Artificial sweeteners and male infertility: Network toxicology and experimental evidence reveal FGFR1 as the critical target.

Computational screening flagged aspartame, neotame, and sucralose as reproductive toxins targeting FGFR1; validated only in mouse Leydig cells in a dish

Journal: Reproductive Toxicology | Published: 2026-02-17 | Type: Journal Article | PMID: 41713522 Authors: Meng Chunyang et al. (Beijing Anzhen Nanchong Hospital of Capital Medical University; Affiliated Hospital of Southwest Medical University, Luzhou, Sichuan) Funding/COI: Funding not disclosed. Authors declare no competing financial interests.

Summary

This is a network toxicology paper — mostly computational — that screened seven artificial sweeteners for reproductive toxicity signals and landed on FGFR1, a fibroblast growth factor receptor, as the putative target. Aspartame got the most attention. The only wet-lab work involved exposing TM3 mouse Leydig cells to aspartame for 48 hours and measuring gene expression changes. No animals were harmed, no humans were studied, and no protein-level biology was confirmed.

Claims

• Aspartame, neotame, and sucralose scored "high-confidence" reproductive toxicity signals in AdmetSAR 3.0 software
• 91 candidate targets identified from database intersections; FGFR1 ranked as the core gene via machine-learning feature selection
• Molecular docking and 100-ns molecular dynamics simulations showed stable aspartame-FGFR1 binding in silico
• Single-cell RNA-seq (Male Health Atlas dataset) showed FGFR1 expression concentrated in Leydig and vascular-associated testicular cells
• In TM3 mouse Leydig cells at 0.5–2 mM aspartame for 48 h: FGFR1, DUSP6, and SPRY2 were upregulated; STAR (a steroidogenesis regulator) was downregulated

Study Quality

This is hypothesis-generation dressed up as mechanism discovery. Network toxicology — stringing together database queries, target prediction algorithms, and computational docking — is a legitimate first step, but it produces a list of plausible targets, not causal evidence. The authors know this: their conclusion explicitly calls for "in vivo and protein-level validation," which they did not do. The in vitro validation uses TM3 cells, a mouse Leydig cell tumor line, at 0.5–2 mM concentrations; whether those concentrations reflect realistic human exposure to aspartame is never discussed.

The machine-learning feature selection sounds rigorous but operates on small gene lists derived from database overlaps — an analysis highly sensitive to which databases you query and how you set thresholds. Validation against two independent datasets improves credibility slightly, but both datasets were also computational, not experimental.

Red Flags

• No in vivo data — zero animal experiments
• No human data
• TM3 cells are a mouse Leydig cell tumor line, not normal human Leydig cells
• Aspartame concentrations (0.5–2 mM) are not benchmarked against realistic human blood or tissue levels after dietary intake
• No protein-level validation: no Western blot, no immunofluorescence, no functional steroidogenesis assay
• Only RT-qPCR: gene expression shifts measured, not downstream functional effects on testosterone production
• Funding source not disclosed — relevant when evaluating selective reporting risk
• Network toxicology methodology is highly dependent on database completeness and threshold choices, which can be tuned to implicate almost any target

Strengths

• Multi-tool computational convergence: three target-prediction databases, PPI network analysis, machine learning, molecular docking, and MD simulation
• Authors appropriately hedge conclusions and explicitly call for further validation
• Single-cell RNA-seq mapping grounds FGFR1's testicular relevance in real expression data
• 100-ns molecular dynamics provides more binding stability information than static docking alone
• No competing financial interests declared

Verdict

This paper identifies FGFR1 as a candidate target for artificial sweetener reproductive toxicity — and that's about it. The in vitro gene expression data from mouse cells adds a thin layer of experimental support, but there is no protein function data, no animal model, and no established human relevance. The missing exposure-concentration justification is a real methodological gap; if 0.5–2 mM doesn't approximate what Leydig cells actually encounter in someone drinking diet soda, the whole in vitro experiment is academic. If FGFR1 turns out to be a genuine player, this paper will be a footnote in the origin story. Until in vivo work appears, it's a computational hypothesis with one partial confirmation in a petri dish.