Uncovering novel therapeutic targets for premature ejaculation from gut microbiota: A prospective high-throughput sequencing study

Men who responded to dapoxetine had different gut bacteria than non-responders, but the link was weak

Journal: Andrology | Published: 2025-07-22 | Type: Journal Article, Research Support, Non-U.S. Gov't | PMID: 40693335 Authors: Hou Guangdong, Zhang Geng, Zheng Yu, Zhang Siyan (Tangdu Hospital, Fourth Military Medical University, Xi'an); Federici Massimo, Jannini Emmanuele A (University of Rome Tor Vergata); Meng Ping, Wang Fuli, Zhang Bo, Yuan Jianlin (Xijing Hospital, Fourth Military Medical University, Xi'an) Funding/COI: Natural Science Basic Research Program of Shaanxi Province, National Natural Science Foundation of China, Rapid Response Program Project of the Fourth Military Medical University; COI not disclosed in the abstract

Summary

This prospective study sequenced fecal bacteria (16S rDNA) in men with lifelong premature ejaculation before they started dapoxetine, then compared microbial profiles between men who responded to treatment and those who didn't, per Hou et al., 2025. Four bacterial taxa differed between groups and modestly improved a statistical model's ability to predict who would respond. The associations were real but weak, and the study can't say whether the bacteria influence drug response or just happen to track with it.

Claims

• Erysipelotrichaceae_UCG_003, Parabacteroides distasonis, and Prevotella_7_unclassified were more abundant in the effective (responder) group; Collinsella aerofaciens was less abundant.
• These four taxa correlated with CGIC improvement scores at coefficients of 0.331, 0.250, 0.288, and −0.345 respectively — weak-to-moderate associations.
• Individually, each taxon discriminated responders from non-responders with AUCs of 0.654, 0.669, 0.701, and 0.615 — barely-to-modestly better than chance (0.5).
• Adding the four taxa to a clinical prediction model raised its accuracy from AUC 0.738 to 0.796.
• PICRUSt2 functional predictions showed higher inferred gene abundance in the arachidonic acid metabolism pathway in the responder group, proposed as a possible mechanistic link.

Study Quality

The prospective design — collecting stool before treatment rather than after — is a genuine strength, since it avoids the confound of dapoxetine itself reshaping the microbiome before analysis. But the abstract omits the sample size entirely, making it impossible to judge statistical power or how much confidence to put in the LEfSe-selected taxa. The correlation coefficients (0.25–0.35) mean these bacteria explain well under 15% of the variance in treatment response, and the individual AUCs (0.61–0.70) are weak discriminators on their own.

The arachidonic acid pathway finding is two steps removed from direct measurement: PICRUSt2 infers gene content from 16S taxonomic data, it doesn't sequence genes or measure metabolites. That's a prediction built on a prediction, not confirmed pathway activity. There's also no mention of cross-validation or a held-out test set for the improved 0.796 AUC model, which raises the usual overfitting concern when a small number of researcher-selected features are added to a model and evaluated on the same cohort that generated them.

Red Flags

• Sample size not reported in the abstract — power and generalizability can't be assessed
• Correlation coefficients are weak (r = 0.25–0.345); most variance in treatment response is unexplained by these taxa
• 16S rDNA sequencing has limited resolution for species-level calls (e.g., Parabacteroides distasonis, Collinsella aerofaciens); amplicon-based species assignments warrant caution
• Arachidonic acid pathway result is a computational inference (PICRUSt2) from 16S data, not measured metabolomics or metagenomics
• No stated cross-validation or external validation cohort for the improved prediction model, raising overfitting risk
• No placebo arm or drug-free control group — all patients received dapoxetine, so causality between microbiota and response can't be established
• No correction for multiple comparisons mentioned for the LEfSe taxa screen

Strengths

• Prospective, pre-treatment sampling design avoids reverse causation from the drug altering gut flora before analysis
• Uses a standard, validated outcome measure (CGIC) to define treatment response
• Multi-institutional collaboration spanning urology, nuclear medicine, and systems medicine departments
• Government rather than pharmaceutical funding, reducing conflict-of-interest concerns tied to dapoxetine's manufacturer

Verdict

This is a hypothesis-generating pilot, not a practice-changing finding: it's the first study to associate gut bacteria with dapoxetine response, but the correlations are weak, the sample size is undisclosed, the mechanistic pathway claim rests on an inference stacked on an inference, and there's no indication the improved prediction model was validated outside its own dataset. Worth a footnote for researchers watching the gut-microbiome-and-sexual-health space, not a signal that probiotics or microbiome tests belong anywhere near PE treatment yet.