GM-CSF regulates conserved ILC-like transcriptional states and myeloid cell signaling during ulceration in Crohn’s disease

Reviewer #1 (Public review):

The revisions do not resolve my central concerns. Although the authors have added macrophage transcriptomic analyses and moderated the terminology to "ILC1-like" and "ILC3-like," the fundamental limitations remain.

First of all, the data still do not establish that the zebrafish populations represent bona fide ILC1s . The proposed ILC1 identity remains particularly weak, as HSP90AA1, UBB, MCL1 or DOK2 are not ILC1-specific markers, while TBX21 and IFN programs are insufficient to establish lineage identity. Changing the terminology to "ILC-like" acknowledges this limitation but does not resolve it. Consequently, the claims regarding ILC1 differentiation, ILC3 maintenance, and GM-CSF-mediated regulation of ILC fate remain stronger than the data support, since RNA velocity does not by itself demonstrate lineage conversion.

The additional macrophage analyses are interesting but remain correlative. They support a model in which GM-CSF may act through macrophages to create an ILC-supportive environment, but they do not demonstrate that the identified macrophage programs are causally responsible for the observed changes. In particular, the proposed retinoic-acid/HIF mechanism remains speculative.

I also do not think the response adequately addresses the central novelty concern. GM-CSF-mediated protection in DSS intestinal injury is already well established, and the human Xenium data remain largely descriptive and not fully explored. The observation of sparse GM-CSF expression and GM-CSF/pSTAT5 localization does not directly establish the proposed GM-CSF-producing cell-macrophage-ILC circuit. The suggested spatial niche analysis was not actually performed.

Thus, while the revised manuscript presents a more plausible model, it does not establish the claimed mechanism. Unless additional causal evidence is provided, I would recommend substantially reducing the mechanistic claims and presenting the macrophage-ILC pathway as a hypothesis rather than an established mechanism.

Reviewer #2 (Public review):

The authors document that GM-CSF prevents the loss of ILC3 populations and inhibits pro-inflammatory cytokine production during gut inflammation. They combine a preclinical model of gut inflammation in zebrafish with spatial transcriptomic analysis of samples from Crohn's disease patients. The data show that GM-CSF ameliorates gut inflammation by (1) curtailing the differentiation of disease-associated ILC1 and (2) by "boosting" the tissue repair function of ILC3.

The topic of the manuscript is interesting and highly relevant to the question of how intestinal inflammation is initiated or perpetuated. However, there are some limitations. Specifically, it remains unclear through which distinct molecular programs macrophages maintain "beneficial" ILC3 and how they prevent the differentiation or expansion of disease-associated ILC1. Based on previous data, these processes may be interconnected as ILC3 show remarkable plasticity and can convert into ILC1-like cells ("ex-ILC3").

Author response:

The following is the authors’ response to the original reviews.

Public Reviews:

Reviewer #1 (Public review):

This study integrates Xenium spatial transcriptomics of paired inflamed and uninvolved Crohn's disease tissues with functional analyses in a csf2rb-/- larval zebrafish DSS intestinal injury model to investigate the spatial and cell-type-specific roles of GM-CSF. The work is limited mechanistically and adds little to an already disputed field: GM-CSF's role in intestinal inflammation is context-dependent and extensively studied in mice and humans, and this study does not resolve these controversies. The zebrafish appears to be a poor model for these questions: it lacks mammalian intestinal architecture, complex microbiota, and clearly validated functional ILC populations. Putative ILC1s are poorly defined based on stress-response gene modules, while ILC3s are somewhat better characterized, but overall, the system does not allow mechanistic insights into GM-CSF regulation of ILCs. The DSS experiments largely recapitulate the known protective effects of GM-CSF in epithelial injury without clarifying underlying mechanisms.

Figure 1

GM-CSF expression is extremely sparse, rarely exceeding 0.005 frequency even in inflamed regions. The authors should acknowledge this and discuss why. Xenium could be used to characterize the niche around GM-CSF-producing cells, but no new cellular circuit is revealed beyond known myeloid-lymphoid interactions.

We thank the reviewer for this important observation. We agree that GM-CSF-expressing cells represent a very small fraction of all cells within the tissue, even in inflamed regions. We have now explicitly acknowledged this in both the Results (Line 300-304) and Discussion (Line 539-547). The low abundance of GM-CSF-producing cells is consistent with previous studies demonstrating that GM-CSF is not constitutively expressed in the intestinal mucosa but is induced only in activated immune cells during inflammation. Moreover, GM-CSF functions primarily as a locally acting cytokine with paracrine activity rather than as a broadly expressed tissue factor. Consequently, even a small number of GM-CSF-producing lymphocytes can establish localized signaling niches that activate neighboring CSF2RA/CSF2RB-expressing myeloid cells. This localized mode of signaling is also consistent with the low circulating concentrations of GM-CSF reported in both healthy individuals and patients with inflammatory diseases. We have revised the manuscript to better emphasize that the biological significance of GM-CSF lies in its spatially restricted expression and signaling rather than its overall abundance.

Figure 2

Colon length in DSS colitis is not decreased in Csf2rb-/- versus wild-type zebrafish under untreated conditions, suggesting endogenous GM-CSF has minimal impact. In Figure 2E, Tg(mpeg1:mCherry) larvae show staining in vessel- or epithelial-like structures expressing Csf2rb, which does not resemble macrophages and requires clarification. pSTAT5 is upregulated with GM-CSF treatment, but the responding cell types are unclear.

We thank the reviewer for this observation. We agree that intestinal length is comparable between wild-type and csf2rb^-/- larvae under homeostatic conditions. This finding is consistent with the established biology of GM-CSF, which is largely dispensable during steady-state tissue homeostasis but becomes functionally important during inflammatory challenge. Accordingly, the protective effect of GM-CSF became apparent only following DSS-induced intestinal injury, where recombinant human GM-CSF significantly rescued epithelial damage and gut shortening in wild-type but not csf2rb^-/- larvae. We have clarified this point in the Discussion (Line 563-571).

As for 2E comments, we apologize that the previous version of Figure 2 inadvertently omitted the panel labels and annotations for Figure 2E, making it difficult to interpret the image. We have corrected this oversight by restoring the panel label, adding arrows and higher-magnification insets that highlight representative csf2rb+ mpeg1:mCherry+ macrophages in the retina, intestine, and midbody. The purpose of this experiment was to demonstrate that intestinal macrophages express csf2rb, rather than to suggest that csf2rb expression is restricted to macrophages. We agree that additional csf2rb-positive elongated structures are present that do not colocalize with mpeg1:mCherry. This observation is consistent with our human Xenium data, which show that CSF2RB is broadly expressed across multiple cell types, whereas signaling specificity is determined by the more restricted expression of CSF2RA. We have revised both the Results (Line 358-365) and the figure legend (Line 694-699) to clarify this point and avoid implying macrophage-exclusive expression.

Figure 3

Putative ILC1s are defined by stress-response gene modules rather than canonical markers. Overlapping genes with human (HSP90AA1, UBB, MCL1, DOK2) do not indicate ILC1 identity, which is described by IL7R, KLRB1, or TBX21 expression in the human Xenium dataset. ILC2s were not detected, and Ifng expression is broadly distributed, making attribution to ILC1s uncertain. ILC3s are somewhat better defined, but overall, the data do not support mechanistic conclusions about GM-CSF regulation of ILC populations.

We thank the reviewer for this thoughtful comment and agree that annotation of innate lymphoid cell populations in zebrafish remains more challenging than in mammals due to the limited availability of validated lineage-specific markers. We have therefore revised the manuscript to more clearly describe these populations as putative ILC1- and ILC3-like cells identified using the currently available zebrafish ILC markers together with conserved transcriptional programs, rather than implying complete equivalence to mammalian ILC subsets.

Our annotation was not based solely on stress-response genes or IFNG expression. Instead, we first identified the lymphocyte population using established zebrafish ILC markers (il7r, nitr4a, nitr5, nitr9, and ccl20a.3) reported by Hernández et al. (2018), followed by subtype annotation using multiple complementary features. The putative ILC3 cluster expressed canonical regulators of ILC3 identity, including rorc, il23r, tcf7, and mafb, while the putative ILC1 cluster demonstrated enrichment of tbx21, an IFN-associated transcriptional program, and a conserved stress-response module that has likewise been reported in human intestinal ILC1 datasets. IFNG expression was therefore interpreted together with the overall transcriptional program rather than used as a defining marker.

We also agree that the zebrafish model does not currently permit identification of all mammalian ILC subsets. Accordingly, we have revised the Discussion to explicitly acknowledge this limitation and have moderated our conclusions throughout the manuscript (Line 381-388 and Line 588-593). Rather than concluding that GM-CSF directly regulates definitive mammalian ILC subsets, we conclude that GM-CSF regulates conserved ILC-like transcriptional states during intestinal inflammation (Title). Importantly, this interpretation is further supported by independent RNAscope validation of tbx21, rorc, ifng1.1, and il22, together with the concordant expansion of IL7R+KLRB1+TBX21+ cells and contraction of IL7R+KLRB1+RORC+ cells observed in the human Xenium dataset.

Reviewer #2 (Public review):

The authors show that GM-CSF prevents the loss of ILC3 populations and inhibits pro-inflammatory cytokine production during gut inflammation. They combine a preclinical model of gut inflammation in zebrafish with spatial transcriptomic analysis of samples from Crohn's disease patients. The data show that GM-CSF ameliorates gut inflammation by (1) curtailing the differentiation of disease-associated ILC1 and (2) by "boosting" the tissue repair function of ILC3.

The topic of the manuscript is interesting. However, there are various limitations that are summarized below.

We thank the reviewer for these thoughtful comments. We agree that, because GM-CSF receptors are predominantly expressed by myeloid cells rather than ILCs, understanding the myeloid response to GM-CSF is important for interpreting the mechanism by which GM-CSF preserves ILC3 function. In response to the reviewer's suggestions, we performed additional analyses of both our human scRNA dataset (Martin et. al., 2019, PMID: 31474370) and zebrafish macrophage transcriptomes. Together, these new analyses support a model in which GM-CSF acts primarily on macrophages to establish a tissue-repair and ILC3-supportive microenvironment, thereby indirectly maintaining ILC3 identity during intestinal inflammation.

(1) The main finding of the manuscript, that GM-CSF maintains ILC3 populations, is not analyzed in depth. Since the authors' own data and other publications show that the receptors for GM-CSF are expressed in myeloid cells, a better analysis of the transcriptional changes of these populations upon GM-CSF administration is needed.

To better characterize the cellular targets of GM-CSF, we performed differential expression analysis of macrophages and monocytes. In our previous human scRNA-seq dataset (Martin et al.), STAT5-high macrophages were enriched for canonical GM-CSF-responsive genes, including STAT5A, STAT5B, JUNB, and HIF1A, confirming activation of the GM-CSF/STAT5 signaling axis. These macrophages additionally expressed regulatory and tissue-remodeling genes including LGALS1 (Galectin-1), TGFBI, and GPNMB, consistent with an immunoregulatory macrophage phenotype. Notably, LGALS1 has been reported to suppress type 1 inflammatory responses and modulate innate lymphoid cell function, whereas TGFBI is associated with tissue remodeling and immune regulation.

To further investigate the downstream effects of GM-CSF signaling, we analyzed macrophage transcriptomes from zebrafish treated with DSS plus GM-CSF. We observed enrichment of genes involved in retinoic acid and tissue-homeostatic signaling, including rarab, rdh12l, and vhl. Retinoic acid signaling is well established in promoting intestinal immune homeostasis and supporting ILC3 differentiation and maintenance, while HIF signaling has been implicated in intestinal barrier integrity and ILC3 survival. Additional changes in YBX1 and SLC7A7 further suggest modulation of cytokine translation and macrophage immunometabolism that could influence the intestinal immune niche. Consistent with this interpretation, expression of rarab and vhl was reduced in csf2rb-/- macrophages, indicating that activation of this tissue-homeostatic program depends on GM-CSF receptor signaling. Together, these findings support a model in which GM-CSF primarily acts on macrophages to induce a regulatory, tissue-supportive program that creates an environment favorable for maintenance of protective ILC3 responses while limiting excessive type 1 inflammation.

(2) The authors could compare the transcriptome of macrophages and monocytes from inflamed and uninvolved sections in their Xenium dataset. In addition, investigating how zebrafish macrophages change due to the lack of GM-CSF and comparing them with the human findings would add to the data.

We thank the reviewer for this valuable suggestion. While comparison of macrophages and monocytes between inflamed and uninvolved regions using the Xenium dataset is conceptually appealing, the Xenium platform is designed for spatially resolved targeted transcriptomics and is limited to a predefined panel of 468 genes. In addition, transcript detection from FFPE tissue is inherently sparser than whole-transcriptome single-cell RNA sequencing due to probe-based hybridization and RNA fragmentation. Consequently, Xenium provides excellent spatial resolution but is less suited for comprehensive transcriptional profiling of macrophage and monocyte states.

To more comprehensively characterize myeloid transcriptional programs, we therefore analyzed our previously generated human scRNA-seq dataset from inflamed Crohn's disease intestine, which captures the full transcriptome at single-cell resolution. This analysis identified a STAT5-high macrophage population enriched for genes associated with GM-CSF signaling (STAT5A, STAT5B, JUNB), immune regulation (LGALS1, TGFBI, GPNMB), and tissue remodeling, providing a more complete characterization of macrophage responses than could be achieved with the targeted Xenium panel.

In addition, as suggested by the reviewer, we investigated the consequences of impaired GM-CSF signaling in zebrafish by comparing macrophages from WT and csf2rb-/- animals following GM-CSF treatment. These analyses demonstrated that loss of csf2rb attenuated the induction of retinoic acid- and tissue-homeostatic programs, including rarab, rdh12l, and vhl, supporting an evolutionarily conserved role for GM-CSF signaling in establishing a macrophage program associated with intestinal homeostasis. Together, the human scRNA-seq and zebrafish datasets provide complementary evidence that GM-CSF primarily acts through macrophages to promote a regulatory, tissue-supportive phenotype. All these results have been summarized in Results (line 503-513) and Supplementary Figure 7.

(3) Since the authors developed a novel mutation in zebrafish that is predicted to affect myeloid populations, a detailed characterization of the myeloid immune compartment in these organisms is missing.

We thank the reviewer for this important suggestion. We agree that characterization of the myeloid compartment is essential for understanding the consequences of disrupting GM-CSF signaling. To address this, we performed additional transcriptional analyses of macrophages isolated from WT and csf2rb-/- zebrafish following GM-CSF treatment. These analyses revealed that loss of csf2rb markedly attenuated the GM-CSF-dependent macrophage program, including genes involved in retinoic acid signaling (rarab, rdh12l), HIF-associated tissue homeostasis (vhl), cytokine regulation (ybx1), and macrophage immunometabolism (slc7a7). In addition, genes associated with cytoskeletal remodeling, migration, and metabolic adaptation were also reduced in csf2rb-/- macrophages, indicating that GM-CSF receptor signaling is required for full macrophage activation.

These findings demonstrate that the csf2rb mutation does not simply alter individual downstream genes but broadly impairs the establishment of a macrophage program associated with intestinal tissue homeostasis. Together with our human scRNA-seq analyses, these data support a conserved role for GM-CSF signaling in programming regulatory macrophage states that create an environment favorable for maintenance of protective ILC3 responses. We have incorporated these additional analyses into the revised manuscript to provide a more comprehensive characterization of the myeloid compartment.

(4) Niche analysis in the Xenium slides could provide direct evidence on how macrophages close to ILC3 are different from those closer to other cell types, like ILC1.

We thank the reviewer for this insightful suggestion. We agree that spatial niche analysis comparing macrophages associated with ILC3 versus ILC1 would provide valuable mechanistic insight into local immune cell interactions. However, this analysis is technically challenging in our Xenium dataset for several reasons. First, the Xenium platform uses a targeted panel of 468 genes and therefore provides substantially lower transcriptomic depth than whole-transcriptome scRNA-seq. Second, transcript detection in FFPE tissue is inherently sparse due to probe-based hybridization and RNA fragmentation. Finally, ILC populations are relatively rare within the intestinal tissue, limiting the number of macrophages located in close proximity to ILC3 or ILC1 and reducing the statistical power required for robust niche-specific differential expression analyses.

To instead address the reviewer's mechanistic question, we integrated complementary datasets that are better suited for transcriptional profiling. Specifically, our human scRNA-seq analysis identified a STAT5-high macrophage population enriched for regulatory and tissue-remodeling genes, including LGALS1, TGFBI, and GPNMB, consistent with a macrophage phenotype capable of supporting intestinal homeostasis. Furthermore, zebrafish macrophage transcriptomic analyses demonstrated that GM-CSF signaling induces a conserved tissue-homeostatic program characterized by rarab, rdh12l, and vhl, implicating retinoic acid and HIF-associated pathways that have been linked to ILC3 maintenance and mucosal immunity. Together, these complementary datasets provide molecular evidence that GM-CSF programs macrophages toward an ILC3-supportive phenotype, although we acknowledge that direct spatial comparison of macrophages neighboring ILC3 versus ILC1 will require future studies using higher-complexity spatial transcriptomic approaches.

Recommendations for the authors:

Reviewer #2 (Recommendations for the authors):

(1) Overall/Statistics: The statistical analysis described in the method section (Mann-Whitney-Wilcoxon Test) is not suitable for comparison of more than 2 populations.

We thank the reviewer for identifying this oversight. We agree that the original description of the statistical analyses was incomplete, as the Mann–Whitney U (Wilcoxon rank-sum) test is appropriate only for comparisons between two groups. We have revised the Quantification and Statistical Analysis section to accurately describe the statistical methods used throughout the manuscript. Specifically, two-group comparisons were analyzed using the Mann–Whitney U (Wilcoxon rank-sum) test, whereas experiments involving more than two groups were analyzed using the Kruskal–Wallis test followed by Dunn's multiple-comparisons test. We have also clarified that differential gene expression analyses were performed using the Wilcoxon rank-sum test implemented in Seurat, and ligand–receptor analyses were performed using permutation testing with Benjamini–Hochberg false discovery rate correction. These revisions ensure that the statistical methods accurately reflect the analyses performed throughout the study.

(2) Figure 1

A. A zoomed-in image of the regions expressing GM-CSF and G-CSF could help to visualize the changes between uninvolved and inflamed sections.

We thank the reviewer for this helpful suggestion. To better visualize the spatial distribution of the CSF cytokines, we have added representative high-magnification Xenium images of paired uninvolved and inflamed ileal tissue to Supplementary Figure 2C. These enlarged views more clearly demonstrate the focal enrichment of GM-CSF (CSF2) within inflamed tissue, whereas CSF1 remains broadly distributed and CSF3 exhibits comparatively sparse expression with minimal changes between uninvolved and inflamed regions. We have also added a reference to this new supplementary panel in the Results section (line 317-320).

F. Since the expression of the CSF receptors is detectable, it would be interesting to see an analysis of their spatial location.

We thank the reviewer for this valuable suggestion. We agree that spatial visualization of the CSF receptors provides important context for interpreting ligand–receptor interactions. Accordingly, we have expanded Supplementary Figure 2C to include representative Xenium images showing the spatial localization of CSF1R, CSF2RA, and CSF2RB in paired uninvolved and inflamed ileal tissue sections. These images complement our human scRNA-seq analysis by demonstrating the tissue distribution of the receptors within the intestinal microenvironment. Together with the ligand–receptor interaction analysis, these data further support macrophages and dendritic cells as the principal cellular targets of GM-CSF signaling during intestinal inflammation.

(3) Figure 2

A. The experimental setup depicted in the figure does not match the description given in the text (323-324) or the methods (188-191). Importantly, the recovery phase is not detailed in the figure scheme, which complicates result interpretation.

We thank the reviewer for identifying this inconsistency. We have revised Figure 2A to accurately reflect the experimental design described in both the Methods and Results sections. Specifically, the schematic now clearly distinguishes the two experimental paradigms: (i) co-treatment with DSS and GM-CSF during active intestinal injury, and (ii) the recovery protocol in which DSS was removed prior to GM-CSF treatment. We have also expanded the recovery phase in the schematic to indicate the timing of DSS removal, GM-CSF administration, and sample collection. These revisions provide a clearer representation of the experimental workflow and facilitate interpretation of the corresponding results.

D. Statistical significance should be revised using ANOVA or a non-parametric test that can compare more than 2 groups.

We thank the reviewer for this important comment. We agree that comparisons involving more than two groups require an appropriate omnibus statistical test. Accordingly, we have reanalyzed the data using the Kruskal–Wallis test, followed by Dunn's multiple-comparisons test for post hoc pairwise comparisons. The revised statistical analyses have been incorporated into Figure 2D and throughout the manuscript where applicable. In addition, the Quantification and Statistical Analysis section (line 274-284) has been updated to accurately describe the statistical methods used. These revisions do not alter the interpretation or conclusions of the study.

(4) Figure 3

C. It is unclear which data were used to generate the dot plot. The data source should be written in the text or the figure panel.

We thank the reviewer for pointing out this lack of clarity. We have revised the Results section (line 382-385) and Figure 3 legend (line 706) to explicitly state that the dot plot in Figure 3C was generated from the scRNA-seq dataset of FACS-isolated intestinal lck: GFP+ lymphocytes from larval zebrafish. The dot plot shows the expression of established zebrafish ILC markers and conserved ILC-associated genes across the identified lymphocyte clusters and was used, together with additional transcriptional features, to support annotation of the proliferative ILC, putative ILC1-like, and putative ILC3-like populations.

D. While the heatmap is informative to identify signature genes, it fails to highlight the differentially expressed genes between experimental conditions. The authors should present the DEG for each of the ILC clusters. Doing a pathway enrichment analysis could highlight how ILCs are responding to DSS and GM-CSF treatments. In addition, it would be important to test experimentally how the immune compartment (ILC and myeloid populations) is affected by CSF2RB deficiency.

We thank the reviewer for these valuable suggestions. We agree that differential expression analyses provide additional insight into how GM-CSF modulates immune cell states during intestinal inflammation. To address this, we have expanded our analyses in the revised manuscript.

First, we performed differential gene expression analyses of macrophages and incorporated these data into the revised manuscript. Comparative analyses of human STAT5-high macrophages and zebrafish GM-CSF-responsive macrophages identified conserved transcriptional programs associated with regulatory macrophage function, tissue remodeling, and establishment of an ILC3-supportive niche. Furthermore, comparison of WT and csf2rb-/- zebrafish macrophages demonstrated that disruption of GM-CSF receptor signaling attenuates these macrophage programs, providing a molecular characterization of the myeloid compartment in the mutant animals.

Second, pathway enrichment analyses of ILC1 and ILC3 marker genes were already included in the original manuscript (Figure 3F) using Ingenuity Pathway Analysis (IPA), demonstrating conserved transcriptional regulators and signaling pathways between zebrafish and human ILC populations. Rather than performing pathway analysis across treatment conditions, our study focused on identifying conserved lineage-defining programs and experimentally validating the functional consequences of GM-CSF signaling through changes in ILC abundance, lineage marker expression, RNA velocity, and effector cytokine production.

Finally, to directly address the reviewer's concern regarding the impact of csf2rb deficiency on the immune compartment, we have added additional analyses characterizing macrophage transcriptional responses in WT and csf2rb-/- zebrafish following GM-CSF treatment. Together with our existing analyses of ILC population dynamics and lineage-specific marker expression in WT and csf2rb-/- animals, these data provide a more comprehensive characterization of how loss of GM-CSF signaling alters both the myeloid and ILC compartments during intestinal inflammation.

(5) Figure 4

D. The genes used to define the "ILC1 gene module" are not detailed. Importantly, maintaining the scale of the violin plots would allow a better interpretation of the results.

We thank the reviewer for pointing out this lack of detail. We have revised the Figure 4 legend (line 724-727) to explicitly describe the genes included in the putative ILC1-like gene module. This module was derived from differentially expressed genes enriched in the zebrafish putative ILC1-like cluster identified in our scRNA-seq dataset, rather than from IFNG expression alone. The module includes genes associated with the ILC1-like transcriptional state identified in our dataset (Figure 3E). As discussed above, because several of these genes represent a conserved stress-response program rather than canonical lineage-defining ILC1 markers, we have revised the terminology throughout the manuscript from “ILC1 gene module” to “putative ILC1-like gene module” and have moderated our interpretation accordingly.

F-H. The naming of RNA signal is not consistent: Il22 particles vs. Ifng1-1 foci.

We thank the reviewer for noting this inconsistency. We have revised the terminology throughout the manuscript and figure legends to use consistent nomenclature when referring to RNAscope signals. Specifically, all RNAscope transcript signals are now described uniformly as RNA foci.

(6) Figure 5

C. It is unclear if the color code and the y-axis are 2 different measurements or just 2 ways of showing the same. A similar calculation on the ILC3 and ILC1 subsets could give insights into their differentiation potential.

We thank the reviewer for pointing out this lack of clarity. The color coding and y-axis provide complementary representations of the RNA velocity analysis. The y-axis quantitatively represents the proportion of unspliced relative to spliced transcripts, whereas the color scale was included to facilitate visualization of the relative transcriptional/differentiation state of individual cells. We have revised the figure legend (line 739) to clearly distinguish these two representations and retained the color scale because it provides an intuitive visualization of changes in cellular transcriptional state.

D. The authors do not describe how the 'Relative ILC Ratio' was calculated.

We thank the reviewer for pointing out this omission. We have revised the Methods (line 148-153) and the corresponding figure legend to clarify how the relative ILC ratio was calculated. Briefly, we first calculated the densities (cells/μm²) of ILC1s (IL7R⁺TBX21⁺) and ILC3s (IL7R⁺RORC⁺) from the Xenium dataset. The relative ILC ratio was then determined by calculating the proportion of each ILC subset relative to the total IL7R⁺ ILC population identified in each sample following single-cell clustering and cell-type annotation. Finally, these proportions were normalized to the control condition to enable comparison across treatment groups.

(7) Figure 6

B. The colors used obscure CD14 expression. Changing the background intensity (as done later in 5G) could fix this problem.

We thank the reviewer for pointing out this potential source of confusion. In Figure 6B, the green cells represent epithelial cells and are therefore not part of the ulcer exudate/macrophage aggregates being highlighted in this analysis. The macrophage-rich ulcer exudate is identified by the accumulation of CD14-positive cells adjacent to the disrupted epithelial compartment. We have revised the figure legend (line 746-749) to more clearly define the color coding and distinguish the epithelial compartment from the CD14-positive macrophage aggregates.

C. Since the Xenium slides group were selected by inflammation status, is it expected to have any ulcer in an 'Uninvolved' tissue section?

We thank the reviewer for raising this important point. The classification of Xenium sections as inflamed or uninvolved was determined by pathological evaluation of the resection specimens, and the inflammation status of the analyzed regions was confirmed by a pathologist. Importantly, this classification is distinct from the morphological criteria used to identify ulceration. Histologically, an ulcer is defined primarily by disruption or loss of the epithelial barrier rather than by the presence of inflammatory immune-cell infiltration. Therefore, focal epithelial breaks can occasionally be identified even within tissue sections classified as uninvolved, without the prominent inflammatory infiltrates characteristic of ulcerated regions in inflamed tissue. This distinction explains the small number of epithelial disruptions identified in the uninvolved sections. We have revised the Results (line 473-480) and figure legend (line 750-751) to clarify the distinction between pathologist-defined inflammation status and the morphological identification of ulcerated/epithelial-disrupted regions.

E. The macrophage density measurement contradicts the previous panel (6E), which shows no macrophage infiltration in uninvolved ulcers. The authors should clarify how both analyses differ.

We thank the reviewer for pointing out this potential confusion. The two analyses quantify different aspects of macrophage distribution and therefore are not directly comparable. Figure 6D quantifies the number of pro-inflammatory macrophage aggregates associated with ulcer exudate, which are present only in inflamed tissue and absent from uninvolved sections because uninvolved tissues do not contain ulcers. In contrast, Figure 6E measures the overall density of CD14⁺ macrophages within the defined tissue compartments, independent of whether the macrophages form inflammatory aggregates. Thus, Figure 6D evaluates the presence of organized inflammatory macrophage clusters, whereas Figure 6E quantifies the abundance of macrophages throughout the analyzed tissue regions. We have revised the figure legend (line 748-753) to clearly distinguish these complementary analyses and avoid confusion.

In addition to cytokine (6G) localization and signaling (6J-K) colocalization, analysis of ILC-Macrophage colocalization in the Xenium slides or by immunofluorescence could highlight the point that macrophages and ILC interactions change based on disease activity.

We thank the reviewer for this thoughtful suggestion. We agree that direct visualization of ILC–macrophage interactions would further strengthen the mechanistic framework of the study. However, because ILCs are relatively rare within the Xenium dataset and the targeted 468-gene panel provides limited transcriptomic depth, robust quantitative colocalization analyses between ILC subsets and neighboring macrophages were not feasible. Instead, we combined complementary approaches to address this question. Xenium spatial transcriptomics established the localization of GM-CSF signaling within macrophage-rich inflammatory niches, while our human scRNA-seq and zebrafish macrophage transcriptomic analyses demonstrated that GM-CSF induces conserved regulatory and tissue-homeostatic macrophage programs associated with maintenance of protective ILC3 responses. We have expanded the Result (line 503-513) and Discussion (line 575-581) to acknowledge that future studies employing higher-complexity spatial transcriptomic or multiplex imaging approaches will be valuable for directly defining macrophage–ILC niche interactions during intestinal inflammation.

(8) Figure 7

The graphical abstract does not clearly depict the interactions that occur in homeostasis. ILCs and the GM-CSF signaling are lacking. It is also unclear if the left part (Crohn's disease inflammation) depicts the inflamed situation or the consequences of GM-CSF addition in the inflammatory environment. It is also unclear that GM-CSF is acting on macrophages (and other myeloid subsets) and that the ILC3 maintenance and inhibition of ILC1 proliferation are indirect consequences.

We thank the reviewer for this thoughtful suggestion. We agree that the original graphical summary did not sufficiently distinguish the homeostatic and inflammatory states or clearly illustrate the mechanism by which GM-CSF regulates intestinal immunity. We have therefore revised Figure 7 to better depict the proposed model.

Specifically, the revised schematic now illustrates: (i) the homeostatic state, in which ILC3-derived GM-CSF supports resident macrophage homeostasis through CSF2RA/CSF2RB signaling; (ii) the inflammatory state in Crohn's disease, characterized by expansion of ILC1s, contraction of ILC3s, and increased pro-inflammatory cytokine production; and (iii) the protective effects of GM-CSF during inflammation. Importantly, we now explicitly indicate that GM-CSF acts primarily on macrophages and other CSF2RA/CSF2RB-expressing myeloid cells, rather than directly on ILCs. The revised model illustrates that maintenance of ILC3s, suppression of disease-associated ILC1 expansion, and enhanced IL-22 production are indirect consequences of GM-CSF-mediated macrophage reprogramming toward a regulatory, tissue-supportive phenotype. These revisions more accurately reflect the mechanistic conclusions supported by our human spatial transcriptomic, human scRNA-seq, and zebrafish studies.