GzmB was assessed in the skin of 10 KS (2M/8F, mean age 32), 10 HS (6M/4F, mean age 48), and 6 healthy controls (1M/5F, mean age 45) by immunohistochemistry (patient information available in Table 1 ). Compared to healthy skin, the number of GzmB-positive (GzmB⁺) cells were significantly higher in the upper ( Figures 1A, B ) and lower ( Figures 1C, D ) dermis of HS. While the number of GzmB⁺ cells in KS appeared non-significant compared to healthy skin controls, a trend in the accumulation of GzmB⁺ cells was identified in the lower dermis (p=0.09, Figure 1D ). No significant differences in the number of GzmB⁺ cells were observed between KS and HS.

Based on the localization of GzmB in dermal areas characterized by immune cell infiltration, we next investigated the cell source of the protease by staining immune cell markers. Few CD8⁺ T-cells were detected in the upper and lower dermis of KS and HS ( Supplementary Figure 1A ), with no statistical differences compared to healthy skin controls ( Supplementary Figures 1B, C ). While a weak correlation was observed between the number of CD8⁺ and GzmB⁺ cells in the upper dermis of KS and HS (r=0.45, p=0.045, Supplementary Figure 1D ), no correlation was observed in the lower dermis (r=0.17, p=0.49, Supplementary Figure 1E ). This result indicates that CD8⁺ T-cells may not contribute to GzmB production in KS and HS.

In contrast, mast cells (defined by expression of mast cell tryptase, or MCT) significantly accumulated in the lower dermis of HS compared to healthy skin controls ( Figures 2A–C ). The numbers of MCT⁺ and GzmB⁺ cells in the upper (r=0.92, p>0.001, Figure 2D ) and lower (r=0.88, p>0.001, Figure 2E ) dermis of KS and HS (n=20) were significantly positively correlated, indicating mast cells as a major source of GzmB in these conditions. Co-immunofluorescence ( Figure 3A ) and tyramide signal amplification (TSA) ( Supplementary Figure 2 ) assays revealed strong co-localization between MCT and GzmB, confirming mast cells as the main producer of the protease in KS and HS.

Since some of the mast cells in KS and HS exhibit active granule releas ( Figure 3B ), we immunostained skin dermis for Substance P, a neuropeptide previously identified as mediator of human mast cell degranulation (58). Increased presence of Substance P⁺ cells was observed in the dermis of KS and HS compared to healthy skin controls ( Figure 3C ), suggesting that mast cells release GzmB into the ECM of KS and HS through Substance P-mediated degranulation.

Next, we investigated possible targets of mast-cell derived GzmB in KS and HS and first focused on decorin, an established extracellular GzmB substrate (57). In contrast to the abundance of decorin in healthy skin controls ( Figure 4A ), the intensity of decorin staining was significantly reduced in KS and HS ( Figure 4B ). While still partially detectable in the papillary dermis, decorin was almost completely absent from the reticular dermis of KS and HS ( Figures 4A, B ).

As decorin content was particularly low around the dermal epidermal junction (DEJ) of KS and HS, we investigated the levels of LTBP1, a large ECM protein located at the DEJ (59) and involved in the extracellular anchoring of latent TGF-β (9). Compared to healthy skin controls, reduced LTBP1 was observed at the DEJ of KS and HS ( Figure 5A ), with a more pronounced reduction in HS (absent from 8 out of 10 stained samples, Supplementary Figure 3 and Table 1 ) than in KS (absent from 6 out of 10 stained samples, Supplementary Figure 4 and Table 1 ). When pathological skin sections were divided into two groups based on the presence of LTBP1 at the DEJ (LTBP1⁺ versus LTBP1^-), the average number of GzmB⁺ ( Figure 5B ) and MCT⁺ ( Figure 5C ) cells was significantly higher in the upper dermis of LTBP1^- skin lesions. Consequently, we hypothesized that GzmB may cleave LTBP1, resulting in its reduction at the DEJ of KS and HS.

Using the GrabCas software, a bio-informatic tool used to predict the cleavage of protease with aspase-like activities (60), we identified 7 potential GzmB cleavage sites in the LTBP1 amino-acid sequence ( Figure 6 ). Those cleavage sites, mostly located on the central Ca²⁺ binding Epidermal Growth Factor (EGF)-domains of LTBP1, are compatible with the release of a C-terminal fragment containing the SLC binding site ( Figure 6 , red arrowhead).

Next, we investigated GzmB cleavage of LTBP1 in a cell free cleavage assay. Using CHO cells stably transfected with a plasmid encoding LTBP1 (CHO-LTBP1 (61)), we generated LTBP1-enriched conditioned media (LTBP1 CM). After protein precipitation, LTBP1 CM was incubated with GzmB for different time points followed by western blot analysis of the digestion products. Compared to non-digested controls, we identified a reduced LTBP1 molecular weight in the presence of GzmB ( Figure 7A ), indicating that GzmB can cleave LTBP1 from conditioned medium in vitro. A morphological switch towards fibroblastic-like features was also observed in CHO-LTBP1 cultured in presence of 100 nM GzmB ( Supplementary Figure 5A , arrowheads). This partial epithelial to mesenchymal transition (EMT) phenotype was not observed in CHO-LTBP1 treated with PBS or with a lower GzmB concentration (50 nM). Consequently, we hypothesized that GzmB may modulate latent TGF-β bioavailability through the cleavage of LTBP1, thereby regulating the induction of EMT in epithelial cells ( Supplementary Figure 5A ).

To characterize the functional consequences of GzmB-dependent LTBP1 cleavage on latent TGF-β activation, LTBP1 CM was digested with GzmB and subsequently used to treat human keratinocytes for 1 h ( Figure 7B , optimized in Supplementary Figure 5B ). Treatments with active TGF-β1 or GzmB alone were used as positive and negative controls respectively, and Smad2 phosphorylation was assessed in keratinocyte lysates as a readout for activated TGF-β1 signaling ( Figure 7B ). To ensure that the observed effect was dependent on LTBP1 cleavage and independent of decorin degradation, the absence of decorin from LTBP1 CM was confirmed by western blot analysis ( Supplementary Figure 5C ).

In contrast to the faint Smad2 phosphorylation observed in untreated or GzmB-treated keratinocytes ( Figure 7C ), treatment of HaCaT keratinocytes with GzmB-digested LTBP1 CM resulted in an increased Smad2 phosphorylation compared to the non-digested LTBP1 CM, with a 1.7-fold increase for 24h and 2.8-fold for 48h digestion ( Figure 7C ). To confirm the involvement of TGF-β in this process, we performed the same experiments in the presence of a pan-TGF-β inhibitor or an isotype match control antibody ( Figure 7D ). While Smad2 phosphorylation was unaffected by the control antibody, treatment with the pan-TGF-β inhibitor completely abolished Smad2 phosphorylation in all conditions ( Figure 7D ). This observation confirmed that GzmB can promote TGF-β dependent Smad2 phosphorylation through the generation of LTBP1 fragments. At the used concentration (200 nM), GzmB alone does not promote Smad2 phosphorylation in keratinocytes ( Figures 7C, D ).

To mitigate the effect of growth factors in the culture medium, we produced and purified recombinant human (rh)LTBP1 from conditioned medium of stably transfected CHO cells ( Supplementary Figure 6A ). Using cell free cleavage assay, we confirmed the ability of GzmB to cleave purified rhLTBP1 in vitro ( Supplementary Figure 6A ). Keratinocytes treated with rhLTBP1 for 1h demonstrated an increased and dose dependent Smad2-phosphorylation compared to non-treated controls ( Supplementary Figure 6B ). Pre-digestion of rhLTBP1 by GzmB resulted in a 2-fold increased Smad2 phosphorylation compared to the non-digested rhLTBP1 condition in keratinocytes ( Supplementary Figure 6C ), thus confirming the results obtained using CHO-derived LTBP1 CM ( Figure 7 ).

As an additional indicator for TGF-β1 activity, p38 and p44/42 non-canonical TGF-β signaling pathways were also assessed in epithelial cells (62). Compared to the non-treated condition, LTBP1 CM promoted the phosphorylation of MAPK p38 ( Supplementary Figure 7A ) and p44/42 ( Supplementary Figure 7B ). Despite the ability of GzmB to directly activate both MAPK pathways, we observed no differences in p38 phosphorylation ( Supplementary Figure 7A ) as well as a reduction of p44/42 phosphorylation ( Supplementary Figure 7B ) for LTBP1 CM pre-digested with GzmB. Additionally, the LTBP1/GzmB-mediated MAPK activation was still detectable in the presence of the pan-TGF-β inhibitor ( Supplementary Figures 7C, D ), suggesting that this effect may be independent of TGF-β and directly mediated by LTBP1.

Based on the fibroproliferative nature of KS and HS and because fibroblasts are the primary source of ECM molecules in fibrosis, the same experiments were performed using primary human dermal fibroblasts. While only a small increase in Smad2 phosphorylation was observed in primary dermal fibroblasts cultured with GzmB-digested LTBP1 CM compared with non-digested LTBP1 CM ( Figure 8A ), a 2-fold increase in Smad2 phosphorylation was observed when primary dermal fibroblasts were treated with GzmB-digested rhLTBP1 compared to the non-digested protein ( Supplementary Figure 8 ). Together, these findings show that GzmB-dependent LTBP1 cleavage activates TGF-β signaling in dermal fibroblasts.

Unexpectedly, and in opposition to what was previously observed in keratinocytes ( Figures 7C, D ), a marked Smad2 phosphorylation was detectable in primary dermal fibroblasts treated with 200 nM GzmB alone ( Figure 8A ). This phosphorylation was completely abolished in the presence of the pan-TGF-β inhibitor ( Figure 8B ), indicating that GzmB itself can directly activate latent TGF-β secreted by dermal fibroblasts.

We next investigated whether this direct activation of the TGF-β/Smad signaling pathway was regulated by GzmB-dependent proteolysis of secreted proteins. To do so, primary human dermal fibroblasts were starved for 24h to generated fibroblast-derived CM enriched in secreted proteins. In a cell free system, fibroblast CM was incubated for 1h with GzmB to generate protein fragments. Then, and in order to mitigate the direct effect of GzmB, the protease was inhibited using its specific extracellular inhibitor VTI-1002 ( Figure 9A ). The digested fibroblast medium, thus containing protein fragments and inactivated GzmB, was used to treat dermal fibroblasts or keratinocytes for 1h followed by western blot analysis of the TGF-β/Smad signaling pathway ( Figure 9A ). A no inhibitor control, as well as a VTI-1002 alone control, were also included.

While VTI-1002 inhibited GzmB proteolytic activity as demonstrated by the nullification of fibroblast growth factor receptor 1 (FGFR1) cleavage, a strong Smad2 phosphorylation in primary dermal fibroblast was still detectable after GzmB inhibition ( Figure 9B ). Similar results were obtained using keratinocytes, with a 2- to 3-fold increased Smad2 phosphorylation when cultured in the presence of primary dermal fibroblast CM pre-digested with GzmB compared to non-digested controls ( Figure 9C ). As previously reported ( Figures 7C, D ), GzmB alone does not promote Smad2 phosphorylation in keratinocytes ( Figure 9C ).

Using dermal fibroblast CM digested for 1h with GzmB (cell free system), we confirmed the ability of GzmB to cleave fibroblast-derived LTBP1 by immunoblotting ( Figure 9D , left panel). We also identified the presence of two newly generated degradation products in the CM of fibroblasts incubated with GzmB ( Figure 9D , right panel, black arrows). As the intensity and molecular weight of the LAP-β1 pro-peptide (a component of the SLC) remained the same after 1h incubation with GzmB ( Figure 9D , left panel), these results suggest a mechanism whereby GzmB promotes indirect latent TGF-β activation from fibroblast medium through the cleavage of one (or more) unidentified secreted protein(s).

Use of human biospecimens for this study was approved by the University of British Columbia (Clinical research ethics board no: H23-01301). Skin biopsies from healthy skins as well as keloid and hypertrophic scars were collected and provided by Dr. Richard Crawford (Department of Dermatology, University of British Columbia, Vancouver, BC, Canada). Patient information is available in Table 1 .

Immunostaining was performed as previously established (47, 49). Briefly, after dewaxing and rehydration, antigen retrieval was performed in 10 mM sodium citrate buffer (Sigma) containing 0.05% Tween-20 (Sigma-Aldrich) pH 6 for 20 min in a steamer. Sections were then cooled down for 20 min at room temperature and endogen peroxidases were quenched by 3% H₂O₂ (v/v). Non-specific sites were then blocked for 30 min at RT followed by primary antibody incubation overnight at 4°C (diluted in blocking solution). The next day, after 30 min to 1h incubation with the secondary antibody (diluted in blocking solution), HRP was coupled with the biotinylated secondary antibody for 30 min using the VECTASTAIN Elite ABC-Peroxidase kit (Vector laboratories).

The protein sequence of human LTBP1 was obtained from Uniprot (assession no.: Q14766-1) and assessed for potential GzmB cleavage sites using GraBCas (v1.0) software with a cutoff of 0.1 (60). Cleavage sites were mapped to their respective regions in the full-length LTBP1 according to the domain organization available on Uniprot and displayed in (9).

Kera,nocytes (HaCaTs) and primary human dermal fibroblasts were maintained in complete Dulbecco modified Eagle’s medium (DMEM) containing 10% (v/v) Fetal Bovine Serum (FBS, Gibco) and 1% (v/v) Penicillin-Streptomycin (PS, Sigma). Chinese Hamster Ovary (CHO) cells stably transfected with the plasmid encoding LTBP1 (hereafter named CHO-LTBP1) was a gift from Boris Hinz (University of Toronto, Ontario, Canada) and were maintained in DMEM containing 10% (v/v) FBS, 1% (v/v) PS and selected with 1 mg/mL Geneticin (G418 sulfate, Agilent technology). To generate LTBP1-conditioned medium, CHO-LTBP1 were cultured to over-confluency for 3 weeks under serum deprivation and culture medium was collected 3 times a week. Conditioned media enriched in LTBP1 were cleared of cellular debris by centrifugation at 300 g for 10min at room temperature and stored at -20°C. Presence of LTBP1 in CM was assessed by immunoblotting.

CHO cells transfected with the pSecTag/FRT/V5-His-TOPO vector encoding LTBP1-EGFP (61) were donated from Boris Hinz (University of Toronto, Ontario, Canada) and were maintained in complete DMEM containing 10% (v/v) Fetal Bovine Serum (FBS, Gibco), 1% (v/v) Penicillin-Streptomycin (PS, Sigma) and 1 mg/mL Geneticin (G418, Gibco). Stably transfected CHO cells were culture to overconfluence for 2 weeks under serum deprivation. Culture medium was collected 3 times a week, centrifuged at 300g for 5 minutes to remove cellular debris and stored at -20°C. Prior purification, LTBP1-EGFP enriched conditioned medium was dialyzed 3 times (twice for 2 hours, then overnight) at 4°C in PBS using 6-8 kDa dialysis membrane (SpectrumLabs). LTBP1-EGFP conditioned medium was then directly applied on HisTrap HP His Tag protein purification columns (5 mL, Cytiva Life Sciences) using the ÄKRA Start Chromatography System (Cytiva Life Sciences). Column was washed using PBS containing 20 mM imidazole and eluted using 300 mM imidazole diluted in PBS. Fractions enriched in LTBP1-EGFP were analyzed by Brilliant Blue Staining (MiliporeSigma) and immunoblotting, pulled together, and dialyzed 3 times overnight at 4°C in PBS, PBS containing 0.2% (v/v) chloroform, and PBS before storage at -80°C. LTBP1-EGFP concentration was determined using the Pierce BCA protein Assay Kit (Thermo Fisher Scientific).

1,000,000 HaCaTs or 500,000 primary dermal human fibroblasts were seeded in 6 well plates (Corning) and cultured for 24h in complete medium before being serum starved for 16 to 24h. For medium transfer experimentations, the supernatant was removed post starvation and cells were cultured in the presence of LTBP1 CM (from the same set of harvest) predigested with 200 nM GzmB for 24 or 48h at 37°C, or with primary human dermal fibroblast CM predigested with 200 nM GzmB for 1h at 37°C. TGF-β1 (from 5 to 10 ng/mL, PeproTech) or GzmB (280 ng or 200 nM, depending on the experimental design and the digestion volume) stimulations were used as both positive and negative controls. After 1h stimulation, cells were washed twice with cold PBS and total proteins were extracted using radioimmunoprecipitation assay (RIPA) extraction buffer containing EDTA-free protease inhibitor and phosphatase inhibitors (Roche) for 20min on ice. After centrifugation at 13,200g for 10min at 4°C, supernatant containing solubilized proteins was stored at -20°C prior to immunoblotting. Protein concentration was determined using the Pierce™ BCA Protein Assay Kit (Sigma-Aldrich).

For TGF-β neutralization experiments, mouse monoclonal IgG anti-TGF-β1,2,3 (MAB1835, Clone 1D11) antibody was purchased from R&D Systems and isotype-matched mouse Kappa Globulin (clone p3.6.2.8.1) used as negative control was purchased from Invitrogen (14-4714-85). In both cases, cells (or LTBP1 CM) were pre-incubated for 10 min at 37°C in the presence of 2.5 μg/mL of either anti pan-TGF-β or isotype-matched antibody before stimulation. For GzmB inhibition experiments, VTI-1002 (viDA Therapeutics, Vancouver, Canada, BC) diluted in DMSO was used at a final concentration of 1 or 2 μM.

500,000 CHO-LTBP1 cells were seeded in a 6-well plates for 24h in complete medium before being serum starved overnight. The next day, medium was replaced with fresh FBS free DMEM and cells were treated with GzmB (50 or 100 nM) or an equivalent volume of PBS. After 48h treatment, media was replaced by fresh FBS free medium, and cells were treated another time with the same amount of GzmB or PBS. Cells were cultured for another 5 days in the presence of the protease (total duration of the treatment: 7 days) followed by morphological analysis. Phase contrast pictures were taken using a digital inverted microscope (EVOS FL, Life Technology) at 10X magnification.

20 μg of total proteins extracted from keratinocytes or primary human dermal fibroblasts were analyzed by SDS-PAGE in Tris-Glycine buffer (TG, composed of Tris-base from Fisher Bioreagents and Glycine from Fisher chemical) containing Sodium-Dodecyl-Sulfate (SDS, BIO-RAD) before being transferred onto a Polyvinylidene Fluoride membrane (PVDF 0.2 μm pore size, BIO-RAD) at 0.4 A for 2h in TG buffer containing 10% Ethanol (VWR). Membranes were blocked using 10% (w/v) non-fat dry milk diluted in Tris-buffered saline (TBS, Tris-base and NaCl from Fisher Bioreagents) containing 0.1% (v/v) Tween-20 at room temperature for 1h and incubated overnight at 4°C with primary antibody (orbital shaker). The next day, HRP-conjugated secondary antibody was incubated for 1h at RT and revelation was performed using the SuperSignal™ West Pico PLUS chemiluminescent Substrate (Thermo Scientific). Images were acquired on LI-COR Odyssey Fc system (LI-COR Bio-sciences, Lincoln, NE) and analyzed/quantified using the FIJI software (SciJava). Primary and secondary antibodies dilution are detailed in Table 2 . For Supplementary Figure 3C , 200 ng of recombinant human decorin was used as control and purchased from Abnova.

Quantification of the number of GzmB⁺, MCT⁺ and CD8⁺ cells in the upper and lower dermis has been performed according to the following methods: each immunobiological section has been pictured at 10 different areas (20X zoom from the 20X magnification) and ranked from one to ten. Then, 5 pictures were randomly selected, and the number of positive cells was counted. The mean was then calculated as the average number of positive cells from these 5 pictures. For decorin quantification, images were deconvoluted on FIJI (SciJava) using the “deconvolution” plugin and measured using a 60/240 threshold.

The R software (and its graphical user interface R Commander, T Studio Team, PBC, Boston, MA) and Prism software (GraphPad Software, San Diego, CA, USA) were used for graphical representations of data and statistical analyses. Mean comparison between healthy skin, keloid scars and hypertrophic scars was performed using the Kruskal-Wallis test (non-parametric equivalent of ANOVA) followed by Dunn’s multiple comparison (vs. healthy skin). Correlation coefficients r between GzmB and MCT, as well as GzmB and CD8, were determined using the Spearman rank-order (non-parametric equivalent of the Pearson product-moment) correlation coefficient. Comparison of the average number of GzmB⁺/MCT⁺ cells in the upper dermis of LTBP1⁺ and LTBP1^- lesions were performed using unpaired Student’s t-test (parametric) after normality (Shapiro-Wilk test) and homoscedasticity (Fisher’s exact test) validation. p values < 0.05 were considered as statistically significant and are indicated in the figure legends.