Targeting lysosomal protease CTSL promotes anti-tumor immunity and sensitizes HNSCC to PD-1 blockade by stabilizing PDK1 and activating Akt–PD-L1 axis

Abstract

Cathepsin L (CTSL) is expressed in head and neck squamous cell carcinoma (HNSCC), yet its role in immune escape is unclear. Here we show that CTSL directly binds PDK1, blocks its ubiquitin and restrains NEDD4L-mediated ubiquitination, thereby stabilizing PDK1, sustaining AKT phosphorylation, and increasing PD-L1 on tumor cells. This establishes a non-proteolytic scaffolding function, and suppresses tumor growth in xenograft and immunocompetent mouse models; these effects synergize with anti-PD-1 therapy. Clinically, high CTSL expression correlates with increased PD-L1, scarce CD8+ T-cell infiltration, and poor prognosis in multiple HNSCC cohorts. Collectively, our data identify CTSL as a key driver of PD-L1-dependent immune evasion through the CTSL–PDK1–AKT axis and highlight CTSL inhibition as a promising therapeutic strategy and predictive biomarker for PD-1/PD-L1 blockade in HNSCC.

Introduction

Head and neck squamous cell carcinoma (HNSCC) ranks among the most prevalent malignant tumors worldwide and is a major cause of cancer morbidity and mortality [1]. Although current management combines surgery, radiotherapy and platinum-based chemoradiotherapy, long term survival remains poor because locoregional relapses, distant spread and treatment resistance are common. Immune-checkpoint inhibitors that target PD-1, such as nivolumab and pembrolizumab, have delivered clinical benefit for only a minority of patients, and durable responses are uncommon [[2], [3], [4]]. These limitations highlight the need to discover actionable biomarkers and to elucidate the molecular drivers of immune escape in HNSCC.

Cathepsin L (CTSL) is a lysosomal cysteine protease of the papain-like cathepsin family, best known for its endopeptidase activity within acidic compartments such as lysosomes [5]. Besides intracellular protein turnover, CTSL contributes to extracellular-matrix (ECM) remodeling, antigen processing, and cytokine signaling and therefore shapes both innate and adaptive immunity [[6], [7], [8]]. During inflammation, elevated CTSL activity skews macrophages toward an M2-like phenotype and modulates antigen presentation pathways [6]. In adaptive immune responses, it facilitates major histocompatibility complex (MHC) epitope trimming [9] and shapes T cell repertoire selection [10]. Aberrant CTSL expression has been reported in a variety of cancers, including osteosarcoma [11], lung [12,13] and head and neck cancers [14,15], where it promotes tumor invasion [16], metastasis [17], and resistance to therapy [5,[18], [19], [20], [21], [22]]. Whether CTSL directly promotes tumor immune evasion, however, is only beginning to be explored [23]; recent work suggests that CTSL can bolster immunosuppressive myeloid programs within the tumor micro-environment [6], but the precise molecular mechanisms remain unclear.

Programmed cell death ligand 1 (PD-L1) is a critical immune checkpoint molecule that facilitates tumor immune evasion by impairing cytotoxic T cell-mediated tumor cell clearance [[24], [25], [26], [27]]. Engagement of PD-L1 on tumor cells with PD-1 on T cells transmits an inhibitory signal that attenuates T cell proliferation, cytokine production, and cytotoxic function, ultimately leading to T cell exhaustion and tumor persistence. As such, PD-1/PD-L1 blockade has emerged as a frontline immunotherapy across multiple solid tumor types [28], including head and neck squamous cell carcinoma (HNSCC) [[29], [30], [31]]. Nevertheless, only a subset of patients experiences durable clinical benefit [27,28]. Given the pivotal role of PD-L1 expression in determining the efficacy of immune checkpoint blockade (ICB), elucidating its upstream regulatory mechanisms is essential for developing predictive biomarkers and designing rational combinatorial immunotherapies [26].

PD-L1 expression is regulated by a combination of transcriptional and post-translational mechanisms. Transcription factors such as STAT3 [32], NF-κB [33], and HIF-1α [34] directly promote CD274 transcription, while the PI3K/AKT signaling axis contributes to PD-L1 upregulation through multiple pathways, including activation of downstream transcription factors, enhancement of mTOR-mediated protein synthesis, and inhibition of PD-L1 degradation processes [35,36], collectively resulting in increased PD-L1 abundance. Although cathepsin family proteases have been implicated in the modulation of signaling pathways that regulate immune checkpoints [37], direct evidence linking CTSL to PD-L1 expression via the PI3K/AKT pathway has not been documented. In preliminary immunoprecipitation–mass spectrometry (IP–MS) screens, we identified a robust interaction between CTSL and the serine/threonine kinase PDK1 (3-phosphoinositide–dependent protein kinase-1), which is a master regulator of AKT activation [38,39]. This observation prompted us to hypothesize that CTSL may regulate PD-L1 expression and tumor immune evasion through a CTSL–PDK1–AKT signaling axis [40].

In this study, we integrated single-cell and bulk transcriptomic analyses, biochemical interactome mapping, gain- and loss-of-function experiments, and immunocompetent mouse models to elucidate the role of CTSL in HNSCC. We demonstrate that CTSL is enriched in malignant epithelial cells, is associated with diminished infiltration of cytotoxic CD8+ T cells and poor clinical outcomes, and enhances PD-L1 expression via both transcriptional and post-translational mechanisms. Mechanistically, we show that CTSL stabilizes PDK1 by limiting its ubiquitin-mediated degradation. In particular, our data identify the E3 ubiquitin ligase NEDD4L as the principal mediator of PDK1 ubiquitination, and CTSL binding interferes with this process, thereby sustaining AKT activation and PD-L1 upregulation. Genetic deletion or pharmacological inhibition of CTSL suppresses tumor growth and synergizes with PD-1 blockade, while low CTSL protein expression in patient biopsies predicts improved response to ICB. Together, our findings identify a novel CTSL–PDK1–AKT axis as a key mechanism of immune evasion and establish CTSL as a dual-action therapeutic target for overcoming immunotherapy resistance in HNSCC.

Results

Single-cell analysis reveals CTSL expression patterns and links to the tumor microenvironment in HNSCC

To investigate the role of CTSL in the tumor ecosystem of head and neck squamous cell carcinoma (HNSCC), we integrated publicly available single-cell RNA sequencing (scRNA-seq) datasets from paired tumor and adjacent normal tissues. After correcting for batch effects (Supplementary Fig. 1A), unsupervised clustering identified multiple transcriptionally distinct cell populations (Fig. 1A), which were annotated into epithelial, immune, and stromal compartments based on canonical marker expression (Fig. 1B and E). To visualize lineage-specific transcriptional programs, we generated a Circos plot displaying representative differentially expressed genes (DEGs) across major cell types (Fig. 1F).

Fig. 1.

Single-cell transcriptomic landscape of CTSL in HNSCC. (A, B) t-SNE plots visualizing integrated scRNA-seq data, depicting (A) unsupervised clustering to identify distinct cell groups, and (B) annotation of major cell types. (C) Stacked bar plot illustrating the proportion of cells expressing different CTSL levels between Normal and Tumor samples. (D) Stacked bar plot comparing proportions of major immune cell populations between CTSL-High and CTSL-Low groups. (E) Heatmap demonstrating the expression proportions of known marker genes across annotated cell types. (F) Circos plot highlighting differentially expressed genes (DEGs) across major cell types. (G) Contour plots depicting GZMB (top) and NKG7 (bottom) expression intensities. (H) t-SNE plots of CD8+ T cells, colored by subcluster identities (left) and functional annotation (right). (I) Box plot comparing GZMB expression between CTSL-High and CTSL-Low CD8+ T cell populations. (J) Stacked bar plot displaying proportions of CD8+ T cell functional subsets in CTSL-Low and CTSL-High groups. (K) Contour plots showing EPCAM (top) and CDH1 (bottom) expression. (L) t-SNE plots showing epithelial subclusters (left) and their classification as Normal or Malignant (right). (M) Box plot comparing CD274 (PD-L1) expression in malignant epithelial cells between CTSL-Low and CTSL-High groups. (N) Stacked bar plot comparing proportions of Normal versus Malignant epithelial cells between CTSL-Low and CTSL-High samples.

CTSL expression was markedly upregulated in tumor samples compared with matched normal mucosa (Fig. 1C), suggesting its preferential enrichment in the malignant tissue context. Further immune profiling revealed that tumors with high CTSL expression exhibited a reduced proportion of CD8+ T cells and a relative enrichment of B cells and macrophages (Fig. 1D), indicating a shift toward an immunosuppressive microenvironment.

We next assessed the CD8+ T-cell compartment. To validate our annotation of cytotoxic lymphocytes, we examined their expression of GZMB and NKG7, which was confirmed by flow-density (contour) plots showing a distinct cytotoxic profile (Fig. 1G). Subclustering of CD8+ T cells revealed their stratification into several major functional states (Fig. 1H). In CTSL-high tumors, CD8+ T cells displayed significantly lower expression of GZMB (Fig. 1I) and a reduced proportion of effector-like subsets (Fig. 1J), indicating a functional impairment within the cytotoxic T-cell pool.

Within the epithelial compartment, EPCAM and CDH1 expression verified epithelial cell identity (Fig. 1K). We applied copy number variation (CNV) inference to distinguish malignant from normal epithelial cells and subsequently evaluated their differentiation status using CytoTRACE scores (Supplementary Fig. 1 B and C). These two metrics together facilitated accurate stratification of epithelial subclusters, which was clearly visualized on the t-SNE projection (Fig. 1L). Among malignant epithelial cells, those derived from CTSL-high tumors exhibited significantly elevated PD-L1 (CD274) expression compared to CTSL-low counterparts (Fig. 1M). Moreover, stratification of all epithelial cells by CTSL expression revealed that CTSL-high tumors contained a larger proportion of malignant versus normal epithelial cells (Fig. 1N), indicating a potential role of CTSL in promoting epithelial transformation. Consistent with this malignant bias, high CTSL expression was associated with unfavorable clinical outcomes: multivariate Cox regression confirmed that CTSL-high status independently predicted poorer overall survival in HNSCC patients (Supplementary Fig. 1 D).

In summary, elevated CTSL expression in HNSCC is associated with a reduction in functional CD8+ T-cell infiltration and a malignant epithelial phenotype marked by increased PD-L1 expression. These results suggest that CTSL may simultaneously modulate tumor-intrinsic oncogenic programs and reshape the tumor microenvironment to favor immune evasion.

High CTSL expression predicts poor prognosis and positively regulates PD-L1 expression in HNSCC

We next examined the clinical significance of CTSL expression in HNSCC. Kaplan–Meier survival analyses using the TCGA-HNSCC cohort demonstrated that patients with higher CTSL mRNA levels exhibited significantly poorer overall survival (OS, Fig. 2A) and disease-free survival (DFS, Fig. 2B). Consistent with these unfavorable prognostic findings, CTSL expression was markedly elevated in tumor tissues compared to adjacent normal mucosa (Fig. 2C). Moreover, we observed a positive correlation between CTSL and CD274 (PD-L1) mRNA expression across HNSCC samples from the TCGA dataset (Fig. 2D), implicating a potential role for CTSL in the regulation of PD-L1.

Fig. 2.

Correlation of CTSL expression with prognosis and PD-L1 regulation in HNSCC. (A, B) Kaplan–Meier curves linking CTSL expression to overall survival (OS) using TCGA-HNSCC data (A) and to disease-free survival (DFS) using UCSC Xena data (B). (C) Box plot comparing CTSL mRNA levels between tumor and normal tissues. (D) Scatter plot showing correlation between CTSL and PD-L1 (CD274) mRNA expression. (E) Western blot evaluating endogenous CTSL protein across HNSCC cell lines. (F) Western blot analysis of PD-L1 in Cal27 and SAS cells after CTSL knockdown. (G) Flow cytometry evaluating surface PD-L1 expression following CTSL knockdown. (H) Western blot assessing PD-L1 protein following CTSL overexpression. (I) Flow cytometry quantifying PD-L1 surface expression post CTSL overexpression. (J) Western blot detecting PD-L1 levels after treatment with the CTSL inhibitor Z-FY-CHO. (K) Flow cytometry assessing membrane PD-L1 expression post Z-FY-CHO treatment.

To identify suitable models for functional validation, we first profiled endogenous CTSL expression across a panel of HNSCC cell lines and selected representative lines for further manipulation (Fig. 2E; Supplementary Fig. 1E). Subsequent functional assays revealed that CTSL knockdown by shRNA in Cal27 and SAS cells significantly decreased PD-L1 protein levels in both total lysates (Fig. 2F) and on the cell surface, as confirmed by flow cytometry (Fig. 2G). Conversely, CTSL overexpression in HN8 and cells led to a pronounced increase in PD-L1 expression at both total protein (Fig. 2H) and cell surface levels (Fig. 2I). In line with these findings, pharmacological inhibition of CTSL using the inhibitor Z-FY-CHO resulted in a clear reduction of PD-L1 protein levels (Fig. 2J) and decreased its membrane expression in Cal27 and SAS cells (Fig. 2K). Additionally, qPCR analyses confirmed that genetic modulation or pharmacological inhibition of CTSL consistently altered CD274 transcriptional levels, further supporting the regulatory effect of CTSL on PD-L1 expression (Supplementary Fig. 2A–F).

Together, these clinical and experimental results indicate that high CTSL expression is associated with poor clinical outcomes and simultaneously drives PD-L1 up regulation, thereby providing a potential mechanistic basis for CTSL-mediated immune evasion in HNSCC.

CTSL promotes proliferation, migration, invasion, and tumor growth in HNSCC cells

We further investigated the functional role of CTSL in regulating malignant behaviors of HNSCC cells, including proliferation, migration, and invasion. Using syngeneic mouse models with Meer cells expressing CTSL-targeting sgRNAs (sg#CTSL), we further confirmed that CTSL deficiency profoundly suppressed tumor growth (Fig. 3A–D). Flow cytometry analyses demonstrated significantly increased infiltration of CD8+ and GZMB+ T cells into CTSL-deficient tumors, accompanied by reduced percentages of exhausted PD-1+CD8+ T cells (Fig. 3E–G). Consistently, multiplex immunohistochemistry (mIHC) further validated increased densities of CD8+ and GZMB+ T cells in the tumor microenvironment upon CTSL knockout (Fig. 3H–K).

Fig. 3.

Functional effects of CTSL modulation in HNSCC cells. (A) Schematic diagram of the syngeneic tumor model experimental design using Meer cells transduced with negative control (sg#NC) or Ctsl-targeting sgRNAs (sg1#CTSL, sg2#CTSL). (B-D) Meer tumor model results showing (B) mouse body weights throughout the experiment, (C) tumor growth curves and (D) final tumor weights across the three treatment groups (#1: sg#NC, #2: sg1#CTSL, #3: sg2#CTSL). (E-G) Flow cytometry quantification showing the proportions of (E) CD3+CD8+ T cells, (F) GZMB+CD8+ T cells, and (G) PD1+CD8+ T cells among tumor-infiltrating lymphocytes across the three treatment groups. (H-K) Multiplex immunofluorescent staining (mIHC) analysis of Meer tumors showing (H) representative images (DAPI, CTSL, CD8, GZMB), and corresponding quantification of the density of (I) CTSL+ cells, (J) CD8+ T cells, and (K) GZMB+ T cells across the three treatment groups.

Moreover, CCK-8 proliferation assays confirmed the impact of CTSL modulation on cell growth kinetics (Supplementary Fig. 2 G and H). Additionally, consistent with CCK-8 results, EdU incorporation assays showed that CTSL knockdown markedly suppressed the proliferation of Cal27 cells, whereas CTSL overexpression significantly promoted proliferation in HN8 cells (Supplementary Fig. 3A and B). Quantitative analyses confirmed a significant reduction in the percentage of EdU-positive cells following CTSL knockdown and an increase upon CTSL overexpression (Supplementary Fig. 3C and D). Wound-healing assays demonstrated that CTSL depletion impaired, while CTSL overexpression enhanced, the migratory capacity of HNSCC cells. Transwell invasion assays validated these observations, revealing significantly decreased invasion upon CTSL knockdown in Cal27 cells and markedly increased invasion after CTSL overexpression in HN8 cells (Supplementary Fig. 3E–L).

We next assessed the effect of CTSL expression on tumor growth using xenograft mouse models (Supplementary Fig. 4A). Complementary analysis showed no obvious differences in mouse body weights, indicating no significant toxicity from these manipulations (Supplementary Fig. 4B and C). Cal27 cells with the in vitro phenotype, CTSL knockdown produced smaller tumors with slower growth kinetics (Supplementary Fig. 4D–G). In this context—where adaptive immunity and checkpoint-mediated T-cell effects are largely absent—the growth difference most likely reflects tumor cell–intrinsic programs downstream of CTSL. In line with the AKT–PD-L1 axis defined in vitro, these data are consistent with a cell-intrinsic contribution of PD-L1 to proliferation/invasion.

CTSL drives PD-L1 expression and tumor malignancy via the AKT signaling pathway

To identify signaling pathways regulated by CTSL, we stratified TCGA-HNSCC bulk RNA-seq samples into CTSL-high and CTSL-low cohorts and performed differential-expression analysis. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment of the resulting differentially expressed genes (DEGs) revealed a prominent enrichment of PI3K–AKT signaling components (Fig. 4A and B). Western blotting confirmed these findings: CTSL knockdown in Cal27 cells markedly decreased AKT phosphorylation and concomitantly reduced PD-L1 protein expression (Fig. 4C). Treatment with the AKT activator SC-79 restored AKT phosphorylation and rescued PD-L1 expression in CTSL-deficient cells (Fig. 4D). Conversely, CTSL over-expression in HN8 cells increased AKT phosphorylation and PD-L1 abundance, and these effects were reversed by the AKT inhibitor MK-2206 (Fig. 4E and F).

Fig. 4.

CTSL regulates PD-L1 expression and malignant phenotypes through the AKT pathway. (A, B) Bubble plots illustrating pathway enrichment of DEGs derived from an integrated analysis of the TCGA-HNSCC, GSE178537, GSE127165, and GSE227919 cohorts, stratified by CTSL expression: (A) top Gene Ontology biological processes and (B) top KEGG pathways. (C, D) Western blots of CTSL, total and phosphorylated AKT, and PD-L1 in Cal27 cells with CTSL knockdown alone (C) or knockdown combined with the AKT activator SC-79 (D). (E, F) Western blots of the same proteins in HN8 cells with CTSL overexpression (E) or overexpression combined with the AKT inhibitor MK-2206 (F). (G) Transwell invasion assay images and quantification for Cal27 cells comparing SH#NC, SH2#CTSL, and SH2#CTSL + SC-79 groups. (H) Transwell invasion assay images and quantification for HN8 cells comparing Vector, OE#CTSL, and OE#CTSL + MK-2206 groups. (I) Quantification of wound-healing assays in Cal27 cells after CTSL knockdown and SC-79 rescue. (J) CCK-8 proliferation curves for Cal27 cells in the same groups. (K) Quantification of wound-healing assays in HN8 cells after CTSL overexpression and MK-2206 treatment. (L) CCK-8 proliferation curves for HN8 cells in the same groups.

Functional assays further corroborated the role of the CTSL–AKT pathway in tumor aggressiveness. In Cal27 cells, SC-79 significantly rescued the invasion and migration deficits caused by CTSL knockdown (Fig. 4G and I), whereas MK-2206 suppressed the heightened invasive and migratory capacities induced by CTSL overexpression in HN8 cells (Fig. 4H and K). Proliferation assays showed the same AKT-dependent pattern: SC-79 partially restored proliferation of CTSL knockdown cells, while MK-2206 attenuated the proliferative advantage conferred by CTSL overexpression (Fig. 4J and L). Consistently, qPCR analyses demonstrated that SC-79 and MK-2206 respectively restored or blocked the CTSL-dependent changes in CD274 mRNA (Supplementary Fig. 4 H and I).

Collectively, these findings indicate that CTSL promotes PD-L1 expression and drives malignant behaviors in HNSCC predominantly through activation of the CTSL–AKT signaling axis.

CTSL interacts with and stabilizes PDK1 to activate AKT signaling and PD-L1 expression

To decipher how CTSL activates the AKT pathway, we first mapped its interactome by Flag-IP followed by LC-MS/MS (Fig. 5A). Among several candidates, the serine/threonine kinase PDK1—an obligate upstream activator of AKT—stood out. Immunofluorescence confirmed cytoplasmic co-localization of endogenous CTSL and PDK1 in HN8 cells (Fig. 5B; Supplementary Fig. 4L). Reciprocal co-immunoprecipitation of Myc-tagged CTSL and Flag-tagged PDK1 in HEK293T cells verified mutual binding (Fig. 5C and D), which was further validated at endogenous levels in HN8 cells (Fig. 5E). Direct binding was demonstrated with purified proteins in a GST pull-down assay (Fig. 5F).

Fig. 5.

CTSL interacts with and stabilizes PDK1 to activate the AKT axis and upregulate PD-L1. (A) Experimental workflow of Flag-CTSL immuno-precipitation followed by LC-MS/MS identification of interacting proteins. (B) Immunofluorescence colocalization of endogenous CTSL (green) and PDK1 (red) in HN8 cells with DAPI nuclear stain (blue). (C, D) Reciprocal co-immunoprecipitation in HEK293T cells co-transfected with CTSL-Myc and PDK1-Flag showing mutual binding. (E) Endogenous co-IP of CTSL with PDK1 in HN8 cells. (F) GST pull-down assay demonstrating direct binding between purified His-CTSL and GST-PDK1. (G, H) Western blots showing the effect of CTSL overexpression (HN8) or knockdown (Cal27) on PDK1 protein abundance. (I) Western blots revealing that PDK1 inhibitor GSK2334470 abrogates CTSL-induced activation of AKT, and PD-L1 up regulation in HN8 cells. (J) Western blots showing that AKT activator PS48 rescues AKT phosphorylation and PD-L1 expression suppressed by CTSL knockdown in Cal27 cells. (K, L) Immunoprecipitation of PDK1 followed by ubiquitin blotting indicating CTSL overexpression decreases, while CTSL knockdown increases, PDK1 ubiquitination. (M) Schematic of PDK1 domains and Flag-tagged truncations (FL, P1, P2) with co-IP mapping CTSL binding to the catalytic PHOS domain fragment (P1). (N, O) Cycloheximide chase assays showing accelerated PDK1 degradation upon CTSL knockdown (Cal27) and prolonged half-life upon CTSL overexpression (HN8).

The downstream consequences align with a post‑translational mode of control. Cycloheximide‑chase experiments first demonstrated that CTSL depletion accelerated, while CTSL overexpression prolonged, the half‑life of PDK1 protein (Fig. 5N and O). In line with this, immunoprecipitation followed by anti‑ubiquitin immunoblotting showed that CTSL overexpression decreased, whereas CTSL knockdown increased, PDK1 ubiquitination (Fig. 5K and L); and across CTSL perturbations (overexpression, knockdown, inhibitor treatment, and rescue), the poly‑ubiquitin smear on immunoprecipitated PDK1 varied inversely with CTSL levels—higher CTSL, lower ubiquitination; lower CTSL, higher ubiquitination—while input PDK1 levels tracked with CTSL status (Supplementary Fig. 5 A and B). Finally, domain mapping indicated that CTSL binds the PHOS domain of PDK1 (Fig. 5M).

At the pathway and phenotypic levels, the PDK1 inhibitor GSK2334470 and the PDK1 activator PS48 reciprocally modulated AKT phosphorylation and PD‑L1 expression downstream of CTSL (Fig. 5I and J), and partially restored invasion/migration phenotypes consistent with CTSL–PDK1–AKT dependency (Supplementary Fig. 6A–D). PS48 and GSK2334470 likewise rescued CTSL‑dependent changes in CD274 mRNA (Supplementary Fig. 6E and F).

Transcript‑level analyses showed no correlation between CTSL and PDK1 mRNA in HNSCC cohorts (Supplementary Fig. 4J and K), underscoring the post‑translational nature of this regulation. AlphaFold modeling predicted a plausible CTSL–PDK1 interface (Supplementary Fig. 4M).

Finally, having established a ubiquitination‑based stabilization of PDK1 by CTSL, we asked whether this control requires protease activity and which E3 ligase is involved. Because Z‑FY‑CHO is an active‑site–directed aldehyde that both suppresses CTSL enzymatic activity and reduces cellular CTSL protein abundance, pharmacologic inhibition alone cannot disentangle activity‑dependent from protein‑availability effects. We therefore engineered a catalytically inactive CTSL mutant (CTSL‑C25A) to isolate activity‑independent functions and compared it with Z‑FY‑CHO. In these assays, CTSL‑C25A maintained PDK1 stabilization and suppressed PDK1 ubiquitination despite lacking protease activity (Supplementary Fig. 5A and B). Under Z‑FY‑CHO treatment, CTSL protein was reduced in whole‑cell lysates and PDK1 protection was lost (Supplementary Fig. 5A and B), indicating that stabilization does not require catalytic activity and instead depends on adequate CTSL protein availability. To identify which E3 ligase acts on PDK1 in this setting, we intersected UbiBrowser 2.0 predictions with proteins recovered in PDK1 co‑immunoprecipitation/mass‑spectrometry, which prioritized NEDD4L (Supplementary Fig. 5C and D). Consistent with this assignment, siNEDD4L increased PDK1 protein levels (Supplementary Fig. 5D), and in denaturing PDK1‑ubiquitination IPs reduced PDK1 polyubiquitination (Supplementary Fig. 5E). Moreover, co‑immunoprecipitation of PDK1 and NEDD4L under CTSL‑WT/CTSL‑C25A/control conditions showed that higher CTSL levels are associated with a weaker PDK1–NEDD4L association, whereas CTSL loss enhances it (Supplementary Fig. 5J and K). Collectively, these results support a non‑proteolytic mechanism in which CTSL binds and stabilizes PDK1 to sustain AKT signaling and enhance PD‑L1 expression.

Genetic knockout of CTSL inhibits tumor growth and enhances the efficacy of anti-PD-1 therapy in syngeneic mouse models

Given the immunomodulatory role of CTSL, we evaluated the impact of CTSL knockout on tumor progression and PD-1 blockade in immunocompetent models. Western blot confirmed efficient CTSL ablation in Meer and Hepa1-6 cells (Fig. 6B). Meer cells transduced with negative control sgRNA (sg#NC) or CTSL-targeting sgRNA (sg#CTSL) were implanted into C57BL/6 mice, followed by treatment with isotype IgG or anti-PD-1 antibody (Fig. 6A). CTSL knockout alone markedly suppressed tumor outgrowth; combination with anti-PD-1 produced the most pronounced anti-tumor effect, as evidenced by visibly smaller tumors (Fig. 6C), delayed growth kinetics (Fig. 6D), and reduced final tumor weights (Fig. 6F), without affecting body weight (Fig. 6E), indicating minimal toxicity.

Fig. 6.

Genetic knockout of CTSL inhibits tumor growth and enhances anti-PD-1 therapy in syngeneic mouse models. (A) Schematic diagram of the syngeneic tumor model experimental design and analysis plan. (B) Western blot detection of CTSL protein levels in Meer and Hepa1-6 cells transduced with negative control sgRNA (sg#NC) or Ctsl-targeting sgRNA (sg#CTSL). (C-F) Meer tumor model results showing (C) representative excised tumors, (D) tumor growth curves, (E) mouse body weights, and (F) final tumor weights across four treatment groups (#1: sg#NC+IgG, #2: sg#NC+αPD1, #3: sg#CTSL+IgG, #4: sg#CTSL+αPD1). (G-I) Representative images of multiplex immunofluorescent staining (mIHC) showing expression of CTSL, CD8, and GZMB in implanted Meer tumors from the four treatment groups (G), with quantification of CD8+ T cell density (H) and GZMB+ T cell density (I). (J-L) Representative flow cytometry plots showing analysis of tumor-infiltrating lymphocytes from implanted Meer tumors across the four treatment groups, illustrating gating for (J) CD3+CD8+ T cells, (K) GZMB+CD8+ T cells, and (L) PD1+CD8+ T cells. (M-O) Quantification by flow cytometry showing proportions of (M) CD3+CD8+ T cells, (N) GZMB+CD8+ T cells, and (O) PD1+CD8+ T cells among tumor-infiltrating lymphocytes in the Meer tumor model across the four treatment groups. (P-R) Quantification by flow cytometry showing proportions of (P) CD3+CD8+ T cells, (Q) GZMB+CD8+ T cells, and (R) PD1+CD8+ T cells among tumor-infiltrating lymphocytes in the Hepa1-6 tumor model across the four treatment groups.

Multiplex immunohistochemistry revealed that CTSL-deficient tumors harbored denser infiltrates of CD8+ and granzyme B (GZMB)+ T cells, an effect amplified by anti-PD-1 co-treatment (Fig. 6G–I). Flow-cytometric profiling of tumor-infiltrating lymphocytes corroborated these findings, showing higher frequencies of CD3+CD8+ T cells (Fig. 6J and M), increased cytotoxic GZMB+CD8+ subsets (Fig. 6K and N), and reduced PD-1+ exhausted CD8+ cells (Fig. 6L and O) in sg#CTSL tumors.

Parallel experiments in the Hepa1-6 syngeneic model yielded comparable results. CTSL knockout, alone or combined with anti-PD-1, curtailed tumor growth and was accompanied by elevated CD8+ and GZMB+ T-cell infiltration and diminished PD-1+CD8+ exhaustion (Fig. 6P–R; Supplementary Fig. 5I–N).

Collectively, these findings demonstrate that genetic deletion of CTSL reshapes the tumor immune landscape, restricts tumor progression, and sensitizes tumors to PD-1 blockade, underscoring CTSL as a compelling target to overcome resistance to immune-checkpoint therapy in HNSCC (Supplementary Fig. 7).

Pharmacological inhibition of CTSL suppresses tumor growth and enhances response to anti-PD-1 therapy in vivo and correlates with clinical outcomes

Given the potent anti-tumor effects observed upon genetic ablation of CTSL, we next investigated whether pharmacological targeting of CTSL could recapitulate these effects. Using a syngeneic Meer tumor model, mice were treated with the CTSL inhibitor Z-FY-CHO, either alone or in combination with anti-PD-1 antibody (Fig. 7A). Z-FY-CHO monotherapy significantly impaired tumor growth, and its combination with anti-PD-1 led to a markedly enhanced anti-tumor response, as indicated by delayed tumor growth kinetics (Fig. 7B) and significantly reduced final tumor weights (Fig. 7C). No significant differences in mouse body weight were observed among treatment groups, indicating minimal treatment-related toxicity (Fig. 7D).

Fig. 7.

Pharmacological inhibition of CTSL suppresses tumor growth and correlates with immunotherapy response in HNSCC. (A) Schematic diagram of the syngeneic Meer tumor model experimental design involving treatment with CTSL inhibitor Z-FY-CHO and/or anti-PD-1 antibody (αPD1). Four treatment groups: #1 DMSO+IgG, #2 DMSO+αPD1, #3 Z-FY-CHO+IgG, #4 Z-FY-CHO+αPD1. (B-D) Meer tumor model results showing (B) tumor growth curves, (C) final tumor weights, and (D) mouse body weights across the four treatment groups. (E, F) Quantification from multiplex immunohistochemistry (mIHC) analysis showing the density of (E) CD8+ T cells and (F) GZMB+ T cells in Meer tumors across the four treatment groups. (G-I) Flow cytometry quantification showing the proportions of (G) CD3+CD8+ T cells, (H) GZMB+CD8+ T cells, and (I) PD1+CD8+ T cells among tumor-infiltrating lymphocytes in the Meer tumor model across the four treatment groups. (J) Representative multiplex immunofluorescent staining (mIHC) images of patient-derived HNSCC tumor samples. (K) Waterfall plot showing the best percentage change in tumor size from baseline for individual HNSCC patients (n = 21) receiving immunotherapy, stratified by baseline tumor CTSL expression level (Low vs High). (L) Spearman correlation analysis of baseline tumor CTSL expression (H-score) versus best percentage change in tumor diameter for patients treated with immunotherapy (n = 21).

To assess immunologic mechanisms underlying the observed therapeutic synergy, multiplex immunohistochemistry (mIHC) revealed increased intertumoral infiltration of CD8+ and GZMB+ cytotoxic T cells in mice receiving combination therapy compared to either monotherapy (Fig. 7E and F). Flow cytometry further confirmed these findings, demonstrating increased proportions of CD3+CD8+ T cells (Fig. 7G), an elevated frequency of GZMB+CD8+ cytotoxic subsets (Fig. 7H), and a concomitant decrease in exhausted PD-1+CD8+ T cells (Fig. 7I).

Clinicopathological assessment (Supplementary Table 1) indicated that CTSL-high tumors were associated with more advanced pathological features, consistent with an aggressive phenotype. To test clinical relevance, we analyzed baseline biopsies from a cohort of HNSCC patients treated with immune-checkpoint inhibitors. Patients with low CTSL expression experienced greater objective tumor shrinkage on multiplex IHC (Fig. 7J), and waterfall-plot analysis confirmed superior responses in the CTSL-low subgroup (Fig. 7K). Spearman analysis revealed an inverse correlation between tumor CTSL H-score and percentage change in lesion diameter (Fig. 7L), underscoring the value of CTSL as a negative predictive biomarker.

Collectively, these preclinical and clinical findings demonstrate that pharmacological inhibition of CTSL not only suppresses tumor growth but also reprograms the immune microenvironment to enhance anti–PD-1 responsiveness.

Discussion

Cathepsin L (CTSL), traditionally recognized as a lysosomal cysteine protease, emerges here as a regulator of tumor progression and immune suppression. Across head and neck squamous cell carcinoma (HNSCC) and a murine Hepa1‑6 model, our data support a conserved oncogenic role. Integrating single‑cell and bulk transcriptomes, we find CTSL selectively enriched in malignant epithelial cells and independently predictive of poor survival [9,32]. Tumors with high CTSL display reduced CD8⁺ T‑cell infiltration—particularly effector‑like GZMB⁺ subsets—and enrichment of exhausted PD‑1⁺CD8⁺ cells [15,9,33], consistent with an immunosuppressive microenvironment. Clinically, high tumor CTSL associates with inferior response to immune‑checkpoint blockade (ICB). Taken together, these findings establish CTSL as a driver of an immunosuppressive tumor microenvironment in HNSCC.

Because Z-FY-CHO is an active-site–directed aldehyde that suppresses CTSL enzymatic activity and lowers cellular CTSL protein abundance [41], pharmacologic inhibition alone cannot disentangle activity-dependent from protein-availability effects. We therefore used a catalytically inactive CTSL mutant (CTSL-C25A) to isolate activity-independent functions. CTSL-C25A still stabilized PDK1 and reduced its ubiquitination. By contrast, Z-FY-CHO treatment reduced CTSL protein and coincided with loss of PDK1 protection. Together, these data indicate that CTSL-mediated stabilization of PDK1 does not require catalytic activity; whether catalytically active or inactive, CTSL stabilizes PDK1 provided sufficient CTSL protein is available.

Mechanistically, we define a CTSL–PDK1–AKT axis that links a lysosomal protease to PD‑L1 accumulation via non‑proteolytic stabilization of PDK1. CTSL binds the PHOS domain of PDK1 [34], limiting its ubiquitination and prolonging its half‐life. Stabilized PDK1 maintains constitutive AKT phosphorylation, resulting in increased PD-L1 levels on the tumor cell surface and further impairing T-cell effector function [42]. This post-translational mechanism is complemented by modest transcriptional upregulation of CD274 [43]. Finally, disruption of any component of the CTSL–PDK1–AKT pathway suppresses PD-L1 expression and abrogates the enhanced migration/invasion phenotypes: PDK1 is a known hub for cell motility and invasion [44], and PD-L1 overexpression itself promotes tumor cell migration and invasion [45]. These findings establish the CTSL–PDK1–AKT axis as a central node linking lysosomal function to the coordination of two key hallmarks of cancer: immune evasion and cellular invasion.

A key inference from our perturbation series is that PDK1 stabilization by CTSL is activity‑independent but availability‑dependent. The catalytically inactive CTSL‑C25A mutant retains PDK1 stabilization, whereas the active‑site inhibitor Z‑FY‑CHO, an aldehyde that suppresses enzymatic activity and lowers cellular CTSL protein, coincides with loss of PDK1 protection in our assays. Multiple studies have reported reduced CTSL protein after peptide‑aldehyde inhibitor exposure [39]. Thus, pharmacologic data are best interpreted together with genetic controls (CTSL‑C25A and rescue), which isolate a binding‑mediated, activity‑independent mode of stabilization.

Guided by the ubiquitination phenotype, we asked which E3 ligase targets PDK1 in this setting. Intersecting UbiBrowser predictions with proteins detected by PDK1 co‑IP/MS prioritized NEDD4L; functionally, siNEDD4L increased PDK1 protein and reduced PDK1 polyubiquitination, and CTSL abundance inversely associated with PDK1–NEDD4L engagement. These observations support a competitive/access model in which CTSL binding limits NEDD4L access to PDK1 and attenuates ubiquitin‑mediated degradation. Regarding previously reported PDK1‑ubiquitinating E3 ligases, we adopt a conservative interpretation. For example, SPOP predominantly acts as a nuclear adaptor for CUL3‑based complexes, limiting opportunities to engage cytosolic PDK1 in our system [46], whereas SMURF1 has been implicated in non‑canonical or context‑dependent routes [47,48]. By contrast, in our HNSCC models, functional perturbations and co‑IP data point to NEDD4L as the principal ligase; nevertheless, we do not exclude contributions from additional E3s under alternative cellular states.

These mechanisms translate in vivo. In immunodeficient nude mice, genetic CTSL loss constrained tumor growth, consistent with tumor‑intrinsic effects. In immunocompetent syngeneic models (HNSCC and Hepa1‑6), CTSL ablation or pharmacologic inhibition not only suppressed tumor growth but also increased intratumoral CD8⁺/GZMB⁺ infiltration and enhanced responsiveness to anti–PD‑1, without systemic toxicity [40]. Patient biopsies further indicated that lower baseline CTSL associates with greater tumor shrinkage and improved clinical benefit from ICB [12], suggesting CTSL as a negative predictive biomarker.

More broadly, our findings also illuminate functional diversity within the cathepsin family. While cathepsin B has been predominantly linked to extracellular matrix remodeling [49] and cathepsin S to antigen presentation, CTSL operates primarily through stabilizing a cytosolic kinase hub [50]. This unique mode of action implies that CTSL-selective inhibition could suppress immune evasion without compromising essential immune functions such as MHC class II processing. As multiple cathepsin inhibitors have entered clinical testing for non-oncology indications [51,52], our study provides a compelling rationale for the clinical development of selective CTSL inhibitors, to be used as a strategy to overcome ICB resistance.

Limitations include the need for next‑generation CTSL inhibitors with optimized selectivity/pharmacokinetics for human trials [51]; structural insights based on AlphaFold and domain mapping require high‑resolution validation; single‑cell analyses derived from public datasets carry heterogeneity; and while CTSL upregulation spans multiple cancers [7], we focused on HNSCC and one liver cancer model. our in-vivo work focused on HNSCC and one liver cancer model. Because the in-vivo studies used constitutive CTSL knockdown and nude xenografts, we interpret tumor-growth differences primarily as tumor-intrinsic and do not regard them as proof of checkpoint dependence. Future work will deploy doxycycline-inducible CTSL suppression after tumor establishment in immunocompetent syngeneic models to separate tumor-intrinsic from immune-mediated effects, and will map the CTSL–PDK1 interface at residue resolution, define how CTSL occupancy alters NEDD4L recognition, and deploy inducible CTSL perturbations in vivo to dissect cell‑intrinsic versus immune‑mediated effects.

In conclusion, we identify CTSL as a key modulator of tumor immune evasion through stabilization of PDK1 and activation of AKT, leading to PD‑L1 accumulation and impaired T‑cell immunity. By jointly promoting malignant behavior and suppressing anti‑tumor immunity, CTSL acts as a dual driver of aggressiveness. The non‑proteolytic CTSL–PDK1–NEDD4L axis provides a mechanistic basis for combining CTSL targeting with PD‑1/PD‑L1 blockade in HNSCC.

Materials and methods

Ethics approval

All human tissue studies were approved by the Ethics Committee of Xiangya Hospital, Central South University (Approval no XY20240711004). The requirement for informed consent was waived due to the use of de-identified archived specimens, in accordance with the Declaration of Helsinki. All animal procedures were approved by the Institutional Animal Care and Use Committee of Xiangya Hospital and were conducted in compliance with institutional and national guidelines for laboratory animal care.

Data sources and clinical samples

Public Genomic Data: Bulk transcriptomic data and clinical information for head and neck squamous cell carcinoma (HNSCC) were obtained from The Cancer Genome Atlas (TCGA) via the Genomic Data Commons, and from the Gene Expression Omnibus (GEO) under accession numbers GSE178537, GSE127165, and GSE227919. Single-cell RNA sequencing (scRNA-seq) datasets from six independent HNSCC cohorts (GEO accession numbers: GSE206332, GSE188737, GSE182227, GSE181919, GSE206038, GSE227156) were retrieved from the Gene Expression Omnibus. For scRNA-seq analysis, data were processed in R (v4.4.1) using Seurat (v5.2.1). Low-quality genes and cells were filtered out (excluding genes detected in <5 cells; excluding cells with <200 or >7,500 genes, >25 % mitochondrial content, or >1 % hemoglobin gene expression). Data were log-normalized (scale factor 10,000) and the top 3,000 highly variable genes were selected. Datasets were integrated using Harmony (on 50 principal components) to correct batch effects. Principal component analysis (PCA) was performed for dimensionality reduction, followed by clustering (Louvain algorithm, resolution 0.8) and visualization with t-distributed stochastic neighbor embedding (t-SNE) based on the Harmony embeddings. Cell types were annotated by a combination of automated labeling (SingleR with the Human Primary Cell Atlas), canonical marker genes, and manual curation. Malignant epithelial cells were distinguished from non-malignant cells using inferCNV (with normal epithelial references) and confirmed with CytoTRACE scores. Differentially expressed genes were identified using the Wilcoxon rank-sum test (Seurat FindMarkers, log₂ fold change >0.25, adjusted P < 0.05).

Patient Tumor Specimens: Archival tumor tissues were collected from 83 patients with newly diagnosed, histologically confirmed HNSCC treated at Xiangya Hospital (2014–2022). All patients had adequate formalin-fixed, paraffin-embedded (FFPE) tumor material and complete clinicopathological data, and none received anti-cancer therapy prior to biopsy or surgery. Patients with concurrent malignancies or immunosuppressive conditions were excluded. Tumor histology and grade were confirmed by two pathologists, and staging was reassessed according to the 8th Edition AJCC TNM system. Key clinicopathologic variables (age, tumor stage, nodal status, differentiation, Ki-67 index, etc.) were recorded for each case (see Supplementary Table 1). Cathepsin L (CTSL) expression in these tumors was evaluated by immunohistochemistry (IHC) and quantified using an H-score (0–300). Patients were stratified into “CTSL-low” and “CTSL-high” groups based on the median H-score for outcome analyses. A subset of 21 patients from this cohort who later developed recurrent/metastatic disease and received anti–PD-1 therapy (nivolumab, pembrolizumab, or toripalimab) was identified for exploratory analysis. Pretreatment biopsy specimens from these patients were analyzed by multiplex immunofluorescence to assess spatial expression of CTSL, PD-L1, CD8, and granzyme B, and clinical responses were evaluated by RECIST v1.1 criteria (complete/partial response vs. stable/progressive disease).

Cell lines and genetic manipulation

Cell Lines and Culture: Human HNSCC cell lines (CAL27, FaDu, SAS) were obtained from ATCC, and HN8 was obtained from iCell Bioscience (Shanghai, China). The mouse HNSCC cell line (Meer, C57BL/6 background) was from iCell, and the mouse hepatoma cell line Hepa1-6 was from ATCC. HEK293T cells were used for viral packaging and transient transfection. All cells were maintained in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10 % fetal bovine serum and 1 % penicillin–streptomycin, at 37°C in a humidified 5 % CO₂ incubator. Cell line identities were verified by short tandem repeat profiling, and all cultures tested negative for mycoplasma. Experiments were performed on cells within 20 passages of thawing. CTSL-C25A (Cys25Ala) was generated by site-directed mutagenesis on pcDNA3.1-CTSL-Flag and sequence-verified.

Plasmids and Constructs: A human CTSL overexpression plasmid was constructed by cloning the full-length CTSL cDNA into pcDNA3.1(+) with a C-terminal Flag tag (pcDNA3.1-CTSL-Flag); the empty pcDNA3.1 vector served as control. For mapping protein interactions, full-length PDK1 and two truncated PDK1 mutants (N-terminal kinase domain and C-terminal pleckstrin homology domain) were cloned into expression vectors with N-terminal Flag tags. Three distinct short hairpin RNA sequences targeting human CTSL (shCTSL-1, −2, −3; see Supplementary Table 2 for target sequences) were cloned into the pLKO.1-GFP-puro lentiviral vector under the U6 promoter; a non-targeting shRNA (shNC) in the same vector was used as a negative control. For CRISPR/Cas9-mediated knockout of CTSL, a single-guide RNA targeting mouse Ctsl (targeting exon 3, sequence in Supplementary Table 2) was cloned into lentiCRISPRv2, and a scramble sgRNA was used as control. Additionally, two small interfering RNAs targeting human PDK1 (siPDK1-1 and siPDK1-2) and a non-targeting siRNA (siNC) were synthesized by the institutional core facility for transient knockdown experiments. NEDD4L knockdown used siNEDD4L-1/−2 (50 nM; Supplementary Table 2), transfected with Lipofectamine 3000 for 48 h.

Stable Cell Line Generation: Lentiviruses for shRNA knockdown and CRISPR constructs were produced by co-transfecting HEK293T cells with the transfer vectors and packaging plasmids psPAX2 and pMD2.G. Collected viral supernatants were used to transduce target cells in the presence of 8 µg/mL polybrene. For stable CTSL knockdown, CAL27 and SAS cells were infected with shCTSL (or shNC) lentiviruses and, 48 hours post-infection, were placed under puromycin selection (2 µg/mL) for ∼10 days to establish polyclonal stable knockdown lines. Stable CTSL overexpressing lines were generated by transfecting HN8 and FaDu cells with pcDNA3.1-CTSL-Flag (or empty vector) using Lipofectamine™ 3000, followed by G418 selection (600 µg/mL) for 1–2 weeks to isolate resistant clones. For CTSL knockout in murine cells, Meer and Hepa1-6 cells were transduced with lentiCRISPRv2-sgCTSL and selected with puromycin (1 µg/mL) for 7–10 days; single-cell clones were expanded from the CTSL knockout pool. All stable cell line modifications were verified by Western blotting (confirming loss or overexpression of CTSL) and, for CRISPR clones, by Sanger sequencing of the targeted locus. PDK1 knockdown in cell culture was achieved by transient transfection of siPDK1 duplexes using Lipofectamine 3000, with analyses performed 48 hours after transfection.

Animal studies

Mouse Models: Female BALB/c nude mice (6 weeks old, T cell–deficient) and immunocompetent C57BL/6 mice (6 weeks old) were purchased from Hunan SLK JD Laboratory Animal Co. Ltd (Hunan, China) and housed in a specific pathogen-free facility. Mice were maintained under standard conditions (12-hour light/dark cycle, controlled temperature and humidity, with food and water provided ad libitum).

Tumor Xenografts: For human tumor xenograft studies, HNSCC cell lines with stable CTSL modulation were injected subcutaneously into BALB/c nude mice. Specifically, CAL27 cells with CTSL knockdown (shCTSL) or control shRNA (shNC), and HN8 cells overexpressing CTSL (OE-CTSL) or vector control, were suspended in PBS and implanted into the flank (5 × 10^6 cells in 0.1 mL per mouse). Tumor growth was measured every 2–3 days with digital calipers, and volumes were calculated as (length × width²)/2. Mice were observed for up to 21 days, after which they were euthanized and tumors were harvested and weighed.

Syngeneic Tumor Models and Treatments: For immunocompetent models, murine HNSCC (Meer) and hepatoma (Hepa1-6) cells with or without Ctsl knockout were injected subcutaneously into C57BL/6 mice (1 × 10^6 cells in 0.1 mL PBS). Once tumors became palpable (typically ∼50–100 mm³), mice were randomly assigned to treatment groups. In one experiment, cohorts of mice bearing Ctsl-knockout or control tumors were treated with an anti–PD-1 monoclonal antibody (clone RMP1-14, 200 µg intraperitoneally, twice per week) or an isotype-matched IgG control to assess response to immune checkpoint blockade. In a separate experiment, mice bearing established Meer tumors (∼100 mm³) were treated with a selective CTSL inhibitor (Z-FY-CHO, 10 mg/kg intraperitoneally, three times per week for 2 weeks) in combination with or without anti–PD-1 therapy. For this combination study, animals were allocated into four groups (n = 5–8 per group): vehicle + IgG, vehicle + anti–PD-1, Z-FY-CHO + IgG, and Z-FY-CHO + anti–PD-1. Tumor size and body weight were monitored twice weekly throughout the treatment period. Mice were humanely sacrificed approximately 21 days after tumor inoculation (or earlier if meeting humane endpoints).

Tissue Harvest and Processing: At the experimental endpoint, tumors were excised and processed for downstream analyses. Portions of each tumor were fixed in 10 % neutral-buffered formalin and paraffin-embedded for histology and IHC, while other portions were snap-frozen in liquid nitrogen for RNA/protein extraction or mechanically/enzymatically dissociated into single-cell suspensions for flow cytometry. Xenograft tumors derived from human cell lines were minced with sterile scalpels and gently dissociated (e.g., using a gentleMACS™ tissue dissociator), then filtered through a 70 µm cell strainer to obtain single cells. Murine syngeneic tumors were subjected to enzymatic digestion in RPMI-1640 containing collagenase IV, hyaluronidase, and DNase I (final concentrations ∼2 mg/mL, 0.5 mg/mL, and 50 U/mL, respectively) for 30–60 minutes at 37°C with agitation. The digested tissues were then triturated and filtered (100 µm then 70 µm mesh). Red blood cells were lysed with ammonium chloride buffer as needed. Viable leukocytes and tumor cells were pelleted, counted (trypan blue exclusion), and either immediately used for flow cytometric analysis or stored appropriately for later assays.

Flow cytometry

Cell-Surface PD-L1 in Cell Lines: Flow cytometry was used to quantify cell-surface PD-L1 levels on cultured tumor cells. Briefly, cells (including CAL27 and SAS with shCTSL or shNC, HN8 with CTSL overexpression or vector, and cells treated with CTSL inhibitor Z-FY-CHO (10 µM) or DMSO vehicle for 24 h) were harvested at ∼80 % confluence. After washing with ice-cold PBS + 2 % FBS, 1 × 10^6 cells were resuspended in staining buffer (PBS + 2 % FBS) and blocked with 5 % bovine serum albumin for 15 min on ice. Cells were then incubated for 30 min on ice in the dark with a PE-conjugated anti–PD-L1 antibody (clone 29E.2A3, BioLegend; 1:50 dilution) or an isotype-matched PE-conjugated control antibody. After staining, cells were washed twice and counterstained with 7-AAD viability dye (to exclude non-viable cells). Data were acquired on a FACSCanto II flow cytometer (BD Biosciences), collecting at least 10,000 viable cell events per sample. Live, single cells were gated based on forward/side scatter and 7-AAD negativity. The percentage of PD-L1–positive cells and median fluorescence intensity (MFI) were analyzed using FlowJo software (v10.8).

Tumor-Infiltrating Lymphocyte Analysis: Single-cell suspensions from dissociated syngeneic tumors (Meer or Hepa1-6 grafts) were analyzed by flow cytometry to characterize T cell populations. Cell suspensions were first incubated with an anti-mouse CD16/32 Fc-block (TruStain FcX™, BioLegend) for 10 min on ice to prevent non-specific Fc binding. Cells (approximately 1 × 10^6 per sample) were then stained for 20 min at 4°C with fluorochrome-conjugated antibodies against mouse immune markers: APC—Cy7 anti-CD45, PerCP-Cy5.5 anti-CD3ε, APC anti-CD8α, and BV421 anti–PD-1 (all from BioLegend, 1:100 dilutions). A fixable viability dye (Zombie NIR, BioLegend) was included to exclude dead cells. For intracellular granzyme B detection, cells were subsequently fixed and permeabilized (Cyto-Fast kit, BioLegend) and stained with PE-conjugated anti–Granzyme B (BioLegend, 1:100) for 30 min at room temperature. Stained cells were washed and acquired on a FACSCanto II flow cytometer, and at least 10,000 live CD45⁺ events were collected per sample. Data were analyzed with FlowJo, applying sequential gating to isolate single, live lymphocytes and then gating on CD45⁺CD3⁺CD8⁺ T cells. The frequencies of CD8⁺ T cells expressing PD-1 and the fraction producing granzyme B were determined. All flow cytometry experiments were performed with appropriate isotype and single-color controls for gating and compensation. Data shown represent at least three independent experiments.

Western blotting and immunoprecipitation

Cells were lysed on ice in RIPA buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 % NP-40) supplemented with protease and phosphatase inhibitors. After 30 min lysis, lysates were clarified by centrifugation (≈12,000 × g, 15 min, 4°C). The supernatants were collected, and protein concentrations were determined by a BCA assay. For Western blotting, equal amounts of protein (20–40 µg per sample) were mixed with 2 × SDS sample buffer, boiled for 5 min, and separated by SDS–PAGE on 8–12 % polyacrylamide gels. Proteins were transferred onto polyvinylidene difluoride (PVDF) membranes, which were then blocked for 1 hour at room temperature with 5 % non-fat milk in TBST (TBS + 0.1 % Tween-20). Membranes were incubated overnight at 4°C with primary antibodies against the proteins of interest, including CTSL, PD-L1, AKT (total and phosphorylated Ser473), PDK1, and loading controls (e.g., α-tubulin or β-actin). Primary antibodies were used at 1:1,000 dilution (unless specified otherwise) and were sourced from Cell Signaling Technology or Abcam. After washing in TBST, membranes were probed with HRP-conjugated secondary antibodies (anti-mouse or anti-rabbit IgG, 1:3,000) for 1 hour at room temperature. Immunoreactive bands were visualized using an enhanced chemiluminescence (ECL) substrate and documented with a Bio-Rad ChemiDoc imaging system. Densitometric quantification of band intensities was performed using ImageJ software, with protein levels normalized to the chosen loading control.

For co-immunoprecipitation (co-IP) assays, cell lysates containing 1–1.5 mg total protein were first pre-cleared by incubating with Protein A/G agarose beads for 1–2 hours at 4°C. The cleared supernatant was then incubated overnight at 4°C with 3 µg of the appropriate capture antibody or with negative control IgG. Depending on the experiment, antibodies used for IP included anti-Flag M2 (Sigma-Aldrich), anti-Myc (Cell Signaling), anti-CTSL (Abcam), or anti-PDK1 (Cell Signaling), chosen to pull down either epitope-tagged constructs or endogenous proteins. Immune complexes were recovered by adding fresh Protein A/G beads and rotating for an additional 2–3 hours at 4°C. Beads were then washed five times with cold lysis buffer (or NP-40/TBST buffer) to remove non-specific proteins. Bound complexes were eluted by boiling the beads in 2 × SDS loading buffer for 5 minutes. The immunoprecipitated samples were analyzed by SDS–PAGE and Western blotting as described above, probing for the relevant interaction partners. These co-IP experiments were performed in both stable cell lines (with or without CTSL overexpression) and transiently transfected HEK293T cells, to validate protein–protein interactions under different expression conditions.

Cycloheximide Chase Assay: To evaluate protein stability, a cycloheximide (CHX) chase experiment was performed for PD-L1. Cells (CTSL-knockdown CAL27 vs. shNC, and CTSL-overexpressing HN8 vs. vector control) were treated with CHX (50 µg/mL) to block new protein synthesis and were harvested at 0, 4.5, 9, and 18 hours post-treatment. PD-L1 levels at each time point were measured by Western blot. Band intensities were quantified by ImageJ and normalized to β-tubulin. The remaining PD-L1 (% of time 0) was plotted to estimate the protein half-life under each condition.

Mass spectrometry for protein interactome analysis

To identify CTSL-binding proteins, co-immunoprecipitation coupled with mass spectrometry was performed. HN8 cells stably overexpressing Flag-tagged CTSL were lysed and subjected to anti-Flag immunoprecipitation as described above. Parallel immunoprecipitations with a non-specific IgG served as negative controls. The resulting protein complexes were resolved on a 10 % SDS–PAGE gel, and the gel was stained with Coomassie Brilliant Blue. Each entire lane was excised and cut into ∼1 mm³ pieces for in-gel tryptic digestion. Peptides were extracted and analyzed by liquid chromatography-tandem mass spectrometry (LC–MS/MS) using a Thermo Fisher Q Exactive™ Plus Orbitrap mass spectrometer coupled to an EASY-nLC 1200 nanoLC system. Peptide mixtures were loaded onto a C18 trap column and then separated on a C18 analytical column with a linear acetonitrile gradient. The mass spectrometer was operated in data-dependent acquisition mode, acquiring full MS scans (m/z 300–1800) at high resolution, followed by MS/MS fragmentation (HCD) of the top precursor ions. Raw data files were processed using Proteome Discoverer software (v2.2) and searched against the UniProt human reference proteome database. Trypsin specificity was assumed (up to 2 missed cleavages), and a 1 % false discovery rate (FDR) cutoff was applied at both peptide and protein levels. Proteins identified in the CTSL–Flag IP samples but not detected in IgG negative control IPs were considered candidate CTSL interactors. High confidence interacting proteins were defined as those present in at least 2 of 3 biological replicates of the CTSL IP (and absent in negative controls), with at least one unique peptide and a minimum protein score above a predefined threshold. These candidate interacting proteins were subjected to further validation and bioinformatic analysis.

GST pull-down assay

A glutathione S-transferase (GST) pull-down assay was conducted to test direct binding between CTSL and PDK1. Full-length human CTSL cDNA was cloned into the pET-28a expression vector to produce a His-tagged CTSL protein, and full-length human PDK1 cDNA was cloned into pGEX-4T-1 to produce a GST-tagged PDK1 fusion protein. Both constructs were expressed in E. coli BL21(DE3) cells. Recombinant His–CTSL was purified from bacterial lysates using nickel–NTA agarose, and GST–PDK1 was purified using glutathione agarose according to the manufacturers’ protocols. For the pull-down assay, ∼20 µg of purified GST–PDK1 (bound to glutathione agarose beads) was incubated with ∼20 µg of purified His–CTSL in pull-down buffer (20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM DTT, 0.1 % NP-40) at 4°C overnight with gentle rotation. GST alone (without PDK1) incubated with His–CTSL was included as a negative control. After incubation, the beads were washed five times with cold PBS to remove unbound proteins. Bound proteins were eluted by boiling the beads in 2 × SDS sample buffer for 5 min and analyzed by SDS–PAGE. The presence of CTSL and PDK1 in the pull-down complexes was confirmed by Western blotting with anti-His and anti-GST antibodies. This assay verified whether a direct physical interaction between CTSL and PDK1 occurs in vitro.

Cellular functional assays

EdU Incorporation (DNA Synthesis) Assay: Cellular DNA synthesis/proliferation was evaluated using a 5-ethynyl-2′-deoxyuridine (EdU) incorporation assay. CAL27 cells with stable CTSL knockdown (shCTSL) versus negative control (shNC), and HN8 cells stably overexpressing CTSL versus vector control, were seeded on sterile glass coverslips in 24-well plates. When cells reached ∼60 % confluence, they were incubated with EdU (10 µM final concentration) for 2 hours at 37°C. Cells were then fixed with 4 % paraformaldehyde for 15 min and permeabilized with 0.2 % Triton X-100. Incorporated EdU was detected using a Click-iT™ EdU Alexa 594 Imaging Kit (Thermo Fisher Scientific) following the manufacturer’s protocol (azide-conjugated fluorescent dye labeling of EdU). Cell nuclei were counterstained with DAPI. Images of at least five random fields per coverslip were captured using a fluorescence microscope (Olympus IX73). EdU-positive (red fluorescent) nuclei and total (DAPI-stained) nuclei were counted using ImageJ. The percentage of EdU⁺ cells was calculated for each sample as an index of DNA synthesis activity.

Transwell Migration and Invasion Assays: Transwell chamber assays (8 µm pore size inserts, Corning) were used to assess cell migratory and invasive capabilities in vitro. For the migration assay, HNSCC cells (CTSL-silenced CAL27 or negative control, and CTSL-overexpressing HN8 or negative control) were serum-starved for 6 hours, then 2 × 10^4 cells in serum-free DMEM were plated in the upper chamber of a 24-well Transwell insert. For invasion assays, the insert membranes were pre-coated with 50 µL of diluted Matrigel (1:8 dilution of growth factor–reduced Matrigel) to form a reconstituted basement membrane layer, and 5 × 10^4 cells in serum-free medium were added to each upper chamber. In both assays, the lower wells were filled with DMEM containing 10 % FBS to serve as a chemoattractant. After incubation (24 h for migration, 48 h for invasion) at 37°C, non-migrated cells on the top side of the insert were gently removed with a cotton swab. Cells that had migrated or invaded to the underside of the membrane were fixed in 4 % paraformaldehyde for 15 min and stained with 0.1 % crystal violet solution for 10 min. Five random fields per insert were imaged under a microscope at 200 × magnification, and the number of stained cells was counted using ImageJ. For mechanistic experiments, the AKT pathway was modulated during these assays by adding the AKT activator SC-79 (4 µg/mL) or the AKT inhibitor MK-2206 (5 µM) to both the upper and lower chambers (or adding DMSO as vehicle control) immediately after seeding; these reagents were present throughout the assay to evaluate AKT-dependent effects on cell migration/invasion.

Cell Proliferation (CCK-8) Assay: Cell proliferation rates were measured using the Cell Counting Kit-8 (CCK-8, Dojindo), which assesses metabolic activity as a proxy for viable cell number. CAL27 (shCTSL vs shNC) and HN8 (CTSL overexpression vs vector) cells were plated in 96-well plates at an initial density of 2 × 10^3 cells per well (in 100 µL of culture medium). Cells were allowed to grow under standard conditions, and proliferation was monitored over 5 days. Every 24 hours (starting at 24 h after seeding), 10 µL of CCK-8 reagent was added to the designated wells, followed by incubation for 1 hour at 37°C. The absorbance at 450 nm was then measured with a microplate reader. Each time point was performed in triplicate wells, and background subtraction was done using day 0 readings. Growth curves were plotted as the mean ± SD of absorbance values over time. In select experiments, SC-79 (4 µg/mL) or MK-2206 (5 µM) was added to the culture medium of some wells to determine the influence of AKT activation or inhibition on cell proliferation.

Gene expression analysis (qRT-PCR)

Quantitative reverse-transcription PCR (qRT-PCR) was used to quantify mRNA expression of key genes. Cells from various experimental conditions (CAL27 and SAS with shCTSL or shNC; HN8 and FaDu with CTSL overexpression or vector; and CAL27/SAS treated with 10 µM Z-FY-CHO or vehicle for 24 h) were harvested for RNA extraction. Total RNA was isolated using TRIzol® reagent (Thermo Fisher Scientific) according to the manufacturer’s instructions. One microgram of RNA from each sample was reverse-transcribed into cDNA using the GoScript™ Reverse Transcription System (Promega) with oligo(dT) primers. The resulting cDNA was used as template for real-time PCR, which was performed on an ABI 7500 Fast instrument using a SYBR Green detection kit (QuantiFast SYBR Green PCR Kit, Qiagen). Reactions (20 µL volume) were run in technical triplicates for each sample. Gene-specific primers for CD274 (PD-L1), CTSL, PDK1, and the reference gene ACTIN (internal control) were used at 0.5 µM; primer sequences are listed in Supplementary Table 3. Thermal cycling conditions consisted of an initial denaturation (95°C for ∼5 min) followed by 40 amplification cycles (95°C for 10 s, 60°C for 30 s). The specificity of amplification was confirmed by melt-curve analysis. Relative mRNA expression levels were calculated using the 2^-ΔΔCt method, normalizing each target gene to ACTIN and comparing experimental groups to the appropriate negative control group. Each experiment included at least three biological replicates, and data are presented as mean ± SD of those replicates.

Immunofluorescence and immunohistochemistry

Immunofluorescence in Cultured Cells: To visualize subcellular localization and co-localization of proteins, immunofluorescence staining was performed on HNSCC cells grown on coverslips. Cells were fixed with 4 % paraformaldehyde for 15 min at room temperature and permeabilized with 0.2 % Triton X-100 for 10 min. After blocking with 5 % BSA in PBS for 30 min, cells were incubated with primary antibodies against the proteins of interest (for example, rabbit anti-CTSL and mouse anti-PDK1) at 4°C overnight. The next day, slides were washed and incubated with species-specific secondary antibodies conjugated to fluorophores (e.g., DyLight 594 anti-rabbit IgG and DyLight 488 anti-mouse IgG, Vector Laboratories) for 1 hour at room temperature in the dark. Nuclei were counterstained with DAPI, and coverslips were mounted onto slides using an anti-fade mounting medium. Fluorescent images were captured with a Leica TCS SP8 confocal microscope and processed using ImageJ. Co-localization and distribution of CTSL and PDK1 in cells were assessed qualitatively from merged channels.

Multiplex Immunofluorescence on Tissue Sections: Frozen or FFPE tissue sections (4 µm thick) from mouse tumors (syngeneic Meer or Hepa1-6 grafts) and from human biopsy samples were subjected to multiplex immunofluorescence staining to examine tumor and immune cell markers. FFPE sections were first deparaffinized and underwent heat-induced antigen retrieval (in sodium citrate buffer, pH 6.0). All sections (FFPE or frozen) were then blocked with normal serum (e.g., 5–10 % appropriate serum in PBS) to reduce nonspecific binding. Primary antibodies against combinations of targets were applied simultaneously overnight at 4°C. For example, adjacent serial sections were stained with: (1) rabbit anti-CTSL plus rat anti-CD8α; (2) mouse anti–PD-L1 plus rat anti-CD8α; or (3) rabbit anti–granzyme B plus rat anti-CD8α (typical dilution 1:100–1:300 for each primary). After washing, slides were incubated with a cocktail of secondary antibodies labeled with distinct fluorochromes (e.g., DyLight 488 anti-rat IgG, and DyLight 594 anti-rabbit or anti-mouse IgG) for 1 hour at room temperature. Nuclei were counterstained with DAPI, and slides were mounted with ProLong™ Gold Antifade reagent. Negative control slides (omitting primary antibodies) were included to confirm specificity of staining. Stained tissue sections were examined under a fluorescence microscope, and five high-power fields per sample were analyzed. Image analysis (using ImageJ or similar software) was performed to count CTSL⁺ tumor cells and to quantify CD8⁺ T cells co-expressing PD-L1 or granzyme B in the tumor microenvironment.

Immunohistochemistry (IHC): Traditional IHC was carried out on FFPE tissue sections from human HNSCC samples and mouse tumors to detect CTSL and PD-L1. Tissue sections (4 µm) were deparaffinized in xylene and rehydrated through graded ethanol series. Antigen retrieval was performed by heating sections in 10 mM sodium citrate buffer (pH 6.0) at ∼95°C for 15 min. After cooling, endogenous peroxidase activity was quenched by 3 % H₂O₂ for 10 min. Sections were then blocked with 5–10 % normal goat serum for 20 min at room temperature. Primary antibodies against CTSL (rabbit polyclonal, Abcam, 1:400 dilution) or PD-L1 (rabbit monoclonal, Abcam, 1:200) were applied to sections and incubated overnight at 4°C in a humidified chamber. The following day, slides were washed and incubated with a biotinylated secondary antibody (goat anti-rabbit IgG, 1:400) for 30 min, followed by incubation with an avidin–biotin peroxidase complex (Vectastain Elite ABC kit) for another 30 min. The immune complexes were visualized using 3,3′-diaminobenzidine (DAB) as the chromogenic substrate, yielding a brown precipitate at sites of antigen localization. Finally, sections were counterstained with hematoxylin, dehydrated in ethanol, cleared in xylene, and mounted with coverslips.

IHC Scoring: Stained slides were evaluated independently by two board-certified pathologists blinded to the experimental groups. CTSL IHC expression was quantified using the H-score method: each sample’s score was calculated as (percentage of cells at intensity 0 × 0) + (percentage at intensity 1+ × 1) + (percentage at intensity 2+ × 2) + (percentage at intensity 3+ × 3), yielding a total range of 0–300. Tumor samples with H-scores above the cohort median were classified as CTSL-high, and those below the median as CTSL-low for subsequent analyses. PD-L1 expression in tumor cells was scored by the tumor proportion score (TPS), defined as the percentage of viable tumor cells showing positive membranous staining. Consistent with clinical criteria, a TPS >1 % was considered PD-L1 positive. Any discrepancies between the two primary reviewers (which occurred in <5 % of cases) were resolved by joint review and consensus.

Statistical analysis

All quantitative data are presented as the mean ± standard deviation (SD) from at least three independent experiments (or biological replicates), unless stated otherwise. Statistical analyses were performed using GraphPad Prism 9.5 (GraphPad Software) and IBM SPSS Statistics 26. Before hypothesis testing, data were checked for normal distribution (Shapiro–Wilk test) to guide the choice of parametric or non-parametric tests. For comparisons between two groups, a two-tailed unpaired Student’s t-test was used if data were normally distributed (with approximately equal variances), and a Mann–Whitney U test was used for non-normally distributed data. For comparisons of more than two groups, one-way analysis of variance (ANOVA) was applied, followed by Tukey’s multiple-comparison post hoc test if data were parametric; the Kruskal–Wallis test followed by Dunn’s post hoc test was used for non-parametric data. Two-way repeated-measures ANOVA was employed to analyze longitudinal tumor growth curves across treatment groups, with appropriate post hoc testing (e.g., Tukey’s test) to compare time-point differences between groups. Categorical variables (such as clinical response rates or contingency tables) were compared using the Chi-square test or Fisher’s exact test, as appropriate. Correlations between continuous variables were evaluated using Pearson’s correlation coefficient for parametric data or Spearman’s rank correlation for non-parametric data. In analyses stratifying samples by gene expression level (e.g., “CTSL-high” vs “CTSL-low” tumors), groups were defined by median value unless specified otherwise. Survival analyses (for example, using the TCGA cohort) were conducted by generating Kaplan–Meier curves and comparing outcomes with the log-rank test. Where optimal cut-off values for high vs low expression were required (e.g., for TCGA gene expression), the threshold was determined based on maximizing the separation of survival curves (maximally selected rank statistic). A two-sided P value < 0.05 was considered statistically significant for all tests.

CRediT authorship contribution statement

Yaodong Ding: Writing – review & editing, Writing – original draft, Software, Resources, Methodology, Formal analysis, Data curation, Conceptualization. Haoyu Zhang: Writing – review & editing, Methodology, Formal analysis, Data curation. Xueying Wang: Writing – review & editing, Methodology, Investigation, Data curation. Jiaqi Tan: Writing – review & editing, Data curation. Minghao Wang: Methodology, Data curation. Yuhan Chen: Validation, Data curation. Imadoudini Hassimi Safia: Writing – review & editing, Data curation. Gangcai Zhu: Validation, Supervision, Funding acquisition. Xin Zhang: Supervision, Resources, Funding acquisition. Yong Liu: Visualization, Validation, Supervision.

Declaration of competing interest

The authors declare no competing interests.

Data availability

The datasets supporting the conclusions of this article are included within the article and its additional files.