Accuracy of Genomic Prediction of Yield and Sugar Traits in Saccharum spp. Hybrids
, Lifang Qin 1,4, Alexander E. Lipka 5, Sushma Sood 1, James Todd 6 and Marcus Olatoye 7
Round 1
Reviewer 1 Report
Accuracy of Genomic Prediction of Yield and Sugar Traits in Saccharum spp. Hybrids
Md. Islam et al.
The manuscript is well written, and the information is of interest to sugarcane breeders around the world with particular interest in the Florida sugar industry. The manuscript would be acceptable after major revision.
Some specific comments are below:
Abstract: The authors use adjective rather than results from the study to describe effects. Words/phrases such as “modestly improved”, “acceptable level”. I do not think that the data support the conclusion of the last sentence.
Page 1 Line 43 Refer to sugar and yield, explain better and use the terms you define further into the paper.
Page 2 line 77-79 Factors affecting prediction accuracy are listed. The authors exclude phenotyping of the training population. Here is where a review of previously reported H2 and h2 from previous work would be useful as well as GxE information. This information would be useful in designing the training population trial (TP) and determining the size of the TP.
Page 3 Plant Materials, it was difficult to understand much about the TP since it was referenced in a supplementary table. I could not find the information. It seems odd that 18 cultivars and experimental clones from Louisiana were included. The germplasm base of Florida and Louisiana sugarcane breeding programs are unique. You want your TP to most accuracy reflect the target breeding program, which I assumed to be Florida. The authors should provide planting dates and harvest dates for the trial(s).
Page 3 Line 145, that is not the formula used by the Legendre (32) reference. Makes me wonder if the Legendre reference is appropriate on page 4 as well. Author infer the sugar yield but should this be sucrose content.
Page 5 line 243, I like the trait-assisted genomic selection approach to predict among crops.
Page 6, line 287, again I could not find the data in “S” files
Page 6, line 298, the heritability values seem quite low based on previous reports and experience.
Page 7, line 305, I wonder what the results would look like if used on the bottom 20% of the population? With low prediction accuracy, some suggest selecting off the bottom of the population. Also, in this paragraph are vague conclusions such as, (trend, line 311), (more or less similar, line 311), (ADE was better, …, although all models performed statistically similar (line 313). The differences among prediction accuracy values are numerical mainly and those significantly different not of any practical significance. Same comments for the regression results in Lines 318-323
Page 8, when I view the prediction accuracy values, they are quite low. My first instinct is to know more about the phenotyping of the data from the field trial. What is the CV% for each trait in each trial? The authors allude to problems in phenotyping later on in the discussion.
Page 10, is figure 3 referenced inside of the paper? For all box and whisker plots, the authors should explain how to interpret the data in the Material and Methods. I forgot what the dots in the graphs mean. What is the dashed line in each figure?
Page 11, line 387, with such low prediction accuracy (PA) values (likely from problems associated with the quality of the field trial??), I am not sure of the usefulness of discussion of minimum number of SNPs to use in this study. At these PA values, any slight improvement would be beneficial.
Page 13, line 405, training population size. Based on work done by Lian and Bernardo (2014), optimum population size could be determined:
Page 15 line 440, current breeding programs should be referred to as phenotypic recurrent selection programs. The authors state that the single breeding cycle could take more than 10 years – this is inaccurate. Variety release may take 10-13 years, but the breeding cycle is less. In year 5-6 of the recurrent selection program, new clones are cycled back to the crossing house in one-half of the time frame indicated by the authors. Line 441-442 indicates yield plateau with a reference. I am familiar with the work – the authors do not discuss the fact that crop cycle (number of crops grown from a single planting) has lengthened over time. Sugar industries may sacrifice some annual yield in favor of increasing yield for the entire crop cycle. This discussion should move to the introduction because the results in the paper do not support shortening the breeding cycle, only the potential to do so. The statement on lines 448-451 makes the reader question why the Louisiana clones were added to the trial.
Much of the discussion section focuses on potential of the technology, small numeric differences among models, problems with the field trial. The discussion about nonadditive genetic variation is useful. Why wouldn’t the ADE model help with this situation? A section in the introduction should be devoted to past studies of gene action (additive vs nonadditive) in sugarcane.
Page 17, lines 553-557, confusing conclusion. Nonadditive genetic variation was important, and ADE and RKHS models would help with this. However, these models did not statistically differentiate themselves from other models. Conclusions of this paper need to be scaled back quite a bit.
Comments for author File: 
Comments.docx
Author Response
Response to reviewer’s comments
Reviewer 1
The manuscript is well written, and the information is of interest to sugarcane breeders around the world with particular interest in the Florida sugar industry. The manuscript would be acceptable after major revision.
Some specific comments are below:
Abstract: The authors use adjective rather than results from the study to describe effects. Words/phrases such as “modestly improved”, “acceptable level”. I do not think that the data support the conclusion of the last sentence.
Response: We have made some changes to the abstract. However, we believe that the last sentence is good enough to conclude the utilization of this study.
Page 1 Line 43 Refer to sugar and yield, explain better and use the terms you define further into the paper.
Response: Updated
Page 2 line 77-79 Factors affecting prediction accuracy are listed. The authors exclude phenotyping of the training population. Here is where a review of previously reported H2 and h2 from previous work would be useful as well as GxE information. This information would be useful in designing the training population trial (TP) and determining the size of the TP.
Response: Thanks a lot. We have included that information.
Page 3 Plant Materials, it was difficult to understand much about the TP since it was referenced in a supplementary table. I could not find the information. It seems odd that 18 cultivars and experimental clones from Louisiana were included. The germplasm base of Florida and Louisiana sugarcane breeding programs are unique. You want your TP to most accuracy reflect the target breeding program, which I assumed to be Florida. The authors should provide planting dates and harvest dates for the trial(s).
Response: Thanks a lot for pointing this out. We have confirmed that all the supplementary files have been uploaded to the journal submission system. Louisiana sugarcane breeding programs are using a Florida breeding facility for making crosses in Canal Point, FL. Thus, the parental lines from Louisiana are also being used by the Florida breeding program for getting several beneficial traits (Freeze tolerant, disease resistance, ratoonability, and early sugar). We also did principle component analysis to find out the relationship among the tested clones (as below). This figure is only for review purposes. The Louisiana clones (red dots in the figure) are not genetically unique. So, we believe that including 18 germplasm in the TP from Louisiana would make it more representative of the breeding parental line. The planting date was November 2016 (line# 123) and the harvesting dates were February 2017, 2018, and 2019 (line# 131).
Page 3 Line 145, that is not the formula used by the Legendre (32) reference. Makes me wonder if the Legendre reference is appropriate on page 4 as well. Author infer the sugar yield but should this be sucrose content.
Response: We checked again the Legendre (1992) [32] report very carefully. Legendre (1992) reported how to correct the Brix and Pol from the raw data and convert them to factory measurement (sucrose content, CRS, and TRS). We wrote that “SC were measured from the corrected BRIX and pol using the following formula according to Legendre (1992)”. The formula was created by us based on the suggestion given by the author of that paper. We have used this formula in all of our variety release manuscripts (one of those has been cited now). We are attaching the referred paper with this response for the reviewer’s information.
Page 5 line 243, I like the trait-assisted genomic selection approach to predict among crops.
Response: Thanks a lot.
Page 6, line 287, again I could not find the data in “S” files
Response: Thanks again for pointing this out. We have uploaded all the supplementary files to journal submission systems. We have uploaded again to ensure the supplementary files are present.
Page 6, line 298, the heritability values seem quite low based on previous reports and experience.
Response: We also agree with the reviewer. We discussed this matter in the discussion (line# 461-77). There have also been some reports that agreed with our results [51-53] (in the reference list).
Quentin: The reviewer seems to be very concerned with the low heritability and low prediction accuracy. I am assuming the two are related. The capacity of the GS approach to predict is limited by the quality of the genomic signal. If most of the variation in the phenotype is environmental then it is not possible to attain good prediction accuracy. I would respond to all these comments about low heritability and low prediction accuracy by reiterating that no GS study in sugarcane has achieved results that are that much better than what we achieved. However I would recommend scaling back some of the statements to be more realistic about the performance we achieved.
Another point we can raise is that even if the absolute values of heritability are low, we still documented a trend where heritability decreases quite a lot in the later crop cycles (1st and 2nd ratoon crops). It is still important to document that trend.
Page 7, line 305, I wonder what the results would look like if used on the bottom 20% of the population? With low prediction accuracy, some suggest selecting off the bottom of the population. Also, in this paragraph are vague conclusions such as, (trend, line 311), (more or less similar, line 311), (ADE was better, …, although all models performed statistically similar (line 313). The differences among prediction accuracy values are numerical mainly and those significantly different not of any practical significance. Same comments for the regression results in Lines 318-323
Response: Thanks a lot for pointing this. In this manuscript, we have only included the results of prediction accuracy. We did not make any selection of either topmost or bottommost clones. The prediction accuracy already integrates over the entire population, not just the top 20%, so we already have that result to confirm that the coincidence index would probably also yield relatively poor performance if we selected off the bottom of the population. We have included some numbers in that section. We didn’t do any statistical analysis of the performance of GS models. Thus, we have deleted the statistically similar and updated them accordingly. However, we believe that lines 318-323 represent the results of this study very well. Thus, we didn’t make any changes. Please note that so far there have five GS studies have been reported. The prediction accuracies were low to moderate in all of the previous studies. It might be normal for sugarcane.
Page 8, when I view the prediction accuracy values, they are quite low. My first instinct is to know more about the phenotyping of the data from the field trial. What is the CV% for each trait in each trial? The authors allude to problems in phenotyping later on in the discussion.
Response: Thanks a lot for raising this concern. We have included the CV value for all the traits across all crop cycles in Table S2. Most of the CV looks very good and few of the traits have higher CVs, especially those that are complex secondary traits (estimated from other traits using formula). The stalk weight (SW) and stalk population (SP) have the comparatively higher CV and higher prediction accuracy. Thus we believe that CVs are not practically relevant. CVs are higher due to causes related to other factors like low PA.
Page 10, is figure 3 referenced inside of the paper? For all box and whisker plots, the authors should explain how to interpret the data in the Material and Methods. I forgot what the dots in the graphs mean. What is the dashed line in each figure?
Response: Thanks a lot. We have included in the M&M section “In these box-and-whisker plots, the thick central line indicates the median value from 25 iterations of model fitting, the colored box represents the central 50% quantile interval, the lines extending from the box encompass the largest value no further than 1.5 times the interquartile range from the edge of the box, and any points outside that range are plotted individually.The dashed line represents the null expectation for each metric of model performance (0 for prediction accuracy and 0.167 for coincidence index).”
Page 11, line 387, with such low prediction accuracy (PA) values (likely from problems associated with the quality of the field trial??), I am not sure of the usefulness of discussion of minimum number of SNPs to use in this study. At these PA values, any slight improvement would be beneficial.
Response: Thanks a lot for raising the concern. We are also aware of this. Our conclusion for this part is a range of SNP markers (4000 to 5000) could give us an optimum level of prediction accuracy based on the model and traits. If we could use fewer SNPs and get similar PA then we could save a lot of money associated with the SNP genotyping. The main objective of this section is how many SNPs could sufficiently cover the LD of the sugarcane genome. Even there will have small changes in PA would not affect the coverage of SNPs in the sugarcane genome. Our results are also supported by other sugarcane genomic studies. Again the low prediction accuracies in sugarcanwereas also reported by the other researchers.
Write to save the cost. Max PA 0.9 and trend is shallow. Number of SNPs related to PA is not surprising.
Page 13, line 405, training population size. Based on work done by Lian and Bernardo (2014), optimum population size could be determined:
Response: Lian and Bernardo (2014) conducted the genomic selection study using two sets of bi-parental population. Thus, their strategy is out of scope in this study. The paragraph pasted above was copied from the introduction section of that study. That is not their work. We believe that our approach was fine since many previous studies reported a similar way.
Page 15 line 440, current breeding programs should be referred to as phenotypic recurrent selection programs. The authors state that the single breeding cycle could take more than 10 years – this is inaccurate. Variety release may take 10-13 years, but the breeding cycle is less. In year 5-6 of the recurrent selection program, new clones are cycled back to the crossing house in one-half of the time frame indicated by the authors.
Response: Thanks a lot for raising this concern. Each breeding cycle starts with crossing and ends with releasing new cultivars according to Yadav et al. (2020) [21]. Zhao (2020) reported on the Florida sugarcane breeding program. This report mentioned that the Florida sugarcane breeding program needs more than 10 years to release a cultivar and complete a breeding cycle.
Please check Yadav et al. (2020) [21] wrote that “Because of the biology and management of sugarcane and theextensive phenotypic testing system, it can take more than ten years to complete a breeding cycle and even longer to commercially release a new cultivar.” This work is well familiar to this reviewer as stated in the next line.
Voss-Fels et al. (2021) [30] mentioned that “This means that recombination through crossing, and hence reshuffling of alleles, is only carried out once every 10 years per individual breeding cycle”
Line 441-442 indicates yield plateau with a reference. I am familiar with the work – the authors do not discuss the fact that crop cycle (number of crops grown from a single planting) has lengthened over time. Sugar industries may sacrifice some annual yield in favor of increasing yield for the entire crop cycle. This discussion should move to the introduction because the results in the paper do not support shortening the breeding cycle, only the potential to do so.
Response: Thanks a lot for the suggestion. Yadav et al (2020) [21] mentioned in Figure 1 that there was no substantial gain in cane yield in the top six sugarcane-producing countries in the world. Thus we referred to this work and indicates the yield plateau. We didn’t refer to this report as a part of any crop cycle advantages or disadvantages. We included this in the discussion part for letting the readers know the benefit of our study. Our objective of this study is to experimentally evaluate the prospects of GS in the Florida sugarcane breeding programs. The benefits of retooning are out of the scope of this study. The introduction section has already enough background information for conducting this study. Thus, we decided to keep this in the discussion section.
The statement on lines 448-451 makes the reader question why the Louisiana clones were added to the trial.
Response: Please see above.
Much of the discussion section focuses on potential of the technology, small numeric differences among models, problems with the field trial. The discussion about nonadditive genetic variation is useful. Why wouldn’t the ADE model help with this situation? A section in the introduction should be devoted to past studies of gene action (additive vs nonadditive) in sugarcane.
Response: Thanks a lot for raising the concern. We have already mentioned in the discussion part, the possible reasons that ADE did not perform as expected in lines numbers 498 to 502. The low PA accuracy is not only due to problems with the field trial. The complexity of the sugarcane genome also contributes as recognized by other previous sugarcane GS studies. We have included some statements in the introduction section regarding the non-additive genetic variation.
Page 17, lines 553-557, confusing conclusion. Nonadditive genetic variation was important, and ADE and RKHS models would help with this. However, these models did not statistically differentiate themselves from other models. Conclusions of this paper need to be scaled back quite a bit.
Response: We have changed the sentence.
Reviewer 2
In the results both broad and narrow sense heritabilities for almost all traits were lower than expected values and gradually decreased from PC to RT2. This is an important area to be discussed in detail. When the heritabilities decline upto 50% for many traits in RT2 as compared to PC, then overall application of the results is questionable. In general, sugarcane clones in the selection trial as well as varieties, fair poorly for economic traits in the ratoon trials. That is reflected in the present experiments also. Globally, sugarcane stalks accumulate various pathogens of viruses, phytoplasma and bacteria systemically. These pathogens cause degeneration in the ratoons hence we record poor cane yield and sugar yield. The authors have not addressed this issue and not discussed in discussion. Rather they cite competition among the plants in a row, harvesting time effect, bias in selecting ten stalk samples, and/or other unmeasured environmental variation as contributing factors for the variation yield and sugar traits. But they ignored the major cause of systemic pathogens in reducing the economic yield parameters in ratoons.
Response: Thanks a lot for your suggestion. We have included three lines and references in this aspect in the discussion section.
Author Response File: Author Response.docx
Reviewer 2 Report
In the results both broad and narrow sense heritabilities for almost all traits were lower than expected values and gradually decreased from PC to RT2. This is an important area to be discussed in detail. When the heritabilities decline upto 50% for many traits in RT2 as compared to PC, then overall application of the results is questionable. In general, sugarcane clones in the selection trial as well as varieties, fair poorly for economic traits in the ratoon trials. That is reflected in the present experiments also. Globally, sugarcane stalks accumulate various pathogens of viruses, phytoplasma and bacteria systemically. These pathogens cause degeneration in the ratoons hence we record poor cane yield and sugar yield. The authors have not addressed this issue and not discussed in discussion. Rather they cite competition among the plants in a row, harvesting time effect, bias in selecting ten stalk samples, and/or other unmeasured environmental variation as contributing factors for the variation yield and sugar traits. But they ignored the major cause of systemic pathogens in reducing the economic yield parameters in ratoons.
Author Response
In the results both broad and narrow sense heritabilities for almost all traits were lower than expected values and gradually decreased from PC to RT2. This is an important area to be discussed in detail. When the heritabilities decline upto 50% for many traits in RT2 as compared to PC, then overall application of the results is questionable. In general, sugarcane clones in the selection trial as well as varieties, fair poorly for economic traits in the ratoon trials. That is reflected in the present experiments also. Globally, sugarcane stalks accumulate various pathogens of viruses, phytoplasma and bacteria systemically. These pathogens cause degeneration in the ratoons hence we record poor cane yield and sugar yield. The authors have not addressed this issue and not discussed in discussion. Rather they cite competition among the plants in a row, harvesting time effect, bias in selecting ten stalk samples, and/or other unmeasured environmental variation as contributing factors for the variation yield and sugar traits. But they ignored the major cause of systemic pathogens in reducing the economic yield parameters in ratoons.
Response: Thanks a lot for your suggestion. We have included three lines and references in this aspect in the discussion section.
Round 2
Reviewer 1 Report
I have no further comments - acceptable for publication.
