Salinity tolerance loci revealed in rice using high-throughput non-invasive phenotyping

High-throughput phenotyping produces multiple measurements over time, which require new methods of analyses that are flexible in their quantification of plant growth and transpiration, yet are computationally economic. Here we develop such analyses and apply this to a rice population genotyped with a 700k SNP high-density array. Two rice diversity panels, indica and aus, containing a total of 553 genotypes, are phenotyped in waterlogged conditions. Using cubic smoothing splines to estimate plant growth and transpiration, we identify four time intervals that characterize the early responses of rice to salinity. Relative growth rate, transpiration rate and transpiration use efficiency (TUE) are analysed using a new association model that takes into account the interaction between treatment (control and salt) and genetic marker. This model allows the identification of previously undetected loci affecting TUE on chromosome 11, providing insights into the early responses of rice to salinity, in particular into the effects of salinity on plant growth and transpiration.

phenotyping" is nicely written paper with far-reaching conclusion that are not well supported using concrete evidences. Authors have used automated HTP non-evasive system for the TR, TUE and RGR but the output of this study does not make it very novel in terms of significance and impact on advancement for the development of salinity tolerant rice genotypes with high yield. My comments on the paper are : 1. Authors claimed that "This study provides new insights into .........and will contribute to crop improvement". The study involved 24 days old seedlings when they start gaining tolerance and the degree of discrimination between tolerant and sensitive becomes less comparing when 7-10 days old seedling is used for screening. 2. Another point which is also related with extrapolated claim for crop (rice) improvement, rice is extremely sensitive at early seedling stage and reproductive stage (especially about 10 days + and -to booting), but not very sensitive at other stages. Tolerance at seedling stage does not translate into grain yield (ultimate for rice improvement) but tolerance to reproductive stage definitely translates into grain yield. There are numerous studies that explain conclusively that there is very poor association of seedling stage salinity tolerance vs reproductive stage salinity tolerance or with grain yield. There are different set of genes / QTLs that govern seedling and reproductive stage salinity tolerance. What I want to emphasize that there is not much importance of this study to the scientists in this field. 3. Authors tried to over-emphasize the importance of osmotic component of salt stress in rice with non-conclusive evidences or without concrete evidences at many places in [320][321][322][323]. Indeed osmotic stress depends much more upon species the tolerance level of genotypes. In rice, the osmotic phase is very short lived and overtakes by ionic toxicity with few hours. There are plenty of studies with hourly and daily salt uptake in rice that show the shortlived osmotic component of stress. But in this study, authors did not use few buffer plants to show that their leaves did not have the salt concentration until 2-6 days. So suggestive evidences are not good-enough for claiming the contrary findings. I don't feel that this paper will influence thinking or make any big impact in the relevant field and has merit to be published in Nature communications.

Reviewer #3 (Remarks to the Author):
The present study utilized high-throughput phenotyping technology and GWAS to analyze rice plant salinity tolerance. The paper provides a straight-forward to follow and at the same time very detailed description of the utilized methods. Of high interest to the research community is likely the consideration of complex to analyze traits such as relative growth rates, transpiration rates and transpiration use efficiency throughout the plant development under different treatments. The presented data and methodology seems to be valid, the cleansing of the data is explained in sufficient detail and the overall quality of the presentation is high. Utilized statistical methods are referenced and standard deviations, confidence intervals etc. are tabulated as needed. The 'new salinity tolerance loci' seem to be complex and only significant at certain time periods. Detailed follow-up experiments or replications of the experiment are eased by the in-depth explanations, e.g. on exact pot design, soil covering using gravel and plastic mats and plant placement regime. To improve the paper it is suggested to give more details on how and why six derived indices (lines 190 and following, and 466 and following) have been chosen out of the 24 in suppl. table 3). It is not detailed, how the plant shoot area (PSA) is exactly considered. Is it the sum of the projected areas from side and top, or is it a volume estimation? As this data and information is a essential for the analysis, it should be shortly explained in the paper and not only available from reference literature. In line 173 a 'box plot' in figure 2e is mentioned, but figure 2e is a table (should be included as a table). Figure 2c and d should use the same scaling for the y axis, to ease the comparison of the two genotypes. In the paper a 'new' association model (lines 218 and 517) is explained. It was not clear, if this model has been created newly for the paper or if it has been just recently used in related studies, why and what is new? In lines 297 to 303, the advantages of the spline-model are explained in detail, but no disadvantage was mentioned. The advantage of no a priori assumption on the shape of the curve may in some situation also be a disadvantage, as it makes model-based predictions on the data more difficult. The advantage of having just a few clearly defined curve determinants describing the overall growth dynamic e.g. in the logistic growth model are not mentioned. On the experimental procedures, it was not clear if there was a larger biomass variability, and if this affects the weighing data and thus the watering regime (lines 388 and following). As transpiration is largely affected by illumination intensity, it is suggested to think about considering in the future sensor-based illumination and temperature data for the analysis. The provided link to the image data (line 430) did not work in my tests. Before publication this problem should be fixed. Overall the paper is clearly written and considers a wide range of appropriate reference literature. After minor revision it is recommended for publication.
Reviewers' comments on manuscript entitled "New salinity tolerance loci revealed in rice using high-1 throughput non-invasive phenotyping"-NCOMMS-16-03546-T 2 3 We thank the referees for their comments to which we have replied to below. 4 5 Reviewer #1 (Remarks to the Author): 6 7 In this manuscript, a GWAS for salinity tolerance in a panel consisting of indica and aus rice genotypes is 8 conducted. There are two novel aspects of this work: i.) smoothing splines are used to model the 9 phenotypes across 13 different time points, and ii.) the GWAS model includes additional fixed effects for 10 treatment and the interaction between treatment and markers. Some novel loci for transpiration use 11 efficiency are detected using this approach. 12 Overall, I think that this work is solid, especially with respect to the statistical analysis that was 13 conducted. Additionally, I think that the manuscript is generally well written (although it would not hurt 14 to go through another round of revisions). My main constructive criticisms of the manuscript are as 15 follows: 16 17 1.) I disagree the thresholds that were used to determine statistical significance. I strongly suggest using 18 the Benjamini-Hochberg procedure that controls the false discovery rate at 5%. 19 The suggestive threshold of p = 10 −5 as a cut-off for significance has recently been used in other rice 20 studies such as Crowel et al. analysis using the conventional MLM approach with TASSEL software, just as we did. Therefore, we 25 believe that the suggestive significant threshold of p = 10 −5 is an appropriate level for the conventional 26 MLM we used in this study. Nevertheless, we followed the referee's suggestion and calculated the 27 Benjamini-Hochberg critical value for each p-value to control the false discovery rate at the 5% level for 28 the conventional MLM model using TASSEL. The Benjamini-Hochberg cut-off value to determine 29 significance of a SNP was taken as the largest p-value, where the p-value was less than the Benjamini-30 Hochberg critical value. In addition, we investigated the significance of a SNP using the False Discovery 31 Rate (FDR) and Bonferroni correction. We found that neither the Benjamini-Hochberg, nor the FDR, nor 32 the Bonferroni thresholds showed significant p-values for the conventional MLM. For our (more 33 powerful) proposed interaction model, which was the main focus of the paper, we used the Bonferroni 34 threshold at 5% significance to determine SNP significance and these results are the ones presented in 35 the manuscript. To be consistent with the significance testing of both the conventional and interaction 36 models, we have now included the Bonferroni results in the text for both models and clarified the 37 manuscript text about the difference in results found using the two models specifically by showing that 38 the interaction model provided a higher statistical power. 39 40 2.) I felt that there were too many acronyms in the manuscript, some of which ( invalidates this study, and does not reduce its potential to make significant contributions to increasing 144 salinity tolerance. We imposed salinity on 29 day-old plants that were at approximately at the 5-leaf 145 stage, and we have clarified this in the methods section (line 417). Salt-stress was imposed at this stage 146 because we specifically wanted to quantify transpiration rate and transpiration use efficiency. To obtain 147 rates of transpiration that were significantly greater than the evaporation of water from the soil surface, 148 we needed to impose salinity on bigger plants (older plants) to collect individual plant water loss data. It 149 is known that yield is also affected when plants are exposed to salinity at the vegetative stage, affecting Euphytica 127: 235-245). We acknowledge that the main vegetative stage is less sensitive to salinity 153 compared to the early seedling stage (with 2 to 3 leaves); however, our data clearly shows a significant 154 effect of salinity on plant growth and transpiration and significant variation in responses to salinity (as 155 presented in Figures 1 and 2). We have introduced a new paragraph at the end of the discussion to 156 address the referee's concern (line 341-352  195 > We disagree with the referee's assertion that the osmotic phase is very short-lived and is overtaken by 196 the ionic phase within a few hours. To our knowledge, we do not know studies where these kinds of 197 measurement, especially in rice, have been undertaken. In fact, there is no evidence that the so-called 198 osmotic phase is "overtaken" at all. We suspect this will depend on many factors, such as the 199 concentration of salts in the external solution, transpiration rates, and the effects of other stresses on 200 the plant. The ionic effects do not require simply the accumulation of ions -the ions have to accumulate 201 to a sufficient extent and for a sufficiently long time for the ionic effects to become evident. Given our 202 study is focusing on the effects of salinity on the production of new leaves, and given the ionic phase is 203 often defined by the effects of salt accumulation on acceleration of the senescence of the older leaves, 204 the referee's argument could rapidly become a semantic one. We could, if the editors want, formulate a 205 new term to ensure that we define that the processes being studied in the current manuscript are 206 clearly focusing on the growth of new leaves and not on the death of old leaves. However, we think this 207 semantic is unlikely to be helpful, and, to us, at least, it is self-evident that we are focusing on a process 208 related to early salinity response, which is clearly distinct from the one related to the longer-term 209 premature senescence of older leaves (the process generally accepted in the literature as being the one 210 primarily related to the so-called "ionic phase"). We have also added four lines of text addressing this 211 issue in the Discussion, lines 340-343. 212 213 I don't feel that this paper will influence thinking or make any big impact in the relevant field and has 214 merit to be published in Nature communications. 215 > We think that this referee has misunderstood some of the primary points of this work and disregarded 216 the novelty of the methods presented in the manuscript, as evidenced by all three of his/her comments. 217 We hope that our explanations above are sufficiently clear to enable a discounting of the final opinion 218 proffered by this referee. 219 220 Reviewer #3 (Remarks to the Author):

222
The present study utilized high-throughput phenotyping technology and GWAS to analyze rice plant 223 salinity tolerance. Overall the paper is clearly written and considers a wide range of appropriate reference literature. After 295 minor revision it is recommended for publication. 296

Reviewer #1 (Remarks to the Author)
In my opinion, the authors did a great job of addressing the majority of my (and the other reviewer's) comments in this manuscript. I have only two further suggestions (one of them is extremely minor) about the statistical aspects of the paper: 1.) For the "suggestive" thresholds (of P = 1.0 x 10^-5) used to designate "significant" GWAS results (particularly in the section between lines 176-248). I completely disagree with the argument of "this threshold has been used in previous studies so therefore it is ok to use it for our study". In general, just because an approach is used in the literature does not mean that it is correct. Moreover, equating this suggestive threshold to statistical significance (as done in Line 217) is both misleading to the reader and statistically incorrect. Please consistently use a significance threshold that controls either a type I error rate (e.g., Bonferroni) or a false discovery rate (e.g., the Benjamini-Hocbherg procedure) for both the traditional MLM and the new statistical model incorporating marker by treatment interactions; at the very least, an "apples to apples" comparison can then be made between the two models. Also, please indicate in the captions to Table 2 and Figure 3 what criterion (e.g., Bonferroni) was used to determine statistical significance.
From a quantitative genetics perspective, I think that a lot of impressive work was done in this manuscript. Please do not diminish this study by using a significance threshold that is not based on well-grounded statistical theory.
2.) Lines 111-112: r^2 is actually the squared Pearson correlation coefficient. Please make this clear in the text.

Reviewer #2 (Remarks to the Author)
This is the revised MS submitted by the authors and i am satisfied by the response for the first two points but i still standby with my third comment on over-emphasis on osmotic component of salt stress. There is no concrete evidence that it last for a long. I am attaching one classical work by Yeo et al. 1991 (JXB) that clearly showed the short lived osmotic response in rice overcome by ionic component of stress using the real-time experiment on leaf elongation and ionic uptake. The displacement transducer very nicely showed the time-course of leaf elongation under culture solution with NaCl, KCl and mannitol.