Intratumour heterogeneity (ITH) fosters the vulnerability of RNA expression-based biomarkers derived from a single biopsy to tumour sampling bias, and is regarded as an unaddressed confounding factor for patient precision stratification using molecular biomarkers. This study aimed to identify an ITH-free predictive biomarker in hepatocellular carcinoma (HCC). We interrogated the confounding effect of ITH on performance of molecular biomarkers and quantified transcriptomic heterogeneity utilising three multiregional HCC transcriptome datasets involving 142 tumoural regions from 30 patients. A de novo strategy based on the heterogeneity metrics was devised to develop a surveillant biomarker (a utility gadget using RNA; AUGUR) using three datasets involving 715 liver samples from 509 patients with HCC. The performance of AUGUR was assessed in seven cross-platform HCC cohorts that encompassed 1,206 patients. An average discordance rate of 39.9% at the level of individual patients was observed applying 13 published prognostic signatures to classify tumour regions. We partitioned genes into four heterogeneity quadrants, from which we developed and validated a reproducible robust ITH-free expression signature AUGUR that showed significant positive associations with adverse features of HCC. High AUGUR risk increased the risk of disease progression and mortality independent of established clinicopathological indices, which maintained concordance across seven cohorts. Moreover, AUGUR compared favourably to the discriminative ability, prognostic accuracy, and patient risk concordant rates of 13 published signatures. Finally, a well-calibrated predictive nomogram integrating AUGUR and tumour-node-metastasis (TNM) stage was established, which generated a numerical probability of mortality. We constructed and validated an ITH-free AUGUR and nomogram that overcame sampling bias and provided reliable prognostic information for patients with HCC. Intratumour heterogeneity (ITH) is prevalent in hepatocellular carcinoma (HCC), and is regarded as an unaddressed confounding factor for biomarker design and application. We examined the confounding effect of transcriptomic ITH in patient risk classification, and found existing molecular biomarkers of HCC were vulnerable to tumour sampling bias. We then developed an ITH-free expression biomarker (a utility gadget using RNA; AUGUR) that overcame clinical sampling bias and maintained prognostic reproducibility and generalisability across multiple HCC patient cohorts from different commercial platforms. Furthermore, we established and validated a well-calibrated nomogram based on AUGUR and tumour-node-metastasis (TNM) stage that provided an individualised prognostic information for patients with HCC.

Introduction: Intra-tumoral heterogeneity is a prominent characteristic of hepatocellular carcinoma (HCC). However, it remains unexplored whether intra-tumoral transcriptomic differences can capture crucial information regarding HCC evolution and be utilized to derive a predictive signature for patient’s clinical trajectories.Methods: We quantified transcriptomic heterogeneity using four multiregional HCC cohorts comprising 172 samples from 37 patients, and validated transcriptomic heterogeneity and spatial dynamics using multiregional single-cell transcriptomic profiling of 110,817 cells from 34 liver specimens. The HCC evolutionary signature (HCCEvoSig) was developed and assessed across six cross-platform HCC cohorts.Results: Genes exhibiting high intra- and inter-tumor expression variation were significantly enriched in a gene set associated with HCC prognosis, from which we developed and validated a reproducible and robust transcriptomic signature, HCCEvoSig. Multiregional single-cell data confirmed the high intra- and inter-tumoral heterogeneity of HCCEvoSig genes across different cell types, and importantly, demonstrated that the dysregulation of HCCEvoSig genes exhibited a geospatially gradual transition from the non-tumor region to the tumor border and tumor core, as well as from non-malignant to malignant epithelial cells. HCCEvoSig showed significant positive associations with adverse features of HCC, and a high HCCEvoSig risk score predicted increased risks of disease progression and mortality, independent of established clinicopathological indices. Furthermore, HCCEvoSig outperformed 15 published signatures in discriminative ability and prognostic accuracy, particularly regarding 1-year survival rates. Notably, HCCEvoSig demonstrated predictive utility for responses to immunotherapy and trans-arterial chemoembolization. Additionally, we established a well-calibrated predictive nomogram that integrates HCCEvoSig and TNM stage to generate an individualized numerical probability of mortality.Conclusion: Our study reveals that regional transcriptional heterogeneity within tumors is substantial enough to capture survival signals, and the constructed and validated HCCEvoSig provides reliable prognostic information for HCC patients.

Background: Programmed cell death protein-1 (PD-1) blockade has shown promising clinical efficacy in hepatocellular carcinoma (HCC), yet the underlying immunological mechanisms governing response and resistance remain unclear. This study aimed to delineate the temporal and spatial dynamics of T-cell clonotypes and their relationship with pathological response in neoadjuvant PD-1 blockade therapy in HCC. By integrating T-cell receptor (TCR) repertoire analysis with transcriptomic profiling, we sought to elucidate the immune landscape alterations associated with treatment outcomes. Methods: We analyzed longitudinal, matched tumor and blood samples from our previous clinical trial ( NCT05471674 ), encompassing 114 specimens, including pre-nivolumab tumor biopsies, post-treatment resected tumors, adjacent non-tumorous livers, and peripheral blood collected before, after, and during follow-up from 19 treatment-naïve HCC. TCR clonality, diversity, and transcriptomic signatures were assessed to characterize immune responses. Results: Elevated clonality of intratumoral TCR clonotypes (ITCs) post-treatment correlated with increased pathological tumor necrosis, predominantly driven by the most abundant top 1% ITCs. These dominant clonotypes occupied a significantly larger clonal space and demonstrated increased spatial sharing across tumor, non-tumorous liver, and peripheral blood compartments in nivolumab responders (nivo-Rs). Shared clonotypes were positively associated with tumor necrosis extent and cytolytic activity, suggesting active immune engagement. Post-treatment, shared clonotypes exhibited peripheral expansion and greater intratumoral dominance in nivo-Rs, whereas tumor unique intratumoral TCR clonotypes (ITCs) remained prevalent in non-responders (nivo-NRs). These divergent patterns were linked to higher chemokine expression within the tumor microenvironment of nivo-Rs and impaired human leukocyte antigen (HLA) class II antigen presentation in nivo-NRs, indicating distinct immune landscape configurations influencing therapeutic response. Conclusions: Our findings demonstrate that the peripheral infiltration and expansion of high-frequency intratumoral T-cell clonotypes are critical drivers of pathological response to neoadjuvant PD-1 blockade in HCC, highlighting potential immune biomarkers and therapeutic targets to enhance immunotherapy efficacy.

Cancer neuroscience highlights the critical role of neural signaling in tumors, yet a pan-cancer understanding of neuroregulatory dysregulation is lacking. We systematically characterized 130 neurotransmitter receptor (NTR) genes across 9,125 tumors from 33 cancer types. Our analysis revealed heterogeneous NTR mutations, with 17 cancer types showing elevated rates. Notably, amplification of the muscarinic receptor gene CHRM2 exhibited mutual exclusivity with the clinically actionable gene ERBB2, suggesting its potential as a therapeutic target. NTR expression was widely dysregulated and associated with altered DNA methylation, microRNA (miRNA) expression, and patient prognosis. Unsupervised clustering identified five recurrent neuroregulatory subtypes with distinct clinical and molecular features across cancers. Using low-grade glioma and liver cancer as examples, we validated that the S4 and S1 subtypes were consistently correlated with aggressive disease and poor outcomes in these two cancers, respectively. This study establishes a foundational framework for advancing cancer neuroscience and targeting neuroregulatory signaling in cancer therapy.