Differentiating hemorrhagic shock and organophosphate poisoning through integrated skin microbiome–metabolome signatures Authors: Jian Zhao, Qin Su, Shujuan Wang, Qi Li, Ling Chen, Xiaodong Kang, Quyi Xu, Chao Liu, Hu Zhao Published: 2026 ABSTRACT Accurate determination of cause of death and estimation of postmortem interval (PMI) are critical yet challenging tasks in forensic science, particularly in cases with rapid demise and absence of obvious morphological abnormalities. We employed an integrative multi-omics approach to characterize postmortem microbial succession and metabolic alterations on facial skin in mouse models of hemorrhagic shock (HS) and organophosphorus poisoning (OP) across three decomposition stages: bloating (2 days), active decay (8 days), and advanced decay (16 days). Metagenomic profiling revealed significantly reduced α-diversity in HS compared with OP throughout all stages (p < 0.001), accompanied by stage-dependent compositional shifts, including early enrichment of Firmicutes in HS and Proteobacteria in OP. A total of 237 differential taxa were identified, with Providencia and Morganella predominating in OP, whereas Staphylococcus and Corynebacterium dominated bloating stage of HS. Untargeted metabolomics uncovered distinct cause-of-death–linked metabolites, notably elevated 2′-deoxycytidine-5′-diphosphate in early OP and persistent cholic acid/cholate accumulation in HS at later PMI. Functional analysis highlighted histidine and phosphate/phosphonate metabolism as key discriminatory pathways, exhibiting stage-specific oscillations and strong correlations with characteristic taxa. These findings demonstrate that skin-based metagenomic–metabolomic integration provides robust, mechanistically informed biomarkers for both PMI estimation and cause-of-death differentiation, offering a minimally invasive and temporally dynamic tool for forensic investigations.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12866-026-05276-1. Introduction Although hemorrhagic shock (HS) and organophosphorus (OP) poisoning can usually be differentiated in clinical or perimortem contexts by trauma evidence or toxicological assays, their distinction becomes markedly more difficult in decomposed remains [1–3]. Decomposition rapidly obscures gross anatomical features, accelerates the breakdown of hemoglobin, and causes degradation or volatilization of chemical poisons, thereby limiting the reliability of conventional autopsy and toxicology [3, 4]. In forensic practice, such cases represent a critical diagnostic gap, as cause-of-death determination must often be made after several days or weeks of postmortem interval (PMI) [5, 6]. To address this, we hypothesized that the skin microbiome and metabolome retain discriminatory, time-resolved signatures that could distinguish HS from OP during decomposition. By integrating multi-omics profiling across defined PMI stages, this study aimed to establish a minimally invasive, mechanistically grounded framework for both cause-of-death differentiation and PMI estimation in challenging forensic scenarios. In recent years, the postmortem microbiome has emerged as a promising avenue for forensic research. Studies have demonstrated that microbial communities undergo predictable succession during decomposition, which can be leveraged to estimate PMI [5, 7–10]. Furthermore, microbial signatures may reflect underlying physiological states of the host prior to death, thereby offering potential clues about the cause of death [11]. Parallel to this, metabolomics has been increasingly applied in forensic investigations to characterize biochemical shifts after death [12, 13]. Alterations in metabolites not only mirror host metabolic disruption but also integrate microbial contributions, offering a comprehensive snapshot of postmortem biochemical processes. Despite these advances, integrative studies combining metagenomic and metabolomic approaches remain limited, particularly in relation to differentiating causes of death [14]. The skin surface provides a unique and underexplored niche for postmortem studies. As the body’s largest organ and a direct interface with the environment, the skin harbors a diverse microbial ecosystem and a wide range of metabolites [15, 16]. While internal organs are often used for microbial and biochemical analyses, sampling them requires invasive procedures, whereas the skin offers a minimally invasive, accessible alternative [17]. Prior work has suggested that skin-associated microbial and metabolic changes can be informative for PMI estimation [6]. Notably, a recent human cadaver study investigating skin microbiome within the first 12 h postmortem revealed distinct microbial signatures tied to early death stage and morgue reception conditions, underscoring skin’s forensic relevance [18]. However, little is known about how these changes vary between different causes of death or across distinct stages of decomposition. In particular, the facial skin, which is highly exposed and densely colonized, may provide sensitive indicators of postmortem alterations. To address these gaps, we established mouse models of hemorrhagic shock and organophosphorus poisoning and systematically analyzed facial skin swabs collected at bloating stage (PMI = 2 days, BS), active decomposition stage (PMI = 8 days, AcD), and advanced decomposition stage (PMI = 16 days, AdD) postmortem intervals. By integrating metagenomic sequencing with untargeted metabolomics, we sought to characterize microbial succession patterns, metabolic alterations, and their interactions under different death causes and decomposition stages. We hypothesized that distinct microbial taxa and metabolic pathways would emerge as markers differentiating hemorrhagic shock from organophosphorus poisoning and that these signatures would vary across temporal stages of decomposition. Our findings provide novel insights into postmortem biology, highlight the forensic potential of skin-based multi-omics approaches, and suggest new avenues for improving cause-of-death determination in challenging forensic cases. Materials and methods Mouse procurement and ethical approval All animal experiments were conducted in strict compliance with the guidelines of the Institutional Animal Care and Use Committee (IACUC) of Sun Yat-sen University (Approval ID: SYSU-IACUC-2021-000678). BALB/c nude mice were purchased from Vital River Laboratories (Beijing, China). Animals were housed with their respective littermates under controlled conditions (12 h light/12 h dark cycle) and had unrestricted access to standard chow and water. Establishment of decomposed cadaver models Only female BALB/c nude mice were used in this study to minimize inter-individual variability associated with aggressive behavior and stress responses that are more common in males under group housing conditions. Female mice of this strain also exhibit relatively stable body weight and consistent skin exposure characteristics, which are important for standardized facial skin sampling. A total of 36 female BALB/c nude mice (6 weeks old) were randomly allocated into two groups: the Hemorrhagic Shock group (HS, n = 18) and the Organophosphorus Poisoning group (OP, n = 18, Fig. 1A). For the HS group, mice were weighed and positioned supine. A 1 cm incision was made over the left carotid artery using a sterile scalpel to achieve > 2% body weight blood loss within 10 min, avoiding injury to the trachea and cervical vertebrae. For the OP group, dichlorvos (an organophosphorus pesticide) was administered via oral gavage at a dose of 90 mg/kg [19]. Following administration, mice were placed in sterile, drape-lined containers until death. Successful poisoning was confirmed by characteristic clinical signs, including muscle tremors, limb convulsions, salivation, lacrimation, and progressive limb weakness. Mice exhibiting dyspnea, nasal discharge, or immediate death were excluded from the study. For both groups, if mice remained alive 10 min after the respective interventions, they were anesthetized with pentobarbital sodium (50 mg/kg, i.p.) and subsequently euthanized by cervical dislocation under anesthesia. Fig. 1Study design and global metagenomic and metabolomic characteristics of facial skin microbiota and metabolome across postmortem intervals (PMI). A Experimental design and sampling scheme at bloating stage (BS), active decay (AcD), and advanced decay (AdD). B Shannon index revealed significantly reduced α-diversity in the hemorrhagic shock (HS) group compared to the organophosphorus poisoning (OP) group at all PMI stages (p < 0.001, Wilcoxon rank-sum test). C Bray–Curtis PCoA demonstrated significant differences in microbial community composition between HS and OP groups across BS, AcD, and AdD (p < 0.001, Adonis test). D PCoA of untargeted metabolomics showed significant separation only at AdD (p < 0.01, Adonis test). The shaded areas in PCoA represent 95% confidence ellipses Study design and global metagenomic and metabolomic characteristics of facial skin microbiota and metabolome across postmortem intervals (PMI). A Experimental design and sampling scheme at bloating stage (BS), active decay (AcD), and advanced decay (AdD). B Shannon index revealed significantly reduced α-diversity in the hemorrhagic shock (HS) group compared to the organophosphorus poisoning (OP) group at all PMI stages (p < 0.001, Wilcoxon rank-sum test). C Bray–Curtis PCoA demonstrated significant differences in microbial community composition between HS and OP groups across BS, AcD, and AdD (p < 0.001, Adonis test). D PCoA of untargeted metabolomics showed significant separation only at AdD (p < 0.01, Adonis test). The shaded areas in PCoA represent 95% confidence ellipses To minimize human-derived contamination and prevent cross-contamination between carcasses, all deceased mice were placed on pretreated soil collected from the foothills of Baiyun Mountain. The soil was sieved (8 mm mesh) and pre-incubated for one week (20.0–21.0 °C; 64–66.5% RH) before use. The burial site was maintained in a controlled incubator to facilitate natural decomposition. Sample collection Six samples were collected from each group at three postmortem intervals (PMIs): 2 days, 8 days, and 16 days, corresponding to the bloating stage (BS), active decomposition stage (AcD), and advanced decomposition stage (AdD), respectively. Facial skin swabs were collected using sterile disposable swabs, which were immediately transferred into pre-labeled cryotubes and stored in liquid nitrogen (–196 °C) until further analysis. DNA extraction and DNA library preparation Total microbial DNA was extracted using the E.Z.N.A.® Stool DNA Kit (Omega Bio-tek, USA). DNA quality was verified by agarose gel electrophoresis, and concentrations were quantified with a UV spectrophotometer. Extracted DNA was eluted in 50 µL of elution buffer and stored at − 80 °C until sequencing. Genomic DNA was fragmented to an average size of 200–500 bp, followed by end repair, 3′ adenylation, and adapter ligation. Adapter-ligated fragments were purified, PCR-amplified, and subjected to final purification, quantification, and quality control to generate sequencing libraries. Metagenomic sequencing and bioinformatic analysis Qualified libraries were sequenced on the Illumina HiSeq 4000 platform using a paired-end 150 bp (PE150) strategy (Hangzhou Lianchuan Biotechnology Co., Ltd.). Raw reads containing adapters or host-derived sequences were removed using cutadapt (v1.9) and bowtie2 (v2.2.0). Reads with > 10% ambiguous bases (N) or > 40% low-quality bases (Q < 20) were discarded. High-quality reads were de novo assembled using IDBA (v1.1.3), and coding sequences (CDSs) were predicted with MetaGeneMark (linux_64). Redundant sequences were removed using cd-hit (v4.6.1) to obtain a non-redundant unigene catalog. Unigenes were taxonomically annotated by aligning against the NCBI NR database (nr_meta_20150426) using DIAMOND (v0.7.12) (E-value < 0.001). Taxonomic classification was performed across seven hierarchical levels (Kingdom–Species) for bacteria, fungi, viruses, and archaea. Alpha diversity (observed OTUs, Shannon index) and beta diversity (Bray–Curtis-based Principal coordinates analysis, PCoA) were calculated with QIIME (v1.9.1). Group differences in alpha diversity were assessed using the Wilcoxon rank-sum test (p < 0.01). Functional annotation was performed using the KEGG database (release 87.1) to assign KEGG Orthology (KO) identifiers. Species with relative abundance > 0.1% in at least one sample were analyzed by LEfSe (QIIME1 implementation) with thresholds |LDA| ≥ 3.5 and p < 0.05. The Kruskal–Wallis test identified taxa with significant intergroup differences, followed by Linear Discriminant Analysis (LDA) to estimate effect sizes. Differentially expressed unigenes between HS and OP groups at different PMIs were identified using the Mann–Whitney U test. Functional enrichment for Gene Ontology (GO) and KEGG pathways was conducted using the clusterProfiler package in R (p < 0.05). Metabolite extraction and LC–MS/MS analysis Samples were mixed with 120 µL of 80% methanol, vortexed thoroughly, and incubated at room temperature for 10 min. Extracts were stored at − 20 °C overnight to precipitate proteins, then centrifuged at 4000 × g for 20 min. Supernatants were transferred to 96-well plates, and pooled quality control (QC) samples were prepared by mixing 10 µL from each sample. All extracts were stored at − 80 °C until LC–MS/MS analysis. Chromatographic separation was performed using a SCIEX UHPLC system equipped with an ACQUITY UPLC T3 column (100 mm × 2.1 mm, 1.8 μm; Waters, UK) at 35 °C, flow rate 0.4 mL/min. Mobile phases: A, water with 1% formic acid; B, acetonitrile with 1% formic acid. Gradient: 0–0.5 min, 5% B; 0.5–7 min, 5–100% B; 7–8 min, 100% B; 8–8.1 min, 100–5% B; 8.1–10 min, 5% B. Mass spectrometry was conducted on a TripleTOF 5600plus (SCIEX, UK) in both positive and negative ion modes using information-dependent acquisition (IDA) with a full scan range of 60–1200 Da (150 ms) and fragmentation of the top 12 ions per cycle. Source parameters: curtain gas, 30 psi; gas 1 and gas 2, 60 psi; temperature, 650 °C; +5000 V (positive) / − 4500 V (negative). Declustering potential was applied with 60 V (+) and − 60 V (−) and collision energy was set as 50 V (+) and − 20 V (−). QC samples were injected every 10 runs, and instrument calibration was performed every 20 samples. Metabolomics data processing and analysis Metabolomics data acquisition and annotation were performed by a commercial metabolomics platform (Hangzhou Lianchuan Biotechnology Co., Ltd.). Raw LC-MS data were processed using Compound Discoverer software (Thermo Fisher Scientific, US) for peak detection, retention time alignment, and feature extraction. The peak extraction parameters were set as follows: minimum peak intensity, 1,000,000; signal-to-noise ratio (S/N) threshold, 3. Retention time alignment was performed using the adaptive curve model with a mass tolerance of 10 ppm and a maximum shift of 2 min. Gap filling was conducted with a mass tolerance of 10 ppm and an S/N threshold of 1.5. Adduct types were specified as [M + H] + and [M − H]−, with charge state z = 1, and the retention time grouping tolerance was set to 0.2 min. Compound identification was performed by searching the mzCloud database. The precursor ion mass tolerance was set to 10 ppm, and the MS/MS fragment ion mass tolerance was set to 20 ppm. The similarity threshold for annotation was set to ≥ 70%, and fragment ion matching was enabled with a mass tolerance of 10 ppm and an S/N threshold of 3. The identification confidence level was set to Level 2. Only identifications meeting an mzCloud MS/MS similarity score of ≥ 0.7 and a fragment ion mass deviation within 20 ppm were retained. To ensure analytical reliability, QC-based filtering was applied using pooled QC samples. Features with a relative standard deviation (RSD) > 30% across QC injections were removed. Metabolites missing in more than 50% of QC samples were removed. In experimental samples, metabolites missing in more than 20% of samples were excluded. Remaining missing values were imputed using K-nearest neighbor (KNN) and normalized with probabilistic quotient normalization (PQN). Metabolites detected in both ion modes were merged based on MS/MS confirmation and retention time alignment to generate a non-redundant list (Table S1). PCA chart showed tight QC clustering (Fig. S1). Metabolites were annotated by matching KEGG and HMDB databases (mass tolerance 10 ppm) and an in-house MS/MS library (similarity > 80%). Differential metabolites were identified using PLS-DA (p < 0.05, absolute log2 fold change > 0.5, VIP > 1) with the metaX R package. Model robustness was evaluated using 6-fold cross-validation to calculate R² and Q² metrics. To assess overfitting, 200 permutation tests were conducted by randomly shuffling class labels and recalculating Q² values to generate a null distribution. Statistical analysis All statistical analyses were performed in R (v4.3.2). Continuous variables were compared using the two-sided Wilcoxon rank-sum test, categorical variables using Pearson’s Chi-square or Fisher’s exact test, and correlations among microbial taxa, metabolites, and KO genes were evaluated using Spearman correlation. Results Distinct α-diversity and community structures between HS and OP groups across PMI stages Significant alterations in facial skin microbiota diversity were observed between the HS and OP groups during the BS, AcD, and AdD. Shannon index–based α-diversity was markedly reduced in HS compared with OP at all postmortem intervals (Fig. 1B). PCoA using Bray–Curtis distances revealed significant differences in community composition between the two groups across all stages (Fig. 1C). For metabolomic profiles, significant separation was only evident at the AdD stage (Fig. 1D). Phylum-level analysis demonstrated temporal compositional shifts: HS samples showed reduced Proteobacteria and increased Firmicutes during BS, but Proteobacteria dominance emerged in HS during AcD and AdD (Fig. S2). Overall, OP exhibited higher α-diversity and distinct microbial distances compared to HS across PMI stages. Differentially abundant taxa reveal stage-dependent microbial signatures A total of 237 differential species were identified between HS and OP groups (Fig. 2A; Fig. S3), with the BS stage containing the greatest number of taxa, accounting for 81.6% of those found in AcD and AdD. Consistently enriched taxa in OP included Providencia rettgeri, Providencia alcalifaciens, Providencia stuartii, Xenorhabdus bovienii, Morganella morganii, Arsenophonus nasoniae, Photorhabdus laumondii, Serratia sp. S1B, Gilliamella apicola, Aliidiomarina sanyensis, and Streptococcus pneumoniae. Several taxa exhibited stage-dependent inversion: Proteus mirabilis, Klebsiella pneumoniae, and Escherichia coli were more abundant in OP at BS but shifted to HS dominance during AcD and AdD. Conversely, Staphylococcus xylosus, Staphylococcus HMSC067F07, and Corynebacterium mastitidis were enriched in HS at BS but predominated in OP at later stages. Fig. 2Differentially abundant taxa and microbial co-occurrence networks across PMI stages. A 237 differentially abundant bacteria species identified between HS and OP groups in each of the three PMI groups are shown in the phylogenetic tree, grouped in the phyla Firmicutes, Proteobacteria, Bacteroidetes, TM7 and Actinobacteria. The innermost circle shows species relative abundances averaged over all samples. In the outer rings, species are marked for significant (p adjusted by Benjamini-Hochberg < 0.05; multivariable regression model) elevation (red) or depletion (blue) in HS group. Bar plots show each group’s abundance log2 fold change of death cause related species. Each plot shows bars in the four groups (BS, AcD, AdD, and the whole samples) from left to right. Fold-change comparisons were group means vs. group means; for inter-group fold calculations there was a single log2FC value (no per-sample error bar, p values < 0.05 are marked with asterisks). B Co-occurrence relationships among species in normal BS, AcD, and AdD groups performed by Spearman’s correlation. Only significant (adjusted p < 0.0005) correlations are shown. Red lines indicate positive correlations (both taxa increase or decrease together); blue lines indicate negative correlations (inverse abundance patterns) Differentially abundant taxa and microbial co-occurrence networks across PMI stages. A 237 differentially abundant bacteria species identified between HS and OP groups in each of the three PMI groups are shown in the phylogenetic tree, grouped in the phyla Firmicutes, Proteobacteria, Bacteroidetes, TM7 and Actinobacteria. The innermost circle shows species relative abundances averaged over all samples. In the outer rings, species are marked for significant (p adjusted by Benjamini-Hochberg < 0.05; multivariable regression model) elevation (red) or depletion (blue) in HS group. Bar plots show each group’s abundance log2 fold change of death cause related species. Each plot shows bars in the four groups (BS, AcD, AdD, and the whole samples) from left to right. Fold-change comparisons were group means vs. group means; for inter-group fold calculations there was a single log2FC value (no per-sample error bar, p values < 0.05 are marked with asterisks). B Co-occurrence relationships among species in normal BS, AcD, and AdD groups performed by Spearman’s correlation. Only significant (adjusted p < 0.0005) correlations are shown. Red lines indicate positive correlations (both taxa increase or decrease together); blue lines indicate negative correlations (inverse abundance patterns) Co-occurrence analysis demonstrated a progressive simplification of microbial interaction networks with increasing PMI (Fig. 2B). Central hubs, including Staphylococcus HMSC067F07 and Serratia marcescens, played key roles in reshaping the interaction landscape. These network alterations suggest stage-dependent ecological restructuring associated with decomposition processes. Metabolomic profiles differentiate HS and OP groups across PMI stages Untargeted metabolomics of facial skin swabs revealed marked metabolic distinctions between HS and OP. Although PCA did not reveal a clear separation between the two groups, PLS-DA showed a marked separation at the BS stage and could still distinguish the AcD and AdD stages, despite some partial overlap (Fig. 3A; Figs. S4 and S5). The Q2 0.42 (BS), 0.32 (AcD), and 0.26 (AdD), indicating modest predictive performance of the PLS-DA models (Fig. S4). Although permutation testing yielded negative Q2 intercepts (BS: -0.55, AcD: -0.85, and AdD: -1.16), suggesting that the models performed better than random classification; however, the predictive ability remained limited and should be interpreted cautiously. Therefore, VIP scores (VIP > 1) were used only as a variable-importance metric to prioritize candidate metabolites and were further combined with univariate criteria (p < 0.05 and absolute log2 fold change > 0.5) to improve the robustness of differential metabolite selection. A total of 21 stage-specific differential metabolites were identified (Fig. 3B; Fig. S6). At BS, 2’-deoxycytidine-5’-diphosphate sodium salt (dCDP) and Niranthin were markedly upregulated in OP, suggesting increased nucleotide biosynthesis and microbial replication potential. Cholic acid was markedly upregulated in HS. At AcD, HS displayed elevated cholic acid and decreased Benzalkonium chloride (C12). At AdD, cholic acid and cholate remained elevated in HS, whereas phenazine-1-carboxylic acid were downregulated compared to OP. Fig. 3Metabolomic distinctions between HS and OP groups across PMI stages. A PLS-DA score plots of metabolite profiles at the BS, AcD, and AdD stages. The circle areas in PLS-DA represent 95% confidence ellipses. The corresponding permutation tests are shown in Fig. S4. B Volcano plot of differential metabolites in each group. p value adjusted by Benjamini-Hochberg < 0.05, absolute log2 fold change > 0.5, and variable importance for the projection (VIP) > 1 are considered differential metabolites. The x axis represents log2(fold change). The y axis represents -log10(p value) Metabolomic distinctions between HS and OP groups across PMI stages. A PLS-DA score plots of metabolite profiles at the BS, AcD, and AdD stages. The circle areas in PLS-DA represent 95% confidence ellipses. The corresponding permutation tests are shown in Fig. S4. B Volcano plot of differential metabolites in each group. p value adjusted by Benjamini-Hochberg < 0.05, absolute log2 fold change > 0.5, and variable importance for the projection (VIP) > 1 are considered differential metabolites. The x axis represents log2(fold change). The y axis represents -log10(p value) Correlation analyses revealed stage-specific associations between selected taxa and differential metabolites (Fig. 4). Notably, Streptococcus pneumoniae was negatively correlated with dCDP in AdD-OP, suggesting its potential role in modulating replication-related microbial activity. These findings highlight unique metabolite–microbiome interactions underlying cause-of-death–specific decomposition. Fig. 4Correlations between 21 key differential taxa and 31 stage-specific differential metabolites. Spearman’s correlation between differential species and metabolites in HS and OP from BS, AcD, and AdD groups. p values < 0.05 are marked with asterisks. ∗p < 0.05, and ∗∗p < 0.01 Correlations between 21 key differential taxa and 31 stage-specific differential metabolites. Spearman’s correlation between differential species and metabolites in HS and OP from BS, AcD, and AdD groups. p values < 0.05 are marked with asterisks. ∗p < 0.05, and ∗∗p < 0.01 Stage-dependent functional pathways in histidine and phosphate/phosphonate metabolism Functional annotation of KEGG orthologs (KOs) revealed significant alterations in histidine metabolism, phosphate/phosphonate metabolism, and thiamine metabolism at BS (Fig. 5A). Genes related to amino acid metabolism (e.g., hisD, hisG, hisC, hisB, hisF, hisIE) and thiamine metabolism (e.g., thiD, thiN, phoA, thiC) were predominantly upregulated in HS, whereas phosphate/phosphonate utilization genes (e.g., phnY, pepM, phnL, phnM, phnJ, phnI, phnH, phnG, phnA) were enriched in OP. At AcD, phosphate/phosphonate metabolism was primarily elevated in HS, whereas at AdD, histidine and phosphate/phosphonate pathways were mainly upregulated in OP (e.g., hisG, hisI, hisE, hisB, hisZ, hisN, priA, phnY, pepM, phnL, phnM, phnJ, phnI, phnH, phnG, phnA). Fig. 5Functional pathways shift in KO genes. A The heatmap shows the integrated meta-analysis that identified significantly changed KO gene expression in each KEGG pathway examined across three PMI groups. The cell color and intensity represent the log2 abundance fold change of KO genes. The cells identified and marked the significant differential KO gene (p adjusted by Benjamini-Hochberg < 0.05). KO genes with absolute log2 fold change > 1 are shown in the figure. B Spearman’s correlation between differential species and KO genes. p value is adjusted by Benjamini-Hochberg. ∗Adjusted p < 0.05, ∗∗adjusted p < 0.01, and ∗∗∗adjusted p < 0.001 Functional pathways shift in KO genes. A The heatmap shows the integrated meta-analysis that identified significantly changed KO gene expression in each KEGG pathway examined across three PMI groups. The cell color and intensity represent the log2 abundance fold change of KO genes. The cells identified and marked the significant differential KO gene (p adjusted by Benjamini-Hochberg < 0.05). KO genes with absolute log2 fold change > 1 are shown in the figure. B Spearman’s correlation between differential species and KO genes. p value is adjusted by Benjamini-Hochberg. ∗Adjusted p < 0.05, ∗∗adjusted p < 0.01, and ∗∗∗adjusted p < 0.001 Integrative analysis of KO genes and taxa demonstrated strong negative correlations between Proteus mirabilis/Escherichia coli and most histidine and phosphate/phosphonate pathway genes, while Staphylococcus xylosus, Corynebacterium mastitidis, Rhizobacter gummiphilus, Cutibacterium acnes, Streptococcus pneumoniae, and Serratia marcescens exhibited strong positive associations (Fig. 5B). These results indicate that distinct microbial consortia drive the metabolic reprogramming of histidine and phosphate/phosphonate pathways during decomposition, providing mechanistic insights into microbiome–metabolome interactions linked to cause-of-death differentiation. Discussion In this study, we systematically characterized the postmortem dynamics of facial skin surface microbiota and metabolome across different PMI in HS and OP groups. Our findings revealed three major features: (i) a consistent reduction in alpha diversity in the HS group compared to OP across all stages, with the most pronounced differences occurring in the early PMI (BS stage); (ii) stage-specific shifts in microbial composition and functional pathways, including histidine metabolism and phosphate/phosphonate metabolism; and (iii) distinct metabolomic signatures, such as elevated dCDP in early OP and progressive cholic acid (and its deprotonated form cholate) enrichment in HS, that corresponded with microbial diversity and functional changes. These observations provide novel mechanistic insights and support the potential forensic application of integrated microbiome–metabolome profiling for PMI estimation and cause-of-death discrimination. Postmortem microbial succession and cause-of-death imprinting The significant decrease in Shannon diversity in the HS group at BS, AcD, and AdD stages (p < 0.001 for all comparisons) suggests that hemorrhagic shock leads to a more restrictive postmortem microenvironment, likely driven by tissue hypoxia and altered pH immediately after death [20–22]. By contrast, the OP group exhibited higher alpha diversity, consistent with reports that organophosphorus exposure disrupts epithelial barriers and promotes colonization by diverse exogenous microorganisms [23, 24]. PCoA further demonstrated significant differences in Bray–Curtis distances between HS and OP groups at all PMI stages, although the effect was most prominent early after death and gradually converged over time, suggesting a time-dependent decay of cause-of-death imprint on microbial communities [6, 25, 26]. At the phylum level, Firmicutes were enriched in the HS group at the BS stage, while Proteobacteria predominated in the OP group, with a subsequent increase in Proteobacteria in both groups during AcD and AdD stages. This phylum-level switch likely reflects the interplay between oxygen availability, nutrient fluxes, and external contamination during decomposition [27–29]. At the species level, 237 differentially abundant taxa were identified, with Providencia rettgeri, Providencia alcalifaciens, Morganella morganii, Serratia sp., Arsenophonus nasoniae, and Photorhabdus laumondii consistently enriched in OP across stages, while Staphylococcus xylosus and Corynebacterium mastitidis were preferentially abundant in HS during early decomposition. Many of these taxa, particularly Providencia and Morganella, have been repeatedly detected in carrion-associated insect vectors and cadaveric fluids, supporting their role as reliable forensic indicators [30–32]. Co-occurrence network analysis revealed a pronounced simplification of bacterial interactions as decomposition progressed, with key nodes such as Staphylococcus HMSC067F07 and Serratia marcescens showing reduced connectivity at later PMI stages. This transition from complex to simplified microbial networks has been described as a conserved feature of decomposition, reflecting resource depletion and niche contraction [20, 33]. Such network-derived metrics may improve the robustness of PMI models across climatic zones by serving as transferable ecological signals rather than taxon-specific markers. Metabolic signatures distinguish cause of death Untargeted metabolomics revealed that chemical profiles mirrored but lagged behind microbial community restructuring. In the BS stage, OP showed a marked elevation of dCDP, indicating increased deoxynucleotide biosynthesis and microbial replication potential, which may underpin its higher alpha diversity [34, 35]. Bacterial replication requires elevated dNTP pools; many bacteria express nucleotide salvage and biosynthesis enzymes (including 5′-nucleotidases and deaminases) that reshape local nucleotide pools during growth [36, 37]. Thus, early postmortem detection of elevated dNDPs/dNTP-derived signals should be interpreted as evidence of microbial proliferation and can be linked to taxa with high growth rates under necrotic nutrient conditions (for example, opportunistic Enterobacteriaceae). In contrast, cholic acid were enriched in HS during AcD and AdD stages, possibly reflecting hemorrhagic shock–induced hepatobiliary dysfunction and subsequent bile leakage [38]. Phenazine-1-carboxylic acid, a redox-active metabolite produced by Pseudomonas and other bacteria, was upregulated in OP at late PMI, potentially promoting oxidative stress and interspecies competition [39, 40]. Correlation analysis between selected differential taxa and metabolites provided functional context. For example, Aliidiomarina sanyensis was negatively correlated with dCDP in AdD-OP, suggesting a potential role in regulating replication potential and diversity, while Proteus mirabilis and Escherichia coli were negatively associated with histidine and phosphonate metabolism genes, in contrast to positive associations for Staphylococcus, Corynebacterium, and Serratia [41–43]. These findings indicate functional redundancy, whereby different taxa can occupy similar metabolic niches, supporting a shift toward function-based rather than taxonomy-based forensic markers. In addition, niranthin is a plant-derived lignan that is not known to participate in endogenous mammalian metabolic pathways; therefore, its detection in decomposing mouse skin is biologically implausible unless attributable to exogenous contamination [44]. Benzalkonium chloride (C12), a widely used quaternary ammonium disinfectant, is a well-recognized laboratory and handling contaminant [45]. Both compounds can be inadvertently introduced during sample collection or processing through skin antiseptics, gloves, laboratory bench surfaces, or other consumables. These observations underscore the necessity for cautious interpretation of metabolomic data, particularly in untargeted analyses where exogenous contaminants may be misannotated as biologically relevant metabolites. Rigorous implementation of procedural blanks, contaminant screening, and validation using reference standards is therefore essential to ensure the reliability of metabolite identification and downstream biological inference. Functional pathway divergence: histidine and phosphonate metabolism Functional pathway analysis at the KO level identified stage-dependent changes in histidine biosynthesis and phosphate/phosphonate metabolism. However, KO enrichment reflects microbial functional potential rather than direct metabolite abundance. Histidine biosynthesis genes (his operon) were positively associated with Firmicutes/Actinobacteria taxa such as Staphylococcus and Corynebacterium in early HS samples. Histidine metabolism can serve multiple roles in decomposing niches: it supplies nitrogen and carbon, contributes to osmoprotection and metal chelation, and in some bacteria is tied to stress-adaptive biofilm behaviors [46, 47]. Upregulation of histidine pathways in HS-associated taxa is compatible with a scenario where, after hemorrhagic collapse, limited oxygen and altered pH create selective pressure for organisms that can synthesize or recycle amino acids (including histidine) to sustain cellular function under nutrient-stressed conditions. Moreover, histidine biosynthesis and salvage pathways have been directly linked to bacterial fitness and virulence in multiple pathogens (such as Staphylococcus), indicating that histidine metabolism can support survival and adaptive phenotypes in hostile niches [48]. Conversely, phosphonate utilization genes (phnY, phnL, pepM) were enriched in OP in the BS stage. During AcD, HS showed higher expression of phosphonate metabolism, whereas at AdD, OP displayed a combined upregulation of histidine and phosphonate genes. Histidine metabolism supports nitrogen turnover, stress adaptation, and virulence, while C–P lyase–mediated phosphonate utilization is activated under phosphate limitation and is often linked to pathogenicity through the Pho regulon [49–51]. The observed oscillation between these pathways likely reflects shifts in phosphorus/nitrogen availability and the chemical milieu imposed by different causes of death. Limitations and future directions From a forensic microbiology perspective, these findings have several major implications. First, for PMI estimation, early-stage indicators should focus on OP-enriched taxa (e.g., Providencia, Morganella) and replication-related metabolites (dCDP), whereas late-stage models may benefit from incorporating cholic acid and amino acid markers for greater temporal stability [52]. Second, for cause-of-death discrimination, the combination of specific taxa clusters with functional modules (histidine and phosphonate metabolism) and metabolite profiles (cholic acid, phenazines) provides a mechanistically grounded biomarker panel [14]. Third, sex-specific differences in microbial colonization and metabolic profiles have been reported in other contexts [53, 54]. Therefore, future studies should validate these findings in male animals and additional strains to assess generalizability. Finally, none of the metabolite identifications were confirmed using authentic reference standards. Because the annotations were based on accurate mass matching, MS/MS library comparisons, and database searches, they should be interpreted cautiously as putative identifications. Conclusion This integrative multi-omics analysis demonstrates that HS and OP exhibit distinct stage-dependent microbial and metabolic signatures on facial skin during decomposition. HS was characterized by persistently reduced microbial diversity and late-stage bile acid accumulation trends, whereas OP displayed early enrichment of nucleotide-related metabolites such as dCDP accompanied by specific Proteobacteria expansion. These findings highlight the potential of skin-based microbiome–metabolome profiling as complementary forensic evidence for cause-of-death differentiation. However, validation across sexes, environmental conditions, and independent cohorts is necessary before forensic application. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2.