Peptides are recognized as multifunctional bioactive ingredients in cosmetic science, as they offer diverse beneficial effects such as skin rejuvenation, anti-aging, and skin barrier enhancement. In this study, we applied a cheminformatics-assisted peptidomimetic design platform to design novel peptides targeting heat shock protein 47 (Hsp47), a collagen-specific molecular chaperone that is downregulated during skin aging. Using molecular fingerprint similarity-based peptide design and protein–peptide docking simulations, five candidate peptides were screened, among which ICP-1225 (TY) emerged as a potent stimulator of Hsp47 and collagen (COL1A1 and COL3A1) expression in dermal fibroblasts. To improve stability and skin penetration, fatty acid-conjugated derivatives of ICP-1225 were synthesized, and acetyl-TY (ICP-1236) demonstrated the most consistent upregulation of Hsp47 and collagen in vitro. Restoration of Hsp47 protein expression and dermal collagen levels in UVB-damaged ex vivo human skin explants was also observed. These findings highlight the potential of cheminformatics-assisted peptide design in the development of next-generation cosmetic actives. ICP-1236 represents a promising anti-wrinkle candidate through the modulation of Hsp47 and collagen pathways, warranting further clinical evaluation.
Peptides are short chains of amino acids with a length and molecular weight ranging from 2 to 100 and from 0.5 to 10 kDa, respectively. In addition to their role in the structure of proteins, the biological roles of peptides are quite diverse, encompassing aspects such as the modulation of various enzyme activities, hormonal activity, antimicrobials, and neurotransmitters. Based on these effects, peptides have been extensively investigated for their potential application as cosmetic ingredients with skin regeneration, anti-wrinkle, antioxidant, skin brightening, and skin-calming effects. In addition to their wide-ranging applications, peptide ingredients have potential advantages over other bioactive ingredients, including increased selectivity, efficacy, safety, and lower toxicity and immunogenicity. Peptides can be isolated from natural sources, including food products, marine organisms, and venom, and they can also be produced by biotechnical manufacturing processes and chemical synthesis.
Recent advances in computational tools, driven by the increased availability of numerous peptide sequences and their functions, and the use of artificial intelligence (AI) in peptide research have sparked the development of advanced methodologies for the development of novel peptides for cosmetic applications. While artificial intelligence (AI)-guided peptide design has recently emerged as a promising strategy for accelerating bioactive peptide discovery, its effectiveness is often limited by the scarcity of comprehensive structure–activity data. To address this, we developed a novel cheminformatics-assisted peptidomimetic design system for cosmetic applications capable of generating structured molecular datasets that support AI learning and the generalization of structure–function relationships. This is a custom-built, cheminformatics-driven screening tool that incorporates a molecular fingerprint-based structural similarity search algorithm developed using the RDKit cheminformatics toolkit. This platform enables ultra-large-scale similarity-based screening across a virtual peptide library containing approximately 67 million unique sequences. Molecular fingerprints are vectorized representations of small molecules that encode their structural features and relevant biochemical properties. After primary candidates were screened using a chemical structure–similarity-based peptidomimetics design, which refers to peptides designed to mimic the chemical structure of reference ligands or agonists/antagonists of the target protein, the potential binding of the designed peptides against the target protein was predicted, and hypothetical binding affinities were calculated with a peptide–protein docking simulating program. As a result, lead peptides with the highest binding properties were selected, synthesized, and subjected to further biological testing. Cheminformatics provides the structured molecular data and descriptors that enable AI to learn and generalize structure–function relationships. Integrating both approaches represents a natural progression toward data-driven, predictive peptide discovery.
Heat shock proteins (Hsp) are a highly conserved family of proteins primarily produced in response to various forms of stress, such as increased temperature (heat shock), cold, ultraviolet light, inflammation, and hypoxia. Hsp47, also known as Serpin H1, is a molecular chaperone required for the correct folding of procollagen in vertebrates. Decreased expression of Hsp47 in the skin of aged mice and increased expression and secretion of type I collagen in dermal fibroblasts by Hsp47 stimulation suggest that the use of Hsp47 could be plausible in anti-wrinkle applications. We explored the applicability of the cheminformatics-assisted peptide design for developing new anti-wrinkle peptide ingredients with Hsp47-stimulating activity. Therefore, in this study, we aimed to design and validate novel peptide ingredients that stimulate Hsp47 in anti-wrinkle applications using our cheminformatics-assisted peptide design system.
In total, 5 peptides designed using the cheminformatics-assisted peptide design system (listed in Figure 3) and acyl-derivatives of ICP-1225 were synthesized using solid-phase peptide synthesis with fluorenylmethyloxycarbonyl chloride (fMOC) methods. After undergoing synthesis and cleavage, the compounds were purified with conventional preparative reverse-phase high-performance liquid chromatography (prep-RP-HPLC), and their purity was further confirmed using the Waters Alliance e2695 HPLC system (Waters Corp., Milford, MA, USA) equipped with an SQ Detector 2.
To investigate the changes in Hsp47 protein levels in human skin, we used an ex vivo human skin explant model. NativeSkin® human skin tissues were purchased from GenoSkin S.A.S. (Toulouse, France). Skin tissues from the photo-protected site (abdomen) were obtained from healthy female donors with no history of allergies or dermatological disorders, according to the proprietary protocol developed by Genoskin. Ethical compliance was enacted and authorized through official authorization from the French Ministry of Research (Protocol AC-2022-4863, 14 October 2022). All studies were performed in accordance with the principles of the Declaration of Helsinki.
After receiving the tissues, the skin models were stabilized in a 12-well culture plate containing 1 mL of maintenance medium (Genoskin) in a 5% CO 2 humidified chamber. After 2 h of stabilization, the tissues were harvested, embedded in paraffin, and sectioned to a thickness of 4 µm. To explore the effects of photodamage on Hsp47 expression, 50 mJ/cm 2 of UVB was irradiated on the epidermal surface of the skin tissue once a day for 2 consecutive days before peptide treatment was initiated. Peptides were topically applied after each irradiation. After 48 h of incubation, UVB-irradiated skin tissues were harvested and subjected to histological assessment. Using immunohistochemical staining, Hsp47 expression was observed, according to a previously reported procedure, with slight modifications. Rabbit anti-Hsp47 (Abcam, ab109117) was used for staining, and fluorescence intensity was analyzed under a fluorescence microscope (Eclipse Ni-U, Intenslight C-HGFI, DS-Ri2, Nikon, Tokyo, Japan) at 400× magnification. To observe dermal collagen, Masson’s trichrome staining was performed according to a previously described protocol. All experiments were performed in duplicate, and histological analysis was performed by randomly selecting ten regions per slide. The expression levels of each protein were quantified by measuring the fluorescence intensity using the ImageJ 1.51s software (NIH, Bethesda, MD, USA).
The structural similarity between the reference molecule and the virtual peptide library of 67,000,000 peptides (2~6 amino acids), which was generated through combinatorial enumeration, was evaluated using the Tanimoto coefficient, which is a widely employed metric for molecular similarity assessment in cheminformatics. Before the screening, pre-filtering was applied based on physicochemical properties (molecular weight < 1 kDA, miLogP < 3, and net charge between −2 to +2) to ensure a tractable candidate. The canonical SMILES representation of the reference molecule (SMILES_R) and each virtual peptide (SMILES_n) were converted into molecular fingerprints (Fingerprint R and Fingerprint n, respectively). Pairwise similarity scores were then computed according to the following equation, as implemented in the RDKit cheminformatics toolkit:
S i m i l a r i t y n=T a n i m o t o S i m i l a r i t y(F i n g e r p r i n t R,F i n g e r p r i n t n)
The similarity value ranged from 0.0 to 1.0, with high scores indicating a greater structural resemblance to the reference molecule. The same procedure was systematically applied to all peptides in the virtual pool. After performing the similarity calculations, peptides with Tanimoto similarity ≥ 0.56 to the reference molecule (glucosyl–nicotinamide) were initially selected. The selected peptides were ranked in descending order, and high-priority peptide candidates were selected for protein–peptide molecular docking simulation using AutoDock (version 4.2). The crystal structure of human HSP47 (PDB ID: 7BDU) was retrieved from the Protein Data Bank (PDB). Protein preparation involved the removal of crystallized collagen and water molecules, the addition of polar hydrogen atoms, and the assignment of Kollman charges. Both the receptor and peptide ligands were converted to PDBQT format using AutoDockTools. The docking grid box was centered at X: −37.8250, Y: −34.5527, and Z: −19.2483, with dimensions of X: 85.7015 Å, Y: 85.4016 Å, and Z: 52.5783 Å, to sufficiently cover the entire putative binding region of the protein. The protein preparation steps have been outlined in detail (e.g., removal of water molecules, addition of polar hydrogens, and assignment of Kollman charges). Crucially, we also set the exhaustiveness parameter to 15 to ensure a thorough search of binding poses and enhance the accuracy of the affinity calculations.
Normal human dermal fibroblasts (hDFs) (passages 5), human fibroblast expansion basal medium with a low serum growth supplement, and gentamicin/amphotericin were purchased from The
