CXCL12 is a chemokine, a small signalling protein that cells use to attract and instruct other cells. In the androgen-driven pathway studied by Epi Biotech, dermal fibroblasts are a key source of CXCL12, although other skin cells, including hair-follicle keratinocytes, can also produce it. A 2022 paper from the Epi Biotech group reported that CXCL12 is high in the resting phases of the hair growth cycle, that adding it to cultured follicles delayed the growing phase and shortened hair, and that blocking CXCL12 or its receptor CXCR4 reversed this. We read the abstract of that paper only.

The 2024 paper added the androgen link. It proposes a loop with two steps, and the first is the surprising one. Androgens act on the androgen receptor in dermal fibroblasts and push them to make more CXCL12. CXCL12 then acts on dermal papilla cells, the signalling cluster at the base of the follicle, and raises their androgen receptor and CXCR4. More receptor means more androgen sensitivity, so the loop feeds itself. This sits alongside the standard DHT pathway rather than replacing it.

So far this is cell and organ-culture work. It describes a plausible circuit, and the mouse experiments below test whether interrupting it changes hair.

In the 2024 paper, mice made hair-poor with testosterone and given a CXCL12-neutralising antibody under the skin twice weekly grew more dorsal hair than untreated mice (7 mice per group). A topical dutasteride solution was included as a comparison. A DHT model and whisker follicle culture pointed the same way. Most results are reported as significance levels rather than effect sizes, so the paper does not let us say how large the regrowth was.

A 2025 single-cell study looked at what the antibody changed in the skin. Using 45 male mice (15 per group: control, testosterone, testosterone plus antibody), the authors found testosterone expanded fibroblasts about 8-fold and myeloid cells about 6.8-fold. Whole-skin androgen receptor, CXCL12 and CXCR4 rose 1.2-, 2.8- and 6.2-fold, with CXCR4 up 12.7-fold in dermal papilla cells, and the antibody reversed these shifts. It also reduced a dermal papilla subcluster and a group of M2-like macrophages that testosterone had enriched. Hair regrowth is described qualitatively in the main text, and skin was pooled per group for sequencing, so the data describe cell populations rather than variation between animals.

The authors also list their own limit: the long-term safety of blocking CXCL12 has not been tested, given its roles in stem cell trafficking and immune balance. The paper does not name the antibody's source or describe an isotype-matched control antibody, which is the usual way to separate a specific effect from an injected-protein effect.

Pattern hair loss is a human condition, and a testosterone-treated mouse is a model of it, not a copy. The Epi Biotech group addressed this directly in 2025, and the finding cuts both ways.

On one hand, the circuit looks more developed in human follicles than in mouse ones, which supports the idea. On the other, it means the mouse efficacy data came from a model whose dermal papilla lacks the co-expression the human cells show, so the mouse benefit cannot be assumed to translate along the same route. Organ culture is closer to human tissue, but it is a few days of cultured follicles without a blood supply, immune cells or the months-long timescale of hair loss in practice.

A related antibody has been tested in a mouse model of alopecia areata, an autoimmune condition. A 2024 Frontiers in Immunology paper used a humanised version of the antibody in mice that developed disease after lymph node cell transfer. CD8+ T cells made up 8.1% of T cells in healthy controls, 68.9% in alopecia areata mice and 37% after antibody treatment. Two of the authors work for Epi Biotech, and the paper notes that the rise in Cxcl12 in diseased skin was not itself statistically significant.

The 2024 International Journal of Molecular Sciences paper reported a similar signal. In mice with existing disease, treated mice lost less hair than controls, but there were only 3 mice per group. In prevention, 7 of 10 control mice had hair loss at week 13 against 1 of 9 treated mice, and the loss was slight. The same paper found interferon-gamma raised CXCL12 through STAT3, and that baricitinib and a STAT3 inhibitor reduced it. That is a link to JAK inhibitors, the one drug class already used for severe alopecia areata, and it suggests CXCL12 could be one downstream piece of what they affect. It is an interesting overlap, and still a mouse observation.

The case for blocking CXCL12 runs into an awkward result from a group with no link to the company.

Cohen and colleagues did not test an antibody, so their paper does not show that EPI-005 would impair regrowth. It does show that CXCL12 signalling is not purely harmful to hair, and the authors note CXCL12 has roles elsewhere, including in bone marrow stem cell and plasma cell niches. Whether a neutralising antibody would hit the helpful keratinocyte-regulatory T cell route, the harmful fibroblast-dermal papilla route, or both, is not something either set of papers answers. That inference is ours, not the authors'.

The Epi Biotech group responded in an April 2026 Perspective in Frontiers in Immunology. It argues the axis is context-dependent, calls the regulatory T cell role a "narrow scenario" that does not refute the harmful role in disease, and describes the trade-off as "minor and manageable". It reports CXCR4 mainly on pro-inflammatory macrophages and dermal papilla cells in diseased scalp, using a public dataset. The Perspective has six authors: three are employed by Epi Biotech and three, including a co-corresponding author, are at Seoul National University. It contains no new experiments and no safety data, so its defence of the approach is an interpretation of existing evidence, not an independent test of the treatment.

Epi Biotech has described EPI-005 in Korean press coverage, and some claims there go further than the journal papers do. In a February 2024 Medifonews report, the company said EPI-005 is a CXCL12-neutralising antibody in preclinical development, selected for a 2022 national drug development project and aimed first at pattern hair loss. It said the drug could suit women and that local injection would limit systemic side effects. No numerical efficacy data were given.

None of the papers we read tests a pig model, local scalp retention or an injection schedule, and the 2024 paper itself says delivery routes and scalp pharmacokinetics were not tested. These are company statements and should be read as unreplicated plans. A February 2026 Hankyung article still lists EPI-005 as a CXCL12-neutralising antibody in the company's pipeline, with no stage or detail added. For other candidates at a similar point, see our pipeline tracker, and for the closest comparison in an antibody approach, ABS-201 and HMI-115, which target prolactin pathways.

This article is provided for educational purposes and does not constitute medical advice. EPI-005 is an investigational preclinical programme and is not available as a treatment. AmpleLab products are cosmetic formulations, are unrelated to EPI-005, and are not intended to diagnose, treat, cure, or prevent any condition. Consult a qualified healthcare professional about hair loss.

AmpleLab.