calcite. We find our assumptions partly confirmed, due to the strong association of ¹³C-depleted calcite with Group 1. The remaining variability could be explained by additional factors, such as the different fluid sources, growth generations, and barite ages in our dataset, as suggested by the reviewer. These alternatives will be emphasized in the revised version. Additionally, The BIC values for one-, two, three, and four-component models and the rejected three-component model will be reported in the Supplement.
• It is certainly an interesting question whether the observed trends are robust when separating the two localities in the GMM. Unfortunately, the small sample size of correlated δ¹⁸O–δ³⁴S analysis in Forsmark barites (n = 21 vs. >200 at Laxemar) prevents us from doing this in a statistically robust way. We disagree with using whole crystals or samples for the GMM, because crystals can exhibit multiple growth generations and samples may contain crystals of different ages. In our opinion, considering them collectively (e.g., average values) would create artifical signals.

1. Quantify the basis for the AOM- and OSR-dominated interpretations.

The wording that Group 1 was “dominantly associated” with AOM and Group 2 was “dominated” by OSR is reasonable. However, because Group 1 also includes barite associated with less strongly ¹³C-depleted calcite, and because OSR-related barite appears to occur in both groups, the basis for “dominantly” should be stated more explicitly. I suggest adding a short summary table or a few sentences indicating how many samples or analyses in each group are associated with strongly ¹³C-depleted calcite, moderately depleted calcite, and non-depleted calcite. It would also be useful to distinguish cases where the calcite is petrographically co-genetic with barite from cases where average δ¹³C_calcite values from the barite-bearing sample were used. This would strengthen the process-level interpretation without requiring any change to the main conclusion.

• Reply: We agree that our current phrasing in this regard lacks clarity. We will provide numbers on how many barite data points per group are associated with strongly ¹³C-depleted calcite. We suggest to use a threshold of < -40‰, since such ¹³C-depleted calcite cannot be explained by organic carbon oxidation alone and requires at least an admixture of carbon from methane oxidation. We will additionally add the group assignments of each data point to the Supplementary Data Table.

1. Present the preservation-bias interpretation more cautiously or support it with a simple calculation.

The proposed explanation that extremely ³⁴S-enriched barite is absent because barite becomes undersaturated during advanced Rayleigh distillation is interesting. However, the current discussion is mostly qualitative. I encourage the authors either to provide a simple first-order calculation showing how sulfate concentration, Ba²⁺ availability and barite saturation state may evolve during Rayleigh distillation, or to slightly soften the wording of this conclusion. In particular, the maximum observed δ³⁴S_barite value of approximately +52‰ should be presented as the highest value recorded in the dataset rather than as a demonstrated saturation threshold. If no quantitative saturation calculation is added, the manuscript should more clearly acknowledge alternative explanations, such as temporal decoupling between pyrite and barite formation, limited Ba availability during late-stage sulfate reduction, post-formation dissolution, open-system mixing, or sampling bias.

• Reply: We agree that +52 ‰ cannot be used as generalized threshold for the degree of ³⁴S-enrichment at which barite becomes undersaturated. Quantitative assessment of these process would be desirable, however, would be based on many assumptions, since the compositions of the palaeofluids are unknown. In our opinion, calculating the evolution of the barite saturation state may add confusion, as it would erronously imply a constancy of fluid compositions over time. Therefore, we will rephrase our statement, highlighting alternative explanations to our suggested model. Particularly, a decoupling of pyrite and barite formation in time, rather than just between different microenvironments (as implied in our previous draft) has to be considered. Barium availability indeed is a critical limitation for barite precipitation in the continental subsurface (Fichtner et al., 2024). Thus, barite precipitation may deplete the dissolved barium pool like MSR depletes sulfate, supporting barite undersaturation as a mechanism for limiting ³⁴S-enrichment. A sampling bias cannot be excluded, which we will discuss, but this risk was mitigated by sampling a large number of barite grains from various fractures at different depths (including those with extremely ³⁴S-enriched pyrite). Thus, we consider sampling biases unlikely to explain the absence of extreme ³⁴S-enrichment in barite alone.

1. Clarify the inference of sulfate reduction rates and sulfate/electron-donor controls.

The interpretation that differences in δ¹⁸O/δ³⁴S slopes reflect variations in sulfate reduction rates, fractionation and sulfate/electron-donor ratios is reasonable and consistent with previous conceptual models. However, the manuscript would benefit from a clearer distinction between directly observed isotope patterns and inferred microbial or hydrogeochemical controls. For example, higher sulfate availability does not necessarily imply higher sulfate reduction rates if electron donors are limited. Similarly, apparent ³⁴ε, sulfate concentration and cell-specific sulfate reduction rate are related but not interchangeable. I suggest revising this section to state more explicitly that SRR differences are inferred from isotope systematics rather than independently measured. This small clarification would improve the logical flow of the discussion.

• Reply: We agree that this paragraph requires some clarification and will revise it as suggested.

1. The comparison with Antler et al. and marine sedimentary systems should be expanded.

The manuscript notes an important discrepancy: the AOM-influenced Group 1 has a steeper δ¹⁸O/δ³⁴S trend than the OSR-dominated Group 2, and the Group 2 trend is shallower than OSR-related trends reported from marine sedimentary systems. This difference is important because it supports one of the manuscript’s broader implications, namely that isotope interpretation frameworks developed for marine barite cannot be directly transferred to deep subsurface fracture-hosted barite. This point deserves a more mechanistic discussion. The authors should compare the deep fracture system with marine SMTZ settings in terms of sulfate concentration, Ba availability, electron-donor limitation, fluid residence time, openness of the sulfate reservoir, temperature, and barite saturation state. A short comparative table would be useful. Such a comparison would help readers understand whether the observed δ¹⁸O–δ³⁴S trends are a predictable consequence of low-sulfate deep subsurface environments, or whether they require modification of existing conceptual models for sulfate oxygen and sulfur isotope evolution during microbial sulfate reduction.

• Reply: We agree that a more mechanistic discussion of the differences between subsurface and marine environments would strengthen our manuscript. Comparison of sulfate concentrations in modern fracture fluids from Laxemar (0.4 g/l) (Drake et al., 2013) with seawater (2.7 g/l) implies relatively rapid sulfate depletion in the subsurface. Additionally, the supply of Ba²⁺ from feldspar dissolution in granitic subsurface environments is relatively slow compared to the supply of Ba²⁺ from degrading organic matter in marine environments. Both aspects may inhibit the precipitation of strongly ³⁴S-enriched barite, leading to steeper δ¹⁸O–δ³⁴S trends. We will expand this discussion accordingly.

Minor comments

Please check isotope notation throughout the manuscript. Symbols such as δ³⁴S, δ¹⁸O and ³⁴ε should be formatted consistently in the abstract, main text, tables, figures and captions.

• Reply: The manuscripts will be checked for internal consistency before resubmission.

Figure 4 would benefit from uncertainty envelopes around the regression lines. The caption should also clarify whether each plotted point represents one SIMS spot with paired δ³⁴S–δ¹⁸O data, one crystal, or one sample.

• Reply: Uncertainty envelopes will be added to the regression lines in Figure 4 and the caption will be clarified.

In the caption of Figure 5, two panels appear to be referred to as “(d)”. Please correct the panel labeling.

• Reply: Corrected.

The BSE images in Figures 5 and 6 are useful but would be easier to interpret with clearer arrows or labels marking the key petrographic relationships discussed in the text. For the interpolated δ³⁴S heatmap in Figure 6, please clarify that the interpolation is illustrative and should not be over-interpreted beyond the density of SIMS analyses.

• Reply: We will enhance our BSE images with additional visual aids and clarify the illustrative nature of the heatmap in Figure 6.

Reference list formatting: The reference list should be revised carefully for consistency with the Biogeosciences style. In particular, please standardize journal-name formatting, as some journals are written in full whereas others are abbreviated. Please also check article-title capitalization, as some titles appear to use inconsistent capitalization styles. In addition, article titles containing isotope notation and chemical formulas should be checked for correct superscript/subscript formatting, for example δ³⁴S, CH₄, and H₂S, rather than plain-text forms such as δ34S, CH4 or H2S.

• Reply: Reference list formatting will be checked before resubmission.

General comments:

The introduction accounts for more than a quarter of the total manuscript length and, as a result, feels overly long and at times difficult to follow. While the background material is generally relevant, several sections could be shortened or streamlined, allowing the authors to allocate more space to methods, results, and interpretation.

• Reply: We agree that the introduction is relatively long. However, we would like to emphasize that our manuscript is dealing with three different isotope systems in three mineral phases, all in the context of the deep biosphere. Making this accessible to the broad readership of Biogeosciences requires substantial background information. Nevertheless, we will do our best to omit unneccessary detail and possibly integrate some relevant concepts into the respective discussion sections.

My primary concern with the manuscript relates to the statistical treatment of the isotope data, specifically the use of Gaussian Mixture Models (GMM) to define discrete groups that form the basis for much of the subsequent discussion. Although the Bayesian Information Criterion (BIC) favours a three-component model, this solution is explicitly rejected with limited justification. This decision appears subjective and is not discussed in sufficient depth. From a statistical standpoint, discarding the preferred model without a clear, objective rationale weakens inferential independence and risks introducing confirmation bias. If the data robustly support three components, the implications of this outcome should be explored rather than dismissed. Negative isotope slopes, for example, could plausibly reflect multiple generations of mineral growth, mixing of fluids with distinct histories, or other processes independent of the authors’ preferred framework. Presenting and discussing all statistically supported solutions would ultimately strengthen the manuscript.

• Reply: Indeed, the justification for this choice is critical to the statistical integrety of our study. We have addressed this concern in our response to reviewer 1's comment 1.

In addition, the observation that barite core analyses dominate one group while rim analyses dominate another may suggest that intra-crystalline processes during barite growth contribute to the overall spread observed in the data. This possibility is potentially important, particularly given the distinct zoning patterns shown in some samples (e.g., Fig. 6d), and deserves more explicit consideration. Because these group definitions underpin much of the later discussion, the current statistical treatment leaves the interpretations on uncertain footing. Addressing this would significantly improve the manuscript. I encourage the authors to reduce the length of the introduction and use the freed space to present a clearer and more transparent explanation of their clustering approach and its limitations. They could additionally explore other clustering methods. In any case, I would advise approaching the interpretation of the preferred statistical analysis with an unbiased mind, as potentially "unexpected" trends might be the result of significant mechanisms.

• Reply: We agree that the zoning into a Group 1 core and Group 2 rim points to differences during barite crystallization over time. This could be due to a shift in environmental conditions favoring one sulfate reduction pathway over another or the introduction of another fluid source during the growth of this particular barite crystal, which would result in trends with different slopes across the core vs. the rim data points (and potentially a different group assignment). This will be further clarified in our revised submission.

Specific comments:

A further concern relates to the strong imbalance between datasets from the two study sites. The sulfur isotope dataset for example is dominated by samples from Laxemar (323 analyses) compared to Forsmark (21 analyses). This imbalance likely affects the performance of the GMM and means that the derived clusters largely reflect the structure of the Laxemar dataset, with Forsmark data contributing relatively little statistical weight. As a result, it is difficult to view this as a truly comparative study of two localities. If additional Forsmark data cannot be obtained, I suggest either analysing sites separately or clearly acknowledging the limited interpretive weight of the Forsmark dataset. Alternatively, removing the Forsmark data altogether may lead to a more internally consistent analysis.

• Reply: It is true that the derived clusters mostly reflect the Laxemar dataset, which is the main reason why we did not attempt a comparative study between both sites. Instead, we compiled data from two sites with known evidence for OSR and AOM (Drake et al., 2017, 2018) to achieve a larger overall subsurface barite sample size. We will clarify the justification for compiling Forsmark and Laxemar. Unfortunately, additional data from Forsmark cannot be obtained in the framework of this study due to limited and cost intensive instrument time. Due to the small sample size of the Forsmark dataset we cannot perform GMM for both datasets seprately. However, we will run GMM for Laxemar alone to test the influence of Forsmark on the overall result. If there is a subtantial difference, the resulting plot will be reported in the supplement to highlight limitations of compiling datasets from different localities. Either way, we will acknowledge in the discussion the clustering is based on a merged dataset and future results may differ if more data from different sites are gathered.

The manuscript generally treats barite as a mineral that undergoes minimal isotope fractionation during precipitation and is subsequently resistant to dissolution. However, several observations appear inconsistent with this assumption, particularly the systematic differences between core and rim compositions and their interpretation in terms of barite dissolution. Potential kinetic isotope effects during precipitation are not considered, despite their relevance to zoning patterns. This raises questions about the assumption that barite isotopes always provide a faithful archive of dissolved sulfate compositions.

• Reply: Sulfur isotope fractionation between dissolved sulfate and barite during precipitation is negligible (<0.4 per mil; Kusakabe and Robinson, 1977) compared to the observed core-to-rim difference (e.g., ca. 12 per mil in sample KLX01_220; Fig. 6d). Thus, these effects cannot be discussed based on our data. Instead, the observed core-to-rim difference must be related to different processes affecting the sulfate reservoir over time or different mixing ratios of sulfate-bearing fluids between core and rim, or combination thereof. We will clarify this in the revised manuscript. Further, barite is relatively resistant towards dissolution over a broad range of Eh/pH conditions, as well as diagenetic overprint, making it a highly useful longterm archive. However, we have not suggested that it could not occur. Indeed, we discuss barite dissolution as a potential reason for the absence of ³⁴S-enriched barite. To avoid speculation on fluid conditions during periods no longer recorded due to barite dissolution, we focus our discussion on the record remaining in our samples.

Related to this point, core-to-rim sulfur isotope trends in pyrite are attributed primarily to Rayleigh distillation during mineral growth. While this is certainly plausible, a simple quantitative model using literature fractionation factors would help demonstrate whether this mechanism can reproduce the observed magnitudes. Other potential explanations, including precipitation kinetics or diagenetic modification, should also be considered and discussed.

• Reply: Extensive discussion of Rayleigh distillation processes during pyrite formation including supporting evidence and model calculations are presented in the original publication that reported the respective pyrite data (Drake et al., 2018). We would prefer to avoid repeating those here.

The treatment of carbon isotopes in calcite is similarly simplified. The use of fixed δ¹³C thresholds to infer dominant processes (e.g., -70 to -40‰ for AOM; > -20‰ for OSR) does not fully account for mixed carbon sources, partial methane oxidation, or other processes that can influence calcite δ¹³C values. This oversimplification is particularly problematic where averaged δ¹³C values are used for barite-bearing samples (Section 3.2, line 211). Given the wide range of observed values, such averaging may obscure meaningful variability and potentially bias subsequent interpretations.

• Reply: We agree that the carbon pool from which our calcite precipitated is likely affected by mixing. However, since we can only analyze the endproduct, mixing of different carbon sources cannot reconstructed in most cases and would be speculative. To avoid this, we have introduced a threshold of below -40‰ for AOM, since only then we can be certain that the observed δ¹³C values cannot be explained by OSR (or inorganic carbon sources) alone. We consider this a more robust approach for identifying samples affected by AOM.

The manuscript argues that oligotrophic conditions in the deep subsurface imply low sulfate reduction rates (SRR). However, the interpretations associated with Group 2 require relatively high SRR, which is reconciled by invoking large sulfur isotope fractionation factors (ε³⁴). While this explanation is possible, it remains largely unconstrained and risks becoming flexible enough to accommodate almost any trend. Independent constraints on SRR, or at least a more explicit discussion of uncertainties and alternative parameter combinations, would make this argument more convincing. 

• Reply: Indeed, our argument around SRR variations is an inference from our data of ancient barite. Unfortunately, we do not have independent constraints on SRR, particularly in the geological past. Thus, we built our interpretation based on the stable isotope patterns we see and how they are commonly interpreted where independent constraints are available (e.g., in modern marine sediments, see Antler et al. 2013-2015). We will address this by better clarifying our observations from interpretations and expanding on the alternative explanations we discuss in section 3.3 (variations in SRR vs. different MSR pathways vs. variations in ε³⁴).

Finally, the absence of isotopically heavy barite is not uniquely constrained by the current discussion. Several alternative explanations could be considered, including sulfate depletion prior to barite saturation, temporal decoupling between pyrite and barite formation, or preferential sampling of late-stage, sulfate-poor conditions. Incorporating barite saturation index calculations or explicitly evaluating these scenarios would strengthen this part of the manuscript.

• Reply: Indeed, our model involving barite undersaturation prior to strong ³⁴S-enrichment is not the only possible explanation. We have addressed this issue in our response to Reviewer 1's comment 3, where we also highlight the difficulties of a more quantitative treatment.

Minor comments:

Several parts of the manuscript, particularly in the introduction, rely on rather vague statements that would benefit from greater precision. For example, at line 45 the authors refer to a “range” of isotope fractionations without clearly defining what that range is. Providing representative values or at least approximate bounds would greatly improve clarity. Similarly, at line 116, Laxemar is described as being “a couple of kilometers” from the Baltic Sea coastline, while Äspö is located on the coast. Reporting the actual distances between these localities would be more informative and avoid ambiguity.

• Reply: We will clarify the indicated statements as suggested.

The manuscript would also benefit from careful attention to language and terminology. There are several grammatical and spelling issues (e.g., line 54: “the slopes (…) is”) that should be corrected. In addition, some statements rely heavily on scientific shorthand or jargon, such as the phrasing at lines 73–74 where AOM and OSR are described as “producing trends.” Rephrasing these statements in clearer, more explicit terms would make the manuscript more accessible to a broader readership. I have not undertaken a comprehensive review of grammar and spelling, as the structure and wording of the manuscript will likely change substantially in response to the major revisions outlined above. Nonetheless, I strongly recommend that the final revised version be carefully edited for language, clarity, and consistency prior to resubmission.

• Reply: We will thoroughly proofread the manuscript for language, claritiy and consistency before resubmission, particurlary paying attention to the highlighted examples.

Finally, regarding QA/QC, I was unable to identify data from standards or reference materials that would substantiate the uncertainty estimates and error propagation described in the methods section. Including these data, or explicitly stating where and how they are reported (e.g., in supplementary materials), would strengthen confidence in the analytical results.

• Reply: We would like to refer the reviewer to the supplementary data table, where all measurements of reference materials and unknowns analyzed in this study (i.e., barite) are listed. Measurements of reference materials for pyrite and calcite are published in the cited original literature.