Modeling Free Proline Accumulation in Four Mediterranean Geophytes: A Comparative Analysis of Pancratium maritimum, Sternbergia lutea, Iris germanica, and Cyclamen graecum

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. Novelty and relationship to prior work (Introduction, Section 1.3, lines 77–95; References 6–8). The corresponding author has already published seasonal proline dynamics for Sternbergia lutea (ref. 6), Pancratium maritimum (ref. 7), and Cyclamen graecum (ref. 8). The manuscript must state explicitly what is new here versus those papers — particularly whether the proline data for these three species are newly generated or re-analysed from the same 2013–2016 sampling. As written, the only clearly novel elements are Iris germanica and the modeling, and this needs to be made transparent.
2. "Stress" is inferred, never measured (Abstract line 18; Section 2.2, lines 122–125; Section 4.2, lines 279–287). Throughout, proline peaks are attributed to "environmental stress conditions," yet no direct stress variable (soil water potential, tissue water status, soil moisture) was recorded. With only temperature and precipitation as proxies, the causal language should be softened to correlation, or actual stress indicators should be added.
3. Confounding of climate predictors with season and site (Section 4.2, lines 279–287; Section 4.5, lines 317–340). Plants were collected from at least three locations (eastern Euboea, central Euboea, Kaisariani forest), but the source and spatial resolution of the temperature/precipitation data are not stated. If a single regional climate series was used, temperature and precipitation are effectively collinear with month, and the regression coefficients for these variables become uninterpretable. Please clarify the climate data source and address this confound.
4. The predictive modeling is largely circular (Section 2.3, lines 133–146; Section 2.4, lines 174–186; Figure 3B). Species identity is reported as the dominant predictor, and Sternbergia lutea alone carries most of the feature importance (Fig. 3B). With only four species one-hot encoded, the models are essentially fitting species means rather than learning generalizable environment–proline relationships. Either reframe the modeling as descriptive partitioning of variance or justify why R² ≈ 0.41–0.49 represents meaningful predictive capacity.
5. Numerical inconsistency in Random Forest performance (Table 1, line 151; Figure 3 caption, line 168). Table 1 reports the Random Forest as R² = 0.409, RMSE = 8.237, whereas the Figure 3 caption states R² = 0.350, RMSE = 8.642. These must be reconciled and a single set of values reported consistently across text, table, and figure.
6. Figure callouts are reversed (Section 2.3, lines 139 and 143). Line 139 refers to the Random Forest model as "Figure 2," and line 143 refers to the linear model as "Figure 3," but Figure 2 is the linear regression and Figure 3 is the Random Forest. Please correct the in-text figure references.
7. Hypothesis 2 is not supported by the data (Section 1.3, lines 87–89; Section 2.1, lines 109–113; Discussion, Section 3.2). Hypothesis 2 predicts higher proline in species from harsher/more variable habitats, yet coastal Pancratium maritimum (salinity, sand burial) showed only intermediate levels statistically indistinguishable from Cyclamen and Iris, while dry-grassland Sternbergia was the outlier. The Discussion should explicitly acknowledge that this hypothesis was not borne out rather than narrating around it.

8. Sample size and data availability (Section 4.3, lines 291–303; Figure 2, n = 40; Data Availability, lines 375–376). It is unclear how 4–5 individuals per species per month over ~40 months reduce to n = 40 in Figure 2 — please define the unit of analysis and how replicates were aggregated. Given this is a modeling/reproducibility paper, the dataset and Python scripts should be deposited in a public repository rather than provided "upon reasonable request."

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript entitled “Modeling Free Proline Accumulation in Four Mediterranean Geophytes: A Comparative Analysis of Pancratium maritimum, Sternbergia lutea, Iris germanica, and Cyclamen graecum” investigates the seasonal dynamics of free proline accumulation in the underground organs of four Mediterranean geophytes and evaluates the performance of linear and machine learning regression models in predicting these physiological patterns based on environmental variables.

The manuscript addresses a relevant topic in plant stress physiology with a clear, logical structure. However, it requires refinement in its statistical justification and a deeper discussion of the biological mechanisms driving the observed interspecific differences to meet the standards of a high-impact journal.

Major Comments

• Statistical Robustness: The study utilizes R² values to evaluate model performance, which range from 0.409 to 0.488. Given these moderate values, I recommend that the authors include additional metrics such as the Akaike Information Criterion (AIC) or Bayesian Information Criterion (BIC) to more rigorously justify the selection of the linear model over the Random Forest model.
• Mechanistic Interpretation: While the manuscript successfully identifies Sternbergia lutea as a high-proline accumulator compared to Iris germanica, the Discussion would benefit from a more explicit physiological hypothesis regarding the metabolic cost of this adaptation. The authors should elaborate on why lutea employs a more rapid biochemical response compared to the apparent "buffering" strategy of other species.
• Confounding Variables: The authors acknowledge that variables such as soil properties and plant age were not included. In the revised manuscript, please provide a more detailed rationale for how the omission of soil water potential or soil chemical properties might influence the specific "predictive capacity" of the models developed.

Line-to-Line Comments

Abstract

• The abstract is concise; however, it should more clearly state that the findings indicate proline accumulation is a lineage-specific trait rather than a universal response.

Introduction

• The ecological importance of geophytes is well-established. To enhance the professional impact, emphasize that the integration of physiological measurements with machine learning creates a framework for predicting resilience under changing climatic conditions.

Results

• Section 2.1: The use of Tukey’s HSD is a correct statistical application to confirm significant differences among species.
• Section 2.3: The presentation of RMSE and MAE values is appropriate. Ensure that the text explicitly defines the "cyclic transformation" method for the month variable earlier, as this is a highlight of the methodological rigor.

Discussion

• Section 3.2: The comparison of metabolic flexibility is excellent. I suggest further contrasting the morphological/structural adaptations of Iris germanica against the biochemical ones of Sternbergia lutea to bolster the argument.
• Section 3.3: This section is appropriately critical of the study's limitations. I recommend framing these limitations as an explicit call for future studies to expand taxonomic sampling to non-geophytic species.

Materials and Methods

• The methodology for proline determination (ninhydrin-based colorimetry) follows established standards.
• The provision of precise GPS coordinates for the collection sites in Euboea and Athens is highly appreciated for reproducibility.

Conclusion

• The conclusion effectively summarizes the broader ecological implications. The final section should specifically highlight how this framework informs climate-resilient conservation strategies.

Figures and Tables

• Table 1 & 2: Tables are well-formatted and provide necessary clarity.
• Figure 1: The seasonal variation plot is clear. Adding standard error bars would improve the presentation of the biological replicates.
• Figures 2 & 3: The scatter plots and feature importance charts are informative. Adding residuals plots to these figures would further demonstrate the diagnostic rigor of the modeling.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The manuscript is well written and has importance in the relevant field.

The following points need to be addressed:

1.The study lacks a clear hypothesis test for the machine learning claim, please justify

2. The R² symbol appears sometimes as R2R2 (with roman R) and sometimes as R2R2 (blackboard bold) in the text. Please standardize to a single format.

3.The authors write: "four to five individual plants were collected during each monthly sampling period." It is not stated whether these were the same individuals resampled each month (repeated measures) or different individuals each time. This distinction affects statistical independence. Please clarify explicitly.

4.Table 2 lists Iris germania (incorrect) in the first column but correctly uses Iris germanica elsewhere. Correct the table entry.

5.please ensure the proper labelling of all figures

6.This is a common challenge in ecological modeling when the number of species is small relative to environmental heterogeneity" please revise this sentance and make clear sence

7.

 

Comments on the Quality of English Language

Carefull profreading is required

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This study investigated seasonal patterns of free proline concentration in the underground organs of four Mediterranean geophytes - Pancratium maritimum, Sternbergia lutea, Iris germanica, and Cyclamen graecum -under natural environmental conditions. The results revealed clear species-specific and seasonal variation in proline accumulation. The findings demonstrate that proline accumulation in Mediterranean geophytes reflects distinct physiological adaptation strategies associated with seasonality and environmental variability.

L. 40. The authors write, "...free proline plays a central role." This is a very bold statement, since among compatible osmolytes, proline does not play a central role. For example, the absolute proline content in tissues exposed to a stressor is tens of times lower than that of other compatible osmolytes, such as sugars. In geophyte tubers, sugar content can exceed 10%.

The study is very interesting, but the main comments relate to the methodological section of the manuscript.
It should be noted that the supplementary materials file does not open or cannot be read.

L. 114. Figure 1. The standard deviation/error of the mean must be included in the figure. This is especially true given the high values ​​and monthly fluctuations for Sternbergia lutea (red line).

L. 148-151. Table 1. In the R² table, the coefficient of determination is below 0.5. This means that the correlation model is weak and no relationships were identified.
It should be noted that RMSE stands for root mean square error. It is measured in the same units as the parameters being studied. MAE stands for mean absolute error.
Why did the authors include this table? The data presented in Table 1 are repeated in Figures 1 and 2.

Figure 2. The data presented in the figure contains large outliers. The inference from the RMSE/MAE ratio is a 2-fold gap. This again indicates outliers (RMSE is calculated using the square of the mean and pulls the values ​​upward).
This suggests that the sample size is small. Given that R² < 0.5, this model only describes half (approximately 50%) of the data, and does not describe the other half.

The model cannot predict the data!!!

L. 162. Figure 3 lacks the MAE; it appears only in Table 1. It can be assumed that the authors omitted this because the data gap is more than twofold.

It is important to note that the manuscript in Table 1 and Figure 3 contains different data: in the table, R² = 0.409, while in the figure, R² = 0.350.

Recommendation to the authors
Submit the manuscript to the journal without data on Sternbergia lutea, as the data on this species distort the R² model.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have addressed my concerns. I have no further comments on this article.

Best of luck to the authors. Thank you.

Author Response

The authors sincerely thank the reviewer for the careful evaluation of the revised manuscript and for acknowledging that the concerns raised during the review process have been satisfactorily addressed. We greatly appreciate the reviewer's constructive comments and valuable feedback, which have helped us improve the quality and clarity of the manuscript.

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript entitled “Modeling Free Proline Accumulation in Four Mediterranean Geophytes: A Comparative Analysis of Pancratium maritimum, Sternbergia lutea, Iris germanica, and Cyclamen graecum” addresses a potentially interesting topic, i.e., seasonal variation in free proline accumulation in Mediterranean geophytes and its relationship with environmental variability. The subject is relevant to plant stress physiology, Mediterranean ecology, and climate-change biology. However, the current manuscript remains scientifically underdeveloped and is not yet suitable for publication in its present form.

Recommendation: Reject / Major revision required before reconsideration

The major problem is that the manuscript makes broad claims about physiological adaptation, plant resilience, predictive modelling, and climate-change relevance, but the data package is too limited to support these claims. The study relies primarily on one biochemical marker, free proline, together with monthly temperature, precipitation, species identity, and seasonal modelling terms. This is not sufficient to infer plant stress mechanisms or to establish a robust predictive framework for physiological resilience.

The manuscript would be much stronger if it were presented as a modest descriptive study of seasonal proline variation. However, in its present form, it is framed as a comparative predictive and ecophysiological modelling study. That framing is not supported by the available evidence.

Major concerns

1. Limited novelty and weak conceptual advance

The manuscript states that three of the four species datasets were derived from previously published studies, whereas only the Iris germanica dataset is newly presented. Combining previously published datasets can be valuable, but only when the synthesis generates a clear new conceptual, mechanistic, or methodological advance. In this manuscript, the integration remains largely descriptive. The modelling component does not provide sufficient novelty or predictive strength to compensate for the limited amount of new experimental data.

The authors should clarify more explicitly what is genuinely new in this manuscript. If the novelty is only the inclusion of Iris germanica and the re-analysis of existing datasets, then the manuscript should be reframed as an exploratory comparative analysis rather than a predictive framework for plant resilience.

2. Overinterpretation of proline as a stress-resilience marker

Free proline is an important osmolyte and stress-associated metabolite, but proline accumulation alone cannot be treated as direct evidence of stress tolerance, ecological resilience, or adaptive superiority. Proline levels can reflect many processes, including tissue age, phenological stage, dormancy, metabolic status, developmental timing, nitrogen metabolism, and species-specific baseline physiology. Without additional physiological or biochemical markers, the interpretation remains speculative.

The manuscript does not include direct measurements of plant water status, soil moisture, soil water potential, salinity, oxidative stress, membrane damage, antioxidant enzyme activity, relative water content, chlorophyll fluorescence, photosynthetic traits, soluble sugars, or growth performance. Therefore, the authors cannot confidently conclude that the observed proline patterns reflect stress tolerance or resilience.

The authors should reduce the strength of their claims throughout the Abstract, Introduction, Discussion, and Conclusions. Terms such as “resilience,” “stress adaptation,” “predictive capacity,” and “forecasting” should be used with much greater caution.

3. Insufficient physiological depth

A plant-stress physiology manuscript should not rely on only one biochemical variable if it aims to discuss stress mechanisms. At minimum, the authors should include or discuss the absence of complementary physiological indicators such as: Relative water content; Soil moisture or soil water potential; Tissue water potential; Malondialdehyde; Hydrogen peroxide; Electrolyte leakage; Antioxidant enzymes such as SOD, CAT, POD, APX, and GPX; Soluble sugars; Chlorophyll concentration; Chlorophyll fluorescence; Growth or biomass traits; and Ion balance, especially for coastal or salinity-exposed species.

Without such variables, the study cannot distinguish whether proline accumulation reflects stress exposure, stress tolerance, phenological regulation, or species-specific metabolism.

4. Confounding between species, site, habitat, and environment

The four species were collected from different habitats and geographical locations. This creates a serious confounding problem. Species identity is not independent from site, habitat type, local soil conditions, microclimate, and environmental history. Therefore, the manuscript cannot confidently attribute differences in proline accumulation to species-specific physiological strategies alone.

For example, a coastal species, a woodland species, and species from dry grassland or shrubland habitats may differ not only because of intrinsic physiology but also because of site-specific soil texture, salinity, moisture availability, nutrient status, light environment, and microclimatic exposure. These variables were not measured or modelled.

The authors should acknowledge this limitation more strongly and avoid language implying that species identity alone explains adaptive strategy.

5. Model performance is modest and does not justify predictive claims

The reported model performance is not strong enough to support claims about forecasting physiological responses. The linear seasonal regression model explained less than half of the variance, and the Random Forest model performed worse. The manuscript should therefore avoid presenting the models as robust predictive tools.

The Random Forest analysis is particularly weak. With a small dataset, strong species effects, and limited predictors, Random Forest regression may produce feature-importance patterns that are difficult to interpret. In the current manuscript, the machine-learning component appears decorative rather than essential.

The authors should either remove the Random Forest analysis or substantially improve its justification by including: Full model-tuning details; Number of trees; mtry or equivalent tuning parameters; Repeated cross-validation; Permutation-based feature importance; Residual diagnostics; Species-stratified prediction errors; Sensitivity analysis excluding species identity; and Comparison against simpler baseline models.

Without these additions, the Random Forest model adds little scientific value.

6. Figure package is insufficient

The manuscript contains too few figures for the scope of claims being made. A climate-only figure, if present, should be moved to Supplementary Materials unless it is directly integrated with biological response data. Climatic data alone do not provide strong biological evidence.

The current figure package should be substantially improved. The authors should add:

1. A multi-panel seasonal proline figure for each species, showing replicate-level points and mean ± SE or confidence intervals.
2. Species-specific boxplots or violin plots showing seasonal distributions of proline.
3. A figure integrating climate variables and proline dynamics, preferably with lagged climate variables.
4. A correlation matrix or heatmap showing relationships among proline, temperature, precipitation, month, species, and any available environmental variables.
5. Model residual plots.
6. Observed vs predicted plots separated by species.
7. Residual vs fitted plots for the linear model.
8. Species-specific prediction-error plots.
9. If available, additional physiological markers such as RWC, MDA, antioxidant enzymes, chlorophyll, soluble sugars, or tissue water status.

The authors’ justification for not adding error bars because of visual clarity is not convincing. Visual clarity can be preserved through faceting, shaded error ribbons, supplementary panels, or separate species-specific plots. Similarly, residual plots should not be excluded from a modelling study merely because they complicate the figure layout.

7. Insufficient diagnostic support for the modelling framework

The manuscript uses modelling language but does not provide enough diagnostic evidence. A study claiming predictive modelling should show more than R², RMSE, and MAE. The authors should include diagnostic plots, residual distributions, residual autocorrelation checks, model assumptions, and sensitivity analyses.

Because the data are seasonal and temporally structured, the authors should also justify whether leave-one-out cross-validation is appropriate. In ecological and time-series contexts, random or leave-one-out procedures can overestimate generalizability if observations are temporally or spatially dependent. The authors should consider blocked cross-validation, leave-one-month-out, leave-one-year-out, or leave-one-species-out approaches, depending on the research question.

8. Discussion remains speculative

The Discussion contains several plausible but insufficiently supported interpretations. For example, the contrast between Sternbergia lutea and Iris germanica is interesting, but claims regarding biochemical versus structural adaptation are not directly tested. The manuscript does not measure metabolic fluxes, carbon/nitrogen allocation, storage carbohydrates, anatomical traits, phenology, tissue hydration, or stress injury. Therefore, such explanations should be framed explicitly as hypotheses rather than conclusions.

The Discussion should be rewritten with a clearer separation between: What the data directly show; What the models suggest; What remains speculative; and What future studies should test.

9. Authors should consult stronger plant-stress papers and improve the manuscript accordingly

The authors should not merely add citations; they should study how stronger plant-stress physiology papers present and interpret data. I recommend that the authors consult the following papers before resubmission:

• Patanè et al. 2022, “Relative Water Content, Proline, and Antioxidant Enzymes in Leaves of Long Shelf-Life Tomatoes under Drought Stress and Rewatering.”
• Wadood et al. 2024, “Unraveling the impact of water deficit stress on nutritional quality and defense response of tomato genotypes.”
• Karagüzel et al. 2025, “Integrative osmotic–antioxidant mechanisms in salinity-stressed Gerbera jamesonii.”

These papers provide better examples of how plant-stress data should be presented, including multi-panel figures, replicate-level variation, standard errors, post-hoc letters, integrated physiological measurements, correlation/PCA analyses, and model diagnostics. The current manuscript should be revised to approach this level of clarity and rigor.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

The authors responded to all inquiries, revised the table, and softened their interpretation of the data after considering the reviewer's remarks and recommendations. The text may thus be approved for publication in this format.

Author Response

The authors sincerely thank the reviewer for the thorough evaluation of our manuscript and for the constructive comments provided throughout the review process. We greatly appreciate the reviewer's recognition of our efforts to address all concerns, revise the table, and refine the interpretation of the results. The insightful suggestions have significantly contributed to improving the clarity, accuracy, and overall quality of the manuscript. We are grateful for the reviewer's positive assessment and recommendation for publication.

Round 3

Reviewer 2 Report

Comments and Suggestions for Authors

The revised manuscript is clearly improved, especially in tone: the authors have moderated some claims, added error bars to Figure 1, clarified that the modeling is exploratory, and acknowledged missing physiological measurements. However, the revision is still not sufficient for publication in its current form, because the authors mainly softened the language instead of adding the analyses, diagnostics, or data requested by the reviewer.

The core problem remains: this is still a single-marker descriptive proline study, but the manuscript continues to use language around “predictive performance,” “stress tolerance,” “adaptive strategies,” “physiological responses,” “forecasting,” and “climate-resilient conservation strategies.” The abstract still presents modeling as a central contribution, even though the linear model explains only about half the variance and Random Forest performs worse.

The paper may be publishable only if reframed as a modest exploratory/descriptive study of seasonal proline variation, not as a predictive plant-stress physiology paper.

Major comments

1. The manuscript still overclaims relative to the dataset; The authors have reduced some claims, but overinterpretation persists. The manuscript still states that the findings are “consistent with differing physiological responses,” that proline is a “species-specific physiological trait,” and that integrating measurements with machine learning provides a framework for describing species-specific patterns. This remains too strong. The study measures one biochemical marker in underground organs, together with monthly temperature, precipitation, species identity, and month. These data can support only the following conclusion: Seasonal free proline concentrations differed among four geophytes, with Sternbergia lutea showing markedly higher values than the other species. They cannot support strong claims about stress tolerance, adaptation, ecological resilience, or physiological strategy.
2. The response to missing physiological data is inadequate

The reviewer specifically asked for additional physiological or biochemical indicators: relative water content, soil moisture, soil water potential, malondialdehyde, hydrogen peroxide, electrolyte leakage, antioxidant enzymes, soluble sugars, chlorophyll, chlorophyll fluorescence, growth traits, biomass, and ion balance. The authors acknowledge that these data are absent, but they do not provide new measurements. Acknowledging missing data is not enough if the manuscript still discusses stress physiology. Without these measurements, the paper cannot distinguish whether proline accumulation reflects stress exposure, developmental stage, dormancy, tissue age, nitrogen metabolism, or species-specific baseline metabolism.

3. Species, site, and habitat remain completely confounded

The revised manuscript acknowledges that species identity is not independent from sampling site and habitat. This is good, but the interpretation still repeatedly treats species identity as biologically meaningful. The manuscript itself states that species and habitat were not independent and that species identity should be interpreted as a combined biological and environmental effect. This is a fatal design limitation. Because each species was collected from a different natural habitat, the study cannot determine whether differences in proline are caused by species identity, soil properties, salinity, water availability, microclimate, phenology, or local environmental history.

4. The modeling component remains weak and somewhat decorative

The model results remain modest: the linear model has R² = 0.488, while the Random Forest model has R² = 0.350 and performs worse than the simpler model. The authors added some Random Forest parameters, including 300 trees, minimum leaf size of two, and LOOCV. However, the important requested diagnostics are still missing: residual plots, residual distribution, residual autocorrelation, leave-one-species-out validation, sensitivity analysis excluding species identity, permutation importance, and species-specific prediction errors. Because the dataset is small and species identity dominates the model, the Random Forest feature importance is not very informative. It risks telling us only that Sternbergia lutea is different from the other three species.

5. Figure package remains insufficient

The authors added error bars to Figure 1, which is an improvement. But the reviewer asked for a stronger figure package, including replicate-level points, species-specific panels, residual plots, observed-versus-predicted plots by species, residual-versus-fitted plots, and a climate–proline integration figure. The authors mostly declined these requests as beyond scope. For a modeling paper, residual diagnostics are not optional decoration. They are basic evidence that the model is interpretable.

6. The conclusion still overreaches

The conclusion still states that the work contributes to understanding osmotic adjustment mechanisms and can support climate-resilient conservation strategies, including prioritization of vulnerable species, restoration taxa selection, and adaptive management under climate change. This is not supported by the dataset. A study of seasonal proline in four species, with three reused datasets and no direct water-status or stress-injury measurements, cannot support conservation prioritization or adaptive management claims.

Minor comments

1. There is a serious editing error in the Introduction: “environmental stress The linear seasonal regression model explained…” This sentence is misplaced and must be corrected.
2. The objective still says “forecast proline dynamics,” which is too strong.
3. The hypothesis that proline dynamics can be “reliably predicted” is contradicted by the modest model performance.
4. The phrase “predictive performance” should be replaced with “descriptive performance” throughout.
5. The ANOVA should be reconsidered because monthly observations are not independent in the ordinary sense, and aggregation of biological replicates into monthly species means discards within-month biological variation.
6. The title should not lead with “Modeling” unless the modeling is strengthened.

Author Response

Please see the attachment

Author Response File: Author Response.pdf