Microbial Memory: test whether metabolite differences persist across generations

Reanalyze public mouse cecal-metabolite measurements to identify whether early-life-stress-associated differences recur in descendants once cage, litter, experimental batch and multiple comparisons are considered. Deliver reproducible effect estimates and a defensible list of candidates for mechanistic follow-up—or a supported conclusion that none can yet be prioritized.

The Microbial Memory project proposes that microbiota-derived metabolites help transmit stress-related adaptations. This task purchases one bounded step toward that hypothesis: quantitative evaluation of measured metabolite differences in an existing two-generation dataset. It does not test ancient ancestry, microbial strain transmission, germline mechanisms or behavioral mediation.

The fixed workbook below contains 153 sample records with generation, age, mouse identifier, treatment, cage, litter and experimental-batch fields, together with metabolite measurements. Include every record and every measurement column from J through AP. Retain totalOrganicAcids and totalAmines as explicitly secondary summary outcomes; do not count them as independent metabolites. Preserve joint analyte labels such as phenylalanine/Leucine rather than pretending to have separate measurements.

Estimate MSUS-versus-control contrasts separately in F1 and F2. Define the effect scale and uncertainty method before reporting results. Fit a model that accounts for supported cage/litter dependence and experimental batch, and report the treatment-by-generation contrast with the accompanying age/design qualification. Generation and measurement age differ in this dataset; their effects cannot be separated by relabeling or unsupported adjustment.

Inspect whether identifiers are globally unique or reused across generations, treatment/batch confounding, numbers of independent cages and litters per comparison, missingness, zero values and model convergence. Explain the grouping structure used and any non-estimable contrasts. An unadjusted mouse-level analysis may be included only as a labeled sensitivity comparator, not treated as independent evidence.

For each individual metabolite, report group summaries, both generation-specific effect estimates and their uncertainty, and the interaction estimate where identifiable. Apply Benjamini–Hochberg adjustment separately to the complete family of individual-metabolite F1 contrasts, the complete family of F2 contrasts, and the complete family of interaction tests. Provide raw and adjusted p-values and the family denominator. Keep the two totals in one separately labeled secondary family containing their F1, F2 and interaction tests. If valid inference cannot be obtained for an endpoint, retain its row and explain why; do not silently reduce the family or replace missing tests with favorable values.

Perform two robustness checks. First, assess sensitivity to zero handling and response transformation with at least two clearly justified choices; do not assume zeros are known nondetects unless source methods establish that interpretation. Second, refit after omitting each cage in turn and summarize how estimates and candidate ordering change. If leaving out a cage destroys estimability, record that directly.

Classify which results, if any, justify prioritizing further mechanistic measurement. Distinguish same-direction estimates, statistically supported within-generation contrasts and evidence of similar effect sizes. A nonsignificant interaction is not evidence of equivalence or persistence. If using an equivalence claim, state and justify its margin and perform an appropriate interval-based assessment; no equivalence claim is required. No particular metabolite or positive result is required.

All required material must be included as file bytes in the Submission; links alone are insufficient.

• analysis_data.csv: the complete in-scope extraction, retaining original sample and metadata fields, units as supported by the methods, and source column names. Include a data dictionary mapping original fields to analysis variables.
• data_audit.md: record counts and exclusions, design/grouping checks, missing/zero patterns, unit provenance, family definitions and the exact models/transformations. Any exclusion needs a source-based or clearly stated analytical reason and a retained audit row.
• effects.csv: every primary and secondary endpoint with descriptive summaries, estimates, intervals, raw/adjusted tests, denominators and status for non-estimable results; include enough identifiers to distinguish contrast and transformation.
• sensitivity.csv: zero/transform and leave-one-cage-out results, including failed/undefined refits and their reasons. Include numeric values underlying every submitted plot.
• Executable R or Python code, dependencies/versions and run instructions that regenerate the extraction, analysis tables and plots from the public workbook or its submitted faithful extraction. Include all required numeric inputs and code files. Freely available ordinary-laptop software must suffice; no raw-read sequencing pipeline, paid API or new animal data is required.
• decision_memo.md with uncertainty plots and a complete endpoint evidence map: state which candidates remain stable enough for follow-up and why, or why no prioritization is warranted. Explain which additional observations would distinguish microbial transmission, host-mediated inheritance and shared environment, without prescribing an animal experiment protocol. Explicitly distinguish a metabolite association from a behavioral or causal mechanism.
• checks.md: extraction reconciliation, model diagnostics/convergence, independent verification of one nontrivial contrast or test, verification of the multiple-testing calculation, and reproduction instructions. Explain material disagreements with the source rather than forcing published significance.

Required original study: Otaru et al., *Transgenerational effects of early life stress on the fecal microbiota in mice*, Communications Biology 7,670 (2024), DOI 10.1038/s42003-024-06279-2. Use the 2024 article with its 18 June 2024 correction, as available on 17 September 2026. The correction adds a funding acknowledgement. Public full-text XML mirror.

Fixed numeric input: original 42003_2024_6279_MOESM4_ESM.xlsx, worksheet Supplementary-Data-1, publisher download, 74,661 bytes, SHA-256 80732964a843340eddbe0e740b0c5f97b573c6ec9a3248cfe2a46f1bc90e74fe. The fixed workbook governs numerical values. The associated article methods govern the study design and measurement interpretation. Document conflicts instead of silently altering the input.

The publisher's Supplementary Information is supporting context for design and published metabolite summaries. Raw sequencing accession PRJEB57336 is outside this task; its download and reprocessing are not required.

Background only: Microbial Memory project. No private project inputs or new data collection are required.

Only fully eligible Submissions rank. Prefer, in order: independently checked numerical reproducibility; justified handling of clusters, confounding, zeros and multiple testing; and decision usefulness, defined as explaining exactly which evidence makes a candidate stable or fragile and what missing information would change that judgment. Positive conclusions and longer candidate lists receive no advantage. Remaining ties go to the earlier submitted eligible Submission. If one qualifies, it wins; if none qualifies, the outcome is no_valid_submission.

No wet-lab or animal work, microbiome engineering, ancient-microbe claims, human trauma claims, behavioral prediction, sequence reprocessing or new participant/animal data. This task does not establish a causal inheritance mechanism.