An asymmetric allelic interaction drives allele transmission bias in interspecific rice hybrids

Hybrid sterility (HS) between Oryza sativa (Asian rice) and O. glaberrima (African rice) is mainly controlled by the S1 locus. However, our limited understanding of the HS mechanism hampers utilization of the strong interspecific heterosis. Here, we show that three closely linked genes (S1A4, S1TPR, and S1A6) in the African S1 allele (S1-g) constitute a killer-protector system that eliminates gametes carrying the Asian allele (S1-s). In Asian–African rice hybrids (S1-gS1-s), the S1TPR-S1A4-S1A6 interaction in sporophytic tissues generates an abortion signal to male and female gametes. However, S1TPR can rescue S1-g gametes, while the S1-s gametes selectively abort for lacking S1TPR. Knockout of any of the S1-g genes eliminates the HS. Evolutionary analysis suggests that S1 may have arisen from newly evolved genes, multi-step recombination, and nucleotide variations. Our findings will help to overcome the interspecific reproductive barrier and use Asian–African hybrids for increasing rice production.

A few questions listed below might help imporve this work. Major criticisms: 1 The interaction of A4 with A6 and TPR is very critical to the proposed working model. However, genetic interaction does not necessarily mean physical interaction of the proteins. The evidence based on the present data for interaction is quite weak. And a further downside is that the work has provided no clue as to how and on what the killer works and the possible cellular and biochemical processes leading to gamete abortion. This left with a feeling that we came to an abrupt end. Some additional work seems necessary for smoother ending.
2 As mentioned in the introduction, both male and female gametes carrying S1-s are selectively aborted. Thus how is the phenotype of the embryo-sac in the hybrid here? Based on Fig. 2b, the author analyzed the segregation ratio based on the assumption that the female gametes with s/-were aborted. Is there any experiment to support the assumption? For Fig. 2a, I am wondering if ss/--is not observed in the segregating population. If yes, to avoid ambiguity, ss/--might be added in the genotype row with zero individual. For Fig. 2c, the green, red, and blue colors are in accordance with the colors in Fig. 2b, which is quite clear. However, the white and grey colors have no correspondence in Fig, 2b, this might cause confusion.
3 I am wondering if the three components at S1 have orthologs in other grass species. The Result said that one ortholog of TPR was identified in foxtail millet. If the orthologs existed in other species, a phylogenetic tree might tell something about the origin and relationship of these genes. In addition, a haplotype network is recommended to trace the stepwise mutational step of different alleles. Is it possible that the most ancestral haplotype contains the A4-TPR-A6, and these genes were either lost or under rapid evolution in other linages? It is not clear to me (or is it clear yet?), based on the presently available resequencing data from all sources, whether there are any sativa accessions that have all three genes (and/or other of the A1 to A6 genes) albeit that they may not be functional. This may have important implication in the evolutionary study.
glaberrima. The authors found three closely linked genes, A4, TPR, and A6 in theS1 locus of O. glaberrima. The TPR encodes a protein containing two trypsin-like peptidase domains and a ribosome biogenesis regulatory domain, and it has been already published by the same group (Molecular Plant 2017). A6 encodes a peptidase protein with similar feature to the TPR peptidase, and it has been already reported by the same group and another group. A4 encodes an unknown protein and shown to interact with A6 and TPR using BiFc assay in this manuscript. This manuscript is significant in the point reporting dual function of TPR: a complex of TPR, A4 and A6 makes up the killer system in sporophyte, and TPR alone has another role in the protection of TPRcontaining gametes.
The hypothesis of the A4-TPR-A6 gamete killer-protector system is very complicated, and hard to understand for me at a glance, but careful reading of this manuscript and the previous paper published in Molecular Plant 2017 allowed me to understand that the working model in Fig. 5 is supported by the sufficient data employing transgenic japonica rice with TPR and CRISPR/Cas9 knockout mutants of NIL.
I have some comments for minor revisions.
(1) In Lines 78 and 79, some more descriptions are needed for OgTPR1 and SSP, including full name of the genes and what they encode. Most readers will not understand what they are until they read the previous papers.

Review
This study dissects the genetic architecture of a major postzygotic reproductive barrier locus between two rice species, Oryza sativa and O. glaberrima. The motivation for this study is two-fold. First, reproductive barriers between rice taxa represent a major obstacle to breeding efforts. Second, detailed studies of reproductive barriers can shed new light on the evolution of species and are therefore of broad evolutionary interest.
In this study, the authors first use a combination of sophisticated experiments to characterize the S1 locus, a cluster of three genes that interact to confer reproductive isolation. Then, the authors characterize genetic variation at the S1 locus in the study species, as well as related species with the aim to reconstruct the evolutionary history of the locus.
The strength of this study is the detailed analysis of the molecular mechanism underlying HS mediated by the S1 locus, which provides interesting new insights into the genetic basis of HS in rice. Unfortunately, the study also suffers from major weaknesses. These include the clarity of the presentation, the conceptual basis motivating the different components of this study, and the evolutionary inferences.

Clarity of the presentation
This study is targeting a general audience and therefore must be understandable also to people who are not deeply into the field of HS in rice and into the details of analysing molecular gene functions and gene interactions. In the present version, the manuscript is very difficult to follow.
For the analysis of the molecular mechanisms, information is necessary on why the chosen transgenic lines were produced and what expectations are for the different crosses performed with these lines. In part, this information is provided, but clearly not for all relevant lines / crosses. Please clarify which lines are hemi-, homo-or heterozygous and what the relevance of this is for your expectations and findings. Again, such information is partly provided but is missing in many instances.
Phenotypes are not well explained. Please provide more information on what organs show evidence for HS, pollen, spikelets, both? What about the female function (e.g. ovules)?
The term 'gametophytic manner' and 'sporophytic manner' appears to be colloquial language. Please use established terminology and provide an explanation upon first usage of these terms such that non-specialist readers can follow.
The legend to Figure 5b should be part of the main text and would provide relevant information to readers.
The extensive naming and renaming of genes (e.g. page 4) is confusing. Please use consistent naming for the genes throughout the manuscript and for example indicate in brackets alternative names for the same genes that have been used in other studies.
Some of the terms used to describe evolutionary processes and scenarios are not adequate and need to be revised. A case in point is the first sentence of the introduction: HS does not 'suppress gene migration' but (for example) 'reduces gene-flow' between populations or species. Evolution in general is not targeted or directional and therefore also HS does not have a purpose, i.e. 'to drive speciation and contribute to ...'.
Use of the term 'gene flows' in this manuscript is wrong and the term 'evolutionary flow' (page 8) does not exist. These parts need to be reformulated and appropriate terminology is essential. Otherwise, readers will not be able to follow the complex scenarios inferred.

Conceptual basis
The paper deals with the analysis of a killer-protector system. Please explain in the introduction what this is and / or refer to suitable recent review papers that provide adequate description.
The explanation provided of the Bateson-Dobzhansky-Muller incompatibility model is imprecise (i.e. 'certain speciation loci') and requires revision. First, it would be of help to be clear on the essence of the Bateson-Dobzhansky-Muller model. The standard description, with which most non-specialist readers may be most familiar, focuses on the origin and fixation of mutations at two (or more) loci in each of two lineages derived from a common ancestor that interact epistatically and have negative effects in the hybrids (e.g [1]). The locus investigated in the present study therefore does not reflect this standard definition because it consists of a single locus (S1) that is composed of three apparently highly linked genes that interact. This one-locus model also fits the Bateson-Dobzhansky-Muller model [2] but is not the standard model and therefore requires introduction.
It remains obscure how the Bateson-Dobzhansky-Muller incompatibility relates to meiotic drive and selfish genetic elements. Please clarify these aspects and / or drop reference to concepts that are not essential for the present study.
Use of the term 'speciation locus' is not warranted in the present study. Rarely, if ever, is speciation a result of very few or a single locus, and the role of the S1 locus in speciation has not been studied. S1 apparently plays an important role in current reproductive isolation between the two studied species but this tells us nothing about its potential role in the speciation process.
The motivation for trying to reconstruct the evolutionary history of the S1 locus is unclear. I presume that the authors tried to establish the point that the alleles evolved independently in two lineages, where they became fixed and cause HS when brought into contact. These points could be considered essential components of Bateson-Dobzhansky-Muller incompatibilities and may explain why so much emphasis has been put on the idea that alleles have become fixed. To clarify this, a better explanation is required of what the key components of a Bateson-Dobzhansky-Muller incompatibility are in your view -and what scenarios / hypotheses you are testing.

Evolutionary inferences
The discussion of the evolution of the S1 locus is very difficult to follow and seems to mix up timescales. The discussion about the role of Gondwanaland in the evolution of Asian and African rice, and O. meridionalis (page 10), is incompatible with dated divergence times between species in the Oryza genus Unfortunately, foxtail millet appears far too diverged to be of much use in the present study. Data from more closely related taxa should be available and included in the analysis and discussion.
Based on the presented data on variation at the S1 locus within and among species, it appears that there is quite substantial variation within species and the common ancestor of O. sativa, glaberrima and meridionalis was most likely polymorphic at S1. Can the variation observed by a consequence of incomplete lineage sorting?
The present discussion entirely neglects the fact that Bateson-Dobzhansky-Muller incompatibilities can be polymorphic within species [1]. This aspect needs to be incorporated and discussed.
It remains unclear whether the 'S1 locus' behaves indeed as a single locus and why. Is there evidence for reduced recombination, maintaining allelic combinations (as evidenced potentially through elevated linkage disequilibrium (LD)) or are recombination rates similar to other genomic regions?

Reviewer #1
This reviewer gave a positive overall comment and raised 4 major suggestions Reviewer 1#: The manuscript dealt with 'An asymmetric allelic interaction drives allele transmission bias in interspecific rice hybrids". As well known, there is huge yield potential in the hybrid rice breeding. It is an important strategy for rice breeder to take advantage of this heterosis. However, hybrid sterility is a major form as reproductive isolation during domestication and hinders the exploitation of the heterosis displayed by inter-specific or inter-subspecific hybrid. In this manuscript, the authors demonstrated that three closely linked genes (S1A4, S1TPR and S1A6) in the African rice S1 allele constitute a gamete killer-protector system. Knockout of anyone of the three genes at this locus can overcome the interspecific reproductive barrier. Evolutionary analysis showed that the S1 loci arose from newly evolved genes and complex reorganization process. The manuscript would be of interest to a wide audience to understand the molecular mechanisms governing the incompatible interactions between the different species.
1. The cloned hybrid sterility loci, such as S5, S7, Sa, Sc, S27/S28, are closely adjacent genetic interaction. Why is the asymmetric genetic interaction localized at the S1 locus? Whether do the interval genes, such as A5 and so on, play a role in the genetic interaction? gamete-killer complex, the ORF5+ protein is encoded by the indica allele, but ORF4+ is encoded by the japonica allele. In the present study, the functional African rice allele S1-g consists of three closely linked active genes; S1A4, S1TPR, and S1A6 (SSP). The gamete-killer function of S1-HS requires the three components (S1A4, S1TPR, and S1A6) only from S1-g, while the protector function depends on solely on S1-g-derived S1TPR; no component from the Asian rice allele is required for this gamete killer-protector system. The S1 gamete killer-protector system is thus determined only by the African allele, representing an asymmetric allelic interaction. This character is distinct from the S5 and Sa systems, which are symmetric allelic interactions involving components from both divergent alleles.
Our results showed that knockout of S1A5 in the S1-g region did not affect pollen development in the NIL-g line, nor the hybrid sterility of the S1 heterozygotes ( Supplementary Fig. 3). On the other hand, the complementation of the japonica parental line RP-s with the three genes S1A4, S1TPR, and S1A6 (without S1A5 or the other putative ORFs from the S1-g region) could cause male and female sterility ( Fig. 1 and Supplementary Fig. 5). We can therefore exclude the possibility that the other genes in this genetic interval are involved in S1-mediated HS.
2. The S1 gene derived from African rice Oryza glaberrima, induced preferential abortion of both male and female gametes possessing its allelic alternative from Asian rice O. sativa only in the heterozygote. In this manuscript, the authors speculated that the co-existence of A4, TPR and A6 might produce a sterility signal in a sporophytic manner to kill gametes (both male and female), moreover, TPR expressed in the microspores and A6 (SSP) was not detected in EM and LM stage (Supplementary Fig.   1). It is better to provide more detailed data for expressions of the three genes in reproduction development?
Response: As suggested, we investigated the expression of the three genes in the anthers and panicles at various developmental stages using qRT-PCR. The new data are presented in Supplementary Fig. 2. 3. The subcellular localization of these three proteins A4, A6 and TPR were localized in the nucleus, however, the authors showed that they produced a sterility signal in a sporophytic manner to kill gametes (both male and female). Is there any indirect result from these three gene interaction?
Response: We investigated the phenotypes of the single-gene knockout mutants s1a4 and s1a6, revealing that six agronomically important traits, including their architecture, grain length, grain width, plant height, panicle length, and grain number per panicle, were not significantly affected; however, their 1000-grain weights were slightly increased, which may be related to the changes in amino acid metabolism hinted at by the transcriptome sequencing analysis. We therefore propose that the interactions of these three genes may have a pleiotropic effect on the seed-filling process in addition to their roles in HS. The transcriptome data hinted that the BCAA deficiency may have a detrimental effect on gamete development; therefore, we propose that the complex of the three S1-g gene products may induce excessive BCAA degradation in F 1 and NIL-g plants, resulting in the sterility signal affecting gamete development. The S1TPR produced in S1-g-type gametes may maintain adequate levels of BCAA via its peptidase function, thereby rescuing the gametes from amino acid deficiency. The S1-s-type gametes in the F 1 hybrids lacking the functional S1TPR would still display a BCAA deficiency however, and fail to be restored to fertility.

These results have been presented in Supplementary
In Asian rice cultivars (S1-sS1-s), the genes involved in BCAA degradation are expressed at low levels, meaning that the gametes have enough BCAA to continue their development. We have added new data in Supplementary Figs. 8-10 and mentioned the proposed mechanism in our revised manuscript (Pages 8-9, lines 238-254; Pages 13-14, lines 382-397).

Reviewer #2
This reviewer gave a positive overall comment and raised 4 major suggestions and 6 minor points This paper identified a killer-protector system at S1 locus responsible for hybrid sterility and segregation distortion between two cultivated rice species. This work substantially expanded our understanding of reproductive isolation. As shown by the work, the S1 locus is an important speciation locus as it affects both male and female sterility in the hybrids. The data obtained from genetic analyses are sound and convincing, revealing a complex interaction among the three genes at this locus that form a killer-protector system in regulating hybrid sterility. Although killer-protector system has been reported previously, a novel and interesting feature is the characterization of one causal gene, TPR, that kills the gamete in hybrids at sporophytic stage and is also responsible for rescuing the gamete in a gameotophytic manner. The subsequent sequence and divergence analysis describes an evolutionary picture with respect to the origin of such reproductive barrier. Together with other reported studies, the results indicate that the killer-protector system may be a general strategy for postzygotic reproductive isolation during the evolution. In addition, this result may serve the first step for a feasible approach to overcome the inter-specific reproductive isolation to facilitate the utilization of distant heterosis in hybrid rice breeding.
1. The interaction of A4 with A6 and TPR is very critical to the proposed working model. However, genetic interaction does not necessarily mean physical interaction of the proteins. The evidence based on the present data for interaction is quite weak. And a further downside is that the work has provided no clue as to how and on what the killer works and the possible cellular and biochemical processes leading to gamete abortion. This left with a feeling that we came to an abrupt end. Some additional work seems necessary for smoother ending.
Response: We agree with the reviewer's comments. We have conducted pull-down assays to confirm the interaction among the three proteins (Fig. 4c,   4d), the results of which indicated that S1A4 physically interacts with S1TPR and S1A6. These data are shown in Fig. 4c and 4d in the revised manuscript.
To probe the possible biochemical processes leading to gamete abortion, we performed a transcriptome sequencing analysis, revealing that the transcript levels of six genes involved in the BCAA degradation pathway were significantly higher in the F 1 and NIL-g plants than in RP-s ( Supplementary Figs. 8-10 Based on these observations, it is possible that the complex of the three proteins may induce excessive BCAA degradation, which would have a detrimental effect on fertility; therefore, we propose that the complex of the three S1-g gene products may induce excessive BCAA degradation in F 1 and NIL-g plants, resulting in the sterility signal affecting gamete development. The S1TPR produced in S1-g-type gametes may maintain adequate levels of BCAA via its peptidase function, thereby rescuing the gametes from amino acid deficiency. The S1-s-type gametes in the F 1 hybrids lacking the functional S1TPR would still display a BCAA deficiency and fail to be restored to fertility. In Asian rice cultivars (S1-sS1-s), the genes involved in BCAA degradation are expressed at

115: E1955-E1962) To answer the reviewer's question about the phenotype of the
embryo-sac, we conducted a histological observation of the embryo sacs of the hybrids derived from the crosses RP-s × NIL-g and S1A4-S1A6 t × S1TPR t . The results were similar to those of the previous studies, indicating that gametes carrying S1-s or lacking the S1TPR t transgene are selectively aborted during mega-gametogenesis (Supplementary Fig. 6).
Based on the genetic evidence presented in Fig. 2a, the co-segregations of the genotypes and phenotypes were consistent in the S1-heterozygotes carrying the transgenic S1TPR t genotypes (TT, T-, --), which strongly supports the notion that the female and male gametes containing s/-(S1-s without S1TPR t ) were aborted. No plants carrying the ss/--genotype were observed in the segregating population; to avoid ambiguity, we accepted the reviewer's suggestion and added a new row for ss/--containing no individuals.
We also appreciated the reviewer's suggestions regarding our figures. To make them clearer for readers, we added color codes in Fig. 2b according to the reviewer's suggestion. The color codes were also applied in Fig. 1f and Fig. 3d to avoid confusion.
3. I am wondering if the three components at S1 have orthologs in other grass species.
The Result said that one ortholog of TPR was identified in foxtail millet. If the orthologs existed in other species, a phylogenetic tree might tell something about the origin and relationship of these genes.
In addition, a haplotype network is recommended to trace the stepwise mutational step of different alleles. Is it possible that the most ancestral haplotype contains the A4-TPR-A6, and these genes were either lost or under rapid evolution in other linages?
It is not clear to me (or is it clear yet?), based on the presently available resequencing data from all sources, whether there are any sativa accessions that have all three genes (and/or other of the A1 to A6 genes) albeit that they may not be functional. This may have important implication in the evolutionary study.
Response: We thank the reviewer for this constructive suggestion. In response to these comments, we used the nucleotide coding sequences of the three components of the S1 locus as a query to perform a BLAST search of the Poaceae. The putative orthologs were used to construct a phylogenetic tree with the maximum likelihood method in MEGA7 (www.megasoftware.net/) and 1,000 bootstrap replications. We found that the nucleotide coding sequences of the Oryza S1TPR orthologs, but not S1A4 and S1A6, were significantly similar to sequences in the genomes of Zea mays, Sorghum bicolor, Brachypodium distachyon, Hordeum vulgare, Aegilops tauschii, Triticum aestivum, and Setaria italica ( Supplementary Fig. 11).
To determine the origin of the most ancestral S1 haplotype, O. officinalis The haplotypes of these species only harbor S1TPR and/or S1TP sequences, which may exclude the possibility of the haplotype carrying S1A4-S1TPR-S1A6 being the most ancestral.
We previously aligned the genomic sequences of the S1-g and S1-s alleles with the publicly available sequences from the wild rice genome project and the 3000 Rice Genomes Project (https://registry.opendata.aws/3kricegenome/). This showed that only the S1TP-type haplotypes (with the premature stop codon), but not the S1A4 and S1A6 sequences, were present in all O. sativa and O. rufipogon

accessions (Supplementary Figs. 12 and 13 and Supplementary Table 8).
4. Since you have no clue at the moment on what and how the killer works, it may be premature to state "this S1 gamete killer-protector system is distinct from other Bateson-Dobzhansky-Muller type…". This is also related to the question in the previous paragraph about the evolutionary analysis.

Response: We have accepted the reviewer's suggestion, and clarified the difference of killer-protector system between S1 locus and other loci in this revision (Page 13, lines 371-381).
Minor points: 1. Because of the complexity of the genes and alleles and interactions among the S1 locus, a more unified and understandable gene/allele name is recommended for ease of reading. For example, OgTPR1, TP, S1TPR, TPR, T, and S1-g were used to describe the same gene or genotype in the manuscript. This is not friendly to the readers. Reorganization of the names is recommended by referring to and observing genetic nomenclature for gene names and also allele designation. Please be very clear with regards to locus (consisting of three genes), genes and alleles, as well as their enumerating. Keep in mind that this may become a classical example in textbooks.
Response: In the revised version, we accepted the reviewer's suggestion and unified the different terms for the same gene or genotype in the main text of manuscript. OgTPR1, S1TPR, and TPR were unified as S1TPR; S1A4 and A4 were unified as S1A4; and S1A6 and A6 were unified as S1A6. Due to space limitations, the abbreviations TPR, A4, A6, s, and g, corresponding to S1TPR, S1A4, S1A6, S1-s, and S1-g, respectively, were still used in the figures, with definitions provided in the legends. To distinguish between the transgenes and the endogenous genes, we added superscript "t" after the name of transgenes.
2. Line 68: "a lot of". So far the number is not a lot yet.

Response: We changed "a lot of" to "several" (Page 3, line 77).
3. Line 76-77: "S1 is therefore the most important…". This claim seems to compare apple with pear.

Response: We have deleted this description to tone down the claim in the revised manuscript (Page 3, line 87).
4. Line 211: Allele 0 in genetics nomenclature in early days usually means a null allele.
Response: We have changed the "allele 0" to "Allele 1" and amended all alleles with the correct numbers in the revised manuscript.
5. Line 229: the result suggested that allele 1-4 was fixed in O. sativa. There seems to be an error here. Is it allele 1-3?
Response: The reviewer is right. We apologize for our carelessness and have adjusted all alleles to reflect the amended order.
6. The writing does not read well. Need further editing and polishing.

Reviewer #3
This reviewer gave a positive overall comment and raised 2 minor points This manuscript reported genes responsible for hybrid sterility between Oryza sativa and Oryza glaberrima. The authors found three closely linked genes, A4, TPR, and A6 in the S1 locus of O. glaberrima. The TPR encodes a protein containing two trypsin-like peptidase domains and a ribosome biogenesis regulatory domain, and it has been already published by the same group (Molecular Plant 2017). A6 encodes a peptidase protein with similar feature to the TPR peptidase, and it has been already reported by the same group and another group. A4 encodes an unknown protein and shown to interact with A6 and TPR using BiFC assay in this manuscript.
This manuscript is significant in the point reporting dual function of TPR: a complex of TPR, A4 and A6 makes up the killer system in sporophyte, and TPR alone has another role in the protection of TPR-containing gametes.
The hypothesis of the A4-TPR-A6 gamete killer-protector system is very complicated, and hard to understand for me at a glance, but careful reading of this manuscript and the previous paper published in Molecular Plant 2017 allowed me to understand that the working model in Fig. 5 is supported by the sufficient data employing transgenic japonica rice with TPR and CRISPR/Cas9 knockout mutants of NIL.
I have some comments for minor revisions.
1. In Lines 78 and 79, some more descriptions are needed for OgTPR1 and SSP, including full name of the genes and what they encode. Most readers will not understand what they are until they read the previous papers.

Response: We provided this detailed information in the revised manuscript (Page 3, lines 88-89).
2. The names of some alleles in the text look wrong, based on the data of Supplementary Fig. 5. In line 229, "Alleles 1-4 and 1-8" should be Alleles 1-3 and 1-8; in line 239, "Alleles1-10 to 1-12" must be alleles 1-9 to 1-11; in line 240, "Alleles alleles 1-14 and 1-15. Line 75 of Figure legend of Supplementary Fig. 6 should be also corrected to "Alleles 1-3 and 1-8" from the current description of "Alleles 1-4 and 1-8" Response: The reviewer is right. We apologize for our carelessness and thank the reviewer for their observation. In current revision, we have amended all alleles with the correct numbers.

Reviewer #4 This reviewer gave a positive overall comment and raised 14 suggestions involved in 3 specific aspects
This study dissects the genetic architecture of a major postzygotic reproductive barrier locus between two rice species, Oryza sativa and O. glaberrima. The motivation for this study is two-fold. First, reproductive barriers between rice taxa represent a major obstacle to breeding efforts. Second, detailed studies of reproductive barriers can shed new light on the evolution of species and are therefore of broad evolutionary interest. In this study, the authors first use a combination of sophisticated experiments to characterize the S1 locus, a cluster of three genes that interact to confer reproductive isolation. Then, the authors characterize genetic variation at the S1 locus in the study species, as well as related species with the aim to reconstruct the evolutionary history of the locus. The strength of this study is the detailed analysis of the molecular mechanism underlying HS mediated by the S1 locus, which provides interesting new insights into the genetic basis of HS in rice.
Unfortunately, the study also suffers from major weaknesses. These include the clarity of the presentation, the conceptual basis motivating the different components of this study, and the evolutionary inferences.
1. Clarity of the presentation 1.1 This study is targeting a general audience and therefore must be understandable also to people who are not deeply into the field of HS in rice and into the details of analysing molecular gene functions and gene interactions. In the present version, the manuscript is very difficult to follow.

Response:
We have carefully revised our manuscript according to the specific suggestions raised by all reviewers, making it more understandable for general audiences. We also asked a professional editor to polish the current version.
1.2 For the analysis of the molecular mechanisms, information is necessary on why the chosen transgenic lines were produced and what expectations are for the different crosses performed with these lines. In part, this information is provided, but clearly not for all relevant lines / crosses. Please clarify which lines are hemi-, homo-or heterozygous and what the relevance of this is for your expectations and findings.
Again, such information is partly provided but is missing in many instances. NIL-g and S1A4-S1A6 t × S1TPR t . Our results were similar to those of previous studies, indicating that the gametes carrying S1-s or lacking S1TPR t are selectively aborted during mega-gametogenesis (Supplementary Fig. 6).

Response: We have supplied this detailed information in
1.4 The term 'gametophytic manner' and 'sporophytic manner' appears to be colloquial language. Please use established terminology and provide an explanation upon first usage of these terms such that non-specialist readers can follow. 1.6 The extensive naming and renaming of genes (e.g. page 4) is confusing. Please use consistent naming for the genes throughout the manuscript and for example indicate in brackets alternative names for the same genes that have been used in other studies.
Response: We have accepted the reviewer's suggestion and unified the names of the genes. OgTPR1, S1TPR, and TPR were unified as S1TPR; S1A4 and A4 were unified as S1A4; and S1A6 and A6 were unified as S1A6 in the revised manuscript. Due to space limitations, abbreviations for the genes and alleles are still used in the figures, with definitions provided in the legends.
1.7 Some of the terms used to describe evolutionary processes and scenarios are not adequate and need to be revised. A case in point is the first sentence of the introduction: HS does not 'suppress gene migration' but (for example) 'reduces gene-flow' between populations or species. Evolution in general is not targeted or directional and therefore also HS does not have a purpose, i.e. 'to drive speciation and contribute to ...'. Use of the term 'gene flows' in this manuscript is wrong and the term 'evolutionary flow' (page 8) does not exist. These parts need to be reformulated and appropriate terminology is essential. Otherwise, readers will not be able to follow the complex scenarios inferred.
Response: We thank the reviewer for sharing their expertise in evolutionary biology and corrected these terms according to the reviewer's suggestion. We rewrote this section of the text using the appropriate terms (Page 2, lines 51-52; 2. Conceptual basis 2.1 The paper deals with the analysis of a killer-protector system. Please explain in the introduction what this is and / or refer to suitable recent review papers that provide adequate description. Response: As suggested, we added a brief description of the killer-protector system and cited a recent review paper (Bravo et al., 2018, Trends Genet., 34: 3, lines 60-67).  [1]). The locus investigated in the present study therefore does not reflect this standard definition because it consists of a single locus (S1) that is composed of three apparently highly linked genes that interact. This one-locus model also fits the Bateson-Dobzhansky-Muller model [2] but is not the standard model and therefore requires introduction.

424-433) in the Introduction section (Page
It remains obscure how the Bateson-Dobzhansky-Muller incompatibility relates to meiotic drive and selfish genetic elements. Please clarify these aspects and / or drop reference to concepts that are not essential for the present study.
Response: We thank the reviewer for this comment. We have now cited these references and reorganized this part of the introduction.
Meiotic drivers, also called 'ultra-selfish' genes, are preferentially transmitted into more than half (sometimes all) of the meiotic products and progeny of heterozygotes by destroying the other viable meiotic products lacking the driver, leading to segregation distortion (Bravo et al., 2018, Trends Genet. 34:   424-433). Similarly, the HS loci act as selfish genetic elements that eliminate competing alleles to gain a transmission advantage. Both result in Bateson-Dobzhansky-Muller incompatibility. We clarified these aspects in the Introduction (Page 2-3, lines 52-60).
2.3 Use of the term 'speciation locus' is not warranted in the present study. Rarely, if ever, is speciation a result of very few or a single locus, and the role of the S1 locus in speciation has not been studied. S1 apparently plays an important role in current reproductive isolation between the two studied species but this tells us nothing about its potential role in the speciation process.
Response: We changed 'speciation locus' to 'HS locus' in the current version.

2.4
The motivation for trying to reconstruct the evolutionary history of the S1 locus is unclear. I presume that the authors tried to establish the point that the alleles evolved independently in two lineages, where they became fixed and cause HS when brought into contact. These points could be considered essential components of 3.2 Unfortunately, foxtail millet appears far too diverged to be of much use in the present study. Data from more closely related taxa should be available and included in the analysis and discussion.
Based on the presented data on variation at the S1 locus within and among species, it appears that there is quite substantial variation within species and the common ancestor of O. sativa, glaberrima and meridionalis was most likely polymorphic at S1. Can the variation observed by a consequence of incomplete lineage sorting? The present discussion entirely neglects the fact that Bateson-Dobzhansky-Muller incompatibilities can be polymorphic within species [1].
This aspect needs to be incorporated and discussed.
Response: We used the nucleotide sequences of the three components of the S1 locus as templates to perform a BLAST search of the Poaceae. We found that the putative ortholog of S1TPR in Setaria italica is similar to S1TPR in African rice ( Supplementary Fig. 11). In response to the reviewer's suggestion, we further  Table 8). We found that the S1A4 and S1A6 orthologs are not present in these species, although all of them had putative S1TPR orthologs containing C at SNP site 7, corresponding to the premature stop codon in S1TP. These results suggest that S1A4 and S1A6 probably arose from newly evolved genes in the AA-genome Oryza species, and that variants of this site 7 SNP in S1TPR and S1TP co-existed in the primitive Oryza gene pool ( Supplementary Figs. 12 and 13 and Supplementary Table 8). We analyzed and described these new data in our revised manuscript (Page 10, lines 271-295).
We agree with the reviewer that the extensive variations in the S1 locus In our previous study, 11,000 individuals were used to finely map S1 to a 29-kb region (at S1-g), which contains these three S1-related genes. Among these individuals, only two recombinants with recombination sites within this 29-kb region were detected in the S1 mapping region (Xie et al., 2017, Mol. Plant 10: 1137-1140). The sequence structures between S1-g and S1-s are so different (see Figure 1a) that the meiotic pairing and recombination between the S1-g and S1-s intervals should be largely suppressed, making this region behave as a single complex locus.

Reviewer #1 (Remarks to the Author):
The manuscript NCOMMS-18-34859A dealt with "An asymmetric allelic interaction drives allele transmission bias in interspecific rice hybrids". In this revision, the authors present a significantly improved version of their manuscript. In my opinion, they have correctly addressed most of the points raised in my original review. They investigated the expression of the three genes in the anthers and panicles at various developmental stages，conducted a transcriptome sequencing analysis， and found that six genes involved in the branched-chain amino acid degradation pathway might have a detrimental effect on gamete development. Based on these results, the authors showed that three closely linked genes (S1A4, S1TPR and S1A6) in the African S1 allele constitute a killer and protector system that eliminates gametes carrying the Asian allele, but S1TPR can rescue S1-g gametes. S1TPR is a killer and also is a protector that is distinguished from the reported kill-protector system. Better to add this in Discussion.
Reviewer #2 (Remarks to the Author): The revised manuscript has well addressed the questions of reviewer 2 and is suitable for acceptance. One additional minor point: line 77-80, a recently cloned locus ESA1 for interspecific hybrid incompatibility might be added to the introduction.
The major concerns of reviewer 4 involved three aspects of the previous manuscript including presentation, explanation of the models, and evolutionary analysis. Because of the complexity of interactions in S1 (which involved a number of transgenes and hybridizations in functional characterization), reviewer 4 asked for a clearer elaboration of the genetic materials and genetic design of the experiments (as well as the expectations, phenotypes and several concepts). Now the revised version is much easier to read and understand. Reviewer 4 also emphasized the interpretation of genetic models (BDM model and killer-protector model). These models are well explained in the background and touched in the results in the revised manuscript. Finally, the evolutionary analyses have been improved in the revised version. More convinced evidences are given for inference of the origin and divergence of S1 locus. Taken together, the comments have been appropriately addressed.

Reviewer #3 (Remarks to the Author):
This is a revised manuscript that I have reviewed previously. The authors revised it incorporating my comments. According to the comments and suggestions by other reviewers, they have also included new data such as expression analysis of the candidate genes in anthers and panicles at different developmental stage, transcriptome analysis of anthers, pull-down assay demonstrating the interaction among the candidate proteins, and discussion of mechanism. This manuscript has been revised satisfactorily.

REVIEWERS' COMMENTS:
Reviewer #1 (Remarks to the Author): The manuscript NCOMMS-18-34859A dealt with "An asymmetric allelic interaction drives allele transmission bias in interspecific rice hybrids". In this revision, the authors present a significantly improved version of their manuscript. In my opinion, they have correctly addressed most of the points raised in my original review. They investigated the expression of the three genes in the anthers and panicles at various developmental stages，conducted a transcriptome sequencing analysis， and found that six genes involved in the branched-chain amino acid degradation pathway might have a detrimental effect on gamete development. Based on these results, the authors showed that three closely linked genes (S1A4, S1TPR and S1A6) in the African S1 allele constitute a killer and protector system that eliminates gametes carrying the Asian allele, but S1TPR can rescue S1-g gametes. S1TPR is a killer and also is a protector that is distinguished from the reported kill-protector system. Better to add this in Discussion.
Response: We thank the reviewer for positive comments and have added the dual function of S1TPR in Discussion (refer to Page 13, Lines 374-377).
Reviewer #2 (Remarks to the Author): The revised manuscript has well addressed the questions of reviewer 2 and is suitable for acceptance. One additional minor point: line 77-80, a recently cloned locus ESA1 for interspecific hybrid incompatibility might be added to the introduction.
The major concerns of reviewer 4 involved three aspects of the previous manuscript including presentation, explanation of the models, and evolutionary analysis.
Because of the complexity of interactions in S1 (which involved a number of transgenes and hybridizations in functional characterization), reviewer 4 asked for a clearer elaboration of the genetic materials and genetic design of the experiments (as well as the expectations, phenotypes and several concepts). Now the revised version is much easier to read and understand.
Reviewer 4 also emphasized the interpretation of genetic models (BDM model and killer-protector model). These models are well explained in the background and touched in the results in the revised manuscript.
Finally, the evolutionary analyses have been improved in the revised version. More convinced evidences are given for inference of the origin and divergence of S1 locus.
Taken together, the comments have been appropriately addressed.
Response: We thank the reviewer for positive comments. As requested, we cited a recent reference and added this information in the Introduction section (refer to Page3, Lines 76-77).

Reviewer #3 (Remarks to the Author):
This is a revised manuscript that I have reviewed previously.
The authors revised it incorporating my comments. According to the comments and suggestions by other reviewers, they have also included new data such as expression analysis of the candidate genes in anthers and panicles at different developmental stage, transcriptome analysis of anthers, pull-down assay demonstrating the interaction among the candidate proteins, and discussion of mechanism. This manuscript has been revised satisfactorily.

Response: We thank the reviewer for positive comments.
Reviewer #4: unavailable.
Response: We thank the reviewer 2 for evaluating the comments raised by reviewer #4.