Integration of Global Lipidomics and Gonad Histological Analysis via Multivariate Chemometrics and Machine Learning: Identification of Potential Lipid Markers of Ovarian Development in the Blue Mussel (Mytilus edulis)
, Stefano Carboni 3,4, Cinzia De Vittor 2, Phillip D. Whitfield 5,6, Mary K. Doherty 5 and Adam D. Hughes 1Round 1
Reviewer 1 Report
Comments and Suggestions for Authors
Generally this research performed on very high level. I have only several notes on this manuscript.
1. First is excessive use statistical treatment of experimental data.
2. I think that for better understanding of the results are necessary simple phospholipid and fatty acid analyses. As is known, identification of individual fatty acids by LC-MS is insufficient.
3. An advantage of complicated (and expensive) LC-MS method over classic histology is not so obvious.
Author Response
Generally this research performed on very high level. I have only several notes on this manuscript.
Comment 1: First is excessive use statistical treatment of experimental data.
Response 1: Dear reviewer 1, thank you for this comment. The present manuscript addresses a heavyweight statistical approach to highlight lipids that specifically change between ripe and not ripe blue mussel ovaries (BMOs). The extensive statistics are required to highlight these lipids from a very complex and convoluted data set where gonadal development was not the main factor being assessed. We also used simpler chemometric methods such as PCA (see lines 303 or Figure A2), but the patterns and results were not very clear for several reasons, e.g. individual variability in the lipid profiles of the gonads, the different mussel groups included in the study, etc. Using the supervised approach, we were instead able to highlight a small subset of lipids that changed between the two gonadal groups.
Comment 2: I think that for better understanding of the results are necessary simple phospholipid and fatty acid analyses. As is known, identification of individual fatty acids by LC-MS is insufficient.
Response 2: Dear reviewer 1, thank you for this comment. The main focus of the work is to assess changes in the whole lipidome between ripe and non-ripe BMOs, so the overview here is far more holistic than a fatty acid or phospholipid profiling approach. LC-MS/MS analysis is capable of providing the full configuration of a lipid molecule (lipid class + fatty acid composition and position of esterification of each of the acyl groups), but we anticipate that this is beyond the scope of this manuscript. The bulk fatty acid composition (sum of carbons and double bonds) of each phospholipid found in BMO can also be found in Supplementary Data 1. We have also performed a fatty acid analysis of the BMO dataset, which was not included in the original version of the manuscript and is now available as further supplementary material (Supplementary Data 2) to the paper. In any case, we plan to refine the method in the future and perform a full quantification of each of the lipids highlighted in the manuscript as well as verify the lipid markers of the panel on different and independent mussel gonad datasets.
Comment 3: An advantage of complicated (and expensive) LC-MS method over classic histology is not so obvious.
Response 3: Dear reviewer 1, thank you for this comment. We agree with you that LC-MS is currently more expensive than gonadal histology. However, please note that the aim of the present work is to show for the first time that specific lipids can be a suitable proxy for ovarian development in the blue mussel. The present work can open a new space for further method development which, leading to cost reduction and analytical efficiency, in time could make of LC-MS a more suitable and efficient alternative to gonad histology.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors
The study conducted by Laudicella and colleagues provides important insights and useful information on an innovative approach that effectively integrates traditional methods, such as gonadal histological analysis, and statistical (multivariate chemometrics) and machine learning methods for identifying potential lipid markers to distinguish ripe from not ripe blue mussel ovaries. Considering the high ecological and commercial value of the analyzed species, the proposed analysis protocol could be considered a valuable tool to better understand the reproductive cycle of M. edulis in order to more appropriately manage and breed such an important resource. The manuscript is well written and structured, providing sufficient background information in the introduction and an accurate description of the adopted multidisciplinary approach in materials and methods section. The results obtained are well presented and discussed in the corresponding paragraphs, supported by clear plot and good quality images. The use of English is fluent and easy to understand. Only a few revisions are suggested, mainly concerning the taxonomic nomenclature of some species mentioned in the text:
- Line 26: It is better to write the full name (Liquid Chromatography Mass Spectrometry) before using the acronym (LC-MS) for the first time
- Line 43: Most of the keywords have already been used in the title
- Line 70-72: I suggest replacing Crassostrea gigas, Chlamis nobilis, Tapes decussatus, Tapes philippinarum with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Magallana gigas, Mimachlamys crassicostata, Ruditapes decussatus, Ruditapes philippinarum respectively
- Line 103: I suggest replacing C. gigas with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), M. gigas
- Line 106: I suggest replacing Donax trunchulus with Donax trunculus (WoRMS - https://www.marinespecies.org/)
- Line 125: I suggest replacing Paeneus marguirensis with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Penaeus merguiensis
- Line 361: The correct name of the R package used should be "ggplot2" instead of "gglot2"
- Line 425: I suggest replacing P. marguensis with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), P. merguiensis
- Line 426: The word "reported" is written twice
- Line 473: I suggest replacing Placopecten magellanus with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Placopecten magellanicus
- Line 511: I suggest replacing Venerupsis decussatus with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Ruditapes decussatus
- Line 568: I suggest replacing P. marguiensis with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), P. merguiensis
Author Response
Comments and Suggestions for Authors
The study conducted by Laudicella and colleagues provides important insights and useful information on an innovative approach that effectively integrates traditional methods, such as gonadal histological analysis, and statistical (multivariate chemometrics) and machine learning methods for identifying potential lipid markers to distinguish ripe from not ripe blue mussel ovaries. Considering the high ecological and commercial value of the analyzed species, the proposed analysis protocol could be considered a valuable tool to better understand the reproductive cycle of M. edulis in order to more appropriately manage and breed such an important resource. The manuscript is well written and structured, providing sufficient background information in the introduction and an accurate description of the adopted multidisciplinary approach in materials and methods section. The results obtained are well presented and discussed in the corresponding paragraphs, supported by clear plot and good quality images. The use of English is fluent and easy to understand. Only a few revisions are suggested, mainly concerning the taxonomic nomenclature of some species mentioned in the text:
Comment 1: Line 26: It is better to write the full name (Liquid Chromatography Mass Spectrometry) before using the acronym (LC-MS) for the first time
Response 1: We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 2: Line 43: Most of the keywords have already been used in the title
Response 2: We thank the reviewer 2 for this comment, we prefer to not change the keywords since we believe that these are the most important themes covered in the manuscript.
- Comment 3: Line 70-72: I suggest replacing Crassostrea gigas, Chlamis nobilis, Tapes decussatus, Tapes philippinarum with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Magallana gigas, Mimachlamys crassicostata, Ruditapes decussatus, Ruditapes philippinarum respectively
response 3: We thank the reviewer 2 for this comment, the text has been amended as requested.
-Comment 4: Line 103: I suggest replacing C. gigas with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), M. gigas
Response 4: We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 5 Line 106: I suggest replacing Donax trunchulus with Donax trunculus (WoRMS - https://www.marinespecies.org/)
Response 5: We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 6 Line 125: I suggest replacing Paeneus marguirensis with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Penaeus merguiensis
Response 6 We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 7 Line 361: The correct name of the R package used should be "ggplot2" instead of "gglot2"
Response 7 We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 8 Line 425: I suggest replacing P. marguensis with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), P. merguiensis
Response 8 We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 9 Line 426: The word "reported" is written twice
Response 9 We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 10 Line 473: I suggest replacing Placopecten magellanus with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Placopecten magellanicus
Response 10 We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 11 Line 511: I suggest replacing Venerupsis decussatus with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), Ruditapes decussatus
Response 11 We thank the reviewer 2 for this comment, the text has been amended as requested.
Comment 12 Line 568: I suggest replacing P. marguiensis with the accepted scientific names according to the World Register of Marine Species (WoRMS - https://www.marinespecies.org/), P. merguiensis
Response 12 We thank the reviewer 2 for this comment, the text has been amended as requested.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for Authors
Summary of the study:
This paper discusses the culmination of machine learning and histological analysis in studying the global lipidomics in the gonadal tissues of blue mussels from Scotland .The study was also extended to identify the lipid biomarkers that are crucial for ovarian development. Mussels are important part of marine ecosystem and has high nutritional value and high demand as food. However the mussel production in Europe has halted because of changes in the climate. Understanding how the mussels adapt to these changes both physiologically and biochemically will aid in the further commercial production of mussels. One of the physiological adaptation of mussels is through energy saving and spawning using reproductive cycle. Reproductive cycle of mussels involve gonadal maturation, oocyte development. Oocytes of blue mussels accumulate lipid and glycogen for embryo development and adaptation to changes in the weather. Lipids mainly triglycerides significantly influence gonad development. Seasonal variations during embryo development causes changes in the lipid composition. The authors used LC-MS and Gonadal histological analysis (GHA) for studying the lipidomics changes in the ripe vs immature ovaries .The blue mussels (n=51) were collected and reared in Scotland and dissected for right mantle lobe and treated with formalin for GHA analysis whereas lipids for LC-MS analysis were extracted using Folchs’s extraction and analysis was performed using C-18 column and detected by high resolution mass spectrometer in both positive and negative ionization. The data was then processed in R by using both unsupervised and supervised machine learning models such as PCA ,Volcano plots, OPLS-DA and Random Forest. 25 significant features were identified by the machine learning models that differentiated between ripe and immature ovaries. Using OPLS-DA 40 lipid correlated between ripe and immature ovaries .Random forest was able to classify the lipid classes as evidenced by the identification of saturated and unsaturated lipids. Using all the statistical machine learning models potential lipid markers such as CerPE, CAEP, LPC, PA, PC, PE, PG and PS were identified. Finally, the functional role of the identified lipid markers was confirmed by LiOn enrichment analysis.
Major Concerns:
1. The study mentioned the use of shell fish paste containing different algae sps. What is the rationale behind using these specific organisms and did the lipids from these diets have affected the lipidomics differences that you observed? Adding this information might provide additional information for reproducing the experiments.
2. The sample size for ripe and not ripe ovaries were different. Was it a typo? If not why did you decide to go with different sample size? Do you think this will affect your data when you apply different statistical models? Please provide this information in the manuscript for clear understanding.
3. I agree with the authors view of GHA limitations such as limited sample size, need of skilled operator but it will be noteworthy to know how this limitations are not affecting the analysis of larger data sets and what steps can be taken to limit these biases in statistical analysis.
4. GHA also suffers from not able to distinguish translational changes in the gonadal development. This affects the reproductive cycle and high throughput nature of the method . The authors can have a section to discuss about GHA limitations and how they overcame these in their analysis.
5. Overfitting of the data using OPLS-DA is a significant problem and hinders the ability to full uncover the important features in gonadal development. Does the cross validation and permutations really help to lower this or is this data dependent? The authors can describe about changes in the data results before and after applying CV and permutations (can be added to supplemental information). It can help researchers working on data sets using OPLS-DA.
Minor Concerns
1. The line 81 mentions historical approaches but I think it should be histological. Please correct it.
Author Response
Reviewer 3
Comments and Suggestions for Authors
Summary of the study:
This paper discusses the culmination of machine learning and histological analysis in studying the global lipidomics in the gonadal tissues of blue mussels from Scotland .The study was also extended to identify the lipid biomarkers that are crucial for ovarian development. Mussels are important part of marine ecosystem and has high nutritional value and high demand as food. However the mussel production in Europe has halted because of changes in the climate. Understanding how the mussels adapt to these changes both physiologically and biochemically will aid in the further commercial production of mussels. One of the physiological adaptation of mussels is through energy saving and spawning using reproductive cycle. Reproductive cycle of mussels involve gonadal maturation, oocyte development. Oocytes of blue mussels accumulate lipid and glycogen for embryo development and adaptation to changes in the weather. Lipids mainly triglycerides significantly influence gonad development. Seasonal variations during embryo development causes changes in the lipid composition. The authors used LC-MS and Gonadal histological analysis (GHA) for studying the lipidomics changes in the ripe vs immature ovaries .The blue mussels (n=51) were collected and reared in Scotland and dissected for right mantle lobe and treated with formalin for GHA analysis whereas lipids for LC-MS analysis were extracted using Folchs’s extraction and analysis was performed using C-18 column and detected by high resolution mass spectrometer in both positive and negative ionization. The data was then processed in R by using both unsupervised and supervised machine learning models such as PCA ,Volcano plots, OPLS-DA and Random Forest. 25 significant features were identified by the machine learning models that differentiated between ripe and immature ovaries. Using OPLS-DA 40 lipid correlated between ripe and immature ovaries .Random forest was able to classify the lipid classes as evidenced by the identification of saturated and unsaturated lipids. Using all the statistical machine learning models potential lipid markers such as CerPE, CAEP, LPC, PA, PC, PE, PG and PS were identified. Finally, the functional role of the identified lipid markers was confirmed by LiOn enrichment analysis.
Major Concerns:
Comment 1 The study mentioned the use of shell fish paste containing different algae sps. What is the rationale behind using these specific organisms and did the lipids from these diets have affected the lipidomics differences that you observed? Adding this information might provide additional information for reproducing the experiments.
Response 1 We thank reviewer 3 for this comment. Shellfish paste was used to rear the mussels, as a large amount of food was needed to keep the mussels during the experiment (which was not possible obtain with the facilities available on site). The shellfish paste (https://reedmariculture.com/products/shellfish-diet?srsltid=AfmBOortzMyXP8NeV8BkUh6xDOqn5LW4AWt7nRK0El2H_19nNZiUXwkX) is a mixture of different strains of microalgae grown in open pond cultures, spun down and collected in a bottle. Although it is not ideal, there are several scientific studies that have successfully used it to rear mussels. Some examples can be found in the following references:
https://doi.org/10.1155/2023/9841172
https://doi.org/10.1016/j.toxicon.2013.01.010 https://doi.org/10.1016/j.fsi.2018.02.014
https://doi.org/10.1186/s43591-023-00052-8
The results presented in this manuscript are not only based on laboratory-reared mussels fed with shellfish paste, but also on a large number of mussels collected from the wild (see Methods supplement 1 for more details). Therefore, we do not consider diet to be a factor that confounds the results described in this paper. In addition, it should be noted that the gonads are a tissue least affected by changes in diet in terms of polar lipid composition (All lipids listed in table 1 are polar) please see Soudant et al. Journal of Experimental Marine Biology and Ecology 1996 Vol. 205 Pages 149-163.
Comment 2 The sample size for ripe and not ripe ovaries were different. Was it a typo? If not why did you decide to go with different sample size? Do you think this will affect your data when you apply different statistical models? Please provide this information in the manuscript for clear understanding.
Response 2 We thank reviewer 3 for this comment. The differences in sample size (26 not ripe versus 25 ripe BMOs) are result of the decision to include the totality BMO samples that we processed and analysed in this paper. As it is difficult with mussels to determine the sex of individuals before they are sacrificed, we were aware of such eventuality. In light of this opted to keep all possible information (BMOs analysed) to enhance as possible our statistical power. We are aware that imbalance between the two groups, in this case difference in just one individual, could have influenced the statistical outputs. However, the unbalancing effect due to just one single individual more in non-ripe BMO group can have limited effects on the statistical outcome, also taking into account that RandomForest can be adjusted to reduce the effects of different sample sizes by stratified sampling, which was implemented by setting the sampsize parameter equal to the smaller of the two groups (ripe). In doing so, 1 random non-ripe BMO was omitted for each bootstrapping, therefore 25 not-ripe BMOs were compared with 25 ripe BMOs, with this action repeated 200000 times during the whole RandomForest run.
Comment 3 I agree with the authors view of GHA limitations such as limited sample size, need of skilled operator but it will be noteworthy to know how this limitations are not affecting the analysis of larger data sets and what steps can be taken to limit these biases in statistical analysis.
Response 3 We thank reviewer 3 for this comment, two short paragraphs have been added in the conclusion section of the manuscript to accommodate these considerations. Lines 606-622
Comment 4 GHA also suffers from not able to distinguish translational changes in the gonadal development. This affects the reproductive cycle and high throughput nature of the method . The authors can have a section to discuss about GHA limitations and how they overcame these in their analysis.
Response 4 We thank reviewer 3 for this comment and would like to point out that the manuscript aims to integrate GHA with LC-MS lipidomics data. We are concerned that a detailed description of the limitations of gonadal histological approaches is beyond the scope of this manuscript, and we are unsure whether we have sufficient observations to attempt to link specific lipids to different atretic and degenerative stages. This will require specific experimental design, where a large number of observations should be considered for each atretic/degenerative phenomenon.
Comment 5 Overfitting of the data using OPLS-DA is a significant problem and hinders the ability to full uncover the important features in gonadal development. Does the cross validation and permutations really help to lower this or is this data dependent? The authors can describe about changes in the data results before and after applying CV and permutations (can be added to supplemental information). It can help researchers working on data sets using OPLS-DA.
Response 5 We thank reviewer 3 for this comment, CV and permutation tests were applied here as a measure of OPLS-DA model performance and fitness, so these two tests may be informative about the reliability of the interpretation of the model results and not in the actual features that are important and selected by the model classification. In terms of performance, the text already reports in line 335 that 10.2% of the variance can be predicted by the model (Figure A3A, p1R2x), while 16.7% of the variance is uncorrelated with the grouping variable (oR2x). R2y instead provides information on the variance within the two groups, which is correlated with the grouping variable (p1R2y = 50%) and uncorrelated with the grouping variable (oR2y = 17%). The model accuracy (goodness of prediction, Q2 = 0.363) is close to the value of 0.4 accepted for metabolomics data and it depends, also, on the number of variables correlated with the grouping factor. Since overall differences between the two ovarian groups were not very large it does not surprise a not large value for this parameter (otherwise it would not have been necessary the large data analysis efforts made in the present manuscript to extract those meaningful information). The permutation test (the two groups are randomly assigned to individual observations and the permutation of the classification model is calculated under the null hypothesis that there are no differences between the two groups in the absence of an effect) showed that the two groups were correctly separated by the model, as the permuted distribution – the true one with the “correct labels” - was significantly different from the permuted one (Figure A3b). For more details on the permutation and CV test, see Westerhuis et al. 2008. Metabolomics DOI 10.1007/s11306-007-0099-6. A brief description of these parameters is provided, as requested, in the caption of Figure A3.
Minor Concerns
Comment 6 The line 81 mentions historical approaches but I think it should be histological. Please correct it.
Response 6 We thank reviewer 3 for this comment and believe that "historically" is the correct word, as we meant to say that in the past gonadal development in mussels was assessed using allometric indices which used the increase in flesh weight as a measure of mussel condition. The sentence has been modified to improve its readability.
Author Response File: Author Response.pdf
