Natural variation at FLM splicing has pleiotropic effects modulating ecological strategies in Arabidopsis thaliana

Investigating the evolution of complex phenotypes and the underlying molecular bases of their variation is critical to understand how organisms adapt to their environment. Applying classical quantitative genetics on a segregating population derived from a Can-0xCol-0 cross, we identify the MADS-box transcription factor FLOWERING LOCUS M (FLM) as a player of the phenotypic variation in plant growth and color. We show that allelic variation at FLM modulates plant growth strategy along the leaf economics spectrum, a trade-off between resource acquisition and resource conservation, observable across thousands of plant species. Functional differences at FLM rely on a single intronic substitution, disturbing transcript splicing and leading to the accumulation of non-functional FLM transcripts. Associations between this substitution and phenotypic and climatic data across Arabidopsis natural populations, show how noncoding genetic variation at a single gene might be adaptive through pleiotropic effects.

In this manuscript, entitled Splicing natural variation at FLM induces 1 synergistic pleiotropy in Arabidopsis thaliana", Hanemian and co-authors report the dissection of a QTL controlling the rosette growth difference shown by Can-0 and Col-0, two A. thaliana genotypes. The fine-mapping of a QTL in chromosome 1, followed by a thorough functional validation (complementation by allelic test and by transformation) reveals that this variation is encoded in the FLM locus. Despite the absence of coding variation, directed mutagenesis demonstrates that a variant affecting splicing causes the modification. That splicing affects function at the FLM locus had already been demonstrated, but the authors work further to show that this natural variant not only affects rosette growth and flowering time, it also alters the components of the leaf economic spectrum. A modification of the function of FLM appears to be sufficient to shift the leaf economic spectrum by a few percent towards increased resourceacquisitive leaf physiology. To my knowledge, this is the first time that a flowering time allele is explicitly related to a phenotypic measure of central ecological relevance. The authors further show that the loss of function allele of Can-0 is shared by a number of other Iberian genotypes, and the distribution of the alleles associates with areas where temperature is relatively less homogeneous. The results presented in this manuscript are excellent, and the association between flowering and LES very novel and convincing.The elaboration on the putative adaptive relevance of the variant is an interesting addition that adds relevance to the work. I therefore strongly recommend this manuscript for publication. I think at this point that the discussion needs some work.
Here are several points that the final version should improve. 1) I am confused about how the authors interpret temperature effects. The QTL at FLM seems to be dependent on mean temperature and the polymorphism seems to associate with temperature heterogeneity. Why should that imply that its action is dependent on the amplitude of temperature fluctuations? Differences between 16 and 23 degrees are well within the realm of the daily temperature fluctuations one expects, no? Would it not be straightforward to validate this hypothesis? 2) Why are Col-0 and flm-3 so far off the spectrum measured within A. thaliana and across species (Fig.4b)? I think there is something unrealistic there. Why should thaliana display the LES range of all herbaceous species (it is not a succulent)? 3) Is it necessary to arm-wave around the role of FLM on WUE? I believe it is more relevant to discuss the link with LES. What does it mean? Why can one believe that this have any ecological relevance? Is LES related to fitness and how? 4) Origin of the Can allele. Which of the two alleles is ancestral? Why not use the other Brassicaceae species to determine the ancestral state? 5) Relict origin: while I understand that the authors are referring to this classification to describe the history of the Can allele, the literature on relics is unclear as to what really defines relicts. A more generic conclusion is that the Can-0 and the Col-0 allele have distinct admixed origins. In this context, it is not straightforward to determine whether the Can-0 allele is adaptive (the association with temperature fluctuations suggests so, but this would have to be experimentally confirmed). 6) The last paragraph must be rewritten. I think the authors are somewhat confused about several concepts. Studies show that mutations with an intermediate pleiotropic effect are more likely to be consistently selected when seasonal conditions fluctuate. The authors show some pleiotropic effects of FLM, but the pattern of selection of each of these traits is unclear. Is flowering time variation independent of the LES, or is the LES a trait integrating variation across molecular processes, at the organ level? What I mean here is that the authors would not describe the effect of the mutation on splicing and flowering time as being pleiotropic. They are causal. As long as the causality on LES is not determined, it is a stretch to highlight the case studies by the authors as an example of adaptive (ie synergistic) pleiotropic variant. Can the two traits evolve independently as well (in which case one could study whether the pleiotropic mutation is more adaptive than a none pleiotropic mutation affecting only FT and not LES or only LES and not FT)? I understand that such follow-up work is not in the realm of the present manuscript, but I think the manuscript would be improved if it were refocused on the significance of pleiotropy, making clear that whether it is causality that links these traits, or a synergistic association between two traits whose genetic basis can otherwise be independent is not known. 7) As a consequence, I don't think the authors bring evidence of synergistic pleiotropy. So the author should either make their point more clearly or change their title.

Miscellaneous comments:
Line 169: what is "higher" growth? Faster growth? Higher final PRA? Line 252-253: I don't think the word "trigger" is appropriate. May be something like "The effect of FLM loss of function on LMA reaches 2-3% of the total variance….". In addition, since LMA and Amass being way out of the range normally reported in plants, it is unclear whether the effect reported here is realistic. Line 283: I appreciate that the authors are being cautious on this conclusion. Their p-value is not massively significant, and such correlation can also arise from outliers of population structure. This hypothesis should be validated with ad-hoc experiment manipulating temperature fluctuations. Line 355-356: not clear why mentioning neo-functionalization adds to the discussion. I think this paragraph is not very useful. Line 363: P5CS1 splicing: please check this work carefully, the alternative splicing affects a transcript that is not necessary for function. Line 401: this sentence is confusing. Please rewrite. Line 413: reciprocal planting experiments? Where? Statement is unclear. I think the authors mean a common garden across environments differing in temperature fluctuations and precipitations. Fig. 4b: why so many points for Col-0/flm3? Are the 30 replicates shown here? Would a mean with error bars be more appropriate? The legend needs to be clarified.
Reviewer #2 (Remarks to the Author): The middle of this paper (Results) describes a nice genetic dissection of some growth traits differences between 2 Arabidopsis lines. Hanemian et al. first grew RILs from the Can-0 and Col-0 cross and performed a QTL analysis. They then watched the major QTL segregate in the progeny of a residually heterozygous RIL, confirming the phenotypic effect. The QTL contained FLOWERING LOCUS M (FLM), so the authors went to nulls for this gene. Next, they performed a quantitative complementation assay in F1 hybrid plants to mutant flm-3. Finally, Hanemian et al. transformed the flm-3 mutant with the indicated sequence from both parental lines. In aggregate, these data show that allelic variation at FLM is responsible for growth and color differences between Can-0 and Col-0 owing to the QTL on chromosome 1. Subsequent work suggests that a SNP creates a splice site difference, causing the phenotype. In general, the work is a compelling demonstration of the power of forward genetics in Arabidopsis.
The introduction needs a complete re-write. Despite much lingo and many broad statements, I could not tell what the paper was actually about until I read the results section (which is very clear).
Regarding the motivation for the study, the fact that a mutation to a MADS-box transcription factor has effects on multiple phenotypes is not remotely surprising. Any mutation that affects growth or allocation will affect numerous plant dimensions through time. However, the fact that the Quantitative Trait Nucleotide (QTN) is a splice variant is notable discovery. While not unprecedented (line 359 onward), the result bears directly on the general question of what sorts of mutations make a QTN. This would be a better focus for the Introduction/Discussion.

Reviewer #3 (Remarks to the Author):
The manuscript by Haneman et al describes how natural variation at the FLOWERING LOCUS M (FLM) gene contributes to pleiotropy in Arabidopsis. The authors mapped QTLs for several growth related traits, which turned out to be mapping to the FLM locus. When they analysed the FLM locus in strains that differed in the alleles, the underlying molecular mechanism hinted at changes in splicing. The authors go to length to try and establish that variation in splicing contributes to this variation across multiple growth related traits. The authors claim that Natural variation in FLM splicing mediates synergistic pleiotropy in Arabidopsis. The authors show that the alleles of interest have an interpretable geographic distribution indicative of adaptive significance. Overall the story is interesting, as it goes beyond flowering and it would be of interest to plant biologists and evolutionary biologists. However the study suffers from few issues, which I have listed below, addressing which would substantially improve this manuscript.
This manuscript suffers severely from a chronological effect of experiments. In 2013/14, there was Nature paper published by Pose et al, which proposed a model based on an assumption that the primary transcripts arising from FLM are FLM-beta and FLM-delta. This assumption was incorrect and this mis-led the field and several researchers on a wrong track. This paper also appears to be one of those mis-led by this story and that is reflected in their interpretation. When I look at the data they present it is very clear, what is really happening. In Col-0 you have primarily FLM-beta. In Can-0, you have a new splice acceptor site that is strong enough to outcompete the normal splice site that would have resulted in FLM-beta. Therefore, with the new splicing instead of AG1 (i.e., splice acceptor that is in front of Exon2) the new splice site is used resulting in a novel transcript, that contains a premature stop codon. The authors are too influenced by the beta/delta model presumably due to historical reasons, and fail to see this. As far as I could see their data clearly shows that the primary effect is loss of the normal (FLM-beta) transcript and formation of a novel transcript with a premature stop codon due to a hot splice acceptor site. There is no need at all to discuss their work in the context of FLM-delta in this paper. This is the primary problem with this paper and it requires re-writing the paper with the current knowledge. The authors presumably are tip-toeing around this issue, which is not very helpful.
I have given some clear suggestions below. rHIF068: The legend says it is fixed for FLM. The genotype panel indicates that orange is Col and Blue is Can. Then what do the colours indicate in rHIF68? Does this mean that rHIF068 that is fixed for Col and Can alleles for FLM do not differ in their phenotypes? Is this correct? Or rHIF68 is fixed for Col allele? This makes not much sense. It think both these figures can be removed. They do not really add any value. It is pretty clear from this work as well as from earlier works that such analysis would lead to misleading conclusions. The authors have done a lot of work. But this does not help either the paper or the story itself. I suggest the authors to directly discuss the sequencing of full-length cDNA clones and remove this from the manuscript. The entire section of lines 194 to 213 can be removed. I believe all these discussions with exon2 vs exon 3 etc are not very useful. The authors can simply state that temperature affects FLM splicing and then present their cloning data. I suggest to make Fig  Lines 231 and 232: We have no idea whether the promotion of growth has got anything to do with these alleles. Should be rephrased to say that there was no differences in PRA were observed between plants harbouring either allele. Lines 327-329: "which is expressed at a substantial level in many Arabidpsis strains"? where is this data from… I do not believe any paper has shown this clearly. I suggest to remove this.
Lines 338: Remove in exon 3 etc.. A premature stop codon is a premature stop codon. End of story.
Lines 344-346 and the full line of argument: This is incorrect to state. Nd-1 and Ei-6 carries full deletion of the FLM locus. I am not sure how appropriate is the line of argument made in this paragraph.

Reviewer #1 (Remarks to the Author):
In this manuscript, entitled Splicing natural variation at FLM induces 1 synergistic pleiotropy in Arabidopsis thaliana", Hanemian and co-authors report the dissection of a QTL controlling the rosette growth difference shown by Can-0 and Col-0, two A. thaliana genotypes. The finemapping of a QTL in chromosome 1, followed by a thorough functional validation (complementation by allelic test and by transformation) reveals that this variation is encoded in the FLM locus. Despite the absence of coding variation, directed mutagenesis demonstrates that a variant affecting splicing causes the modification. That splicing affects function at the FLM locus had already been demonstrated, but the authors work further to show that this natural variant not only affects rosette growth and flowering time, it also alters the components of the leaf economic spectrum. A modification of the function of FLM appears to be sufficient to shift the leaf economic spectrum by a few percent towards increased resource-acquisitive leaf physiology. To my knowledge, this is the first time that a flowering time allele is explicitly related to a phenotypic measure of central ecological relevance. The authors further show that the loss of function allele of Can-0 is shared by a number of other Iberian genotypes, and the distribution of the alleles associates with areas where temperature is relatively less homogeneous. The results presented in this manuscript are excellent, and the association between flowering and LES very novel and convincing.The elaboration on the putative adaptive relevance of the variant is an interesting addition that adds relevance to the work. I therefore strongly recommend this manuscript for publication. I think at this point that the discussion needs some work.
Here are several points that the final version should improve. 1) I am confused about how the authors interpret temperature effects. The QTL at FLM seems to be dependent on mean temperature and the polymorphism seems to associate with temperature heterogeneity. Why should that imply that its action is dependent on the amplitude of temperature fluctuations? Differences between 16 and 23 degrees are well within the realm of the daily temperature fluctuations one expects, no? Would it not be straightforward to validate this hypothesis?

>>> On the one hand, we used the known temperature-dependent impact of FLM on flowering time to validate our QTL (Fig1B) and subsequently on the other hand, we observed an association between the causal polymorphism in the FLM gene and temperature heterogeneity (Fig5C). We didn't mean to imply that FLM action is dependent on the temperature heterogeneity, but that allelic variation at FLM and its distribution may depend on this climatic variable. Still, we think the impact of daily temperature changes should not be confounded with temperature fluctuations across the plant cycle and development. Testing the impact and mode of action of temperature fluctuation consequences either on FLM splicing or flowering time would indeed require more careful experiments especially in growth chambers set at different temperature regimes, but this doesn't seem so straightforward to us as there are multitude of temperature fluctuations that could be tested. We are not totally confident in our understanding of this specific question, so if we have missed the point of the reviewer here, please get back to us.
2) Why are Col-0 and flm-3 so far off the spectrum measured within A. thaliana and across species (Fig.4b)? I think there is something unrealistic there. Why should thaliana display the LES range of all herbaceous species (it is not a succulent)? (Fig. 4B)

. Indeed, plants were grown in the greenhouse under relatively low light intensity (ca. 65 µmol m -2 s -1 PPFD), compared to the natural accessions grown in the PHENOPSIS phenotyping platform under moderate light intensity for instance (ca. 195 µmol m -2 s -1 PPFD; Granier et al., 2006), or the interspecific data collected in natura under much higher (and fluctuating) light intensity. Still, the main point is that the relationship between these traits was conserved across these datasets (p>0.05).
3) Is it necessary to arm-wave around the role of FLM on WUE? I believe it is more relevant to discuss the link with LES. What does it mean? Why can one believe that this have any ecological relevance? Is LES related to fitness and how?

>>> This is a very good suggestion and we had already investigated this idea. Unfortunately, the lack of a clear haplotype underlying FLM-Can allele restricts the investigation to a single SNP position. We also aligned the sequences of the intron2 of FLM from A. lyrata, A. halleri, C. grandiflora, C. rubella, FLM-Can and FLM-Col. The A. thaliana sequences of this intron were much shorter than any of the other sequences making the comparison difficult.
We also drew phylogenetic trees using genomic, coding and intron2 FLM sequences of the same species, but the A. thaliana always clustered altogether. Consequently we decided to describe this as an allele of ancient origin and not the ancestral allele (now L273), as we were unable to characterise it. 5) Relict origin: while I understand that the authors are referring to this classification to describe the history of the Can allele, the literature on relics is unclear as to what really defines relicts. A more generic conclusion is that the Can-0 and the Col-0 allele have distinct admixed origins. In this context, it is not straightforward to determine whether the Can-0 allele is adaptive (the association with temperature fluctuations suggests so, but this would have to be experimentally confirmed).

>>> Thanks for your remark. We have introduced even more caution now as to our use of relicts in the results and discussion and refer to the "generic conclusion" suggested by this reviewer (now L273 & L391). As for the adaptive character of variation at FLM, we have reviewed our text and we think we have now been careful in using this term and indeed we suggest experiments to confirm this.
6) The last paragraph must be rewritten. I think the authors are somewhat confused about several concepts. Studies show that mutations with an intermediate pleiotropic effect are more likely to be consistently selected when seasonal conditions fluctuate. The authors show some pleiotropic effects of FLM, but the pattern of selection of each of these traits is unclear. Is flowering time variation independent of the LES, or is the LES a trait integrating variation across molecular processes, at the organ level? What I mean here is that the authors would not describe the effect of the mutation on splicing and flowering time as being pleiotropic. They are causal. As long as the causality on LES is not determined, it is a stretch to highlight the case studies by the authors as an example of adaptive (ie synergistic) pleiotropic variant. Can the two traits evolve independently as well (in which case one could study whether the pleiotropic mutation is more adaptive than a none pleiotropic mutation affecting only FT and not LES or only LES and not FT)? I understand that such follow-up work is not in the realm of the present manuscript, but I think the manuscript would be improved if it were refocused on the significance of pleiotropy, making clear that whether it is causality that links these traits, or a synergistic association between two traits whose genetic basis can otherwise be independent is not known.

>>> We have modified the text at several instances to keep it clear that we did not determine the causality relationships between traits related to LES and flowering time (now L322-323). We also added to the discussion that genes impacting flowering time were previously associated with LES, although the relationship between these is not resolved (see 2nd and 3rd paragraphs of the new version).
7) As a consequence, I don't think the authors bring evidence of synergistic pleiotropy. So the author should either make their point more clearly or change their title.

Miscellaneous comments:
Line 169: what is "higher" growth? Faster growth? Higher final PRA?

>>> This has been changed for faster growth (now L152).
Line 252-253: I don't think the word "trigger" is appropriate. May be something like "The effect of FLM loss of function on LMA reaches 2-3% of the total variance….".

>>> This has been corrected accordingly (now L225-227)
In addition, since LMA and Amass being way out of the range normally reported in plants, it is unclear whether the effect reported here is realistic.

>>> See answer to point #2)
Line 283: I appreciate that the authors are being cautious on this conclusion. Their p-value is not massively significant, and such correlation can also arise from outliers of population structure. This hypothesis should be validated with ad-hoc experiment manipulating temperature fluctuations.

>>> In this paragraph, we used the same type of statistical analysis as in the previous paragraph (now L240-246) i.e. taking into account the population structure. Regarding the p value, we inverted the 2 last sentences of the paragraph as it could be confusing (now L252-256). Reviewer1 is right, an experiment would be required to validate this hypothesis (now stated in L336-341).
Line 355-356: not clear why mentioning neo-functionalization adds to the discussion. I think this paragraph is not very useful.

>>> The sentence mentioning the neofunctionalization has been removed.
Line 363: P5CS1 splicing: please check this work carefully, the alternative splicing affects a transcript that is not necessary for function.

>>> We double-checked this work from Kesari et al 2012, they showed that intronic/noncoding natural variations are able to alter alternative splicing which in turn changes the amount of functional versus non-functional form of P5CS1 (which is in turn affecting proline content).
Line 401: this sentence is confusing. Please rewrite.

>>> This part about WUE has been removed to refocus the discussion according to your point #3)
Line 413: reciprocal planting experiments? Where? Statement is unclear. I think the authors mean a common garden across environments differing in temperature fluctuations and precipitations.

>>> We have modified this accordingly (now L339-341).
Fig. 4b: why so many points for Col-0/flm3? Are the 30 replicates shown here? Would a mean with error bars be more appropriate? The legend needs to be clarified.

Reviewer #2 (Remarks to the Author):
The middle of this paper (Results) describes a nice genetic dissection of some growth traits differences between 2 Arabidopsis lines. Hanemian et al. first grew RILs from the Can-0 and Col-0 cross and performed a QTL analysis. They then watched the major QTL segregate in the progeny of a residually heterozygous RIL, confirming the phenotypic effect. The QTL contained FLOWERING LOCUS M (FLM), so the authors went to nulls for this gene. Next, they performed a quantitative complementation assay in F1 hybrid plants to mutant flm-3. Finally, Hanemian et al. transformed the flm-3 mutant with the indicated sequence from both parental lines. In aggregate, these data show that allelic variation at FLM is responsible for growth and color differences between Can-0 and Col-0 owing to the QTL on chromosome 1. Subsequent work suggests that a SNP creates a splice site difference, causing the phenotype. In general, the work is a compelling demonstration of the power of forward genetics in Arabidopsis.
The introduction needs a complete re-write. Despite much lingo and many broad statements, I could not tell what the paper was actually about until I read the results section (which is very clear). Regarding the motivation for the study, the fact that a mutation to a MADS-box transcription factor has effects on multiple phenotypes is not remotely surprising. Any mutation that affects growth or allocation will affect numerous plant dimensions through time. However, the fact that the Quantitative Trait Nucleotide (QTN) is a splice variant is notable discovery. While not unprecedented (line 359 onward), the result bears directly on the general question of what sorts of mutations make a QTN. This would be a better focus for the Introduction/Discussion.

Reviewer #3 (Remarks to the Author):
The manuscript by Haneman et al describes how natural variation at the FLOWERING LOCUS M (FLM) gene contributes to pleiotropy in Arabidopsis. The authors mapped QTLs for several growth related traits, which turned out to be mapping to the FLM locus. When they analysed the FLM locus in strains that differed in the alleles, the underlying molecular mechanism hinted at changes in splicing. The authors go to length to try and establish that variation in splicing contributes to this variation across multiple growth related traits. The authors claim that Natural variation in FLM splicing mediates synergistic pleiotropy in Arabidopsis. The authors show that the alleles of interest have an interpretable geographic distribution indicative of adaptive significance. Overall the story is interesting, as it goes beyond flowering and it would be of interest to plant biologists and evolutionary biologists. However the study suffers from few issues, which I have listed below, addressing which would substantially improve this manuscript. This manuscript suffers severely from a chronological effect of experiments. In 2013/14, there was Nature paper published by Pose et al, which proposed a model based on an assumption that the primary transcripts arising from FLM are FLM-beta and FLM-delta. This assumption was incorrect and this mis-led the field and several researchers on a wrong track. This paper also appears to be one of those mis-led by this story and that is reflected in their interpretation.

>>> The reviewer is correct about the chronological effect: when we started our work on FLM, the main paper about this gene was indeed from Pose et al 2013 and we started to study the transcripts of both FLM-beta and FLM-delta isoforms, while it was shown in further articles (Lutz et al 2015 & 2017; Sureshkumar et al 2016) that FLM-beta is the functional isoform, and that the change in splicing toward potentially non-functional isoforms is the mechanism in play. However, we think that our interpretation was right as it is close to what the reviewer writes below (see the paragraph starting with "FLM is a well-known…" L330 in our original version, and now L342 unchanged).
When I look at the data they present it is very clear, what is really happening. In Col-0 you have primarily FLM-beta. In Can-0, you have a new splice acceptor site that is strong enough to outcompete the normal splice site that would have resulted in FLM-beta. Therefore, with the new splicing instead of AG1 (i.e., splice acceptor that is in front of Exon2) the new splice site is used resulting in a novel transcript, that contains a premature stop codon. The authors are too influenced by the beta/delta model presumably due to historical reasons, and fail to see this. As far as I could see their data clearly shows that the primary effect is loss of the normal (FLMbeta) transcript and formation of a novel transcript with a premature stop codon due to a hot splice acceptor site.
There is no need at all to discuss their work in the context of FLM-delta in this paper. This is the primary problem with this paper and it requires re-writing the paper with the current knowledge. The authors presumably are tip-toeing around this issue, which is not very helpful.

>>> Although we think our interpretation was this one already, we agree that our Results section describing our work on other isoforms is not needed anymore regarding the current FLM model (hence we have deleted L198-211 in the Results of the previous version). Thanks to this reviewer's comment, this modification simplifies this section and avoids paying tribute to information which are now outdated.
I have given some clear suggestions below. Fig 1A: rHIF068: The legend says it is fixed for FLM. The genotype panel indicates that orange is Col and Blue is Can. Then what do the colours indicate in rHIF68? Does this mean that rHIF068 that is fixed for Col and Can alleles for FLM do not differ in their phenotypes? Is this correct? Or rHIF68 is fixed for Col allele? This makes not much sense. Table S4. rHIF068 is segregating for a region that does not include FLM (contrary to rHIF099) and, hence, does not differ in phenotypes. The legend has been improved for clarity.

>>> The significance of the Fig1C corresponds to the pvalue (= 2.638e-05) mentioned line 181 (original version). We have added a sentence in the legend for clarity.
Lines: 194 to 195. Rephrase this.. This idea of discussing FLM splicing in the context of FLM beta and delta leads to fundamental confusion in the manuscript. It is better to state that FLM splicing/expression Fig S3. It think both these figures can be removed. They do not really add any value. It is pretty clear from this work as well as from earlier works that such analysis would lead to misleading conclusions. The authors have done a lot of work. But this does not help either the paper or the story itself. I suggest the authors to directly discuss the sequencing of full-length cDNA clones and remove this from the manuscript. The entire section of lines 194 to 213 can be removed. I believe all these discussions with exon2 vs exon 3 etc are not very useful. The authors can simply state that temperature affects FLM splicing and then present their cloning data. I suggest to make Fig S4 as the main fig S2, which can easily drive the point. show that flm has no effect on Amass (pvalue code is not indicated in the legend…)?

>>> Regarding the previous modification/deletion of the paragraph about FLM isoforms, we agree with the reviewer that the fig2 and FigS3 are not relevant anymore. We only kept the relative expression obtained by qRT-PCR with primers spanning the first intron as it is a good proxy for the loss of expression of FLM-beta in Can-0. This result now becomes the Fig2A. We then followed the suggestion of the reviewer by making the initial
1>>> In Fig4, we assessed the effect of variation at FLM on LMA and Amass traits individually, in 2 different genetic backgrounds. In the 4A right-hand side panel, when comparing the mutant flm-3 to its wild type counterpart (Col-0 genetic background), we indeed didn't detect a significant difference for Amass (although there seems to be a trend in the "expected" direction), while we noticed a significant difference using the rHIF099 genetic background. This is now described in more details in the text (L223-225 track changes version). As for the p-value code, we described the basic stats in the material and methods section (paragraph "statistical analysis") to avoid too busy legends. Fig.4B show? That the covariance between LMA and Amass is not changed by FLM or that FLM has an effect on both LMA and Amass ? 2>>> In the Fig4B panel, we compare the covariance between LMA and Amass associated with the FLM function to the covariances observed at the intra-or interspecific level. As discussed in L294-297 (previously submitted version), LMA and Amass are two "core traits of the Leaf Economics Spectrum (LES), a universal trade-off between resource acquisition and resource conservation observed across thousands of plant species. The tight relationship between Amass and LMA is supposed to reflect fundamental physiological constraints among plant species worldwide, shaping their diversification and local adaptation". So, we did not expect FLM to change the covariance between these 2 traits (if any) but rather to contribute to the variation observed along the LES by affecting multivariate phenotypes coordinately. In other words, our answer is "yes" to both points raised here by the reviewer. FLM pleiotropically controls trait variation within a phenotypic space -the LES -that is highly constrained by physiological trade-offs and it has a subtle effect on both LMA and Amass as shown in Fig4A (although the effect on Amass is less clear in the Col-0 genetic background as mentioned in the previous answer).

What does
We want to mention here that we initially analyzed the covariance between LMA and Amass associated with variation at FLM in each genetic background separately. We observed a similar correlation in the Col-0 genetic background when using the flm-3 / WT comparison (standard major axis (SMA): r2 = 0.21, P < 0.001) and in the rHIF099 genetic background when using the rHIF099-Can / rHIF099-Col comparison (SMA: r2 = 0.16, P < 0.01). It is noteworthy that the correlation is stronger in the Col-0 genetic background although only LMA was significant when each trait was analysed individually (further explanation about SMA stats below). As the genetic backgrounds themselves were not associated with the covariance and were in similar phenotypic ranges, we decided to bulk all the data measured in this study in Fig4B (SMA: r2 = 0.17, P < 0.001).
The term "trigger" is confusing, I don't think that it is the genotype that are the cause of the covariance between LMA and Amass, but rather physiological constraints, and these are also perceived in flm-3 and FLM-Can.

3>>>
We agree with the reviewer that "trigger" can be misleading as it suggests a link of direct causality between variables that we cannot demonstrate (discussed in L318-327, previously submitted version), and we thank the reviewer for pointing out this issue. We had indeed previously changed one instance of 'trigger' following this reviewer's suggestion but had overlooked three other instances, which are now modified as well into 'is associated with' (L225 & L345 track changes version and in the figure 4 legend). Then, if the covariance observed in our dataset was only caused by physiological constraints without genetic effect, we should not observe any difference between the pair of genotypes tested for each trait in the Fig4A panel. So our results show that the variations along the LES axis can be partly explained by allelic variations at FLM.
In the text, the authors explain that the covariance between LMA and Amass is the same as in a worldwide sample of A. thaliana. If the slope does not change, and the effect is only significant on LMA, why do the authors conclude that FLM has a substantial effect on plant physiology? It appears to me that the effect of FLM is hardly detectable… 4>>> Despite a slight and non-significant phenotypic variation observed for Amass due to flm-3 mutation, this variation was sufficient to recapitulate the LES correlation between LMA and Amass even when considering only the data of the Col-0 genetic background (see above first answer, SMA: r2 = 0.21, P < 0.001). For this, we used state of the art statistical methods (SMA) to calculate the correlation significance and compare the regression slopes of our FLM dataset with the A. thaliana natural accessions and the species datasets (obtained from literature). SMA tests are commonly used in trait-based ecology to measure regression slopes and test differences between groups (see for