Integrative Single-Cell and Bulk Transcriptomic Analyses with Spatial Validation Identify a Residual Fatty Acid–EMT Subset Driving Chemotherapy Resistance in Triple-Negative Breast Cancer via MIF- and MK-Mediated Ligand–Receptor Signaling

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Doha et al. provided an interesting paper on chemotherapy resistance in TNBC. The authors used single cell RNA-seq to identify a signature, FA-EMT, in TNBC epithelial cells that had been previously treated. They validated this signature in bulk RNA-seq and demonstrated that patients with high expression of this signature have worse overall survival. The authors also suggest that chemoresistant TNBC have different TME and demonstrate increased Tregs and MDSCs and reduced stromal cells. Interaction of the epithelial cells expressing the FA-EMT signature with Tregs and MDSCs was demonstrated by a protein tissue dataset from mouse tumors. Overall, the authors suggest that this FA-EMT signature affects the TME and is associated with chemotherapy resistance.  While this is an interesting study, there are several areas that need to be modified before it can be published. 

Major issues

• The authors make an assumption that the nonTNBC NAC treated cells are sensitive based on KI67 status whereas the TNBC cells are resistant because of high KI67. However, based on the tables from the supplemental data from the original study, two of the nonTNBC patients are said to have had minimal to no chemotherapy response whereas the TNBC patients had a partial response. In addition, the third nonTNBC was not treatment naïve but the treatment was three years prior for an earlier diagnosis. Therefore, labeling these cells (all 3 nonTNBC patients) as sensitive versus resistant appears to be flawed. Moreover, many of the differences in figures 1, 3, and 4 may have more to do with tumor type than chemotherapy resistance. It seems that it would make more sense to compare treatment naïve to NAC samples of only TNBC cells to determine if these signatures are due to chemoresistance (For figures 1, 3, and 4). As there are many samples that are treatment naïve from both TNBC and nonTNBC patients in the dataset that the authors used, it seems reasonable to be able to do these comparisons. This would strengthen their conclusions.

 

Figure 1B should just show the cancer epithelial cells with KI67 overlay as the majority of the cells that are being described related to figure 1B are not epithelial cells but immune cells (and include the current 1B in the supplemental figures). 

The conclusion about the different cell types in the tumors in figure 1 is overstated. The sample size is too small to be able to make as strong a claim as was made. In addition, the authors again focus on the idea of chemosensitive vs resistant but this may be more of a tumor type difference since there are not corresponding tumor naïve samples or samples where there were a clear sensitivity and resistance.  Specifically, the authors mention “This shift — from an immune-rich…” It’s unclear how this can be a shift since the samples are different tumor types. 

The argument for figure 1H would be strengthened by comparison of treatment naïve TNBC vs non-TNBC cells and determining if these genes are not differentially expressed. In addition, the argument would also be strengthened if the treatment naïve TNBCs were compared to these NAC TNBCs and if these genes were enriched in the NAC epithelial cells.

In figure 2, particularly the survival figures, it would be beneficial to separate by chemoresistance and sensitive and within those patients to separate by high and low levels of FA-EMT. There are a number of patients that are chemosensitive but also have a high FA-EMT score.  The authors should also address these patients in their discussion/conclusions. 

Figure 4B, how does this compare to other hallmark signaling pathways? It would be helpful to have other pathways included as a way to put the information in context. 

Minor

The specific samples used for the scRNA-seq analysis should be mentioned in the methods beyond just the dataset since there are many additional samples that were not used for the experiments. This would allow for reproducibility. 

Please provide in the supplemental data a UMAP projection of the cancer epithelial cells labeled by patient (TNBC1, TNBC2, ER1,ER2, Her2ER) to demonstrate that the different clusters are not patient specific. 

Figure 3A should say by tumor type not chemo response

Figure 4A should be redone with the TNBC and nonTNBC on different graphs to be able to interpret the sender receiver interaction better. 

For the description of the mouse data for figure 5, it would be helpful to mention a little bit more about the mouse model and that these are not paclitaxel treated samples, but paclitaxel studies showed only response in the SR cells. The way it is currently written makes it seem that either these were derived to be paclitaxel resistant or that they were treated samples. 

It would be helpful if it is clarified what the metabolic-EMT signature is compared to FA-EMT signature. Perhaps the metabolic-EMT signature name should not be used until it is described in the text as a subset of the FA-EMT signature. 

It is unclear why these cores were chosen as images for figure 5G. It would be helpful to include similar sized cores as well as additional cores as a supplemental figure. 

Lastly, please make sure the figures are of sufficient resolution as they were not in the original manuscript.  

Author Response

Response to Reviewer 1 Comments

We sincerely thank the Reviewer for the positive evaluation of our manuscript and for the constructive and detailed comments, which have significantly improved the clarity, rigor, and presentation of our work. All comments have been carefully addressed as detailed below. All revisions are highlighted red in the revised manuscript.

Point-by-point response to Comments and Suggestions for Authors

Comments 1:  The authors make an assumption that the nonTNBC NAC treated cells are sensitive based on KI67 status whereas the TNBC cells are resistant because of high KI67. However, based on the tables from the supplemental data from the original study, two of the nonTNBC patients are said to have had minimal to no chemotherapy response whereas the TNBC patients had a partial response.

In addition, the third nonTNBC was not treatment naïve but the treatment was three years prior for an earlier diagnosis. Therefore, labeling these cells (all 3 nonTNBC patients) as sensitive versus resistant appears to be flawed.

Moreover, many of the differences in figures 1, 3, and 4 may have more to do with tumor type than chemotherapy resistance. It seems that it would make more sense to compare treatment naïve to NAC samples of only TNBC cells to determine if these signatures are due to chemoresistance (For figures 1, 3, and 4). As there are many samples that are treatment naïve from both TNBC and nonTNBC patients in the dataset that the authors used, it seems reasonable to be able to do these comparisons. This would strengthen their conclusions.

Response: Point 1: Labeling of non-TNBC patients as sensitive based on MKI67 status

We fully agree with the reviewer. Based on Supplementary Table 1 of Wu et al. 2021, the five NAC-treated patients include: TNBC (n = 2): CID4513 ( partial response) and CID4523 (partial response); and non-TNBC (n = 3): CID4066 (minimal response), CID4398 ( minimal response), and CID3963 (treated three years prior for an earlier diagnosis — not concurrent NAC). We acknowledge that labeling these patients as "sensitive" versus "resistant" based on MKI67 alone was an oversimplification that does not accurately reflect the pathological response data. We have accordingly revised the manuscript throughout: all scRNA-seq figures consistently use Chemo_TNBC versus Chemo_non_TNBC as group labels — not "resistant" versus "sensitive" — and Figure 4 has been revised to adopt this TNBC versus non-TNBC labeling for complete consistency.

Point 2: Differences in Figures 1, 3, and 4 may reflect tumor type rather than chemotherapy resistance, and Point 3: Proposed treatment-naïve versus NAC-treated TNBC comparison

We thank the reviewer for raising these important points and want to clarify that the comparison of TNBC versus non-TNBC following NAC is precisely the primary aim of our study. Our central scientific question is: why is TNBC as a molecular subtype disproportionately associated with treatment failure and poor survival relative to non-TNBC? The identification of the FA-EMT subset as selectively enriched in Chemo_TNBC versus Chemo_non_TNBC is therefore not a confound — it represents a TNBC-specific biological driver of treatment failure maintained in the post-chemotherapy residuum of this aggressive subtype. The resistance interpretation of these findings is not based on the scRNA-seq comparison alone, but is formally established through independent bulk transcriptomic validation across 277 TNBC-only patients with pathological response classification (pCR versus RD), where FA-EMT co-expression scores were significantly elevated in resistant tumors (p < 0.001) with dramatically worse overall survival — directly ruling out tumor subtype differences as an explanation for our findings since both groups are TNBC.

Nevertheless, in response to the reviewer's concern that the differences observed in Figures 1, 3, and 4 may reflect tumor type rather than chemotherapy resistance, we performed both the following analyses suggested using the treatment-naïve samples available in the Wu et al. 2021 atlas:

Analysis 1 — Treatment-naïve TNBC versus naïve non-TNBC cancer epithelial cells (Supplementary Figure S1G; Supplementary Dataset S2): The top upregulated and downregulated genes in naïve TNBC versus naïve non-TNBC were markedly different from those observed in Chemo_TNBC versus Chemo_non_TNBC. Critically, S100A6, FABP5 and other key FA-EMT signature genes were not upregulated in naïve TNBC, confirming that the key FA-EMT signature genes are not pre-existing intrinsic TNBC subtype markers in treatment-naïve tumors — directly addressing the reviewer's concern that the differences observed may simply reflect tumor type biology.

Analysis 2 — Chemo-treated versus naïve TNBC cancer epithelial cells (Supplementary Figure S1H; Supplementary Dataset S3): In striking contrast, S100A6 (log₂FC = +1.67) and FABP5 (log₂FC = +1.32) were dramatically and significantly upregulated in Chemo-treated TNBC epithelial cells relative to treatment-naïve TNBC (adjusted p < 0.001). Furthermore, EMT-associated transcription factors HMGA2 (log₂FC = +0.38), SNAI2 (log₂FC = +0.36), and ZEB1 (log₂FC = +0.22) were all significantly elevated in the Chemo-treated residuum, confirming that the FA-EMT transcriptional program is specifically induced or selected by chemotherapy in TNBC rather than representing a pre-existing subtype characteristic.

Together, these two analyses directly confirm that the differences observed in Figures 1, 3, and 4 are not simply a reflection of intrinsic TNBC subtype biology, but are specifically associated with the post-chemotherapy state. This is further supported by the three-platform validation framework employed in this study — scRNA-seq discovery, 277-patient bulk validation with formal pCR/RD classification, and TNBC spatial CyCIF validation — which collectively provide the appropriate evidence hierarchy to establish FA-EMT as a TNBC-specific resistance driver, independently of the limitations inherent in the small scRNA-seq discovery cohort.

Comments 2: Figure 1B should just show the cancer epithelial cells with KI67 overlay as the majority of the cells that are being described related to figure 1B are not epithelial cells but immune cells (and include the current 1B in the supplemental figures).

Response: We thank the reviewer for this suggestion. Figure 1B has been updated to display the MKI67 expression overlay restricted to cancer epithelial cells only, providing a more precise visualization of proliferative activity within the malignant compartment. The original Figure 1B showing all cell types has been moved to Supplementary Figure S1C.

Comments 3: The conclusion about the different cell types in the tumors in figure 1 is overstated. The sample size is too small to be able to make as strong a claim as was made. In addition, the authors again focus on the idea of chemosensitive vs resistant but this may be more of a tumor type difference since there are not corresponding tumor naïve samples or samples where there were a clear sensitivity and resistance.  Specifically, the authors mention “This shift — from an immune-rich…” It’s unclear how this can be a shift since the samples are different tumor types.

Response: We thank the reviewer for this important and valid critique. We fully agree that the small sample size (n = 5) and the absence of matched treatment-naïve TNBC comparators limit the strength of conclusions that can be drawn from the compositional differences observed in Figure 1D–E. We have revised the relevant text to temper the language, acknowledge the tumor type contribution, and remove the term "shift" as it implies a longitudinal comparison that is not supported by the study design.

Revised text added in manuscript and highlighted in red:

"These compositional differences — with T-cell and stromal dominance in Chemo_non_TNBC and Cancer Epithelial and Myeloid expansion in Chemo_TNBC — likely reflect both intrinsic molecular subtype biology and chemotherapy-associated microenvironmental remodeling. While the patient number is limited (n = 5), the dataset encompasses 7,373 high-quality single cells from Chemo_TNBC and 13,287 from Chemo_non_TNBC, providing sufficient cellular resolution to characterize the transcriptional landscape of each compartment at single-cell depth. These observations should therefore be interpreted as hypothesis-generating rather than definitive."

Comments 4: The argument for figure 1H would be strengthened by comparison of treatment naïve TNBC vs non-TNBC cells and determining if these genes are not differentially expressed. In addition, the argument would also be strengthened if the treatment naïve TNBCs were compared to these NAC TNBCs and if these genes were enriched in the NAC epithelial cells.

Response: We thank the reviewer for this valuable suggestion, which we have now directly addressed through two additional analyses using the treatment-naïve samples available in the Wu et al. 2021 atlas.

Analysis 1 — Treatment-naïve TNBC versus naïve non-TNBC cancer epithelial cells (Supplementary Figure S1G; Supplementary Dataset S2); naïve TNBC  [patients: CID4465, CID44971, CID4495, CID44991, CID4515] vs  naïve non-TNBC  [patients: CID3586, CID3838, CID3921, CID3941, CID3948, CID4040, CID4067, CID4290, CID4461, CID4463, CID4471, CID4517-1, CID4530, CID4535])): The top upregulated and downregulated genes in naïve TNBC versus naïve non-TNBC were markedly different from those observed in Chemo_TNBC versus Chemo_non_TNBC (Supplementary Figure S1E and Supplementary Dataset S2). Critically, S100A6 was not upregulated in naïve TNBC (log₂FC = −0.48) and FABP5 did not reach the significance threshold (log₂FC = +0.48), confirming that the key FA-EMT signature genes are not pre-existing intrinsic TNBC subtype markers in treatment-naïve tumors.

Analysis 2 — Chemo-treated versus naïve TNBC cancer epithelial cells (Supplementary Figure S1H; Supplementary Dataset S3); NAC TNBC [CID4513, CID4523] vs Naïve TNBC [CID4465, CID44971, CID4495, CID44991, CID4515]: In striking contrast, the top upregulated genes in chemo-treated versus naïve TNBC epithelial cells closely mirrored those observed in Chemo_TNBC versus Chemo_non_TNBC. S100A6 (log₂FC = +1.67) and FABP5 (log₂FC = +1.32) were among the top 10 significantly upregulated in NAC-treated TNBC epithelial cells relative to treatment-naïve TNBC (adjusted p < 0.001). Furthermore, EMT-associated transcription factors HMGA2 (log₂FC = +0.38), SNAI2 (log₂FC = +0.36), and ZEB1 (log₂FC = +0.22) were all significantly elevated in the NAC-treated residuum, confirming that the FA-EMT transcriptional program is specifically induced or selected by chemotherapy in TNBC rather than representing a pre-existing subtype characteristic.

Taken together, these analyses directly address the reviewer's concern: the FA-EMT signature genes identified in Figure 1H are not simply intrinsic TNBC subtype markers present in naïve tumors, but are specifically enriched in the post-chemotherapy TNBC residuum, supporting their role as chemotherapy-driven resistance markers. These results have been added as Supplementary Figures S1G–H and Supplementary Datasets S2–S3, and the following text has been added to the main manuscript:

"To confirm that this signature reflects chemotherapy-associated transcriptional selection rather than intrinsic TNBC subtype biology, we compared cancer epithelial cells from treatment-naïve TNBC versus non-TNBC samples from the same atlas — in which S100A6 and FABP5 were not upregulated in naïve TNBC— confirming that the key signature genes are not pre-existing intrinsic TNBC subtype markers in treatment-naïve tumors (Supplementary Figure S1E; Supplementary Dataset S2). In contrast, comparison of chemo-treated versus naïve TNBC epithelial cells demonstrated that S100A6 and FABP5 were among the top 10 most significantly upregulated genes in the post-chemotherapy residuum (adjusted p < 0.001), and EMT-associated transcription factors HMGA2, SNAI2, and ZEB1 were all significantly elevated in the Chemo-treated residuum — confirming that the FA-EMT signature is specifically driven by chemotherapy selection pressure in TNBC (Supplementary Figure S1F; Supplementary Dataset S3)."

Comments 5: In figure 2, particularly the survival figures, it would be beneficial to separate by chemoresistance and sensitive and within those patients to separate by high and low levels of FA-EMT. There are a number of patients that are chemosensitive but also have a high FA-EMT score.  The authors should also address these patients in their discussion/conclusions.

Response: We thank the reviewer for this constructive suggestion. As requested, we performed stratified survival analysis separating patients by chemotherapy response (pCR versus RD) and further stratifying within each response group by high versus low FA-EMT score. The resulting survival curves are now shown in the supplementary figures S2A-C.

However, as the plots reveal, the survival curves within each chemotherapy response group are very similar regardless of FA-EMT score level — Resistant/High and Resistant/Low follow nearly identical trajectories, as do Sensitive/High and Sensitive/Low. This pattern indicates that within a given response category, the survival outcome is very similar and close, with insufficient power within each subgroup (n = 31 resistant, n = 76 sensitive) to detect further stratification.

We therefore respectfully submit that the stratification shown in the main Figure 2 — comparing high versus low FA-EMT co-expression across the entire cohort regardless of response — more directly and powerfully answers the clinically relevant question of whether FA-EMT co-expression predicts survival in TNBC patients.

Regarding the reviewer's observation that a number of chemotherapy-sensitive patients carry high FA-EMT scores — this is an important and insightful point that we have now addressed explicitly. In the GSE58812 cohort, some of 76 sensitive (pCR) patients had FA-EMT scores above the cohort median, indicating that FA-EMT elevation is not exclusive to the resistant group and that a subset of chemotherapy-sensitive patients harbor this transcriptional program. This finding has important clinical implications: it suggests that FA-EMT scoring may identify a subset of pCR patients who, despite achieving pathological complete response, carry a latent resistance-associated transcriptional program that may predispose them to future relapse. We have added the following to the result discussion:

"Notably, stratified survival analysis separating patients by chemotherapy response and FA-EMT score level revealed that survival curves were similar between high and low FA-EMT subgroups within each response category, indicating that chemotherapy response itself is the dominant prognostic determinant in this cohort (S2A-C). However, a subset of chemotherapy-sensitive (pCR) patients carried high FA-EMT co-expression scores, suggesting that FA-EMT elevation is not exclusive to clinically resistant tumors and may identify a latent resistance-prone transcriptional state even among patients who initially achieve pathological complete response. These patients may warrant closer surveillance and could potentially benefit from FA-EMT-targeted intervention to prevent disease relapse, a hypothesis that warrants prospective investigation.

Comments 6: Figure 4B, how does this compare to other hallmark signaling pathways? It would be helpful to have other pathways included as a way to put the information in context.

Response: We thank the reviewer for this important suggestion.

To provide the requested context: Using CellChat analysis we identified a total of 27 FA-EMT → T cell signaling pathways (18 in the Chemo-TNBC [T] group and 9 in the Chemo-non-TNBC [N] group; Supplementary Table S8), and a total of 40 FA-EMT → Myeloid signaling pathways (25 in the T group and 15 in the N group; Supplementary Table S10). Figure 4C and 4E display the top pathways with information flow > 0.5 in either group as a diverging bar plot, providing the full comparative context across all major signaling programs — allowing direct visual comparison of T-enriched versus N-enriched versus shared pathways. Within this landscape, structural and ECM pathways — including COLLAGEN (T: 2.89; N: 2.62), LAMININ, and FN1 — showed the highest absolute information flow in both groups, reflecting shared epithelial–immune crosstalk present regardless of molecular subtype. However, the most selectively T-enriched immunomodulatory pathways were MIF (T: 1.92 vs N: 0.65; 2.9-fold), MK (T: 0.80 vs N: absent), and APP (T: 0.59 vs N: absent). Similarly, for FA-EMT → Myeloid signaling, MK showed the most dramatic T-specific enrichment (T: 1.16 vs N: 0.05; 23-fold) and MIF was 2.1-fold enriched in T versus N. The 23-fold enrichment of MK signaling in the T-group myeloid compartment is particularly striking and reinforces MDK–NCL as the most selectively TNBC-associated axis in innate immune suppression. These enrichment ratios establish MDK–NCL and MIF–CD74–CXCR4 as the most specifically T-associated signaling axes within the broader FA-EMT signaling repertoire, justifying their nomination as the principal mechanistically grounded therapeutic targets.

We have updated the text Section 3.4 to contextualize the pathway findings:

" Comparison of signaling information flow (Figure 4C) — which displays pathways with information flow > 0.5 in either group, enabling direct comparison of T-enriched, N-enriched, and shared signaling programs across the full FA-EMT → T cell pathway landscape — revealed that structural and ECM pathways including COLLAGEN and LAMININ showed the highest absolute information flow in both groups, reflecting shared epithelial–immune crosstalk. However, the most selectively Chemo-TNBC (T)-enriched immunomodulatory pathways were MIF (T: 1.92 vs N: 0.65; 2.9-fold), MK (T: 0.80 vs N: absent), and APP (T: 0.59 vs N: absent) — with MIF and MK enriched exclusively or predominantly in the T group against this shared structural background.

Similarly, Figure 4E — displaying FA-EMT → Myeloid pathways with information flow > 0.5 in either group — revealed that MK signaling showed the most dramatic T-specific enrichment across the myeloid compartment (T: 1.16 vs N: 0.05; 23-fold), while MIF was 2.1-fold enriched in T versus N. The 23-fold enrichment of MK signaling in the T-group myeloid compartment is particularly striking and reinforces MDK–NCL as the most selectively TNBC-associated signaling axis in innate immune suppression. These enrichment ratios establish MDK–NCL and MIF–CD74–CXCR4 as the most specifically T-associated axes within the broader FA-EMT signaling repertoire, justifying their nomination as the principal therapeutic targets..

These enrichment ratios establish MDK–NCL and MIF–CD74–CXCR4 as the most specifically resistance-associated signaling axes within the broader FA-EMT signaling repertoire, justifying their nomination as the principal therapeutic targets."

 

 

Minor

Comments 7: The specific samples used for the scRNA-seq analysis should be mentioned in the methods beyond just the dataset since there are many additional samples that were not used for the experiments. This would allow for reproducibility.

Response: We agree and have revised the Methods section accordingly. All individual patient IDs are now explicitly stated for each analytical group, verified against Wu et al. 2021 Supplementary Table 1. The primary Chemo cohort is identified as Chemo_TNBC (CID4513, CID4523) and Chemo_non_TNBC (CID3963, CID4066, CID4398). For the two validation analyses, naïve TNBC cells derive from CID4465, CID44971, CID4495, CID44991, and CID4515; naïve non-TNBC cells derive from all 14 treatment-naïve, non-TNBC cases in the atlas (CID3586, CID3838, CID3921, CID3941, CID3948, CID4040, CID4067, CID4290, CID4461, CID4463, CID4471, CID4517-1, CID4530, CID4535). The revised text now in the methods clarifying the selection logic for full reproducibility.

Comments 8:  Please provide in the supplemental data a UMAP projection of the cancer epithelial cells labeled by patient (TNBC1, TNBC2, ER1,ER2, Her2ER) to demonstrate that the different clusters are not patient specific.

Response: A UMAP projection of cancer epithelial cells colored by patient histology (TNBC_1, TNBC_2, HER2_ER, ER_1, ER_2) has been added as Supplementary Figure S1J. Both TNBC-associated clusters are characterized by elevated FA signaling, with one additionally carrying a higher EMT signature. As is expected and well-established in single-cell studies of primary breast tumors, cancer epithelial cells exhibit substantial inter-patient transcriptional heterogeneity— resulting in partial patient-level separation on the UMAP. This pattern is consistent with published breast cancer scRNA-seq atlases, including the source dataset (Wu et al. 2021). Importantly, the FA-EMT transcriptional program — defined by co-expression of FABP5, S100A6, VIM, and related markers — is detected across both TNBC donors and is significantly upregulated in chemo-treated versus treatment-naïve TNBC (Supplementary Dataset S3), confirming chemotherapy-driven rather than patient-specific induction. Its enrichment in Chemo_TNBC is further validated at the bulk transcriptomic level in two independent cohorts and CyCIF validation, establishing the FA-EMT signature as a reproducible molecular program rather than a single-patient artifact.

Comments 9:  Figure 3A should say by tumor type not chemo response

Response: We thank the reviewer for this correction. Figure 3A has been updated to label groups by molecular tumor type (Chemo_TNBC vs Chemo_non_TNBC) rather than chemotherapy response, consistent with the label corrections applied throughout the revised manuscript.

Comments 10:  Figure 4A should be redone with the TNBC and nonTNBC on different graphs to be able to interpret the sender receiver interaction better.

Response: We thank the reviewer for this suggestion. A UMAP projection of cancer epithelial cells colored by patient histology (TNBC_1, TNBC_2, HER2_ER, ER_1, ER_2) has been added as Supplementary Figure S1J, allowing direct visualization of the patient-of-origin distribution across epithelial subclusters.

Comments 11:  For the description of the mouse data for figure 5, it would be helpful to mention a little bit more about the mouse model and that these are not paclitaxel treated samples, but paclitaxel studies showed only response in the SR cells. The way it is currently written makes it seem that either these were derived to be paclitaxel resistant or that they were treated samples.

Response: We thank the reviewer for this important clarification. The Results section (Figure 5) has been revised to explicitly state that the SP and SR tumor lines represent intrinsically distinct syngeneic lines not derived by paclitaxel treatment, but subsequently shown to exhibit consistent differential paclitaxel response — SP being intrinsically resistant and SR sensitive.

The revised text in manuscript: "In this model, two tumor lines were established with intrinsically distinct stromal phenotypes — stroma-poor (SP; n = 26 cores) and stroma-rich (SR; n = 43 cores) — that were not derived by paclitaxel treatment but were subsequently shown to exhibit a consistent differential response to paclitaxel: SP tumors are intrinsically paclitaxel-resistant while SR tumors remain paclitaxel-sensitive [16]. These two lines therefore represent established paclitaxel-resistant and paclitaxel-sensitive phenotypes, enabling spatial validation of our computational findings at single-cell protein resolution within intact tumor tissue."

Comments 12:  It would be helpful if it is clarified what the metabolic-EMT signature is compared to FA-EMT signature. Perhaps the metabolic-EMT signature name should not be used until it is described in the text as a subset of the FA-EMT signature.

Response: We thank the reviewer for this helpful comment. The Results text has been revised to introduce and define the Metabolic-EMT designation explicitly before use.

 The revised text reads: "Among the epithelial clusters, a discrete population was identified by co-expression of Vimentin, S100A6, pMYC, and Ki67 with low EpCAM and E-cadherin — directly mirroring the protein-level signature of the FA-EMT subset identified transcriptomically in Figure 1, representing EMT identity (Vimentin, low EpCAM/E-cadherin), lipid metabolic stress (S100A6), MYC-driven fatty acid biosynthesis (pMYC), and proliferative activation (Ki67). This population is hereafter designated Epi_Metabolic_EMT, reflecting its protein-level definition as the spatial CyCIF counterpart of the transcriptomic FA-EMT subset."

Comments 13:  It is unclear why these cores were chosen as images for figure 5G. It would be helpful to include similar sized cores as well as additional cores as a supplemental figure.

Response: We thank the reviewer for this comment. Figure 5G has been revised to display size-matched representative SP and SR cores. The cores shown were chosen as representative of the median Metabolic-EMT score within each group. In addition, a new Supplementary Figure (S4) showing a broader panel of SP and SR cores across the full range of core sizes has been added, demonstrating that the enrichment of the Epi_Metabolic_EMT population in SP relative to SR cores is consistent across the TMA. The text revised to reflect that.

Comments 14:  Lastly, please make sure the figures are of sufficient resolution as they were not in the original manuscript.

Response:  All figures in the revised manuscript have been regenerated and exported at a minimum of 300 DPI in TIFF/PDF format, in accordance with IJMS submission guidelines.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The work by Doha et al., integrates single-cell and bulk transcriptomics to identify markers and pathways conferring resistance to chemotherapy in TNBC. Overall, this is an interesting study with attention to detail. I have prepared just a few suggestions to help the authors in providing a revised version of their work for further consideration.

1.The number of single cell RNA-seq datasets is relatively low. Can the authors validate the identified gene signatures in a larger number of available single-cell RNA-seq or spatial transcriptomics datasets?

2.Please provide higher-resolution figures. At least in the pdf version it is difficult to clearly assess them.

3.Please provide more extensive supplementary tables/excels with marker genes and their statistics for all major subpopulations and clusters for interested readers.

Author Response

Response to Reviewer 2 Comments

We sincerely thank the Reviewer for the positive evaluation of our manuscript and for the constructive and detailed comments, which have significantly improved the clarity, rigor, and presentation of our work. All comments have been carefully addressed as detailed below. All revisions are highlighted red in the revised manuscript.

Point-by-point response to Comments and Suggestions for Authors

Comments 1:  The number of single cell RNA-seq datasets is relatively low. Can the authors validate the identified gene signatures in a larger number of available single-cell RNA-seq or spatial transcriptomics datasets?

Response: We thank the reviewer for this important point. We fully acknowledge that the primary scRNA-seq discovery cohort comprises five NAC-treated patients — the complete set available in the Wu et al. 2021 atlas (GSE176078) with chemotherapy treatment annotation. To address the concern of limited sample size, we performed two additional large-scale validation analyses using treatment-naïve samples from the same atlas, substantially increasing the number of cells and donors interrogated:

Analysis 1 (Supplementary Figure S1G; Table S2) compared treatment-naïve TNBC versus naïve non-TNBC cancer epithelial cells across 19 patients (n = 8,389 naïve TNBC cells [5 patients] vs 13,132 naïve non-TNBC cells [14 patients]), demonstrating that FA-EMT markers are not intrinsic to the TNBC subtype in the absence of chemotherapy.

Analysis 2 (Supplementary Figure S1H; Table S3) compared chemo-treated versus treatment-naïve TNBC cancer epithelial cells across 7 TNBC patients (n = 2,225 NAC TNBC cells vs 8,389 naïve TNBC cells), confirming that the key FA-EMT gene signtuers are significantly upregulated specifically in response to chemotherapy. Together with bulk transcriptomic validation in 277 patients across two independent cohorts (GSE25066, GSE58812) and spatial CyCIF validation across 69 tumor cores (565,438 cells), these analyses provide multi-platform, multi-cohort evidence that substantially mitigates the limitation of a small primary scRNA-seq discovery dataset.

Comments 2: Please provide higher-resolution figures. At least in the pdf version it is difficult to clearly assess them.

Response: We apologize for the insufficient figure resolution in the original submission. All figures have been regenerated and resubmitted at a minimum of 300 DPI in PDF format, in accordance with IJMS author guidelines.

Comments 3: Please provide more extensive supplementary tables/excels with marker genes and their statistics for all major subpopulations and clusters for interested readers.

Response: We thank the reviewer for this suggestion. The revised submission includes a comprehensive Supplementary Dataset file (11 tables) providing full marker gene lists with statistics for all major subpopulations and analyses:

Table S1 — Full DGE results for cancer epithelial subclusters (9,906 genes)

Table S2 — DGE: treatment-naïve TNBC vs naïve non-TNBC cancer epithelial cells (11,424 genes; Analysis 1)

Table S3 — DGE: chemo-treated vs naïve TNBC cancer epithelial cells (11,825 genes; Analysis 2)

Table S4 — Full DGE results for fibroblast subclusters

Table S5 — Full DGE results for T cell subclusters

Table S6 — Full DGE results for myeloid subclusters

Table S7 — Bulk transcriptomic validation results (GSE25066; 277 patients)

Table S8 — FA-EMT → T cell signaling pathway information flow, Resistant vs Sensitive (27 pathways)

Table S9 — FA-EMT → T cell ligand–receptor pairs

Table S10 — FA-EMT → Myeloid signaling pathway information flow, Resistant vs Sensitive (40 pathways)

Table S11 — FA-EMT → Myeloid ligand–receptor pairs

All tables 1-7 include gene-level statistics (avg_log2FC, pct.1, pct.2, p_val, p_val_adj) enabling independent interrogation by interested readers.

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have made substantial improvements to the manuscript, helping to clarify that the differences are not due to tumor type. However, there are several areas of new text that would benefit from modification. 

Line 107-108 in the text should clarify the UMAP by cohort is supplemental 1C and by epithelial cells is Figure 1B. As it is currently, the text does not reflect this for Figure 1B. 

Lines 169, 175/6, and 634 state that this study confirms these effects are not due to TNBC but due to chemo. It’s appreciated that the naïve to NAC comparisons were done to rule out the effect being just TNBC. However, since these samples were from different patients and not a study of pre to post treatment in the same patient, it would likely be better to say suggests instead of confirms that it is due to the chemo in these sentences as there is no way to confirm this using the current datasets.

Additionally, several issues with figures should be addressed: 

The resolution of figure 3 needs to be checked as it is lower resolution than the other figures and hard to view. This image does not appear to be the updated version as the rebuttal suggested that the titles of the figure would be modified but it appears to be the same as the first version. The titles for figures 1, 3 being by chemo response are somewhat misleading and may benefit from being TNBC vs non-TNBC (some of the figures are listed this way).  

For the new survival curves in the supplemental data (S2A-C), it is unclear why there is such a difference in survival probability when stratified for the resistant patients compared to the data in figure 2.

Author Response

Response to Reviewer 1 Comments

We sincerely thank the Reviewer for the positive evaluation of our manuscript and for the constructive and detailed comments. All comments have been carefully addressed as detailed below. All revisions are highlighted blue in the revised manuscript Round 2.

Point-by-point response to Comments and Suggestions for Authors

Comments 1:  Line 107-108 in the text should clarify the UMAP by cohort is supplemental 1C and by epithelial cells is Figure 1B. As it is currently, the text does not reflect this for Figure 1B.

Response:

We thank the reviewer for this clarification. The text has been revised to explicitly state that Figure 1B shows MKI67 expression overlaid on cancer epithelial cells only, while Supplementary Figure S1C shows the equivalent split UMAP across all cell types in the full cohort.

Revised text:

Proliferative activity was assessed by overlaying MKI67 expression onto UMAP projections split by cohort, restricted to cancer epithelial cells (Figure 1B); the equivalent split UMAP across all profiled cell types is shown in Supplementary Figure S1C.

 

Comments 2:  Lines 169, 175/6, and 634 state that this study confirms these effects are not due to TNBC but due to chemo. It’s appreciated that the naïve to NAC comparisons were done to rule out the effect being just TNBC. However, since these samples were from different patients and not a study of pre to post treatment in the same patient, it would likely be better to say suggests instead of confirms that it is due to the chemo in these sentences as there is no way to confirm this using the current datasets.

Response:

We agree with the reviewer's important point. Since the naïve and NAC-treated samples derive from different patients rather than matched pre/post-treatment pairs, the use of "confirms" overstates the strength of causal inference. We have revised all three instances to "suggests" and added an appropriate qualifier acknowledging this.

Manuscript edits:

 (Results): "…suggesting that the key signature genes are not pre-existing intrinsic TNBC subtype markers in treatment-naïve tumors…suggesting that the FA-EMT signature is associated with chemotherapy selection pressure in TNBC, though matched pre/post-treatment samples would be required to formally confirm this."

 (Discussion para 1): "…two treatment-naïve validation analyses suggesting that the FA-EMT signature is chemotherapy-driven rather than intrinsic to the TNBC subtype…"

 (Discussion para 2): "Importantly, our validation analyses suggest that this co-expression is not a baseline feature of TNBC but may be specifically induced by chemotherapy exposure, consistent with a treatment-selected transcriptional state rather than a pre-existing subclonal population; prospective matched pre/post-treatment single-cell profiling would be required to formally confirm this interpretation."

Comments 3:  Additionally, several issues with figures should be addressed:  The resolution of figure 3 needs to be checked as it is lower resolution than the other figures and hard to view. This image does not appear to be the updated version as the rebuttal suggested that the titles of the figure would be modified but it appears to be the same as the first version. The titles for figures 1, 3 being by chemo response are somewhat misleading and may benefit from being TNBC vs non-TNBC (some of the figures are listed this way). 

Response: 

We apologize for the resolution issue in Figure 3. The figure was previously regenerated at 300 DPI but was accidentally not included in the prior revision submission. The updated high-resolution version is now included in this submission; all high-resolution figures have been submitted separately. Additionally, the titles for Figures 1 and 3 have been revised to replace "by chemo response" with "Chemo_TNBC vs Chemo_non_TNBC" throughout, consistent with the molecular subtype framing used across the revised manuscript.

Manuscript edit — Figure 1 title:

Figure 1. The post-chemotherapy TME is transcriptionally and compositionally distinct between Chemo_TNBC and Chemo_non_TNBC, defined by a myeloid-enriched landscape and selective expansion of a co-proliferative FA-EMT epithelial subset.

Manuscript edit — Figure 3 title:

Figure 3. Coordinated immunosuppressive remodeling of the stromal, T-cell, and myeloid compartments in Chemo_TNBC versus Chemo_non_TNBC.

 

Comments 4: For the new survival curves in the supplemental data (S2A-C), it is unclear why there is such a difference in survival probability when stratified for the resistant patients compared to the data in figure 2.

We thank the reviewer for this important observation. The difference arises because Figure 2C–E and Supplementary Figure S2A–C use different stratification strategies. Figure 2C–E shows Kaplan–Meier curves stratified by FA-EMT score alone (High vs Low, median split) across all patients in the GSE25066 cohort, whereas Supplementary Figure S2A–C performs a four-group stratification that simultaneously conditions on both chemotherapy response (Sensitive/Resistant) and FA-EMT score (High/Low). In the four-group stratification, Resistant/High and Resistant/Low patients both carry the dominant prognostic weight of the Resistant phenotype — which itself confers markedly worse survival — causing the Resistant subgroups to cluster together at the bottom of the survival curves, while both Sensitive subgroups cluster at the top regardless of FA-EMT score. This explains why the response designation dominates over FA-EMT score within each group in S2A–C, and why the overall survival probability ranges differ from Figure 2 which shows the independent marginal effect of FA-EMT score across all patients.

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have responded adequately to my comments 

Author Response

We thank the Reviewer for their positive assessment and for the constructive feedback provided in the previous round, which has substantially strengthened the manuscript.