Re-Analysis of Public Transcriptomes Identifies a Candidate Histone-Cluster Signal in Eutopic Endometrium: A Hypothesis-Generating Link to Chronic Pelvic Pain

Abstract

Background

Endometriosis is a leading cause of chronic pelvic pain (CPP), but the public transcriptomic datasets analysed here contain no pain phenotype, links to CPP are therefore hypothesis-generating.

Methods

We re-analysed GSE153739 (four endometriosis, three controls) and tested replication in GSE51981 (77 cases, 34 disease-free controls). Non-reproducible differential-expression results from the original submission were removed. Both cohorts were rebuilt from source processed matrices. Histone-cluster scores were modelled with menstrual-cycle and proliferation covariates, and ranked enrichment was tested before and after histone-gene depletion. A separate ion-channel analysis was pre-specified before execution.

Results

Histone scores were higher in endometriosis in GSE153739 (β = 1.108, P = 0.047) and GSE51981 (β = 0.321, P = 0.0025; phase-adjusted β = 0.394, P = 0.00021). In GSE51981 the effect was concentrated in proliferative-phase endometrium (β = 0.665, P = 0.00083), with a disease-by-phase interaction (P = 0.020). S-phase or proliferation-PC1 adjustment attenuated the effect below nominal significance. In phase-adjusted analyses, SLE enrichment had q = 0.038, whereas NET formation was not enriched (q =0.73); removing clustered histones abolished enrichment in GSE51981 and reversed NET enrichment to negative. The pre-specified ion-channel endpoint was positive (phase-adjusted β = 0.240, P = 8 × 10⁻⁶; voltage-gated potassium NES = 1.550, q = 0.013), but the histone and ion-channel scores both covaried with proliferation measures and the same array-level technical axis.

Conclusion

A reproducible, proliferative-phase-concentrated histone-cluster signal is present in eutopic endometrium, but its biological and technical components cannot be separated in these data. Histone-rich KEGG annotations should not be interpreted as independent pathways. The data do not establish extracellular histone release, NETosis, or a causal mechanism for pelvic pain.

Introduction

Endometriosis affects approximately 10% of reproductive-aged women and is a major cause of chronic pelvic pain (CPP), infertility and healthcare burden.¹ Chronic pelvic pain is persistent or recurrent pain perceived in structures related to the pelvis, and endometriosis is an important gynaecological cause.^2,2 Despite the widespread use of gonadotrophin-releasing hormone antagonists, progestins and surgical excision, pain may persist or recur in a substantial proportion of patients.^1,3,4 The gap between anatomical disease control and symptom control suggests that pelvic pain in endometriosis has a molecular basis extending beyond visible implants.

Recent transcriptomic and single-cell studies have shown that endometriosis lesions are heterogeneous, with disease-specific stromal reprogramming, immune-niche remodelling and metabolic rewiring.^5,6 In parallel, work in chronic pain neurobiology has implicated cellular senescence, including p53-mediated senescence and the senescence-associated secretory phenotype, in persistent nociceptive states,^7–9 while microglial remodelling provides an additional mechanism of chronic pain maintenance.¹⁰ Neutrophil extracellular traps (NETs), web-like scaffolds of decondensed chromatin, histones, myeloperoxidase and neutrophil elastase released by activated neutrophils, have been implicated in female reproductive disease.^11–14 A 2025 case–control study of 95 patients with endometriosis reported that serum cell-free DNA, nucleosomes and neutrophil elastase are elevated and correlate with visual analogue scale pain scores.¹³

These observations have motivated interest in chromatin-associated transcriptional signals in endometrial tissue. However, such signals are difficult to interpret from pathway annotation alone. Replication-dependent histone genes are members of several unrelated KEGG gene sets, including systemic lupus erythematosus (SLE), NET formation and Alcoholism; coordinated up-regulation of the histone cluster will therefore illuminate all of these annotations simultaneously without implying that three distinct biological programmes are active. Coordinated histone-cluster up-regulation is also the canonical transcriptional signature of a difference in S-phase fraction or proliferative index between groups. Any report of such annotations must therefore distinguish a histone-cluster signal from three convergent pathways, and must distinguish a disease-associated signal from a proliferation artefact.

A further consideration is long non-coding RNA-mediated transcriptional control. LINC01638 has been characterised in multiple epithelial cancers as a regulator of proliferation and epithelial-to-mesenchymal transition-related programmes,^15–18 and has recently been shown to promote epithelial-to-mesenchymal transition in endometriosis epithelial cells by up-regulating RHOB through HDAC1 suppression, with knockdown producing G1 arrest and reduced proliferation, adhesion, migration and invasion.¹⁹ The dataset used here, GSE281916, derives from that experimental system. It is not a pain model, and its relevance to CPP is indirect.

In this revision we therefore ask three questions that can be answered with the available data. First, is a histone-cluster transcriptional signal present in eutopic endometrium from women with endometriosis? Second, is it reproducible in an independent cohort, and is it explained by proliferative index or menstrual-cycle phase? Third, do the KEGG annotations that carry histone genes as members retain enrichment when those genes are removed? We do not attempt to establish a mechanism for pelvic pain, and we state explicitly that no dataset analysed here contains a pain phenotype.

Materials and Methods

Datasets and Ethical Considerations

This study used only legally obtained, publicly available, de-identified transcriptomic data from the National Center for Biotechnology Information Gene Expression Omnibus (GEO) and involved no recruitment, intervention, collection of new human biological specimens, or access to identifiable private information. This study is therefore exempt from ethics review under national legislation, specifically Article 32, items (1) and (2), of the Measures for Ethical Review of Life Science and Medical Research Involving Human Subjects (Guo Wei Ke Jiao Fa [2023] No. 4), jointly issued on 18 February 2023 by the National Health Commission, the Ministry of Education, the Ministry of Science and Technology and the National Administration of Traditional Chinese Medicine of the People’s Republic of China. Article 32 provides that research using human information data or biological specimens which causes no harm to the human body, involves no sensitive personal information and involves no commercial interests may be exempt from ethics review where it uses legally obtained publicly available data, or data generated through observation that does not interfere with public behaviour (item 1), or where it uses anonymised information data (item 2). The present secondary analysis satisfies both conditions: all data were obtained from the NCBI Gene Expression Omnibus, a public repository, and all records analysed were publicly released and contained no direct participant identifiers accessible to the authors. Ethical approval and informed consent for the original human studies were the responsibility of the investigators who generated the source datasets.

GSE281916 is a perturbation experiment in the 12Z endometriosis epithelial cell line comprising three control small interfering RNA samples and three LINC01638 small interfering RNA samples collected 72 hours after transfection, profiled by 50-base-pair paired-end stranded total RNA sequencing with ribosomal RNA depletion on an Illumina NovaSeq 6000 platform. It is an endometriosis epithelial-cell perturbation model and is not patient tissue, pelvic tissue or pain tissue.

GSE153739 comprises eutopic endometrium from four women with endometriosis and three healthy controls, sequenced on an Illumina NextSeq 500 platform.²⁰ All seven samples were collected in the mid-proliferative phase of the menstrual cycle, so that cycle phase is matched by design. The source study used oligo(dT) poly(A) selection and a Tuxedo/Cufflinks–Cuffdiff processing workflow.

GSE51981 (Affymetrix HG-U133 Plus 2.0)²¹ served as an independent replication cohort. Samples annotated as non-endometriosis uterine pelvic pathology were excluded from the control group to retain disease-free comparators, yielding 77 endometriosis cases (all severities pooled) and 34 healthy controls (n = 111). Pooling all severities was used only for case–control replication and does not preserve the disease-stage structure of the original study.

Data Provenance and Reconstruction

This revision distinguishes source-study processed data from newly generated analyses. Both endometrial cohorts were rebuilt end to end from the source processed matrices within a single auditable workflow. For GSE153739, the complete seven-sample GEO-processed transcript-expression matrix was mapped to GENCODE v44 annotation, aggregated to gene level and transformed as log2(FPKM + 1). For GSE51981, the GEO processed matrix was collapsed deterministically from probes to genes, and array-level quality metrics were recomputed from the same matrix used for all downstream models. For GSE281916, GEO confirms the 12Z cell-line design.

We note explicitly that the gene-level differential-expression statistics reported in the original submission could not be reproduced from a documented count matrix and statistical model within the present audit trail. Rather than restate those values with revised wording, we have removed them from this revision. Every number reported below was regenerated in the reconstruction described here; earlier intermediate results were discarded rather than carried forward.

Scores and Statistical Models

A composite histone score was computed as the mean gene-wise z score across the prespecified replication-dependent histone-cluster genes. On the Affymetrix HG-U133 Plus 2.0 platform, 29 of the 36 prespecified leading-edge histone genes, and 49 of the 60 canonical clustered core histone genes, were represented after deterministic probe-to-gene mapping and collapse; the GSE51981 histone score was therefore calculated from the measurable genes, and the denominators are reported explicitly. In GSE153739 all 36 and all 60 genes were represented.

Proliferation was summarised by S-phase and G2/M scores and by a composite mean z score and by a principal component derived from canonical cell-cycle gene sets²² (42 of 43 S -phase and 52 of 54 G2/M genes represented) from which all replication-dependent histone-cluster genes were excluded, so that proliferation adjustment is not circular with respect to the histone score.

Composite-score disease models were fitted by ordinary linear regression. Gene-wise models used to generate the ranked statistics for enrichment analysis were fitted with limma using empirical Bayes moderation (eBayes, trend = TRUE, robust = TRUE).²³ For GSE51981, histone-score models were fitted crude and with adjustment for menstrual-cycle phase, phase plus S-phase score, phase plus G2/M score, and phase plus proliferation principal component 1; phase models exclude two samples of unknown phase (n = 109). Phase-stratified models and a formal disease-by-phase interaction test were fitted in addition. Because all seven GSE153739 samples were collected in the mid-proliferative phase, cycle phase cannot be modelled in that cohort, and the corresponding models are crude and adjusted for each proliferation measure in turn. Benjamini–Hochberg correction was applied across the model set in each family. Associations between the histone score, the proliferation scores and array-level quality metrics were assessed by Spearman correlation.

Gene Set Enrichment

Ranked enrichment for four prespecified KEGG annotations (SLE, NET formation, Alcoholism, and ATP-dependent chromatin remodeling) was performed with fgseaMultilevel²⁴ using set.seed(20260814), eps = 0, minSize = 10 and maxSize = 500, with genes ranked by the limma moderated t statistic for the disease coefficient. Because fgseaMultilevel uses adaptive multilevel estimation, a fixed user-specified permutation count does not apply. Benjamini–Hochberg q values were computed across the four prespecified pathways within each model.

The same frozen gene sets, the same ranking statistic, and the same enrichment framework and random seed were applied to both cohorts. No other enrichment engine or gene-set snapshot was used for any result reported in this manuscript.

These enrichment P values quantify the non-random positioning of pathway members within an observed ranked gene list. They are computed by gene-set permutation and do not incorporate between-subject sampling variability; accordingly, small enrichment q values in the seven-sample GSE153739 cohort must not be read as sample-level statistical certainty.

Histone-Depletion Sensitivity Analysis

Histone dependence was assessed under two prespecified definitions: the exact 36 histone genes shared by the original leading edges, and the broader set of 60 canonical clustered H2A, H2B, H3 and H4 genes. The first definition is a proper subset of the second, and no other histone gene definition was used. Depletion was implemented by removing these genes from the ranked list before enrichment testing, subject to the platform coverage stated in Scores and Statistical Models, and the same two definitions were applied to both cohorts.

Pre-Specified Ion-Channel Analysis

One further analysis was pre-specified in a document written and archived before execution, which fixed the gene sets, the models, the primary endpoints and the reporting rules for every possible outcome, including the rule that a negative result would be reported in the Results section. The hypothesis was inherited from the potassium-channel observation in the original submission, which arose from the differential-expression pipeline later found not to be reproducible; nothing from that analysis is carried forward as evidence.

Because the hypothesis originated in GSE153739, that cohort cannot confirm it. GSE51981 was pre-specified as the confirmatory cohort and GSE153739 as descriptive only, with an explicit rule that a GSE153739 result may not upgrade or alter the confirmatory conclusion. This is the reverse of the arrangement used for the histone analysis and is deliberate.

The GSE51981 processed matrix used for this analysis was assembled from the 111 official GEO Sample SOFT records rather than from the Series Matrix archive, after repeated failures to retrieve the Series Matrix intact. This departure from the original execution plan is recorded in the archived deviations log. We verified that it has no numerical consequence: the composite histone score, the S-phase, G2/M and composite proliferation scores, and all array-level quality metrics computed from the reassembled matrix are identical, to the limit of floating-point representation, to those computed from the Series Matrix and used in A Histone-Cluster Signal in GSE153739 and Its Sensitivity to Proliferation to Technical Covariation of the Histone and Proliferation Scores. The analyses in this section and in the preceding sections therefore rest on the same underlying data.

Four Gene Ontology gene sets were frozen in advance: inward rectifier potassium channel activity (GO:0005242), voltage-gated potassium channel activity (GO:0005249), voltage-gated sodium channel activity (GO:0005248), and ion channel activity (GO:0005216) as a breadth control. Pre-execution checks confirmed that none of these sets intersects the canonical clustered histone genes or the S-phase and G2/M gene sets. Platform coverage was 27/28, 95/98, 20/20 and 424/442 genes respectively, so no result below is attributable to inadequate coverage. A composite inward-rectifier score was constructed exactly as the histone score was, and modelled through the same five-model sequence; enrichment used the same engine and seed, with Benjamini–Hochberg correction across the four ion-channel sets, computed separately from the four KEGG sets and never pooled with them.

Code and Data Availability

Input matrices, frozen gene sets, analysis scripts with fixed random seeds, exact output tables including negative results, the pre-specification document for the analysis in Pre-Specified Ion-Channel Analysis, R sessionInfo and a SHA256 manifest are archived at Zenodo (DOI: 10.5281/zenodo.22097724). Source data are publicly available from GEO under accession numbers GSE281916, GSE153739 and GSE51981.

Results

Scope of the LINC01638 Perturbation Dataset

GSE281916 is a perturbation experiment in the 12Z endometriosis epithelial cell line rather than pelvic pain tissue or patient material. The associated primary study reports that LINC01638 knockdown reduces proliferation, adhesion, migration and invasion, induces G1 arrest, suppresses RHOB and increases HDAC1.¹⁹ This provides endometriosis-relevant cell-biological context and a plausible explanation for the breadth of the transcriptional response in this dataset, including its metabolic component, which is consistent with growth arrest rather than with a pain-specific programme. We therefore use this dataset only to frame LINC01638-dependent epithelial biology and draw no inference about nociception from it. Generic inflammatory pathway labels arising in this dataset are treated as annotation-level observations and not as evidence of disease-specific inflammatory biology.

A Histone-Cluster Signal in GSE153739 and Its Sensitivity to Proliferation

In GSE153739, the composite replication-dependent histone score was higher in endometriosis than in control endometrium (β = 1.108, standard error [SE] = 0.422, t = 2.62, P = 0.047). Adjustment for each proliferation measure in turn attenuated the estimate by 17–19% and moved it above the nominal significance threshold: S-phase score β = 0.902 (P = 0.112), G2/M score β = 0.917 (P = 0.089), composite proliferation mean β = 0.908 (P = 0.101), proliferation principal component 1 β = 0.921 (P = 0.093). With seven samples and one further degree of freedom consumed by each covariate, these models are uninformative as to whether the effect is proliferation-dependent, and we do not interpret them in either direction; the question is addressed in the larger cohort below.

Within the internally matched enrichment framework, the ranked enrichment of the four prespecified annotations behaved as shown in Table 1 (Figure 1).

Table 1.

Ranked Enrichment of Four Prespecified KEGG Annotations in GSE153739 (n = 7). Values are Normalised Enrichment Scores with Targeted Benjamini–Hochberg q Across the Four Prespecified Pathways

Model	SLE	NET Formation	Alcoholism	ATP-Dependent Chromatin Remodeling
Crude	2.654 (q = 7.3 × 10−15)	2.464 (q = 4.5 × 10−14)	2.372 (q = 3.2 × 10−13)	1.851 (q = 3.5 × 10−5)
+ proliferation mean	3.114 (q = 7.4 × 10−22)	2.449 (q = 1.7 × 10−13)	2.378 (q = 7.0 × 10−13)	1.290 (q = 0.033)
+ proliferation PC1	3.042 (q = 1.4 × 10−21)	2.476 (q = 4.1 × 10−14)	2.376 (q = 5.2 × 10−13)	1.234 (q = 0.107)
+ PC1, 36 histones removed	2.561 (q = 1.4 × 10−10)	1.921 (q = 3.2 × 10−6)	1.777 (q = 8.6 × 10−5)	−0.936 (q = 0.60)
+ PC1, 60 histones removed	2.410 (q = 6.1 × 10−7)	1.675 (q = 4.9 × 10−4)	1.471 (q = 0.008)	−1.127 (q = 0.26)

Abbreviations: SLE, systemic lupus erythematosus; NET, neutrophil extracellular trap; ATP, adenosine triphosphate; KEGG, Kyoto Encyclopedia of Genes and Genomes; PC1, proliferation principal component 1; q, targeted Benjamini–Hochberg adjusted P value.

Figure 1.

Ranked gene set enrichment of four prespecified KEGG annotations in GSE153739 (n = 7), crude model. Enrichment plots are shown for (A) systemic lupus erythematosus, (B) neutrophil extracellular trap formation, (C) Alcoholism and (D) ATP-dependent chromatin remodeling, with genes ranked by the limma moderated t statistic for the disease coefficient. Tick marks below each curve indicate the positions of pathway members in the ranked list; the filled point marks the peak of the running enrichment score. Annotations (A-C) share extensive replication-dependent histone-cluster membership and are not interpreted as independently activated biological programmes.

Proliferation adjustment did not abolish enrichment of the three histone-rich annotations in this cohort. Removal of clustered histone genes attenuated all three and reversed the sign of ATP-dependent chromatin remodeling, indicating that the last annotation was entirely histone-driven. Residual positive enrichment after histone depletion is not, by itself, evidence of histone independence, because ranked enrichment reflects the distribution of all remaining pathway members. We note that after removal of all 60 canonical clustered histones the NET-formation leading edge in this cohort was composed predominantly of neutrophil and complement effectors, including NCF2, NCF4, CYBB, CYBA, C5AR1, C3, ITGB2, ITGAM, ITGAL, FPR1, RAC1, RAC2, CASP1, SELPLG, SYK, GSDMD and ELANE.

Independent Replication in GSE51981

In the independent GSE51981 cohort the composite histone score was higher in endometriosis than in disease-free endometrium (β = 0.321, SE = 0.104, t = 3.10, P = 2.5 × 10⁻³; n = 111). Adjustment for menstrual-cycle phase strengthened the estimate (β = 0.394, SE = 0.103, t = 3.84, P = 2.1 × 10⁻⁴; n = 109). Adjustment for the G2/M score left it little changed (β = 0.330, SE = 0.114, t = 2.89, P = 4.7 × 10⁻³), whereas adjustment for the S-phase score (β = 0.181, SE = 0.118, t = 1.53, P = 0.129) or for proliferation principal component 1 (β = 0.230, SE = 0.118, t = 1.95, P = 0.054) attenuated it below nominal significance (Table 2, Figure 2).

Table 2.

Composite Histone-Score Disease Coefficients in GSE51981 Under Progressive Covariate Adjustment. Phase-Adjusted Models Exclude Two Samples of Unknown Cycle Phase

Model	n	β	SE	P	BH q
Crude	111	0.321	0.104	2.5 × 10−3	0.006
+ menstrual-cycle phase	109	0.394	0.103	2.1 × 10−4	0.001
+ phase + G2/M score	109	0.330	0.114	4.7 × 10−3	0.008
+ phase + proliferation PC1	109	0.230	0.118	0.054	0.068
+ phase + S-phase score	109	0.181	0.118	0.129	0.129

Abbreviations: n, number of samples; β, disease coefficient; SE, standard error; BH q, Benjamini–Hochberg adjusted P value; G2/M, gap 2/mitosis; PC1, proliferation principal component 1.

Figure 2.

Histone-score models and enrichment sensitivity. (A) Composite histone-score disease coefficients with 95% confidence intervals in GSE153739 (crude and adjusted forproliferation principal component 1) and GSE51981 (crude, phase-adjusted, and phase-adjusted with G2/M, proliferation principal component 1 or S-phase score as an additional covariate). (B) Phase-stratified histone-score coefficients in GSE51981 for proliferative, early-secretory and mid-secretory endometrium; the disease-by-phase interaction was F = 4.08, df = 2, P = 0.020. (C) Normalised enrichment scores for the systemic lupus erythematosus annotation in GSE51981 across the five-model sensitivity sequence (the two histone-depleted models are also adjusted for phase and proliferation PC1); bars are shaded by whether the targeted Benjamini–Hochberg q value is below 0.05. (D) Normalised enrichment scores for all four prespecified annotations in GSE51981 before and after removal of all 60 canonical clustered histone genes. Targeted Benjamini–Hochberg q values are shown above the bars in panel C. Negative enrichment is shown in red in (D).

Phase-stratified analysis localised the effect to proliferative-phase endometrium (β = 0.665, SE = 0.186, t = 3.57, P = 8.3 × 10⁻⁴, BH q = 0.002; n = 49, 20 controls and 29 cases), with no detectable difference in early-secretory (β = 0.103, P = 0.46; n = 24) or mid-secretory endometrium (β = 0.089, P = 0.37; n = 36). A formal test of the disease-by-phase interaction confirmed that these estimates differ (F = 4.08, df = 2, P = 0.020), with negative interaction terms for early-secretory (β = −0.562, P = 0.034) and mid-secretory tissue (β = −0.577, P = 0.016). The concentration of the effect in proliferative-phase endometrium is therefore supported by a formal interaction test rather than by subgroup significance alone, and it is consistent with the discovery cohort, in which all seven samples were collected in the mid-proliferative phase (Figure 2B).

Within proliferative-phase samples alone, the pattern of covariate sensitivity seen in the whole cohort was reproduced: adjustment for the G2/M score left the disease coefficient largely intact (β = 0.637, P = 0.013), whereas adjustment for the S-phase score (β = 0.310, P = 0.24) or proliferation principal component 1 (β = 0.388, P = 0.15) attenuated it below nominal significance.

The histone score was negatively correlated with proliferation measures in this cohort: Spearman ρ = −0.276 with the S-phase score (P = 3.3 × 10⁻³), −0.206 with proliferation principal component 1 (P = 0.030), −0.174 with the composite proliferation z score (P = 0.068) and −0.094 with the G2/M score (P = 0.33). Negative correlations of similar magnitude were present within each cycle phase considered separately. The direction of these associations is opposite to that predicted if a coordinated replication-dependent histone signal were a direct reflection of a higher S-phase fraction in cases.

Ranked enrichment in GSE51981 is shown in Table 3. The SLE annotation reached adjusted significance in the phase-adjusted model (NES = 1.419, q = 0.038) and in the phase- and proliferation-adjusted model (NES = 1.397, q = 0.045), but not in the crude model (NES = 1.383, q = 0.070). NET formation showed no enrichment in any model. Removal of clustered histone genes eliminated the significant SLE enrichment; none of the four annotations remained significantly enriched after depletion, and the NET-formation NES became negative (NES = −0.850 and −0.860 after removal of 36 and 60 histone genes respectively).

Table 3.

Ranked Enrichment of Four Prespecified KEGG Annotations in GSE51981. Values are Normalised Enrichment Scores with Targeted Benjamini–Hochberg q Across the Four Prespecified Pathways

Model	SLE	NET Formation	Alcoholism	ATP-Dependent Chromatin Remodeling
Crude	1.383 (q = 0.070)	0.851 (q = 0.83)	1.125 (q = 0.30)	−1.221 (q = 0.27)
+ phase	1.419 (q = 0.038)	0.927 (q = 0.73)	1.228 (q = 0.10)	−1.154 (q = 0.27)
+ phase + PC1	1.397 (q = 0.045)	0.968 (q = 0.59)	1.184 (q = 0.22)	−1.031 (q = 0.54)
+ phase + PC1, 36 histones removed	1.183 (q = 0.40)	−0.850 (q = 0.82)	1.099 (q = 0.40)	−1.080 (q = 0.40)
+ phase + PC1, 60 histones removed	1.202 (q = 0.45)	−0.860 (q = 0.75)	1.108 (q = 0.45)	−1.053 (q = 0.49)

Abbreviations: SLE, systemic lupus erythematosus; NET, neutrophil extracellular trap; ATP, adenosine triphosphate; KEGG, Kyoto Encyclopedia of Genes and Genomes; PC1, proliferation principal component 1; q, targeted Benjamini–Hochberg adjusted P value.

Technical Covariation of the Histone and Proliferation Scores

Because replication-dependent histone transcripts have atypical 3′ end processing and are not uniformly polyadenylated,²⁵ and because both the histone and proliferation scores are derived from the same expression matrix, we examined whether these scores covaried with array-level technical metrics recomputed from that matrix.

The histone score was associated with expression-matrix principal component 1 (Spearman ρ = −0.630, P = 1.3 × 10⁻¹³), with the interquartile range of array intensity (ρ = −0.412, P = 6.9 × 10⁻⁶) and with sample processing order (ρ = −0.367, P = 7.3 × 10⁻⁵). The proliferation scores were associated with the same metrics in the opposite direction: the S-phase score correlated with expression-matrix principal component 1 at ρ = +0.631 and with array interquartile range at ρ = +0.622, and proliferation principal component 1 at ρ = +0.573 and ρ = +0.553 respectively (Figure 3).

Figure 3.

Per-sample composite histone score and its technical covariation in GSE51981 (n = 111). (A) Histone score by disease status; horizontal bars indicate group means with 95% confidence intervals. (B) Histone score by disease status within each menstrual-cycle phase. (C) Histone score against the S-phase score. (D) Histone score and the S-phase score against expression-matrix principal component 1, showing that the two scores load on the same array-level technical axis with opposite sign. Spearman coefficients are shown in panels C and D.

The histone and proliferation scores therefore load on a common array-level technical axis with opposite sign. This has two consequences that we state explicitly rather than resolve. First, the negative correlation between the histone score and the proliferation measures may be technically induced and should not be read as a biological anticorrelation. Second, adjustment for a proliferation score in this cohort is in part adjustment for array-level variation, so the attenuation of the disease coefficient after S-phase or proliferation-principal-component adjustment cannot be attributed to proliferative biology alone. Proliferation-related and technical contributions to the histone-cluster signal cannot be separated in these data.

Pre-Specified Ion-Channel Analysis

The potassium-channel observation reported in the original submission arose from a pipeline that was subsequently found not to be reproducible. Rather than discard the hypothesis or restate it, we tested it under a pre-specification written and archived before execution (Pre-Specified Ion-Channel Analysis), with GSE51981 as the confirmatory cohort.

The pre-specified primary endpoint was met. In GSE51981 the composite inward-rectifier score was higher in endometriosis than in disease-free endometrium (crude β = 0.254, P = 2 × 10⁻⁶; phase-adjusted β = 0.240, P = 8 × 10⁻⁶), and enrichment of the voltage-gated potassium channel gene set reached adjusted significance in the phase-adjusted model (NES = 1.550, targeted BH q = 0.013), together with the breadth-control ion channel activity set (NES = 1.243, q = 0.013). However, the composite score was strongly associated with array-level technical structure (Spearman ρ = −0.675 with expression-matrix PC1, BH-adjusted P = 2.9 × 10⁻¹⁵; ρ = −0.567 with array interquartile range, BH-adjusted P = 2.6 × 10⁻¹⁰; ρ = −0.283 with sample processing order, BH-adjusted P = 5.2 × 10⁻³; and ρ = +0.238 with array median and with array median absolute deviation, BH-adjusted P = 0.014, in the opposite direction) and with the proliferation measures (ρ = −0.724 with the S-phase score; ρ = −0.700 with proliferation PC1; ρ = −0.639 with the G2/M score), and it was markedly attenuated by adjustment for the S-phase score (β = 0.054, P = 0.30) or proliferation PC1 (β = 0.065, P = 0.21), though less so by adjustment for the G2/M score (β = 0.117, P = 0.022). The signal therefore meets the pre-specified criterion for support while remaining subject to the same confounding structure as the histone signal, and we report the two facts together (Table 4).

Table 4.

Pre-Specified Ion-Channel Analysis in GSE51981 (n = 111; Phase-Adjusted Models n = 109). Composite Inward-Rectifier Score Models are Shown Above, Ranked Enrichment of the Four Pre-Specified Gene Sets in the Phase-Adjusted Model Below

A. Composite inward-rectifier score model
	β	P	BH q
Crude	0.254	2 × 10−6	1.0 × 10−5
+ menstrual-cycle phase (primary endpoint)	0.240	8 × 10−6	1.9 × 10−5
+ phase + G2/M score	0.117	0.022	0.037
+ phase + proliferation PC1	0.065	0.21	0.26
+ phase + S-phase score	0.054	0.30	0.30
B. Pre-specified gene-set enrichment (phase-adjusted)
	Genes measured	NES	Targeted BH q
Voltage-gated potassium channel activity	95/98	1.550	0.013
Ion channel activity (breadth control)	424/442	1.243	0.013
Inward rectifier potassium channel activity	27/28	1.116	0.36
Voltage-gated sodium channel activity	20/20	0.930	0.56

Abbreviations: β, disease coefficient; P, P value; BH q, Benjamini–Hochberg adjusted P value; NES, normalised enrichment score; G2/M, gap 2/mitosis; PC1, proliferation principal component 1.

Three features of this result constrain its interpretation and are reported because the pre-specification required it. First, the composite score and the enrichment analysis do not point to the same gene set: the score was built from the inward-rectifier term, which was the specific observation in the original submission, yet that term did not reach adjusted significance in the ranked analysis, and the enrichment instead appeared in the broader voltage-gated potassium set and in the breadth-control set, which are substantially nested. Second, the voltage-gated sodium channel set, which motivated the inclusion of a sodium-channel term because of the reported role of NaV1.7 in endometriosis-associated pain, showed no enrichment (NES = 0.930, q = 0.56); this literature is therefore retained only as the stated motivation for the gene-set choice and as a future direction, and is not supported by these data. Third, although the effect was numerically strongest in proliferative-phase samples (β = 0.322, P = 3.4 × 10⁻⁵, BH q across phases = 1.0 × 10⁻⁴; early secretory β = 0.197, P = 0.088; mid secretory β = 0.117, P = 0.24), the formal disease-by-phase interaction was not significant (P = 0.215), so the phase concentration demonstrated for the histone score does not extend to this signal.

In the descriptive GSE153739 cohort the composite score was directionally consistent (crude β = 0.301, P = 0.022; β between 0.236 and 0.246, P between 0.025 and 0.035, across the proliferation-adjusted models), but the ranked analysis did not support the potassium sets (inward rectifier NES = 1.292, q = 0.19; voltage-gated potassium NES = −1.035, q = 0.38), and the voltage-gated sodium set showed strong negative enrichment (NES = −2.233, q = 1.9 × 10⁻⁵) in the direction opposite to the confirmatory cohort. Under the pre-specified rules, these results are descriptive and do not modify the conclusion; they are reported because they are part of the archived output.

Table 5 sets out the full correlation structure. Both composite scores and all three proliferation measures load on the same array-level axis: the histone and ion-channel scores negatively, the proliferation scores positively, with absolute coefficients between 0.41 and 0.68 against expression-matrix principal component 1 and against array interquartile range. The two composite scores share no genes with each other and no genes with the cell-cycle sets, yet all five quantities are organised by this common structure.

Table 5.

Spearman Correlations of the Composite Histone Score, the Composite Ion-Channel Score and the Proliferation Scores Against Array-Level Quality Metrics and Against One Another in GSE51981 (n = 111). P Values in Panel B Are Nominal; Benjamini–Hochberg Adjustment Was Applied Correlation Family as Archived

A. Correlations with array-level quality metrics
	Array PC1	Array IQR	Array median/MAD		Sample order
Histone score	−0.630	−0.412	+0.004		−0.367
Ion-channel score	−0.675	−0.567	+0.238		−0.283
S-phase score	+0.631	+0.622	−0.257		+0.393
G2/M score	+0.445	+0.444	−0.176		+0.267
Proliferation PC1	+0.573	+0.553	−0.212		+0.337
B. Between-score correlations
		Spearman ρ		P
Ion-channel score vs histone score		+0.458		4.2 × 10−7
Ion-channel score vs S-phase score		−0.724		2.8 × 10−19
Ion-channel score vs proliferation PC1		−0.700		1.2 × 10−17
Histone score vs S-phase score		−0.276		3.3 × 10−3

Abbreviations: Array PC1, principal component 1 of the expression matrix; proliferation PC1, principal component 1 derived from canonical cell-cycle gene sets (histone genes excluded); IQR, interquartile range; MAD, median absolute deviation; ρ, Spearman correlation coefficient; P, nominal P value.

Finally, the ion-channel and histone scores were themselves correlated at the sample level (Spearman ρ = +0.458, P = 4.2 × 10⁻⁷), despite sharing no genes. The two signals are therefore not independent readouts of the same tissue, a point developed in What These Data Support.

Discussion

What These Data Support

Three findings are supported. First, a replication-dependent histone-cluster transcriptional signal is higher in eutopic endometrium from women with endometriosis in two independent cohorts, in crude and cycle-phase-adjusted models. Second, this signal is concentrated in proliferative-phase endometrium: the effect was detected in proliferative-phase samples and not in secretory-phase samples, and a formal disease-by-phase interaction test confirms that the phase-specific estimates differ. We describe the effect as concentrated rather than restricted, because absence of a detectable difference in the secretory phases is not evidence that the true effect there is zero. This is a tested finding rather than a subgroup observation, and it explains why the signal was detectable in a seven-sample discovery cohort in which all samples were mid-proliferative. It also implies that cycle phase must be matched or modelled in any attempt to replicate this observation.

Third, and most securely, the KEGG annotations previously reported as co-activated are not three convergent biological programmes. Removal of clustered histone genes abolished enrichment of all four prespecified annotations in the 111-sample cohort and reversed the NET-formation annotation to negative. Coordinated up-regulation of replication-dependent histone genes is a recognised source of apparent co-enrichment across unrelated KEGG gene sets, and we have tested that possibility directly rather than acknowledged it narratively. The manuscript is therefore framed around a histone-cluster signal rather than around any KEGG label that carries histone genes as members.

A fourth observation emerged from the pre-specified ion-channel analysis and, in our view, is the most generalisable result of this work. Two transcriptional signals that share no genes—the replication-dependent histone cluster and an inward-rectifier ion-channel module—were both elevated in endometriosis, were correlated with each other at the sample level (ρ = +0.458), were both attenuated by S-phase adjustment, and both covaried strongly with the same array-level technical axis, in the same direction as each other and opposite to the proliferation scores. Analysed separately and annotated through different ontologies, they would read as two independent biological findings. Analysed together, they behave as different projections of one underlying structure in which disease status, proliferative state and assay-level variation are not separable.

The correlation structure in Table 5 makes the point concretely. Five quantities—two composite scores that share no genes with each other or with the cell-cycle sets, and three proliferation measures—all load on the same array-level axis, the two composite scores in one direction and the proliferation measures in the other. Whichever gene set is chosen, the resulting score carries this structure with it.

This has a practical implication beyond the present datasets. Gene sets that carry the replication-dependent histone cluster as members, and composite scores built from co-regulated modules, will both respond to that structure. Where cycle phase, proliferative index and assay chemistry are not controlled by design, apparently convergent pathway results in endometrial transcriptomes should be tested for shared membership and for common technical loading before they are interpreted as convergent biology. We provide both tests, and the archived code to reproduce them, as the concrete contribution of this re-analysis.

What These Data Do Not Support, and What Remains Unresolved

These transcriptomic data do not demonstrate extracellular histone release, citrullinated-histone NETosis, protein arginine deiminase 4 activation, or any causal relationship to pain. No dataset analysed here contains a pain phenotype, a pain-stratified contrast or a nociception readout. The NET-formation annotation, which framed the original submission, showed no enrichment in the independent cohort in any model and became negative after histone depletion. GSE281916 is an epithelial-cell perturbation experiment whose published phenotype is growth arrest and modulation of epithelial-to-mesenchymal transition; it provides no evidence about nociception.

The origin of the histone-cluster signal itself also remains unresolved, and we consider it important to state the limits of what the present analysis can decide. The disease coefficient was attenuated below nominal significance by adjustment for the S-phase score and for proliferation principal component 1, although it was largely unchanged by adjustment for the G2/M score. A proliferation-related contribution therefore cannot be excluded. At the same time, the interpretation of those adjusted models is itself constrained: the histone and proliferation scores load on a common array-level technical axis with opposite sign, so adjusting for a proliferation score in this cohort is in part adjusting for array-level variation. The observed negative correlation between the histone and proliferation scores is consistent with that technical structure and is opposite in direction to what a straightforward S-phase-fraction explanation would predict.

The same reasoning applies to the pre-specified ion-channel result. It met its pre-specified criterion for support, and we report it as such; but the composite score was more strongly coupled to the proliferation measures and to the array-level technical axis than the histone score was. Its score-level and enrichment-level results identified different gene sets, and its apparent phase concentration was not confirmed by a formal interaction test. It is a signal that is present and reproducibly measurable, not a mechanism that has been identified.

We therefore do not claim that the signal is proliferation-independent, and we do not claim that it is a proliferation artefact. Both proliferation-related and technical contributions remain possible, and they cannot be separated from disease status in these data. Resolving this requires cohorts in which cycle phase, proliferative index and library or array chemistry are controlled by design rather than by covariate adjustment after the fact.

Relationship to Chronic-Pain Biology Is Analogy, Not Evidence

The senescence and neuroimmune literature cited here derive from spinal microglia and dorsal root ganglia,^8–10 and the ion-channel literature from an experimental endometriosis model.²⁶ These systems are tissue-mismatched to eutopic endometrium. They are used to motivate testable hypotheses and are explicitly not offered as evidence that the same mechanisms operate in the samples analysed here. Figure 4 is presented as a hypothesis map; the datasets were analysed independently and are not statistically linked, and none of the relationships depicted is tested by the present data.

Figure 4.

Hypothesis-generating map integrating independently observed transcriptomic themes across the three datasets. Dashed arrows denote proposed relationships requiring direct experimental validation and should not be interpreted as demonstrated causal links. No pain phenotype, protein-level assay of neutrophil extracellular trap formation, or causal test was performed in this re-analysis.

Strengths and Limitations

The strengths of this study are the end-to-end reconstruction of both cohorts within a single auditable workflow, using one ranking statistic, one frozen gene-set snapshot, one enrichment framework and one random seed; an independent replication cohort of 111 samples; a formal test of the disease-by-phase interaction rather than reliance on subgroup comparison; direct testing of the proliferation and annotation-artefact explanations rather than their narrative acknowledgement, including quantification of the technical covariation that limits those tests; one analysis specified in full, including its reporting rules for every outcome, before it was run; and the archiving of inputs, scripts and outputs, including negative results, for independent verification.

The limitations are substantial. First, the GSE153739 discovery cohort comprises seven samples; enrichment statistics conditional on a ranked gene list do not substitute for subject-level precision, and every covariate-adjusted model in that cohort is uninformative in either direction. Second, proliferation-related confounding cannot be excluded, as set out in What These Data Do Not Support, and What Remains Unresolved. Third, both the histone and proliferation scores covary with array-level technical metrics, so technical confounding cannot be excluded either, and the two cannot be separated. Fourth, GSE153739 libraries were prepared by oligo(dT) poly(A) selection; replication-dependent histone messenger RNAs are non-polyadenylated and terminate in a 3′ stem-loop,²⁵ and in poly(A)-selected libraries they are captured only through the minor oligo-adenylated fraction, whose abundance is sensitive to RNA integrity and 3′ end processing. Fifth, the Affymetrix platform used for GSE51981 represented 29 of the 36 prespecified leading-edge histone genes and 49 of the 60 canonical clustered core histones, so the replication histone score is computed from incomplete coverage of the prespecified set. Sixth, both endometrial cohorts profile eutopic rather than ectopic tissue, and the GSE51981 replication pools all disease severities. Seventh, bulk transcriptomics cannot identify the cell population generating the histone signal. Eighth, the pre-specified ion-channel analysis met its primary endpoint but is subject to the same and, on the measured correlations, somewhat stronger confounding structure; its score-level and enrichment-level results do not converge on the same gene set, and the sodium-channel set that motivated its inclusion was negative. Ninth, gene-level differential-expression results from the discovery cohorts are not reported in this revision because they could not be reproduced from a documented count matrix and statistical model; any future report of such results should regenerate them from an explicitly specified pipeline.

Future Directions

Validation should precede therapeutic nomination. The priorities are cycle-phase-matched and cell-cycle-matched endometrial cohorts profiled with total RNA rather than poly(A)-selected libraries; measurement of extracellular nucleosomes and histones in tissue, peritoneal fluid or menstrual effluent; citrullinated histone H3, myeloperoxidase and neutrophil elastase assays if NETosis is specifically hypothesised; and single-cell or spatial profiling to identify the cellular source of the signal. Protein arginine deiminase 4 inhibition, deoxyribonuclease I, senolytic and ion-channel strategies remain conditional on such validation and are not proposed here as therapeutic conclusions.

Conclusions

This re-analysis identifies a replication-dependent histone-cluster transcriptional signal in eutopic endometrium from women with endometriosis. The signal is present in two independent cohorts and is concentrated in proliferative-phase endometrium, with a significant disease-by-phase interaction. The systemic lupus erythematosus, neutrophil extracellular trap formation and Alcoholism KEGG annotations previously reported as co-activated reflect shared histone-cluster membership: after histone depletion in the larger cohort none of the four prespecified annotations remained significantly enriched, and the neutrophil extracellular trap annotation became negative. The origin of the histone signal itself is not resolved by these data. The disease effect was attenuated by S-phase and proliferation-principal-component adjustment though not by G2/M adjustment, and the histone and proliferation scores load on a common array-level technical axis with opposite sign, so proliferation-related and technical contributions cannot be excluded or separated. A separately pre-specified analysis met its primary endpoint for an inward-rectifier ion-channel signal, but that signal shows the same confounding structure and, despite sharing no genes with the histone score, is correlated with it at the sample level. The most generalisable result of this work is therefore methodological: two signals that would read as independent discoveries if analysed separately behave as projections of one structure in which disease status, proliferative state and assay-level variation are not separable, and pathway findings in endometrial transcriptomes should be tested for shared gene membership and common technical loading before being interpreted as convergent biology. These data do not establish extracellular histone release, canonical NETosis, or any causal mechanism for chronic pelvic pain; none of the datasets analysed contains a pain phenotype. The findings are hypothesis-generating and define a specific target for validation in cohorts where cycle phase, proliferative index and assay chemistry are controlled by design.

Funding Statement

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Data Sharing Statement

All datasets analysed in this study are publicly available from the NCBI Gene Expression Omnibus under accession numbers GSE281916, GSE153739 and GSE51981. Analytical scripts, frozen gene sets, the pre-specification document and all output tables are archived as described in Code and Data Availability.

Ethics Approval

This study used only legally obtained, publicly available, de-identified transcriptomic data from the NCBI Gene Expression Omnibus and involved no recruitment, intervention, collection of new human biological specimens, or access to identifiable private information. This study is therefore exempt from ethics review under national legislation, specifically Article 32, items (1) and (2), of the Measures for Ethical Review of Life Science and Medical Research Involving Human Subjects (Guo Wei Ke Jiao Fa [2023] No. 4), jointly issued on 18 February 2023 by the National Health Commission, the Ministry of Education, the Ministry of Science and Technology and the National Administration of Traditional Chinese Medicine of the People’s Republic of China. Article 32 provides that research using human information data or biological specimens which causes no harm to the human body, involves no sensitive personal information and involves no commercial interests may be exempt from ethics review where it uses legally obtained publicly available data, or data generated through observation that does not interfere with public behaviour (item 1), or where it uses anonymised information data (item 2). The present secondary analysis satisfies both conditions: all data were obtained from the NCBI Gene Expression Omnibus, a public repository, and all records analysed were publicly released and contained no direct participant identifiers accessible to the authors. Ethical approval and informed consent for the original human studies were the responsibility of the investigators who generated the source datasets.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors report no conflicts of interest in this work.