Cytokine-driven CD38 + HLA-DR + CD8 + T cells define a bystander program predicting poor prognosis in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) exhibits poor immunotherapy responses despite CD8 + T cell infiltration, suggesting critical gaps in understanding tumor-reactive versus bystander immunity. While CD8 + T cells can mediate anti-tumor effects through TCR engagement, their effector function may be subverted by immunosuppressive programs. Here, we identified a distinct CD38 + HLA-DR + CD8 + T cell population whose expansion associates with adverse HCC outcomes independent of tumor-specific immunity. We integrated data from ex vivo cytokine (IL-12/15/18)-stimulated CD8 + T cells, bulk RNA-seq and CITE-seq of matched peripheral blood (PB), non-tumor liver (NT), and tumor (T), and TCR clonotype mapping from an HCC cohort. We further projected cytokine-driven versus mutation-associated neoantigen (MANA) signatures onto TCGA-LIHC and integrated these with clinical flow cytometry, serum cytokines, and survival modeling in this cohort. IL-12/15/18 maximally drove bystander CD8⁺ proliferation and effector function. Bulk RNA-seq distinguished cytokine- from TCR-driven activation, with strong MKI67 upregulation and lower exhaustion transcripts ( PDCD1, CTLA4, LAG3 ) in the former. Co-expression of CD38 and HLA-DR reliably identified these IL-12/15/18-induced bystander-activated CD8⁺ T cells. Comparative profiling indicated that cytokine-activated CD38⁺HLA-DR⁺ CD8⁺ T cells adopt an NK-like killing program. CITE-seq across PB/NT/T confirmed an in-vivo CD38⁺HLA-DR⁺ subset, whose state shifts from circulating bystander-like to tumor-adapted effector; TCR mapping showed extensive PB-tumor clonotype sharing. Gene-set analyses revealed an E2F-dominated proliferation/DNA-repair program in CD38⁺HLA-DR⁺ CD8⁺ T cells (“risk” genes), versus an IRF-driven TCR/IL2-STAT5/IFN-gamma effector program in MANA-specific “protective” genes. In TCGA-LIHC, the CD38⁺HLA-DR⁺ signature associated with higher IL12A/IL15/IL18 expression and immunosuppressive signatures (Foxp3 high Tregs, MDSCs), indicating a cytokine-rich suppressive milieu. Clinically, peripheral CD38⁺HLA-DR⁺ CD8⁺ T cells were enriched in HCC versus healthy donors and were highest in tumors; concomitant elevation of serum IL-12p70, IL-15, and IL-18 tracked with increased CD38⁺HLA-DR⁺ frequencies. Higher peripheral CD38⁺HLA-DR⁺ proportions associated with worse survival and remained an independent prognostic factor after adjustment for tumor burden and liver function. Together, cytokine-driven bystander activation generates a transcriptionally and functionally distinct CD38⁺HLA-DR⁺ CD8⁺ population with NK-like features. Unlike MANA-specific T cells linked to favorable outcomes, this cytokine-driven program aligns with proliferative, immunosuppressive networks, leading to poor prognosis and nominating CD38⁺HLA-DR⁺ CD8⁺ T cells as a biomarker and potential therapeutic target in HCC.