Metabolic Trans-Omic Analysis Reveals Key Regulatory Disruption of Energy Metabolism in Alzheimer’s Disease

Article activity feed

1. 

   eLife

   Jul 13, 2026

   eLife Assessment

   This study presents a useful application of trans-omic network analyses to existing human brain datasets, generating systems-level insights into metabolic dysregulation in Alzheimer's disease. Overall, the authors' analytical choices are solid, with appropriate use of existing data and methods, and with many of their results confirming previous findings. However, some of the authors' key claims, related to previously unknown details of regulatory relationships, are only partially supported due to limitations in dataset cell-type resolution and network robustness, as well as a lack of functional validation. This work will be of interest to cellular or systems neurobiologists studying Alzheimer's disease and could serve as a helpful starting point for future work.

   [Editors' note: this paper was reviewed by Review Commons.]

   Read the original source
2. 

   eLife

   Jul 13, 2026

   Reviewer #1 (Public review):

   The authors aim to reconstruct a multi-layer metabolic regulatory network in Alzheimer's disease by integrating transcriptomic, proteomic, and metabolomic datasets from human brain tissue. By linking transcription factors, enzyme expression, metabolic reactions, and metabolites, the study seeks to provide a systems-level understanding of disease-associated metabolic dysregulation.

   The revised manuscript has improved in clarity and includes additional analyses in response to prior comments, particularly the incorporation of cell-type proportion estimates derived from matched single-cell datasets. This represents a meaningful step toward addressing concerns about the interpretation of bulk tissue and strengthens the study's descriptive rigor.

   However, several key limitations remain. First, although cell-type …

   Reviewer #1 (Public review):

   The authors aim to reconstruct a multi-layer metabolic regulatory network in Alzheimer's disease by integrating transcriptomic, proteomic, and metabolomic datasets from human brain tissue. By linking transcription factors, enzyme expression, metabolic reactions, and metabolites, the study seeks to provide a systems-level understanding of disease-associated metabolic dysregulation.

   The revised manuscript has improved in clarity and includes additional analyses in response to prior comments, particularly the incorporation of cell-type proportion estimates derived from matched single-cell datasets. This represents a meaningful step toward addressing concerns about the interpretation of bulk tissue and strengthens the study's descriptive rigor.

   However, several key limitations remain. First, although cell-type proportions are now estimated, this information is not integrated into downstream analyses. As a result, it remains unclear whether the observed metabolic changes reflect cell-intrinsic regulation or shifts in cellular composition. This distinction is critical for interpreting the inferred regulatory network and limits the strength of the conclusions.

   Second, the biological conclusions remain largely confirmatory. The reported downregulation of energy-related pathways, including the TCA cycle and oxidative phosphorylation, is consistent with prior literature. The trans-omic framework provides a structured representation of these changes, but the manuscript does not convincingly demonstrate that this approach yields new mechanistic insight beyond existing knowledge. In particular, the network is not sufficiently leveraged to identify novel regulatory relationships or generate testable hypotheses.

   Third, while the framework integrates multiple molecular layers, the contribution of upstream transcriptional regulation remains unclear. The most compelling findings appear to arise from protein and metabolite layers, and the manuscript does not clearly demonstrate how transcription factors or mRNA-level changes contribute to the interpretation of metabolic dysregulation. This weakens the claim that the study provides a fully integrated trans-omic perspective.

   Fourth, concerns regarding network robustness remain. The analysis relies on partially overlapping cohorts across omics modalities, and although this limitation is acknowledged, no formal sensitivity or robustness analyses are presented. This reduces confidence in the stability and generalizability of the inferred network structure.

   Finally, the handling of covariates remains limited. While the authors justify this based on data availability and prior studies, the lack of consistent adjustment for known confounders introduces uncertainty in attributing observed differences specifically to disease-related biology.

   Overall, the study presents a technically sound application of a trans-omic integration framework and provides a coherent overview of metabolic dysregulation in Alzheimer's disease. However, the findings are primarily descriptive and confirmatory, and the current analyses do not fully demonstrate that the approach yields novel biological insight. The work will be of interest to researchers in systems biology and multi-omics integration, but its impact on advancing understanding of disease mechanisms is likely to be moderate.

   Read the original source
3. 

   eLife

   Jul 13, 2026

   Author response:

   General Statements:

   We appreciate the reviewers for the critical review of the manuscript and the valuable comments. We have carefully considered the reviewer’s comments and have revised our manuscript accordingly.

   Point-by-point description of the revisions:

   Reviewer #1 (Evidence, reproducibility and clarity):

   Major comments

   (1) This study leaves out lipid metabolism as a major energy metabolism pathway relevant to AD. The authors themselves cite the significance of acylcarnitines and CPT1A in AD (pg. 3, lines 32-33, pg. 4, lines 1-2). Lipid metabolism and homeostasis is known to be disrupted in AD1. Fatty acid oxidation is a known energy source in the prefrontal cortex2 and will also generate acetyl coA, which this study reveals is a significant decreased metabolite in AD. Furthermore, sphingomyelin emerges as one of …

   Author response:

   General Statements:

   We appreciate the reviewers for the critical review of the manuscript and the valuable comments. We have carefully considered the reviewer’s comments and have revised our manuscript accordingly.

   Point-by-point description of the revisions:

   Reviewer #1 (Evidence, reproducibility and clarity):

   Major comments

   (1) This study leaves out lipid metabolism as a major energy metabolism pathway relevant to AD. The authors themselves cite the significance of acylcarnitines and CPT1A in AD (pg. 3, lines 32-33, pg. 4, lines 1-2). Lipid metabolism and homeostasis is known to be disrupted in AD1. Fatty acid oxidation is a known energy source in the prefrontal cortex2 and will also generate acetyl coA, which this study reveals is a significant decreased metabolite in AD. Furthermore, sphingomyelin emerges as one of the major decreased DEMs as well. Thus, lipid metabolism should be highlighted in Figure 3 and discussed throughout the manuscript; otherwise its omission should be clearly stated and justified.

   We appreciate the reviewer’s insightful comment regarding a critical role of lipid metabolism in AD. We recognize that lipid metabolism is a metabolic pathway deeply involved in AD pathology (Baloni et al., 2022, 2020; Varma et al., 2021). Accordingly, we have revised the Limitations section to more strongly emphasize its role as a vital energy source (pg. 13, lines 15-17). Regarding the visualization of lipid metabolism, we extracted lipid-related pathway from the trans-omic network but found that the regulatory relationships among DEPs and DEMs were excessively complex and interconnected. Thus, interpreting this regulatory network seemed to be more challenging compared to the other energy production pathways presented in our manuscript. Therefore, we have concluded that the pathway analysis in our trans-omic network may not be suitable for deeply elucidating the lipid dysregulation in AD. We have added a statement acknowledging this as a limitation of our current methodology in the revised manuscript (pg. 13, lines 13-22).

   (2) The covariates used for differential analysis should be discussed and justified. Notably, age is used as a covariate for transcriptomic analysis but not proteomic and metabolomic analysis, with no justification. Additionally, given the known importance of lipid metabolism in AD and the putative role of APOE in lipid homeostasis3, APOE genetic status should be considered as a covariate, or its omission should be justified.

   We appreciate the reviewer’s comment regarding the included covariates in differential analyses of our study. The reason we did not include other variables, such as age at death and RIN, is that these data were not available for each sample. Thus, we referred to the original research articles from which proteomic or metabolomic datasets used in our study were derived. Regarding the metabolomic dataset, in the original article (Batra et al., 2023), only two metabolites, 1-methyl-5-imidazoleacetate and N6-carboxymethyllysine, were significantly associated with age. In addition, no metabolites were significantly associated with sex, BMI, and years of education. Regarding the proteomic dataset, in the original article (Johnson et al., 2020), age at death, PMI, and sex were included as covariates in the analyses, though these variables were not found to strongly influence the data (Extended Data Fig.2 in (Johnson et al., 2020)).

   (3) The authors make a conclusion statement that suggests intervention: "Collectively, our data suggests that preserving or improving the ability to produce ATP and early intervention in the process of nitrogen metabolism are candidates for the prevention and treatment of dementia" (pg. 12, lines 12-14). This claim is not well-supported by the evidence provided in the study. There are a few limitations: (a) This was an observational, not interventional study; (b) The study did not establish whether the metabolic disruptions are causes or effects in AD; and (c) ATP or other bioenergetic indicators were not directly measured. Therefore, any statements about potential interventions should be removed or qualified as highly speculative.

   We agree with the reviewer that the statement regarding potential interventions was not sufficiently supported by our analyses. Accordingly, we have removed the sentence regarding prevention and treatment from the revised manuscript (e.g., we have deleted final paragraph of the previous manuscript).

   (4) In conjunction with the last point, the main conclusion of the study is that energy production is down in AD. The data presented in Figure 3 are consistent with this conclusion, but it is far from definitive due to limitations stated above in comments 3a and 3b. The authors should offer additional support for this conclusion: experimental follow-up, flux modeling, analysis of alternative datasets with ATP measurement, causal inference.

   We sincerely thank the reviewer for this valuable and constructive suggestion. Regarding flux modeling, we agree that metabolic flux analysis could provide important mechanistic insight. Indeed, previous studies have applied flux modeling in the context of lipid metabolism in Alzheimer’s disease (Baloni et al., 2022). We also attempted to perform flux modeling focusing on energy metabolism. However, we found it difficult to obtain biologically meaningful and robust results and therefore decided not to include these analyses in the current manuscript.

   With respect to ATP measurements, we fully agree that direct evidence of altered ATP levels would further strengthen our conclusion. However, to the best of our knowledge, there are currently no publicly available large-scale datasets that directly measure ATP levels in human postmortem brain tissues. This limitation makes it challenging to incorporate validation in the present study.

   Regarding experimental follow-up, we agree that functional validation is essential to confirm the mechanistic implications of our findings. We are actively considering follow-up experimental studies. However, we consider the present work to be a multi-omic integrative analysis aimed at identifying key molecular alterations and generating biologically important hypotheses. We have revised the Limitation section to more clearly position this manuscript as an observational systems-level analysis (pg. 13, lines 20-22).

   (5) The validation analysis did not sufficiently show the generalizability of this study's results. The authors demonstrated a correlation of 0.53 to the MSBB transcriptomics data and 0.60 to the AMP-AD DiverseCohorts proteomics data. Beyond these correlation coefficients, no meaningful comparison between the datasets is offered. How concordant are the differentially expressed features (or pathways) between the datasets? How robust would the trans-omic network be if incorporating the alternate datasets? Is the main conclusion (energy metabolism is down in AD) supported by the validation datasets? We think this analysis should be expanded and described in the main text.

   Although the results for external metabolomics datasets are reported in Fig S2C, correlation coefficients with the external data are not reported. The authors state, "Note that each study used different definitions for AD and CT groups, had variations in measurement methods and brain regions analyzed." We appreciate these limitations. However, the external data should be re-analyzed using the same definitions of AD and CT, if possible. The limitations and results (which DEMs are shared between datasets) should be discussed in the main text.

   We thank the reviewer for this important comment regarding the generalizability of our findings. In the revised manuscript, we have expanded the validation analyses and summarized the results in Figure S2. First, at the transcriptomic level, Figure S2B and S2C show the overlap between up- and downregulated genes in AD identified in our ROSMAP-derived analyses and those reported in a previously published large-scale meta-analysis of 2,114 postmortem samples across seven brain regions (Wan et al., 2020). A substantial proportion of DEGs were shared, supporting cross-cohort and cross-region robustness to some extent. At the proteomic level, Figure S2E shows a comparison between the ROSMAP and the AMP-AD DiverseCohorts datasets. We highlighted the subset of enzymes involved in the energy metabolism analysis shown in Fig. 3 and calculated a separate correlation coefficient for this subset (Pearson coefficient = 0.86, p-value = 1.5e-7), further supporting our main conclusion. In addition, to assess the concordance between the two datasets in a threshold-independent manner, we additionally performed Rank-Rank Hypergeometric Overlap (RRHO) analysis (Figure S2E). RRHO analysis (Cahill et al., 2018; Plaisier et al., 2010) enables the comparison of ranked protein lists without relying on arbitrary differential expression cutoffs and has been used for cross-dataset comparison in several previous studies (Fröhlich et al., 2024; Maitra et al., 2023). The RRHO heatmaps demonstrated significant enrichment in the concordant quadrants, confirming systematic agreement between datasets beyond simple correlation coefficients. For metabolomics, Figure S2G shows RRHO analyses comparing the ROSMAP metabolomic data with other datasets measured by the same UPLC-MS/MS platform (Batra et al., 2024; Novotny et al., 2023), demonstrating significant concordance in ranked metabolite changes in AD.

   (6) The glycolysis analysis and discussion needs more development. Glycolysis and gluconeogenesis share many of the same enzymes, but they are not the same pathway and should not be discussed as such. To make a claim about the overall influence of enzyme and metabolite levels on glycolysis, the authors should focus on the energetically committing steps of glycolysis (hexokinase, phosphofructokinase, pyruvate kinase) in Figure 3A, and include the full/current version of the figure in the supplement. Gluconeogenesis-specific enzymes (pyruvate carboxylase, PEPCK) are not mentioned at all - are they among the DEPs/DEGs?

   We appreciate the reviewer’s comment regarding the distinction between glycolysis and gluconeogenesis pathway. Among the gluconeogenesis-specific enzyme proteins, G6PC1, FBP1, PC, and PCK2 were measured in our dataset, but none of them were identified as DEPs. In addition, gluconeogenesis is a process that occurs primarily in the liver and kidney rather than the brain. Given this biological context and the lack of significant changes in relevant enzymes, we have revised the terminology throughout the manuscript, replacing “glycolysis/gluconeogenesis pathway” with “glycolysis pathway” in the revised version.

   (7) Given that there wasn't good concordance between the DEGs and DEPs, did including the mRNA and transcription factor layers in the network really add anything useful? It seems like the main conclusions of the manuscript were driven by the protein and metabolite layers only. How many of the DE metabolic enzymes were coregulated at the transcript and protein level? It would be useful to include the 5-layer trans-omic network in the supplement to display these results. Given your network, at what level does it appear that energy metabolism is regulated?

   It is true that our primary conclusion regarding the regulation of energy metabolism is driven by the changes in protein and metabolite abundance. However, we consider the low concordance between mRNA and protein expression itself to be an important feature of AD pathology, as also reported in previous studies (Johnson et al., 2022; Tasaki et al., 2022). Although we did not perform a further analysis of this discordance, we believe that including the TF and mRNA layers into the metabolic trans-omic network strengthens a system-wide view of metabolic dysregulation in AD.

   Regarding the mRNA changes corresponding to the DEP enzymes, please refer to Figure S7A.

   (8) Comment further on the results from Figure 2D. What can be learned from identifying metabolites with the greatest degree centrality? What pathways other than energy metabolism are highlighted by the trans-omic network?

   We assume that some energetic indicators, including AMP and acetyl-CoA, and nitrogen metabolism-related metabolites, Glu, 2-oxoglutarate, and urea, can be potential key regulators of dysregulated metabolism in AD.

   (9) (Suggestion) We suggest the authors leverage their trans-omic network in additional ways beyond giving a snapshot of a few energy metabolism pathways. The analysis of top DEMs could go further. What pathways are impacted beyond energy metabolism? Among the metabolic reactions allosterically regulated by top DEMs, what metabolic pathways are enriched?

   We identified the enriched metabolic pathways that were allosterically regulated by DEMs in AD using Fisher’s exact test. Alanine, aspartate, and glutamate metabolism pathways were significantly enriched in 2-oxoglutarate, glutarate, alanine, and glutamate-regulating metabolic reactions. Arginine and proline metabolism pathway was enriched in N-methyl-L-arginine and putrescine-regulating metabolic reactions. Arginine biosynthesis pathway was enriched in arginine-regulating metabolic reactions. Glycerophospholipid metabolism pathway was enriched in CDP-ethanolamine-regulating metabolic reactions. Glycine, serine, and threonine metabolism pathway was enriched in serine-regulating metabolic reactions. Purine metabolism pathway was enriched in AMP-regulating metabolic reactions. Pyrimidine metabolism pathway was enriched in deoxyuridine and thymidine-regulating metabolic reactions. Sphingolipid metabolism pathway was enriched in sphingosine-regulating metabolic reactions. However, this analysis did not yield sufficiently valuable insights into the regulatory relationships among biomolecules in AD. Thus, we did not include these results in the revised manuscript.

   (10) (Suggestion) Figure 3 shows that most differential signal in AD points to lower energy production due to the combination of differentially expressed metabolites and enzymes, but we are not given much context about the strength of these among all the differential signals. We would suggest including volcano plots where the features of interest, i.e. DE enzymes and metabolites, are colored differently (or a similar figure).

   We thank the reviewer for this constructive suggestion. To provide better context regarding the importance of the differential signals, we have added volcano plots for mRNAs, proteins, and metabolites in Figure S4A, B, and C.

   (11) (Suggestion) The PPI network could be better leveraged to understand metabolic changes in AD. If nodes are grouped into subnetworks (e.g. by Louvain / Leiden clustering) and tested for pathway enrichment, could you find functional subnetworks of coordinately up- and down- regulated metabolic enzymes? This could yield some pathways of interest beyond the energy metabolism pathways already highlighted.

   We appreciate the reviewer’s suggestion to utilize the PPI network for subnetwork analysis. However, it is important to note that the proteomic dataset analyzed in this study is derived from the original work of (Johnson et al., 2020). In that paper, the authors already performed a Weighted Gene Co-expression Network Analysis (WGCNA) across several datasets to identify co-expressed modules and functional pathways.

   Given this, we assumed that applying additional clustering methods to the same dataset would be unlikely to yield significant biological insights beyond the established findings.

   Minor comments

   (1). "All genes" and "all metabolites" should not be the background for the proteomic and metabolic pathway enrichment analysis by Metascape and MetaboAnalyst. The background should be limited to the proteins and metabolites that were measured.

   We fully agree with the reviewer that using “all gene” or “all metabolites” as a background is not suitable for enrichment analyses. As suggested, we have revised the enrichment analyses using the measured proteins and metabolites as a background in both Metascape and MetaboAnalyst (Fig. S4D).

   (2) Highlight the metabolic enzymes in Fig S2B. Calculate a separate correlation coefficient for the enzymes extracted in the energy metabolism analysis from Fig 3.

   We appreciate the reviewer’s suggestion to refine the correlation analysis. As requested, we have revised Fig. S2D to explicitly highlight the subset of enzymes involved in the energy metabolism analysis shown in Fig. 3. We calculated a separate correlation coefficient for the subset (Pearson coefficient = 0.86, p-value = 1.5e-7).

   (3) Use a multiple hypothesis adjusted p-value or q-value in Figure S3.

   We agree with the reviewer regarding the necessity of correcting for multiple comparisons. Accordingly, we have revised Fig. S4D using q-values.

   (4) Describe the methods used to calculate the logFC values from the validation dataset.

   We have revised the Methods to include a detailed description of the procedure used to calculate the log2FC values for the validation datasets (pg. 21, lines 13-15).

   (5) It is difficult to read Figure 3. We would recommend really emphasizing to the reader to refer to Fig S7B as a "key" to this figure. The description of the red/blue arrows and nodes in the methods section (pg. 24, lines 21-36, pg 25, lines 1-4) were also helpful, but very lengthy. We recommend putting an abridged version of this description into the Fig S7 figure legend.

   We appreciate the feedback regarding the readability of Fig. 3. As recommended, we have revised the manuscript to explicitly direct readers to Fig. S8B as an essential “key” for interpreting the network visualization (pg. 8, lines 28). Furthermore, we have added an abridged description of the network elements to the legend of Fig. S8B.

   (6) The S7 figure legend should refer to panels A and B, not E and F.

   We apologize for this oversight. We have corrected the legend of Fig. S8.

   (7) (Suggestion) Are any of the differentially expressed metabolites allosteric regulators of the DE transcription factors? This could be interesting to discuss.

   We appreciate the reviewer’s insightful suggestion about the potential allosteric regulation of the DETFs by DEMs. We conducted an extensive literature search to identify any reports related to this perspective. However, to the best of our knowledge, no such direct interactions have been reported to date.

   Reviewer #1 (Significance):

   The study's strength lies in leveraging three omics modalities across large patient cohorts (n ~ 150-240) to identify coherent signals between transcriptomics, proteomics, and metabolomics in postmortem DLPFC tissue. It was encouraging to see that the main result, showing downregulation for TCA, oxidative phosphorylation, and ketone body metabolism, emerged from consistent signals across both proteomics and metabolomics. This result was consistent with previous findings in other models cited by the author4,5 and other studies 6,7 demonstrating deficiency in energy-producing pathways in AD.

   Another strength of the study is the application of thoughtful methodology to connect differentially expressed proteins and metabolites via an intermediate data layer of metabolic reactions. The authors leverage the KEGG and BRENDA databases and apply sound logic to estimate the effects of enzyme level and metabolite level on pathway activity, with metabolites serving as substrate, product, or allosteric regulator for reactions. This trans-omic network methodology was developed in previous studies cited by the author8,9.

   However, as written, this study is limited in its contribution of new knowledge to the AD research field. The main conclusion (energy production is down in AD, due to regulatory disruption of energy metabolism) is not strongly supported (see comments 1, 3, and 4 for elaboration). The evidence could be improved by orthogonal approaches: further experimentation, further integration of external datasets, causal modeling, or flux modeling. Alternatively, even in the absence of new experimental and computational approaches, the story could be made more complete by further leveraging the trans-omic network to provide insights into (a) the regulation of energy metabolism; and (b) the impacts of key disrupted metabolites (see comments 7-9).

   The study is also limited in its demonstrating the power of these methodologies to provide integrative insights. As mentioned above, the integration of enzyme levels and metabolite levels is clearly useful (Figure 3). In contrast, the utility of the mRNA and transcription factor layers was not evident. The study did not appear to improve or expand upon trans-omic network methodology described in the previous works. Finally, the various analyses (analyzing the trans-omic network for nodes with the highest degree centrality, the PPI analysis, and viewing the energy metabolism pathways in the network) provided disparate results that were only tenuously connected in the discussion section.

   Reviewer #2 (Evidence, reproducibility and clarity):

   Summary

   This manuscript integrates public transcriptomic, proteomic, and metabolomic datasets from ROSMAP DLPFC samples to construct a multi-layer metabolic trans-omic network in Alzheimer's disease. By linking transcription factors, enzyme mRNAs, proteins, metabolic reactions, and metabolites, the authors report coordinated downregulation of the TCA cycle, oxidative phosphorylation, and ketone body metabolism, along with mixed regulatory signals in glycolysis/gluconeogenesis. They interpret these patterns as indicative of broad energetic dysfunction and alterations in amino-acid/nitrogen metabolism in AD. While the framework is conceptually appealing, much of the analysis remains descriptive, and several biological interpretations extend beyond what the data can robustly support. The reliance on bulk tissue without accounting for cell-type composition, limited covariate adjustment, and the absence of validation or sensitivity analyses reduce confidence in the mechanistic conclusions. Overall, the study provides a preliminary systems-level overview, but additional rigor is needed before the proposed trans-omic regulatory insights can be considered convincing.

   Major Comments

   (1) Interpretation requires more cautious phrasing, and validation is essential. The manuscript frequently asserts that specific pathways are "inhibited" or that energetic deficits are "compensated," but these conclusions extend beyond what the descriptive, bulk-level data can support. Because no metabolic flux, causality, or direct functional measurements are included, the results should be framed as putative regulatory shifts, not confirmed impairments. Critically, key claims about pathway inhibition would require flux modeling, perturbation analyses, or experimental validation to be convincing. Without such validation, the mechanistic interpretations remain speculative.

   We thank the reviewer for this crucial comment. We fully agree that, given the descriptive and bulk-level nature of our analysis, mechanistic interpretations must be made with caution. In the absence of direct metabolic flux measurements or experimental validation, our findings should be interpreted as putative regulatory shifts rather than confirmed functional impairments. Accordingly, we have revised the manuscript to temper mechanistic claims. We have replaced definitive statements with more speculative phrasing (e.g., “Our analysis revealed a putative coordinated downregulation …” instead of “Our analysis revealed a coordinated downregulation …” in Abstract section; “we demonstrate the systems-level view of the potential dysregulated energy production …” instead of “we demonstrate the systems-level view of the dysregulated energy production …” in pg. 10, lines 25-26).

   (2) Although the authors acknowledge this in the limitations, bulk-level differences may primarily reflect altered proportions of neurons, astrocytes, microglia, and oligodendrocytes rather than true within-cell-type regulation. Incorporating a cell-type deconvolution or performing a sensitivity analysis would substantially improve interpretability. This issue also impacts the trans-omic network: if the molecules included originate from different cell types, the inferred regulatory relationships may not reflect true intracellular processes.

   We appreciate the reviewer’s point that bulk-level differences can reflect altered proportions of different brain cell types, subsequently affecting the inferred trans-omic network analysis. To assess the changes in cell type proportions of the samples that we used in our study, we additionally used public single-cell transcriptomic datasets, which were obtained from DLPFC tissue of 465 subjects in the ROSMAP cohort (Green et al., 2024). For each omic data that we used in our analyses, we matched the same subjects and calculated the following cell type proportions, astrocytes, excitatory neurons, inhibitory neurons, microglias, oligodendrocytes, and OPCs. Then, we statistically compared the cell type proportions between control subjects and patients with AD (Fig. S3). In the transcriptomic data, we confirmed that the proportion of inhibitory neurons in the AD group was smaller than in the CT group, and that the proportion of oligodendrocytes in the AD group was larger than in the CT group. In the proteomic data, we did not observe any statistically significant changes in the cell type proportion between the two group. In the metabolomic data, we found that the proportion of inhibitory neurons in the AD group was smaller than in the CT group (pg. 6, lines 8-11).

   (3) Differential analysis covariates. For the differential expression analyses, only gender and PMI were included as covariates. Additional variables, such as age at death, RIN, neuropathological measures, and comorbidities, can strongly influence molecular profiles and should be considered to ensure that the observed differences reflect AD-related biology rather than confounding pathological or technical factors.

   We appreciate the reviewer’s comment regarding the included covariates in differential analyses of our study. The reason we did not include other variables, including age at death and RIN, is that these data for each sample were not available. Thus, we referred to original research articles from which proteomic or metabolomic datasets used in our study were derived. Regarding the metabolomic dataset, in the original article (Batra et al., 2023), only two metabolites, 1-methyl-5-imidazoleacetate and N6-carboxymethyllysine, were significantly associated with age. In addition, no metabolites were significantly associated with sex, BMI, or education. Regarding the proteomic dataset, in the original article, age at death, PMI, and sex were included as covariates in the analyses, though these variables were not found to strongly influence the data (Extended Data Fig.2 in (Johnson et al., 2020)).

   (4) Network stability and sample non-overlap. Proteomic, transcriptomic, and metabolomic data come from partially overlapping individuals. The authors should test whether the reconstructed network is robust to: different significance thresholds, restricting analyses to overlapping samples and alternative definitions of AD vs control.

   We appreciate the reviewer’s comment for the trans-omic network stability. In our study, the number of individuals for whom all omic modalities were measured was relatively small (n=25 in CT and n=35 in AD). This limited overlap reduces statistical power and can affect the downstream network construction. We have acknowledged this limitation in the revised manuscript and clarified that the reconstructed networks should be interpreted with caution regarding reproducibility and generalizability (pg. 13, lines 13-23).

   Minor Comments

   (1) Some TF enrichment and regulatory inferences lack explicit mention of multiple-testing correction.

   We apologize for the lack of clarity in our original description. We have corrected for multiple-testing for the TF inference. Thus, we have revised the Methods section to explicitly describe the correction method used and the threshold applied (pg. 23, lines 23-24).

   (2) The limitations section is strong but should explicitly discuss the influence of postmortem interval on metabolite levels.

   We appreciate the reviewer’s comment about the effect of postmortem interval on changes in metabolite levels. Accordingly, we have added the description of this perspective in our revised manuscript (pg. 13, lines 1-5).

   Reviewer #2 (Significance):

   The study extends a trans-omic integration framework, originally applied to metabolic disease, into the context of Alzheimer's pathology. Although the biological findings largely confirm known alterations in mitochondrial and energy metabolism, the network-based approach offers a structured way to view cross-layer regulatory changes. Its main advance is conceptual rather than biological, providing a unified framework rather than uncovering fundamentally new mechanisms. This work will primarily interest researchers in neurodegeneration and systems biology, as well as computational groups developing multi-omics integration methods.

   Reviewer #3 (Evidence, reproducibility and clarity):

   This study leverages existing transcriptomic, metabalomic and proteomic datasets from prefrontal cortex (PFC) to assess metabolic dysregulation in Alzheimer's disease (AD). They found a downregulation of multiple metabolic pathways, including TCA cycle, oxidative phosphorylation, and ketone metabolism, that may explain bioenergetic alterations in AD.

   The study used matching ROSMAP omics datasets from the DLPFC that have allowed more robust data integration. However, the datasets are all generated using bulk tissue, which makes data interpretation difficult. For example, the AD changes they observed may be due to shifts in cell type proportion with disease (e.g. cell death, neuron inflammation). Did the authors account for any potential shifts in cell type proportion in their analysis?

   If the assumption is that the changes in AD are cell intrinsic, which cell types are likely to be impacted? Can the authors integrate any existing single-cell analysis to infer which cell types may be driving the signals they detect, and whether this accounts for some of the antagonistic regulatory effects that were detected?

   We thank the reviewer for their insightful comments. We agree that the use of bulk tissue datasets cannot account for cell-type heterogeneity. As noted in our Limitations section (pg. 12, lines 24-27), we recognize that previous studies have found that the Braak stage is correlated positively with microglia and astrocyte proportions and negatively with oligodendrocyte proportion (Hannon et al., 2024; Shireby et al., 2022). Regarding the integration of single-cell analysis, we have referenced recent snRNA-seq findings (Mathys et al., 2024) in our Limitations section (pg. 12, lines 28-32) to deconvolve our bulk signatures.

   Furthermore, in our revised manuscript, we additionally used public single-cell transcriptomic datasets, which were obtained from DLPFC tissue of 465 subjects in the ROSMAP cohort (Green et al., 2024). For each omic data that we used in our analyses, we matched the same subjects and calculated the following cell type proportions, astrocytes, excitatory neurons, inhibitory neurons, microglia, oligodendrocytes, and OPCs. Then, we statistically compared the cell type proportions between control subjects and patients with AD (Fig. S3). In the transcriptomic data, we confirmed that the proportion of inhibitory neurons in the AD group was smaller than in the CT group, and that the proportion of oligodendrocytes in the AD group was larger than in the CT group. In the proteomic data, we did not observe any statistically significant changes in the cell type proportion between the two groups. In the metabolomic data, we found that the proportion of inhibitory neurons in the AD group was smaller than in the CT group (pg. 6, lines 8-11).

   Reviewer #3 (Significance):

   The manuscript provides multimodal insight into metabolic dysregulation in AD in the PFC. Given that metabolic dysfunction is likely to play a major in disease pathogenesis, this is a study of importance. However, the findings lack granularity at the cell type level, which limits the impact of the study.

   Reference

   (1) Baloni, P., Arnold, M., Buitrago, L., Nho, K., Moreno, H., Huynh, K., Brauner, B., Louie, G., Kueider-Paisley, A., Suhre, K., Saykin, A. J., Ekroos, K., Meikle, P. J., Hood, L., Price, N. D., Alzheimer’s Disease Metabolomics Consortium, Doraiswamy, P. M., Funk, C. C., Hernández, A. I., … Kaddurah-Daouk, R. (2022). Multi-Omic analyses characterize the ceramide/sphingomyelin pathway as a therapeutic target in Alzheimer’s disease. Communications Biology, 5(1), 1074.

   (2) Baloni, P., Funk, C. C., Yan, J., Yurkovich, J. T., Kueider-Paisley, A., Nho, K., Heinken, A., Jia, W., Mahmoudiandehkordi, S., Louie, G., Saykin, A. J., Arnold, M., Kastenmüller, G., Griffiths, W. J., Thiele, I., Alzheimer’s Disease Metabolomics Consortium, Kaddurah-Daouk, R., & Price, N. D. (2020). Metabolic Network Analysis Reveals Altered Bile Acid Synthesis and Metabolism in Alzheimer’s Disease. Cell Reports. Medicine, 1(8), 100138.

   (3) Batra, R., Arnold, M., Wörheide, M. A., Allen, M., Wang, X., Blach, C., Levey, A. I., Seyfried, N. T., Ertekin-Taner, N., Bennett, D. A., Kastenmüller, G., Kaddurah-Daouk, R. F., Krumsiek, J., & Alzheimer’s Disease Metabolomics Consortium (ADMC). (2023). The landscape of metabolic brain alterations in Alzheimer’s disease. Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association, 19(3), 980–998.

   (4) Batra, R., Krumsiek, J., Wang, X., Allen, M., Blach, C., Kastenmüller, G., Arnold, M., Ertekin-Taner, N., Kaddurah-Daouk, R., & Alzheimer’s Disease Metabolomics Consortium (ADMC). (2024). Comparative brain metabolomics reveals shared and distinct metabolic alterations in Alzheimer’s disease and progressive supranuclear palsy. Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association, 20(12), 8294–8307.

   (5) Cahill, K. M., Huo, Z., Tseng, G. C., Logan, R. W., & Seney, M. L. (2018). Improved identification of concordant and discordant gene expression signatures using an updated rank-rank hypergeometric overlap approach. Scientific Reports, 8(1), 9588.

   (6) Fröhlich, A. S., Gerstner, N., Gagliardi, M., Ködel, M., Yusupov, N., Matosin, N., Czamara, D., Sauer, S., Roeh, S., Murek, V., Chatzinakos, C., Daskalakis, N. P., Knauer-Arloth, J., Ziller, M. J., & Binder, E. B. (2024). Single-nucleus transcriptomic profiling of human orbitofrontal cortex reveals convergent effects of aging and psychiatric disease. Nature Neuroscience, 27(10), 2021–2032.

   (7) Green, G. S., Fujita, M., Yang, H.-S., Taga, M., Cain, A., McCabe, C., Comandante-Lou, N., White, C. C., Schmidtner, A. K., Zeng, L., Sigalov, A., Wang, Y., Regev, A., Klein, H.-U., Menon, V., Bennett, D. A., Habib, N., & De Jager, P. L. (2024). Cellular communities reveal trajectories of brain ageing and Alzheimer’s disease. Nature, 633(8030), 634–645.

   (8) Hannon, E., Dempster, E. L., Davies, J. P., Chioza, B., Blake, G. E. T., Burrage, J., Policicchio, S., Franklin, A., Walker, E. M., Bamford, R. A., Schalkwyk, L. C., & Mill, J. (2024). Quantifying the proportion of different cell types in the human cortex using DNA methylation profiles. BMC Biology, 22(1), 17.

   (9) Johnson, E. C. B., Carter, E. K., Dammer, E. B., Duong, D. M., Gerasimov, E. S., Liu, Y., Liu, J., Betarbet, R., Ping, L., Yin, L., Serrano, G. E., Beach, T. G., Peng, J., De Jager, P. L., Haroutunian, V., Zhang, B., Gaiteri, C., Bennett, D. A., Gearing, M., … Seyfried, N. T. (2022). Large-scale deep multi-layer analysis of Alzheimer’s disease brain reveals strong proteomic disease-related changes not observed at the RNA level. Nature Neuroscience, 25(2), 213–225.

   (10) Johnson, E. C. B., Dammer, E. B., Duong, D. M., Ping, L., Zhou, M., Yin, L., Higginbotham, L. A., Guajardo, A., White, B., Troncoso, J. C., Thambisetty, M., Montine, T. J., Lee, E. B., Trojanowski, J. Q., Beach, T. G., Reiman, E. M., Haroutunian, V., Wang, M., Schadt, E., … Seyfried, N. T. (2020). Large-scale proteomic analysis of Alzheimer’s disease brain and cerebrospinal fluid reveals early changes in energy metabolism associated with microglia and astrocyte activation. Nature Medicine, 26(5), 769–780.

   (11) Maitra, M., Mitsuhashi, H., Rahimian, R., Chawla, A., Yang, J., Fiori, L. M., Davoli, M. A., Perlman, K., Aouabed, Z., Mash, D. C., Suderman, M., Mechawar, N., Turecki, G., & Nagy, C. (2023). Cell type specific transcriptomic differences in depression show similar patterns between males and females but implicate distinct cell types and genes. Nature Communications, 14(1), 2912.

   (12) Mathys, H., Boix, C. A., Akay, L. A., Xia, Z., Davila-Velderrain, J., Ng, A. P., Jiang, X., Abdelhady, G., Galani, K., Mantero, J., Band, N., James, B. T., Babu, S., Galiana-Melendez, F., Louderback, K., Prokopenko, D., Tanzi, R. E., Bennett, D. A., Tsai, L.-H., & Kellis, M. (2024). Single-cell multiregion dissection of Alzheimer’s disease. Nature, 632(8026), 858–868.

   (13) Novotny, B. C., Fernandez, M. V., Wang, C., Budde, J. P., Bergmann, K., Eteleeb, A. M., Bradley, J., Webster, C., Ebl, C., Norton, J., Gentsch, J., Dube, U., Wang, F., Morris, J. C., Bateman, R. J., Perrin, R. J., McDade, E., Xiong, C., Chhatwal, J., … Harari, O. (2023). Metabolomic and lipidomic signatures in autosomal dominant and late-onset Alzheimer’s disease brains. Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association, 19(5), 1785–1799.

   (14) Plaisier, S. B., Taschereau, R., Wong, J. A., & Graeber, T. G. (2010). Rank-rank hypergeometric overlap: identification of statistically significant overlap between gene-expression signatures. Nucleic Acids Research, 38(17), e169.

   (15) Shireby, G., Dempster, E. L., Policicchio, S., Smith, R. G., Pishva, E., Chioza, B., Davies, J. P., Burrage, J., Lunnon, K., Seiler Vellame, D., Love, S., Thomas, A., Brookes, K., Morgan, K., Francis, P., Hannon, E., & Mill, J. (2022). DNA methylation signatures of Alzheimer’s disease neuropathology in the cortex are primarily driven by variation in non-neuronal cell-types. Nature Communications, 13(1), 5620.

   (16) Tasaki, S., Xu, J., Avey, D. R., Johnson, L., Petyuk, V. A., Dawe, R. J., Bennett, D. A., Wang, Y., & Gaiteri, C. (2022). Inferring protein expression changes from mRNA in Alzheimer’s dementia using deep neural networks. Nature Communications, 13(1), 655.

   (17) Varma, V. R., Wang, Y., An, Y., Varma, S., Bilgel, M., Doshi, J., Legido-Quigley, C., Delgado, J. C., Oommen, A. M., Roberts, J. A., Wong, D. F., Davatzikos, C., Resnick, S. M., Troncoso, J. C., Pletnikova, O., O’Brien, R., Hak, E., Baak, B. N., Pfeiffer, R., … Thambisetty, M. (2021). Bile acid synthesis, modulation, and dementia: A metabolomic, transcriptomic, and pharmacoepidemiologic study. PLoS Medicine, 18(5), e1003615.

   (18) Wan, Y.-W., Al-Ouran, R., Mangleburg, C. G., Perumal, T. M., Lee, T. V., Allison, K., Swarup, V., Funk, C. C., Gaiteri, C., Allen, M., Wang, M., Neuner, S. M., Kaczorowski, C. C., Philip, V. M., Howell, G. R., Martini-Stoica, H., Zheng, H., Mei, H., Zhong, X., … Logsdon, B. A. (2020). Meta-Analysis of the Alzheimer’s Disease Human Brain Transcriptome and Functional Dissection in Mouse Models. Cell Reports, 32(2), 107908.

   Read the original source
4. 

   Version published to 10.7554/elife.111228.1 on eLife

   Jul 13, 2026
5. 

   Version published to 10.7554/elife.111228 on eLife

   Jul 13, 2026
6. 

   Review Commons

   Mar 5, 2026

   Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Reply to the reviewers

   Manuscript number: RC-2025-03242R

   Corresponding author(s): Shinya Kuroda

   1. General Statements

   We appreciate the reviewers for the critical review of the manuscript and the valuable comments. We have carefully considered the reviewer's comments and have revised our manuscript accordingly.

   The reviewers' comments in this letter are in Bold and Italics.

   2. Point-by-point description of the revisions

   Response to Reviewer #1's Comments

   Evidence, reproducibility and clarity:

   Major comments

   * This study leaves out lipid metabolism as a major energy metabolism pathway relevant to AD. The authors themselves cite the significance of acylcarnitines and …

   Note: This response was posted by the corresponding author to Review Commons. The content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Reply to the reviewers

   Manuscript number: RC-2025-03242R

   Corresponding author(s): Shinya Kuroda

   1. General Statements

   We appreciate the reviewers for the critical review of the manuscript and the valuable comments. We have carefully considered the reviewer's comments and have revised our manuscript accordingly.

   The reviewers' comments in this letter are in Bold and Italics.

   2. Point-by-point description of the revisions

   Response to Reviewer #1's Comments

   Evidence, reproducibility and clarity:

   Major comments

   * This study leaves out lipid metabolism as a major energy metabolism pathway relevant to AD. The authors themselves cite the significance of acylcarnitines and CPT1A in AD (pg. 3, lines 32-33, pg. 4, lines 1-2). Lipid metabolism and homeostasis is known to be disrupted in AD1. Fatty acid oxidation is a known energy source in the prefrontal cortex2 and will also generate acetyl coA, which this study reveals is a significant decreased metabolite in AD. Furthermore, sphingomyelin emerges as one of the major decreased DEMs as well. Thus, lipid metabolism should be highlighted in Figure 3 and discussed throughout the manuscript; otherwise its omission should be clearly stated and justified.*

   * *

   We appreciate the reviewer's insightful comment regarding a critical role of lipid metabolism in AD. We recognize that lipid metabolism is a metabolic pathway deeply involved in AD pathology (Baloni et al., 2022, 2020; Varma et al., 2021). Accordingly, we have revised the Limitations section to more strongly emphasize its role as a vital energy source (pg. 13, lines 15-17). Regarding the visualization of lipid metabolism, we extracted lipid-related pathway from the trans-omic network but found that the regulatory relationships among DEPs and DEMs were excessively complex and interconnected. Thus, interpreting this regulatory network seemed to be more challenging compared to the other energy production pathways presented in our manuscript. Therefore, we have concluded that the pathway analysis in our trans-omic network may not be suitable for deeply elucidating the lipid dysregulation in AD. We have added a statement acknowledging this as a limitation of our current methodology in the revised manuscript (pg. 13, lines 13-22).

   * The covariates used for differential analysis should be discussed and justified. Notably, age is used as a covariate for transcriptomic analysis but not proteomic and metabolomic analysis, with no justification. Additionally, given the known importance of lipid metabolism in AD and the putative role of APOE in lipid homeostasis3, APOE genetic status should be considered as a covariate, or its omission should be justified.
   *

   We appreciate the reviewer's comment regarding the included covariates in differential analyses of our study. The reason we did not include other variables, such as age at death and RIN, is that these data were not available for each sample. Thus, we referred to the original research articles from which proteomic or metabolomic datasets used in our study were derived. Regarding the metabolomic dataset, in the original article (Batra et al., 2023), only two metabolites, 1-methyl-5-imidazoleacetate and N6-carboxymethyllysine, were significantly associated with age. In addition, no metabolites were significantly associated with sex, BMI, and years of education. Regarding the proteomic dataset, in the original article (Johnson et al., 2020), age at death, PMI, and sex were included as covariates in the analyses, though these variables were not found to strongly influence the data (Extended Data Fig.2 in (Johnson et al., 2020)).

   *The authors make a conclusion statement that suggests intervention: "Collectively, our data suggests that preserving or improving the ability to produce ATP and early intervention in the process of nitrogen metabolism are candidates for the prevention and treatment of dementia" (pg. 12, lines 12-14). This claim is not well-supported by the evidence provided in the study. There are a few limitations: (a) This was an observational, not interventional study; (b) The study did not establish whether the metabolic disruptions are causes or effects in AD; and (c) ATP or other bioenergetic indicators were not directly measured. Therefore, any statements about potential interventions should be removed or qualified as highly speculative. *

   * *

   We agree with the reviewer that the statement regarding potential interventions was not sufficiently supported by our analyses. Accordingly, we have removed the sentence regarding prevention and treatment from the revised manuscript (e.g., we have deleted final paragraph of the previous manuscript).

   *In conjunction with the last point, the main conclusion of the study is that energy production is down in AD. The data presented in Figure 3 are consistent with this conclusion, but it is far from definitive due to limitations stated above in comments 3a and 3b. The authors should offer additional support for this conclusion: experimental follow-up, flux modeling, analysis of alternative datasets with ATP measurement, causal inference.
   *

   We sincerely thank the reviewer for this valuable and constructive suggestion. Regarding flux modeling, we agree that metabolic flux analysis could provide important mechanistic insight. Indeed, previous studies have applied flux modeling in the context of lipid metabolism in Alzheimer's disease (Baloni et al., 2022). We also attempted to perform flux modeling focusing on energy metabolism. However, we found it difficult to obtain biologically meaningful and robust results and therefore decided not to include these analyses in the current manuscript.

   With respect to ATP measurements, we fully agree that direct evidence of altered ATP levels would further strengthen our conclusion. However, to the best of our knowledge, there are currently no publicly available large-scale datasets that directly measure ATP levels in human postmortem brain tissues. This limitation makes it challenging to incorporate validation in the present study.

   Regarding experimental follow-up, we agree that functional validation is essential to confirm the mechanistic implications of our findings. We are actively considering follow-up experimental studies. However, we consider the present work to be a multi-omic integrative analysis aimed at identifying key molecular alterations and generating biologically important hypotheses. We have revised the Limitation section to more clearly position this manuscript as an observational systems-level analysis (pg. 13, lines 20-22).

   *The validation analysis did not sufficiently show the generalizability of this study's results. The authors demonstrated a correlation of 0.53 to the MSBB transcriptomics data and 0.60 to the AMP-AD DiverseCohorts proteomics data. Beyond these correlation coefficients, no meaningful comparison between the datasets is offered. How concordant are the differentially expressed features (or pathways) between the datasets? How robust would the trans-omic network be if incorporating the alternate datasets? Is the main conclusion (energy metabolism is down in AD) supported by the validation datasets? We think this analysis should be expanded and described in the main text. **Although the results for external metabolomics datasets are reported in Fig S2C, correlation coefficients with the external data are not reported. The authors state, "Note that each study used different definitions for AD and CT groups, had variations in measurement methods and brain regions analyzed." We appreciate these limitations. However, the external data should be re-analyzed using the same definitions of AD and CT, if possible. The limitations and results (which DEMs are shared between datasets) should be discussed in the main text. *__

   * *

   We thank the reviewer for this important comment regarding the generalizability of our findings. In the revised manuscript, we have expanded the validation analyses and summarized the results in Figure S2. First, at the transcriptomic level, Figure S2B and S2C show the overlap between up- and downregulated genes in AD identified in our ROSMAP-derived analyses and those reported in a previously published large-scale meta-analysis of 2,114 postmortem samples across seven brain regions (Wan et al., 2020). A substantial proportion of DEGs were shared, supporting cross-cohort and cross-region robustness to some extent. At the proteomic level, Figure S2E shows a comparison between the ROSMAP and the AMP-AD DiverseCohorts datasets. We highlighted the subset of enzymes involved in the energy metabolism analysis shown in Fig. 3 and calculated a separate correlation coefficient for this subset (Pearson coefficient = 0.86, p-value = 1.5e-7), further supporting our main conclusion. In addition, to assess the concordance between the two datasets in a threshold-independent manner, we additionally performed Rank-Rank Hypergeometric Overlap (RRHO) analysis (Figure S2E). RRHO analysis (Cahill et al., 2018; Plaisier et al., 2010) enables the comparison of ranked protein lists without relying on arbitrary differential expression cutoffs and has been used for cross-dataset comparison in several previous studies (Fröhlich et al., 2024; Maitra et al., 2023). The RRHO heatmaps demonstrated significant enrichment in the concordant quadrants, confirming systematic agreement between datasets beyond simple correlation coefficients. For metabolomics, Figure S2G shows RRHO analyses comparing the ROSMAP metabolomic data with other datasets measured by the same UPLC-MS/MS platform (Batra et al., 2024; Novotny et al., 2023), demonstrating significant concordance in ranked metabolite changes in AD.

   __The glycolysis analysis and discussion needs more development. Glycolysis and gluconeogenesis share many of the same enzymes, but they are not the same pathway and should not be discussed as such. To make a claim about the overall influence of enzyme and metabolite levels on glycolysis, the authors should focus on the energetically committing steps of glycolysis (hexokinase, phosphofructokinase, pyruvate kinase) in Figure 3A, and include the full/current version of the figure in the supplement. Gluconeogenesis-specific enzymes (pyruvate carboxylase, PEPCK) are not mentioned at all - are they among the DEPs/DEGs?
   __

   We appreciate the reviewer's comment regarding the distinction between glycolysis and gluconeogenesis pathway. Among the gluconeogenesis-specific enzyme proteins, G6PC1, FBP1, PC, and PCK2 were measured in our dataset, but none of them were identified as DEPs. In addition, gluconeogenesis is a process that occurs primarily in the liver and kidney rather than the brain. Given this biological context and the lack of significant changes in relevant enzymes, we have revised the terminology throughout the manuscript, replacing "glycolysis/gluconeogenesis pathway" with "glycolysis pathway" in the revised version.

   *Given that there wasn't good concordance between the DEGs and DEPs, did including the mRNA and transcription factor layers in the network really add anything useful? It seems like the main conclusions of the manuscript were driven by the protein and metabolite layers only. How many of the DE metabolic enzymes were coregulated at the transcript and protein level? It would be useful to include the 5-layer trans-omic network in the supplement to display these results. Given your network, at what level does it appear that energy metabolism is regulated?
   *

   It is true that our primary conclusion regarding the regulation of energy metabolism is driven by the changes in protein and metabolite abundance. However, we consider the low concordance between mRNA and protein expression itself to be an important feature of AD pathology, as also reported in previous studies (Johnson et al., 2022; Tasaki et al., 2022). Although we did not perform a further analysis of this discordance, we believe that including the TF and mRNA layers into the metabolic trans-omic network strengthens a system-wide view of metabolic dysregulation in AD.

   Regarding the mRNA changes corresponding to the DEP enzymes, please refer to Figure S7A.

   *Comment further on the results from Figure 2D. What can be learned from identifying metabolites with the greatest degree centrality? What pathways other than energy metabolism are highlighted by the trans-omic network?
   *

   We assume that some energetic indicators, including AMP and acetyl-CoA, and nitrogen metabolism-related metabolites, Glu, 2-oxoglutarate, and urea, can be potential key regulators of dysregulated metabolism in AD.

   *(Suggestion) We suggest the authors leverage their trans-omic network in additional ways beyond giving a snapshot of a few energy metabolism pathways. The analysis of top DEMs could go further. What pathways are impacted beyond energy metabolism? Among the metabolic reactions allosterically regulated by top DEMs, what metabolic pathways are enriched?
   *

   We identified the enriched metabolic pathways that were allosterically regulated by DEMs in AD using Fisher's exact test. Alanine, aspartate, and glutamate metabolism pathways were significantly enriched in 2-oxoglutarate, glutarate, alanine, and glutamate-regulating metabolic reactions. Arginine and proline metabolism pathway was enriched in N-methyl-L-arginine and putrescine-regulating metabolic reactions. Arginine biosynthesis pathway was enriched in arginine-regulating metabolic reactions. Glycerophospholipid metabolism pathway was enriched in CDP-ethanolamine-regulating metabolic reactions. Glycine, serine, and threonine metabolism pathway was enriched in serine-regulating metabolic reactions. Purine metabolism pathway was enriched in AMP-regulating metabolic reactions. Pyrimidine metabolism pathway was enriched in deoxyuridine and thymidine-regulating metabolic reactions. Sphingolipid metabolism pathway was enriched in sphingosine-regulating metabolic reactions. However, this analysis did not yield sufficiently valuable insights into the regulatory relationships among biomolecules in AD. Thus, we did not include these results in the revised manuscript.

   *(Suggestion) Figure 3 shows that most differential signal in AD points to lower energy production due to the combination of differentially expressed metabolites and enzymes, but we are not given much context about the strength of these among all the differential signals. We would suggest including volcano plots where the features of interest, i.e. DE enzymes and metabolites, are colored differently (or a similar figure).
   *

   We thank the reviewer for this constructive suggestion. To provide better context regarding the importance of the differential signals, we have added volcano plots for mRNAs, proteins, and metabolites in Figure S4A, B, and C.

   *(Suggestion) The PPI network could be better leveraged to understand metabolic changes in AD. If nodes are grouped into subnetworks (e.g. by Louvain / Leiden clustering) and tested for pathway enrichment, could you find functional subnetworks of coordinately up- and down- regulated metabolic enzymes? This could yield some pathways of interest beyond the energy metabolism pathways already highlighted.
   *

   We appreciate the reviewer's suggestion to utilize the PPI network for subnetwork analysis. However, it is important to note that the proteomic dataset analyzed in this study is derived from the original work of (Johnson et al., 2020). In that paper, the authors already performed a Weighted Gene Co-expression Network Analysis (WGCNA) across several datasets to identify co-expressed modules and functional pathways.

   Given this, we assumed that applying additional clustering methods to the same dataset would be unlikely to yield significant biological insights beyond the established findings.

   __* *____Minor comments __

   *12. "All genes" and "all metabolites" should not be the background for the proteomic and metabolic pathway enrichment analysis by Metascape and MetaboAnalyst. The background should be limited to the proteins and metabolites that were measured. *

   We fully agree with the reviewer that using "all gene" or "all metabolites" as a background is not suitable for enrichment analyses. As suggested, we have revised the enrichment analyses using the measured proteins and metabolites as a background in both Metascape and MetaboAnalyst (Fig. S4D).

   *Highlight the metabolic enzymes in Fig S2B. Calculate a separate correlation coefficient for the enzymes extracted in the energy metabolism analysis from Fig 3.
   *

   We appreciate the reviewer's suggestion to refine the correlation analysis. As requested, we have revised Fig. S2D to explicitly highlight the subset of enzymes involved in the energy metabolism analysis shown in Fig. 3. We calculated a separate correlation coefficient for the subset (Pearson coefficient = 0.86, p-value = 1.5e-7).

   *Use a multiple hypothesis adjusted p-value or q-value in Figure S3.
   *

   We agree with the reviewer regarding the necessity of correcting for multiple comparisons. Accordingly, we have revised Fig. S4D using q-values.

   *Describe the methods used to calculate the logFC values from the validation dataset.
   *

   We have revised the Methods to include a detailed description of the procedure used to calculate the log2FC values for the validation datasets (pg. 21, lines 13-15).

   *It is difficult to read Figure 3. We would recommend really emphasizing to the reader to refer to Fig S7B as a "key" to this figure. The description of the red/blue arrows and nodes in the methods section (pg. 24, lines 21-36, pg 25, lines 1-4) were also helpful, but very lengthy. We recommend putting an abridged version of this description into the Fig S7 figure legend.
   *

   We appreciate the feedback regarding the readability of Fig. 3. As recommended, we have revised the manuscript to explicitly direct readers to Fig. S8B as an essential "key" for interpreting the network visualization (pg. 8, lines 28). Furthermore, we have added an abridged description of the network elements to the legend of Fig. S8B.

   *The S7 figure legend should refer to panels A and B, not E and F.
   *

   We apologize for this oversight. We have corrected the legend of Fig. S8.

   *(Suggestion) Are any of the differentially expressed metabolites allosteric regulators of the DE transcription factors? This could be interesting to discuss.
   *

   We appreciate the reviewer's insightful suggestion about the potential allosteric regulation of the DETFs by DEMs. We conducted an extensive literature search to identify any reports related to this perspective. However, to the best of our knowledge, no such direct interactions have been reported to date.

   Significance:* *

   *The study's strength lies in leveraging three omics modalities across large patient cohorts (n ~ 150-240) to identify coherent signals between transcriptomics, proteomics, and metabolomics in postmortem DLPFC tissue. It was encouraging to see that the main result, showing downregulation for TCA, oxidative phosphorylation, and ketone body metabolism, emerged from consistent signals across both proteomics and metabolomics. This result was consistent with previous findings in other models cited by the author4,5 and other studies 6,7 demonstrating deficiency in energy-producing pathways in AD. Another strength of the study is the application of thoughtful methodology to connect differentially expressed proteins and metabolites via an intermediate data layer of metabolic reactions. The authors leverage the KEGG and BRENDA databases and apply sound logic to estimate the effects of enzyme level and metabolite level on pathway activity, with metabolites serving as substrate, product, or allosteric regulator for reactions. This trans-omic network methodology was developed in previous studies cited by the author8,9. However, as written, this study is limited in its contribution of new knowledge to the AD research field. The main conclusion (energy production is down in AD, due to regulatory disruption of energy metabolism) is not strongly supported (see comments 1, 3, and 4 for elaboration). The evidence could be improved by orthogonal approaches: further experimentation, further integration of external datasets, causal modeling, or flux modeling. Alternatively, even in the absence of new experimental and computational approaches, the story could be made more complete by further leveraging the trans-omic network to provide insights into (a) the regulation of energy metabolism; and (b) the impacts of key disrupted metabolites (see comments 7-9). The study is also limited in its demonstrating the power of these methodologies to provide integrative insights. As mentioned above, the integration of enzyme levels and metabolite levels is clearly useful (Figure 3). In contrast, the utility of the mRNA and transcription factor layers was not evident. The study did not appear to improve or expand upon trans-omic network methodology described in the previous works. Finally, the various analyses (analyzing the trans-omic network for nodes with the highest degree centrality, the PPI analysis, and viewing the energy metabolism pathways in the network) provided disparate results that were only tenuously connected in the discussion section. *

   ---

   Response to Reviewer #2's Comments____* *

   *Evidence, reproducibility and clarity: Summary *

   *This manuscript integrates public transcriptomic, proteomic, and metabolomic datasets from ROSMAP DLPFC samples to construct a multi-layer metabolic trans-omic network in Alzheimer's disease. By linking transcription factors, enzyme mRNAs, proteins, metabolic reactions, and metabolites, the authors report coordinated downregulation of the TCA cycle, oxidative phosphorylation, and ketone body metabolism, along with mixed regulatory signals in glycolysis/gluconeogenesis. They interpret these patterns as indicative of broad energetic dysfunction and alterations in amino-acid/nitrogen metabolism in AD. While the framework is conceptually appealing, much of the analysis remains descriptive, and several biological interpretations extend beyond what the data can robustly support. The reliance on bulk tissue without accounting for cell-type composition, limited covariate adjustment, and the absence of validation or sensitivity analyses reduce confidence in the mechanistic conclusions. Overall, the study provides a preliminary systems-level overview, but additional rigor is needed before the proposed trans-omic regulatory insights can be considered convincing. *

   *Major Comments *

   *1. Interpretation requires more cautious phrasing, and validation is essential. The manuscript frequently asserts that specific pathways are "inhibited" or that energetic deficits are "compensated," but these conclusions extend beyond what the descriptive, bulk-level data can support. Because no metabolic flux, causality, or direct functional measurements are included, the results should be framed as putative regulatory shifts, not confirmed impairments. Critically, key claims about pathway inhibition would require flux modeling, perturbation analyses, or experimental validation to be convincing. Without such validation, the mechanistic interpretations remain speculative. *

   We thank the reviewer for this crucial comment. We fully agree that, given the descriptive and bulk-level nature of our analysis, mechanistic interpretations must be made with caution. In the absence of direct metabolic flux measurements or experimental validation, our findings should be interpreted as putative regulatory shifts rather than confirmed functional impairments. Accordingly, we have revised the manuscript to temper mechanistic claims. We have replaced definitive statements with more speculative phrasing (e.g., "Our analysis revealed a putative coordinated downregulation ..." instead of "Our analysis revealed a coordinated downregulation ..." in Abstract section; "we demonstrate the systems-level view of the potential dysregulated energy production ..." instead of "we demonstrate the systems-level view of the dysregulated energy production ..." in pg. 10, lines 25-26).

   * Although the authors acknowledge this in the limitations, bulk-level differences may primarily reflect altered proportions of neurons, astrocytes, microglia, and oligodendrocytes rather than true within-cell-type regulation. Incorporating a cell-type deconvolution or performing a sensitivity analysis would substantially improve interpretability. This issue also impacts the trans-omic network: if the molecules included originate from different cell types, the inferred regulatory relationships may not reflect true intracellular processes. *

   We appreciate the reviewer's point that bulk-level differences can reflect altered proportions of different brain cell types, subsequently affecting the inferred trans-omic network analysis. To assess the changes in cell type proportions of the samples that we used in our study, we additionally used public single-cell transcriptomic datasets, which were obtained from DLPFC tissue of 465 subjects in the ROSMAP cohort (Green et al., 2024). For each omic data that we used in our analyses, we matched the same subjects and calculated the following cell type proportions, astrocytes, excitatory neurons, inhibitory neurons, microglias, oligodendrocytes, and OPCs. Then, we statistically compared the cell type proportions between control subjects and patients with AD (Fig. S3). In the transcriptomic data, we confirmed that the proportion of inhibitory neurons in the AD group was smaller than in the CT group, and that the proportion of oligodendrocytes in the AD group was larger than in the CT group. In the proteomic data, we did not observe any statistically significant changes in the cell type proportion between the two group. In the metabolomic data, we found that the proportion of inhibitory neurons in the AD group was smaller than in the CT group (pg. 6, lines 8-11).

   *Differential analysis covariates. For the differential expression analyses, only gender and PMI were included as covariates. Additional variables, such as age at death, RIN, neuropathological measures, and comorbidities, can strongly influence molecular profiles and should be considered to ensure that the observed differences reflect AD-related biology rather than confounding pathological or technical factors. *

   * *

   We appreciate the reviewer's comment regarding the included covariates in differential analyses of our study. The reason we did not include other variables, including age at death and RIN, is that these data for each sample were not available. Thus, we referred to original research articles from which proteomic or metabolomic datasets used in our study were derived. Regarding the metabolomic dataset, in the original article (Batra et al., 2023), only two metabolites, 1-methyl-5-imidazoleacetate and N6-carboxymethyllysine, were significantly associated with age. In addition, no metabolites were significantly associated with sex, BMI, or education. Regarding the proteomic dataset, in the original article, age at death, PMI, and sex were included as covariates in the analyses, though these variables were not found to strongly influence the data (Extended Data Fig.2 in (Johnson et al., 2020)).

   *Network stability and sample non-overlap. Proteomic, transcriptomic, and metabolomic data come from partially overlapping individuals. The authors should test whether the reconstructed network is robust to: different significance thresholds, restricting analyses to overlapping samples and alternative definitions of AD vs control. *

   __* __We appreciate the reviewer's comment for the trans-omic network stability. In our study, the number of individuals for whom all omic modalities were measured was relatively small (n=25 in CT and n=35 in AD). This limited overlap reduces statistical power and can affect the downstream network construction. We have acknowledged this limitation in the revised manuscript and clarified that the reconstructed networks should be interpreted with caution regarding reproducibility and generalizability (pg. 13, lines 13-23*).

   *Minor Comments *

   *1. Some TF enrichment and regulatory inferences lack explicit mention of multiple-testing correction. *

   We apologize for the lack of clarity in our original description. We have corrected for multiple-testing for the TF inference. Thus, we have revised the Methods section to explicitly describe the correction method used and the threshold applied (pg. 23, lines 23-24).

   * *

   * The limitations section is strong but should explicitly discuss the influence of postmortem interval on metabolite levels.
   *

   We appreciate the reviewer's comment about the effect of postmortem interval on changes in metabolite levels. Accordingly, we have added the description of this perspective in our revised manuscript (pg. 13, lines 1-5).

   __*Reviewer #2 (Significance (Required)):

   Significance *__

   *The study extends a trans-omic integration framework, originally applied to metabolic disease, into the context of Alzheimer's pathology. Although the biological findings largely confirm known alterations in mitochondrial and energy metabolism, the network-based approach offers a structured way to view cross-layer regulatory changes. Its main advance is conceptual rather than biological, providing a unified framework rather than uncovering fundamentally new mechanisms. This work will primarily interest researchers in neurodegeneration and systems biology, as well as computational groups developing multi-omics integration methods. *

   * *

   Response to Reviewer #3's Comments

   ---

   Evidence, reproducibility and clarity

   This study leverages existing transcriptomic, metabalomic and proteomic datasets from prefrontal cortex (PFC) to assess metabolic dysregulation in Alzheimer's disease (AD). They found a downregulation of multiple metabolic pathways, including TCA cycle, oxidative phosphorylation, and ketone metabolism, that may explain bioenergetic alterations in AD.* __ The study used matching ROSMAP omics datasets from the DLPFC that have allowed more robust data integration. However, the datasets are all generated using bulk tissue, which makes data interpretation difficult. For example, the AD changes they observed may be due to shifts in cell type proportion with disease (e.g. cell death, neuron inflammation). Did the authors account for any potential shifts in cell type proportion in their analysis?__ *

   __If the assumption is that the changes in AD are cell intrinsic, which cell types are likely to be impacted? Can the authors integrate any existing single-cell analysis to infer which cell types may be driving the signals they detect, and whether this accounts for some of the antagonistic regulatory effects that were detected?______

   We thank the reviewer for their insightful comments. We agree that the use of bulk tissue datasets cannot account for cell-type heterogeneity. As noted in our Limitations section (pg. 12, lines 24-27), we recognize that previous studies have found that the Braak stage is correlated positively with microglia and astrocyte proportions and negatively with oligodendrocyte proportion (Hannon et al., 2024; Shireby et al., 2022). Regarding the integration of single-cell analysis, we have referenced recent snRNA-seq findings (Mathys et al., 2024) in our Limitations section (pg. 12, lines 28-32) to deconvolve our bulk signatures.

   Furthermore, in our revised manuscript, we additionally used public single-cell transcriptomic datasets, which were obtained from DLPFC tissue of 465 subjects in the ROSMAP cohort (Green et al., 2024). For each omic data that we used in our analyses, we matched the same subjects and calculated the following cell type proportions, astrocytes, excitatory neurons, inhibitory neurons, microglia, oligodendrocytes, and OPCs. Then, we statistically compared the cell type proportions between control subjects and patients with AD (Fig. S3). In the transcriptomic data, we confirmed that the proportion of inhibitory neurons in the AD group was smaller than in the CT group, and that the proportion of oligodendrocytes in the AD group was larger than in the CT group. In the proteomic data, we did not observe any statistically significant changes in the cell type proportion between the two groups. In the metabolomic data, we found that the proportion of inhibitory neurons in the AD group was smaller than in the CT group (pg. 6, lines 8-11).

   Significance

   The manuscript provides multimodal insight into metabolic dysregulation in AD in the PFC. Given that metabolic dysfunction is likely to play a major in disease pathogenesis, this is a study of importance. However, the findings lack granularity at the cell type level, which limits the impact of the study.

   Reference

   1. Baloni, P., Arnold, M., Buitrago, L., Nho, K., Moreno, H., Huynh, K., Brauner, B., Louie, G., Kueider-Paisley, A., Suhre, K., Saykin, A. J., Ekroos, K., Meikle, P. J., Hood, L., Price, N. D., Alzheimer's Disease Metabolomics Consortium, Doraiswamy, P. M., Funk, C. C., Hernández, A. I., ... Kaddurah-Daouk, R. (2022). Multi-Omic analyses characterize the ceramide/sphingomyelin pathway as a therapeutic target in Alzheimer's disease. Communications Biology, 5(1), 1074.
   2. Baloni, P., Funk, C. C., Yan, J., Yurkovich, J. T., Kueider-Paisley, A., Nho, K., Heinken, A., Jia, W., Mahmoudiandehkordi, S., Louie, G., Saykin, A. J., Arnold, M., Kastenmüller, G., Griffiths, W. J., Thiele, I., Alzheimer's Disease Metabolomics Consortium, Kaddurah-Daouk, R., & Price, N. D. (2020). Metabolic Network Analysis Reveals Altered Bile Acid Synthesis and Metabolism in Alzheimer's Disease. Cell Reports. Medicine, 1(8), 100138.
   3. Batra, R., Arnold, M., Wörheide, M. A., Allen, M., Wang, X., Blach, C., Levey, A. I., Seyfried, N. T., Ertekin-Taner, N., Bennett, D. A., Kastenmüller, G., Kaddurah-Daouk, R. F., Krumsiek, J., & Alzheimer's Disease Metabolomics Consortium (ADMC). (2023). The landscape of metabolic brain alterations in Alzheimer's disease. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 19(3), 980-998.
   4. Batra, R., Krumsiek, J., Wang, X., Allen, M., Blach, C., Kastenmüller, G., Arnold, M., Ertekin-Taner, N., Kaddurah-Daouk, R., & Alzheimer's Disease Metabolomics Consortium (ADMC). (2024). Comparative brain metabolomics reveals shared and distinct metabolic alterations in Alzheimer's disease and progressive supranuclear palsy. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 20(12), 8294-8307.
   5. Cahill, K. M., Huo, Z., Tseng, G. C., Logan, R. W., & Seney, M. L. (2018). Improved identification of concordant and discordant gene expression signatures using an updated rank-rank hypergeometric overlap approach. Scientific Reports, 8(1), 9588.
   6. Fröhlich, A. S., Gerstner, N., Gagliardi, M., Ködel, M., Yusupov, N., Matosin, N., Czamara, D., Sauer, S., Roeh, S., Murek, V., Chatzinakos, C., Daskalakis, N. P., Knauer-Arloth, J., Ziller, M. J., & Binder, E. B. (2024). Single-nucleus transcriptomic profiling of human orbitofrontal cortex reveals convergent effects of aging and psychiatric disease. Nature Neuroscience, 27(10), 2021-2032.
   7. Green, G. S., Fujita, M., Yang, H.-S., Taga, M., Cain, A., McCabe, C., Comandante-Lou, N., White, C. C., Schmidtner, A. K., Zeng, L., Sigalov, A., Wang, Y., Regev, A., Klein, H.-U., Menon, V., Bennett, D. A., Habib, N., & De Jager, P. L. (2024). Cellular communities reveal trajectories of brain ageing and Alzheimer's disease. Nature, 633(8030), 634-645.
   8. Hannon, E., Dempster, E. L., Davies, J. P., Chioza, B., Blake, G. E. T., Burrage, J., Policicchio, S., Franklin, A., Walker, E. M., Bamford, R. A., Schalkwyk, L. C., & Mill, J. (2024). Quantifying the proportion of different cell types in the human cortex using DNA methylation profiles. BMC Biology, 22(1), 17.
   9. Johnson, E. C. B., Carter, E. K., Dammer, E. B., Duong, D. M., Gerasimov, E. S., Liu, Y., Liu, J., Betarbet, R., Ping, L., Yin, L., Serrano, G. E., Beach, T. G., Peng, J., De Jager, P. L., Haroutunian, V., Zhang, B., Gaiteri, C., Bennett, D. A., Gearing, M., ... Seyfried, N. T. (2022). Large-scale deep multi-layer analysis of Alzheimer's disease brain reveals strong proteomic disease-related changes not observed at the RNA level. Nature Neuroscience, 25(2), 213-225.
   10. Johnson, E. C. B., Dammer, E. B., Duong, D. M., Ping, L., Zhou, M., Yin, L., Higginbotham, L. A., Guajardo, A., White, B., Troncoso, J. C., Thambisetty, M., Montine, T. J., Lee, E. B., Trojanowski, J. Q., Beach, T. G., Reiman, E. M., Haroutunian, V., Wang, M., Schadt, E., ... Seyfried, N. T. (2020). Large-scale proteomic analysis of Alzheimer's disease brain and cerebrospinal fluid reveals early changes in energy metabolism associated with microglia and astrocyte activation. Nature Medicine, 26(5), 769-780.
   11. Maitra, M., Mitsuhashi, H., Rahimian, R., Chawla, A., Yang, J., Fiori, L. M., Davoli, M. A., Perlman, K., Aouabed, Z., Mash, D. C., Suderman, M., Mechawar, N., Turecki, G., & Nagy, C. (2023). Cell type specific transcriptomic differences in depression show similar patterns between males and females but implicate distinct cell types and genes. Nature Communications, 14(1), 2912.
   12. Mathys, H., Boix, C. A., Akay, L. A., Xia, Z., Davila-Velderrain, J., Ng, A. P., Jiang, X., Abdelhady, G., Galani, K., Mantero, J., Band, N., James, B. T., Babu, S., Galiana-Melendez, F., Louderback, K., Prokopenko, D., Tanzi, R. E., Bennett, D. A., Tsai, L.-H., & Kellis, M. (2024). Single-cell multiregion dissection of Alzheimer's disease. Nature, 632(8026), 858-868.
   13. Novotny, B. C., Fernandez, M. V., Wang, C., Budde, J. P., Bergmann, K., Eteleeb, A. M., Bradley, J., Webster, C., Ebl, C., Norton, J., Gentsch, J., Dube, U., Wang, F., Morris, J. C., Bateman, R. J., Perrin, R. J., McDade, E., Xiong, C., Chhatwal, J., ... Harari, O. (2023). Metabolomic and lipidomic signatures in autosomal dominant and late-onset Alzheimer's disease brains. Alzheimer's & Dementia: The Journal of the Alzheimer's Association, 19(5), 1785-1799.
   14. Plaisier, S. B., Taschereau, R., Wong, J. A., & Graeber, T. G. (2010). Rank-rank hypergeometric overlap: identification of statistically significant overlap between gene-expression signatures. Nucleic Acids Research, 38(17), e169.
   15. Shireby, G., Dempster, E. L., Policicchio, S., Smith, R. G., Pishva, E., Chioza, B., Davies, J. P., Burrage, J., Lunnon, K., Seiler Vellame, D., Love, S., Thomas, A., Brookes, K., Morgan, K., Francis, P., Hannon, E., & Mill, J. (2022). DNA methylation signatures of Alzheimer's disease neuropathology in the cortex are primarily driven by variation in non-neuronal cell-types. Nature Communications, 13(1), 5620.
   16. Tasaki, S., Xu, J., Avey, D. R., Johnson, L., Petyuk, V. A., Dawe, R. J., Bennett, D. A., Wang, Y., & Gaiteri, C. (2022). Inferring protein expression changes from mRNA in Alzheimer's dementia using deep neural networks. Nature Communications, 13(1), 655.
   17. Varma, V. R., Wang, Y., An, Y., Varma, S., Bilgel, M., Doshi, J., Legido-Quigley, C., Delgado, J. C., Oommen, A. M., Roberts, J. A., Wong, D. F., Davatzikos, C., Resnick, S. M., Troncoso, J. C., Pletnikova, O., O'Brien, R., Hak, E., Baak, B. N., Pfeiffer, R., ... Thambisetty, M. (2021). Bile acid synthesis, modulation, and dementia: A metabolomic, transcriptomic, and pharmacoepidemiologic study. PLoS Medicine, 18(5), e1003615.
   18. Wan, Y.-W., Al-Ouran, R., Mangleburg, C. G., Perumal, T. M., Lee, T. V., Allison, K., Swarup, V., Funk, C. C., Gaiteri, C., Allen, M., Wang, M., Neuner, S. M., Kaczorowski, C. C., Philip, V. M., Howell, G. R., Martini-Stoica, H., Zheng, H., Mei, H., Zhong, X., ... Logsdon, B. A. (2020). Meta-Analysis of the Alzheimer's Disease Human Brain Transcriptome and Functional Dissection in Mouse Models. Cell Reports, 32(2), 107908.

   Read the original source
7. 

   Review Commons

   Mar 5, 2026

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #3

   Evidence, reproducibility and clarity

   This study leverages existing transcriptomic, metabalomic and proteomic datasets from prefrontal cortex (PFC) to assess metabolic dysregulation in Alzheimer's disease (AD). They found a downregulation of multiple metabolic pathways, including TCA cycle, oxidative phosphorylation, and ketone metabolism, that may explain bioenergetic alterations in AD.

   The study used matching ROSMAP omics datasets from the DLPFC that have allowed more robust data integration. However, the datasets are all generated using bulk tissue, which makes data interpretation difficult. For example, the AD changes they observed may be due to shifts in cell type …

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #3

   Evidence, reproducibility and clarity

   This study leverages existing transcriptomic, metabalomic and proteomic datasets from prefrontal cortex (PFC) to assess metabolic dysregulation in Alzheimer's disease (AD). They found a downregulation of multiple metabolic pathways, including TCA cycle, oxidative phosphorylation, and ketone metabolism, that may explain bioenergetic alterations in AD.

   The study used matching ROSMAP omics datasets from the DLPFC that have allowed more robust data integration. However, the datasets are all generated using bulk tissue, which makes data interpretation difficult. For example, the AD changes they observed may be due to shifts in cell type proportion with disease (e.g. cell death, neuron inflammation). Did the authors account for any potential shifts in cell type proportion in their analysis?

   If the assumption is that the changes in AD are cell intrinsic, which cell types are likely to be impacted? Can the authors integrate any existing single-cell analysis to infer which cell types may be driving the signals they detect, and whether this accounts for some of the antagonistic regulatory effects that were detected?

   Significance

   The manuscript provides multimodal insight into metabolic dysregulation in AD in the PFC. Given that metabolic dysfunction is likely to play a major in disease pathogenesis, this is a study of importance. However, the findings lack granularity at the cell type level, which limits the impact of the study.

   Read the original source
8. 

   Review Commons

   Mar 5, 2026

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #2

   Evidence, reproducibility and clarity

   Summary

   This manuscript integrates public transcriptomic, proteomic, and metabolomic datasets from ROSMAP DLPFC samples to construct a multi-layer metabolic trans-omic network in Alzheimer's disease. By linking transcription factors, enzyme mRNAs, proteins, metabolic reactions, and metabolites, the authors report coordinated downregulation of the TCA cycle, oxidative phosphorylation, and ketone body metabolism, along with mixed regulatory signals in glycolysis/gluconeogenesis. They interpret these patterns as indicative of broad energetic dysfunction and alterations in amino-acid/nitrogen metabolism in AD. While the framework is …

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #2

   Evidence, reproducibility and clarity

   Summary

   This manuscript integrates public transcriptomic, proteomic, and metabolomic datasets from ROSMAP DLPFC samples to construct a multi-layer metabolic trans-omic network in Alzheimer's disease. By linking transcription factors, enzyme mRNAs, proteins, metabolic reactions, and metabolites, the authors report coordinated downregulation of the TCA cycle, oxidative phosphorylation, and ketone body metabolism, along with mixed regulatory signals in glycolysis/gluconeogenesis. They interpret these patterns as indicative of broad energetic dysfunction and alterations in amino-acid/nitrogen metabolism in AD. While the framework is conceptually appealing, much of the analysis remains descriptive, and several biological interpretations extend beyond what the data can robustly support. The reliance on bulk tissue without accounting for cell-type composition, limited covariate adjustment, and the absence of validation or sensitivity analyses reduce confidence in the mechanistic conclusions. Overall, the study provides a preliminary systems-level overview, but additional rigor is needed before the proposed trans-omic regulatory insights can be considered convincing.

   Major Comments

   1. Interpretation requires more cautious phrasing, and validation is essential. The manuscript frequently asserts that specific pathways are "inhibited" or that energetic deficits are "compensated," but these conclusions extend beyond what the descriptive, bulk-level data can support. Because no metabolic flux, causality, or direct functional measurements are included, the results should be framed as putative regulatory shifts, not confirmed impairments. Critically, key claims about pathway inhibition would require flux modeling, perturbation analyses, or experimental validation to be convincing. Without such validation, the mechanistic interpretations remain speculative.
   2. Although the authors acknowledge this in the limitations, bulk-level differences may primarily reflect altered proportions of neurons, astrocytes, microglia, and oligodendrocytes rather than true within-cell-type regulation. Incorporating a cell-type deconvolution or performing a sensitivity analysis would substantially improve interpretability. This issue also impacts the trans-omic network: if the molecules included originate from different cell types, the inferred regulatory relationships may not reflect true intracellular processes.
   3. Differential analysis covariates. For the differential expression analyses, only gender and PMI were included as covariates. Additional variables, such as age at death, RIN, neuropathological measures, and comorbidities, can strongly influence molecular profiles and should be considered to ensure that the observed differences reflect AD-related biology rather than confounding pathological or technical factors.
   4. Network stability and sample non-overlap. Proteomic, transcriptomic, and metabolomic data come from partially overlapping individuals. The authors should test whether the reconstructed network is robust to: different significance thresholds, restricting analyses to overlapping samples and alternative definitions of AD vs control.

   Minor Comments

   1. Some TF enrichment and regulatory inferences lack explicit mention of multiple-testing correction.
   2. The limitations section is strong but should explicitly discuss the influence of postmortem interval on metabolite levels.

   Significance

   The study extends a trans-omic integration framework, originally applied to metabolic disease, into the context of Alzheimer's pathology. Although the biological findings largely confirm known alterations in mitochondrial and energy metabolism, the network-based approach offers a structured way to view cross-layer regulatory changes. Its main advance is conceptual rather than biological, providing a unified framework rather than uncovering fundamentally new mechanisms. This work will primarily interest researchers in neurodegeneration and systems biology, as well as computational groups developing multi-omics integration methods.

   Reviewer expertise

   My background is in Alzheimer's disease, multi-omics integration, and computational systems biology. I am not a specialist in enzymology or allosteric regulation and therefore cannot fully evaluate the biochemical specificity of those annotations.

   Read the original source
9. 

   Review Commons

   Mar 5, 2026

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #1

   Evidence, reproducibility and clarity

   Major comments

   1. This study leaves out lipid metabolism as a major energy metabolism pathway relevant to AD. The authors themselves cite the significance of acylcarnitines and CPT1A in AD (pg. 3, lines 32-33, pg. 4, lines 1-2). Lipid metabolism and homeostasis is known to be disrupted in AD1. Fatty acid oxidation is a known energy source in the prefrontal cortex2 and will also generate acetyl coA, which this study reveals is a significant decreased metabolite in AD. Furthermore, sphingomyelin emerges as one of the major decreased DEMs as well. Thus, lipid metabolism should be highlighted in Figure 3 and discussed throughout the …

   Note: This preprint has been reviewed by subject experts for Review Commons. Content has not been altered except for formatting.

   Learn more at Review Commons

   ---

   Referee #1

   Evidence, reproducibility and clarity

   Major comments

   1. This study leaves out lipid metabolism as a major energy metabolism pathway relevant to AD. The authors themselves cite the significance of acylcarnitines and CPT1A in AD (pg. 3, lines 32-33, pg. 4, lines 1-2). Lipid metabolism and homeostasis is known to be disrupted in AD1. Fatty acid oxidation is a known energy source in the prefrontal cortex2 and will also generate acetyl coA, which this study reveals is a significant decreased metabolite in AD. Furthermore, sphingomyelin emerges as one of the major decreased DEMs as well. Thus, lipid metabolism should be highlighted in Figure 3 and discussed throughout the manuscript; otherwise its omission should be clearly stated and justified.
   2. The covariates used for differential analysis should be discussed and justified. Notably, age is used as a covariate for transcriptomic analysis but not proteomic and metabolomic analysis, with no justification. Additionally, given the known importance of lipid metabolism in AD and the putative role of APOE in lipid homeostasis3, APOE genetic status should be considered as a covariate, or its omission should be justified.
   3. The authors make a conclusion statement that suggests intervention: "Collectively, our data suggests that preserving or improving the ability to produce ATP and early intervention in the process of nitrogen metabolism are candidates for the prevention and treatment of dementia" (pg. 12, lines 12-14). This claim is not well-supported by the evidence provided in the study. There are a few limitations: (a) This was an observational, not interventional study; (b) The study did not establish whether the metabolic disruptions are causes or effects in AD; and (c) ATP or other bioenergetic indicators were not directly measured. Therefore, any statements about potential interventions should be removed or qualified as highly speculative.
   4. In conjunction with the last point, the main conclusion of the study is that energy production is down in AD. The data presented in Figure 3 are consistent with this conclusion, but it is far from definitive due to limitations stated above in comments 3a and 3b. The authors should offer additional support for this conclusion: experimental follow-up, flux modeling, analysis of alternative datasets with ATP measurement, causal inference..
   5. The validation analysis did not sufficiently show the generalizability of this study's results. The authors demonstrated a correlation of 0.53 to the MSBB transcriptomics data and 0.60 to the AMP-AD DiverseCohorts proteomics data. Beyond these correlation coefficients, no meaningful comparison between the datasets is offered. How concordant are the differentially expressed features (or pathways) between the datasets? How robust would the trans-omic network be if incorporating the alternate datasets? Is the main conclusion (energy metabolism is down in AD) supported by the validation datasets? We think this analysis should be expanded and described in the main text.

   Although the results for external metabolomics datasets are reported in Fig S2C, correlation coefficients with the external data are not reported. The authors state, "Note that each study used different definitions for AD and CT groups, had variations in measurement methods and brain regions analyzed." We appreciate these limitations. However, the external data should be re-analyzed using the same definitions of AD and CT, if possible. The limitations and results (which DEMs are shared between datasets) should be discussed in the main text.

   1. The glycolysis analysis and discussion needs more development. Glycolysis and gluconeogenesis share many of the same enzymes, but they are not the same pathway and should not be discussed as such. To make a claim about the overall influence of enzyme and metabolite levels on glycolysis, the authors should focus on the energetically committing steps of glycolysis (hexokinase, phosphofructokinase, pyruvate kinase) in Figure 3A, and include the full/current version of the figure in the supplement. Gluconeogenesis-specific enzymes (pyruvate carboxylase, PEPCK) are not mentioned at all - are they among the DEPs/DEGs?
   2. Given that there wasn't good concordance between the DEGs and DEPs, did including the mRNA and transcription factor layers in the network really add anything useful? It seems like the main conclusions of the manuscript were driven by the protein and metabolite layers only. How many of the DE metabolic enzymes were coregulated at the transcript and protein level? It would be useful to include the 5-layer trans-omic network in the supplement to display these results. Given your network, at what level does it appear that energy metabolism is regulated?
   3. Comment further on the results from Figure 2D. What can be learned from identifying metabolites with the greatest degree centrality? What pathways other than energy metabolism are highlighted by the trans-omic network?
   4. (Suggestion) We suggest the authors leverage their trans-omic network in additional ways beyond giving a snapshot of a few energy metabolism pathways. The analysis of top DEMs could go further. What pathways are impacted beyond energy metabolism? Among the metabolic reactions allosterically regulated by top DEMs, what metabolic pathways are enriched?
   5. (Suggestion) Figure 3 shows that most differential signal in AD points to lower energy production due to the combination of differentially expressed metabolites and enzymes, but we are not given much context about the strength of these among all the differential signals. We would suggest including volcano plots where the features of interest, i.e. DE enzymes and metabolites, are colored differently (or a similar figure).
   6. (Suggestion) The PPI network could be better leveraged to understand metabolic changes in AD. If nodes are grouped into subnetworks (e.g. by Louvain / Leiden clustering) and tested for pathway enrichment, could you find functional subnetworks of coordinately up- and down- regulated metabolic enzymes? This could yield some pathways of interest beyond the energy metabolism pathways already highlighted.

   Minor comments

   1. "All genes" and "all metabolites" should not be the background for the proteomic and metabolic pathway enrichment analysis by Metascape and MetaboAnalyst. The background should be limited to the proteins and metabolites that were measured.

   2. Highlight the metabolic enzymes in Fig S2B. Calculate a separate correlation coefficient for the enzymes extracted in the energy metabolism analysis from Fig 3.

   3. Use a multiple hypothesis adjusted p-value or q-value in Figure S3.

   4. Describe the methods used to calculate the logFC values from the validation dataset.

   5. It is difficult to read Figure 3. We would recommend really emphasizing to the reader to refer to Fig S7B as a "key" to this figure. The description of the red/blue arrows and nodes in the methods section (pg. 24, lines 21-36, pg 25, lines 1-4) were also helpful, but very lengthy. We recommend putting an abridged version of this description into the Fig S7 figure legend.

   6. The S7 figure legend should refer to panels A and B, not E and F.

   7. (Suggestion) Are any of the differentially expressed metabolites allosteric regulators of the DE transcription factors? This could be interesting to discuss.

   Significance

   The study's strength lies in leveraging three omics modalities across large patient cohorts (n ~ 150-240) to identify coherent signals between transcriptomics, proteomics, and metabolomics in postmortem DLPFC tissue. It was encouraging to see that the main result, showing downregulation for TCA, oxidative phosphorylation, and ketone body metabolism, emerged from consistent signals across both proteomics and metabolomics. This result was consistent with previous findings in other models cited by the author4,5 and other studies 6,7 demonstrating deficiency in energy-producing pathways in AD. Another strength of the study is the application of thoughtful methodology to connect differentially expressed proteins and metabolites via an intermediate data layer of metabolic reactions. The authors leverage the KEGG and BRENDA databases and apply sound logic to estimate the effects of enzyme level and metabolite level on pathway activity, with metabolites serving as substrate, product, or allosteric regulator for reactions. This trans-omic network methodology was developed in previous studies cited by the author8,9. However, as written, this study is limited in its contribution of new knowledge to the AD research field. The main conclusion (energy production is down in AD, due to regulatory disruption of energy metabolism) is not strongly supported (see comments 1, 3, and 4 for elaboration). The evidence could be improved by orthogonal approaches: further experimentation, further integration of external datasets, causal modeling, or flux modeling. Alternatively, even in the absence of new experimental and computational approaches, the story could be made more complete by further leveraging the trans-omic network to provide insights into (a) the regulation of energy metabolism; and (b) the impacts of key disrupted metabolites (see comments 7-9). The study is also limited in its demonstrating the power of these methodologies to provide integrative insights. As mentioned above, the integration of enzyme levels and metabolite levels is clearly useful (Figure 3). In contrast, the utility of the mRNA and transcription factor layers was not evident. The study did not appear to improve or expand upon trans-omic network methodology described in the previous works. Finally, the various analyses (analyzing the trans-omic network for nodes with the highest degree centrality, the PPI analysis, and viewing the energy metabolism pathways in the network) provided disparate results that were only tenuously connected in the discussion section.

   References

   1. Yin F. Lipid metabolism and Alzheimer's disease: clinical evidence, mechanistic link and therapeutic promise. FEBS J. 2023;290(6):1420-1453. doi:10.1111/febs.16344
   2. Ju Y, Li S, Kong X, Zhao Q. Exploring fatty acid metabolism in Alzheimer's disease: the key role of CPT1A. Sci Rep. 2024;14(1):31483. doi:10.1038/s41598-024-82999-z
   3. Sienski G, Narayan P, Bonner JM, et al. APOE4 disrupts intracellular lipid homeostasis in human iPSC-derived glia. Sci Transl Med. 2021;13(583):eaaz4564. doi:10.1126/scitranslmed.aaz4564
   4. Zhang C, Rissman RA, Feng J. Characterization of ATP Alternations in an Alzheimer's Transgenic Mouse Model. J Alzheimers Dis JAD. 2015;44(2):375-378. doi:10.3233/JAD-141890
   5. Jia D, Wang F, Yu H. Systemic alterations of tricarboxylic acid cycle enzymes in Alzheimer's disease. Front Neurosci. 2023;17:1206688. doi:10.3389/fnins.2023.1206688
   6. Sang C, Philbert SA, Hartland D, et al. Coenzyme A-Dependent Tricarboxylic Acid Cycle Enzymes Are Decreased in Alzheimer's Disease Consistent With Cerebral Pantothenate Deficiency. Front Aging Neurosci. 2022;14. doi:10.3389/fnagi.2022.893159
   7. Patel V, Mill J, Okonkwo OC, Salamat S, Li L, Raife T. Global Energy Metabolism Deficit in Alzheimer Disease Brain. J Prev Alzheimers Dis. 2024;11(1):171-178. doi:10.14283/jpad.2023.91
   8. Egami R, Kokaji T, Hatano A, et al. Trans-omic analysis reveals obesity-associated dysregulation of inter-organ metabolic cycles between the liver and skeletal muscle. iScience. 2021;24(3):102217. doi:10.1016/j.isci.2021.102217
   9. Bai Y, Morita K, Kokaji T, et al. Trans-omic analysis reveals opposite metabolic dysregulation between feeding and fasting in liver associated with obesity. iScience. 2024;27(3):109121. doi:10.1016/j.isci.2024.109121

   Read the original source
10. 

    Version published to 10.1101/2025.09.26.678758 on bioRxiv

    Sep 28, 2025