Irisin

A 112-amino-acid protein fragment shed from the muscle membrane protein FNDC5, proposed in 2012 as an exercise-induced myokine that browns white fat; its existence, measurement, and function in humans remain disputed, and it has never been administered to people.

Also appears in: Weight loss · Neuroprotection

What is Irisin?

Irisin is not a short peptide. It is a 112-amino-acid protein fragment, the extracellular fibronectin type III domain of FNDC5 (fibronectin type III domain-containing protein 5), a type I transmembrane protein expressed in skeletal muscle, heart, and brain. In the 2012 paper that named it, Boström and colleagues reported that exercise raises muscle FNDC5 through the transcriptional co-activator PGC-1α, that the ectodomain is cleaved and shed into the blood, and that the shed fragment, christened irisin after the Greek messenger goddess, acts on white fat to switch on the thermogenic protein UCP1 (Boström et al., Nature, 2012, 481(7382):463–468). The unglycosylated polypeptide has a calculated mass near 12.6 kDa; the glycosylated form migrates at about 20 to 22 kDa, and X-ray crystallography shows it folds as a fibronectin type III dimer (Schumacher et al., Journal of Biological Chemistry, 2013, 288(47):33738–33744). The irisin sequence is identical in mouse and human, which the discoverers took as a sign of conserved hormonal function and critics took as a reason to look harder at the human gene.

Irisin appears in the research market because its “exercise in a vial” framing attracted enormous attention: more than a thousand papers, a decade of disputes about whether it circulates in humans at all, and later claims of effects on bone and memory. Recombinant irisin is sold by life-science reagent suppliers, and the name appears sporadically in research-peptide storefronts, but there is no drug product, no compounded preparation, and no human trial in which irisin was given to anyone.

How it works

• PGC-1α → FNDC5 → UCP1 in white fat (the founding model). In mice, muscle PGC-1α induces Fndc5; recombinant FNDC5 at 20 nM induced UCP1 mRNA about 7-fold in cultured primary subcutaneous white adipocytes and doubled their oxygen consumption, and adenoviral expression of FNDC5 in the liver, which raised plasma irisin 3- to 4-fold, increased UCP1 in subcutaneous fat 10- to 20-fold, raised oxygen consumption, and improved glucose tolerance and fasting insulin in high-fat-fed mice (Boström et al., Nature, 2012). The protease that releases irisin from FNDC5 has still not been identified.
• αV integrins as the receptor. Irisin binds αV-class integrins, with αVβ5 the best-characterized partner, and chemical inhibition of αV integrins blocks irisin signaling in osteocytes and adipocytes (Kim et al., Cell, 2018, 175(7):1756–1768.e17). Follow-up work from the same group (A et al., Molecular Cell, 2023, 83(11):1903–1920.e12) proposed a two-step mechanism in which extracellular heat shock protein 90α, released by muscle with exercise, first activates αVβ5 so that irisin can then bind a site distinct from the one used by RGD-motif ligands. Irisin itself lacks an RGD motif.
• Bone. In young male mice, recombinant irisin at a low cumulative weekly dose increased cortical bone mass and bending strength through increased bone formation and fewer osteoclasts, without the adipose browning seen at a 35-fold higher dose (Colaianni et al., Proceedings of the National Academy of Sciences, 2015, 112(39):12157–12162). Kim et al. (2018) added that irisin increases osteocyte survival and sclerostin production, and that deleting Fndc5 prevents the bone loss that follows ovariectomy, a result pointing to irisin as a mediator of bone remodeling rather than simply an anabolic agent.
• Brain. FNDC5 is also expressed in the hippocampus. Knocking down brain FNDC5/irisin impaired long-term potentiation and memory in mice, while peripheral delivery of FNDC5 or swimming prevented the memory deficits caused by amyloid-β oligomers (Lourenco et al., Nature Medicine, 2019, 25(1):165–175). Mice lacking Fndc5 showed impaired cognition with exercise, aging, and in Alzheimer’s models; delivering irisin into the dentate gyrus restored pattern separation, and raising circulating irisin by liver-directed viral delivery improved cognition and neuropathology in two Alzheimer’s mouse models (Islam et al., Nature Metabolism, 2021, 3(8):1058–1070).

Research status

Irisin’s literature is large and split into two camps, and the split concerns facts more basic than efficacy: whether the human gene is translated, whether the protein circulates, and whether anyone has measured it correctly.

• Discovery and immediate challenge. Boström et al. (2012) reported a 2-fold rise in plasma irisin in 8 non-diabetic men after 10 weeks of supervised cycling, measured by Western blot with an antibody against FNDC5. A Nature correspondence the same year questioned whether FNDC5 is an exercise-responsive gene in humans at all (Timmons et al., Nature, 2012, 488(7413):E9–E10), and Erickson’s 2013 retrospective pointed out that the antibody’s epitope lies in the C-terminal region of FNDC5, outside the irisin sequence, that the commercial recombinant protein used in the cell experiments was severely truncated, and that shedding of the ectodomain had never been quantified (Erickson, Adipocyte, 2013, 2(4):289–293).
• The ATA start codon. Raschke et al. (2013) showed that human FNDC5 begins with ATA rather than the canonical ATG, confirmed this by 5′-RACE in human brain and muscle, and found that a human construct with the ATA start produced about 1% of the full-length protein made from an ATG construct in HEK293 cells; FNDC5 mRNA in human muscle biopsies did not change with endurance or strength training, and neither recombinant FNDC5 nor irisin browned human preadipocytes (PLoS One, 2013, 8(9):e73680). Jedrychowski et al. (2015) answered that human irisin is nonetheless translated mainly from the non-canonical start, while Witmer et al. (2024) argued from genomic and reporter data that Fndc5 is translated from an overlooked upstream ATG in both humans and mice, so that the ATA question may have been framed incorrectly by both sides (Cell Metabolism, 2024, 36(5):879–881).
• The measurement dispute. Albrecht et al. (2015) tested four commercial polyclonal antibodies used in irisin ELISA kits, found prominent cross-reactivity with unrelated serum proteins in human and animal samples, and detected no band of the expected size using recombinant glycosylated and unglycosylated irisin as controls; a FNDC5-derived signature near 20 kDa was found in human serum only by mass spectrometry (Scientific Reports, 2015, 5:8889). Jedrychowski et al. (2015) then quantified human irisin by targeted tandem mass spectrometry with isotope-labeled internal standards at about 3.6 ng/mL in 4 sedentary men and about 4.3 ng/mL in 6 men who had completed 12 weeks of aerobic interval training (Cell Metabolism, 2015, 22(4):734–740). The same critics replied that the multi-step sample preparation introduces uncontrolled variation, that available antibodies still bind nonspecific serum proteins, and that “measuring irisin is akin to chasing shadows” (Albrecht et al., Molecular Metabolism, 2020, 34:124–135). Published human irisin concentrations span several orders of magnitude depending on the kit, which is why the hundreds of ELISA-based association studies are generally discounted (Maak et al., Endocrine Reviews, 2021, 42(4):436–456; Perakakis et al., Nature Reviews Endocrinology, 2017, 13(6):324–337).
• Human physiology beyond exercise. Cold exposure and shivering raised circulating irisin in proportion to shivering intensity in a National Institutes of Health study, and FNDC5 treatment increased brown-fat gene expression in human adipocytes in a depot-specific way (Lee et al., Cell Metabolism, 2014, 19(2):302–309); those plasma measurements relied on antibody-based assays of the kind later criticized.
• Later directions. The bone and brain work summarized above (Colaianni 2015, with a correction in PNAS 2015, 112(42):E5763; Kim 2018, with an erratum in Cell 2019, 178(2):507–508; Lourenco 2019; Islam 2021, with an author correction in Nature Metabolism 2021, 3(10):1432) comes entirely from mice and mouse-derived cells. Kim et al. (2018) discloses a pending patent on irisin, Islam et al. (2021) discloses that its senior authors hold an irisin patent and co-founded Aevum Therapeutics, and the 2023 Molecular Cell paper states that the company is attempting to develop irisin as a therapeutic; no clinical program from that effort had been registered as of November 3, 2026.
• Clinical trials. ClinicalTrials.gov lists more than 250 studies mentioning irisin. Every one measures irisin as a biomarker in blood, saliva, cord blood, or gingival fluid, usually in response to exercise, a nutrient, or a drug such as metformin, isotretinoin, or omega-3 fatty acids. None administers irisin or FNDC5 to humans, and no phase 1 study has been registered anywhere. There are no human pharmacokinetic, immunogenicity, or safety data for exogenous irisin.

Common dosage forms

Irisin exists only as a laboratory reagent. It is sold as recombinant protein, lyophilized or in buffered solution, in microgram to 1 mg quantities, expressed in E. coli, HEK293, or CHO cells; many products carry a His or Fc tag, and the glycosylation state differs with the expression system. Research-peptide storefronts list it only occasionally.

Key considerations

• Regulatory status. The Research Market status reflects the absence of any approved product, any compounding pathway, and any registered human trial of irisin administration in any country. All human evidence is observational or exercise-physiology measurement, and much of it was generated with assays whose validity is contested.
• A protein fragment, not a peptide. At 112 residues and roughly 12 kDa before glycosylation, folded as a fibronectin type III dimer, irisin behaves like a cytokine rather than like the short synthetic peptides it is sometimes listed beside. Recombinant preparations vary in expression system, glycosylation, tags, and dimerization, differences that alter receptor binding and immunogenicity and that are rarely declared on product pages.
• Whether human irisin exists in the form and amounts proposed is still argued. The ATA start codon, the cross-reactive antibodies, and the low nanogram-per-milliliter concentrations found by mass spectrometry are each open to competing interpretations, and the 2020 and 2021 assessments from the Albrecht and Maak groups remain the most recent systematic reviews of the assays. Any claim about a person’s irisin level, or about a product raising it, rests on assays the field itself does not trust.
• Effects of exogenous irisin are mouse data, and they are dose- and tissue-dependent. Low doses affected cortical bone without browning fat; higher doses browned fat; the receptor work implies effects on any tissue expressing αV integrins. No tolerability data exist in any species beyond gross pathology in short viral-expression experiments, and a foreign-expressed 12 kDa protein given parenterally could be immunogenic.
• Naming. “Irisin” is the shed ectodomain; “FNDC5” is the full-length membrane protein, originally described in 2002 under the names FRCP2 and PeP before the 2012 renaming. The two are frequently conflated in papers and product listings, and irisin should not be confused with the short synthetic fragments of its sequence, such as isotope-labeled irisin(48-57), that are sold as mass-spectrometry standards and have no biology of their own.