# Deciphering immune heterogeneity in lung adenocarcinoma via machine learning-based Differential Phenotype Immune Score: TPX2 as a key biomarker for immunotherapy resistance

## TL;DR

This study uses machine learning to identify immune subtypes in lung cancer and finds TPX2 as a key factor in immunotherapy resistance.

## Contribution

Introduces DPIS, a machine learning-based immune score, and identifies TPX2 as a novel biomarker for immunotherapy resistance in LUAD.

## Key findings

- Three immune subtypes (Wound Healing, IFN-γ Dominant, Inflammatory) were identified with distinct features and clinical outcomes.
- DPIS effectively stratifies patient survival and is linked to proliferative malignant cells.
- TPX2 promotes tumor growth and immune suppression, making it a potential therapeutic target.

## Abstract

Immune heterogeneity is a major determinant of clinical outcome and immunotherapy responsiveness in lung adenocarcinoma (LUAD). However, the tumor-intrinsic transcriptional programs that drive immune divergence across patients remain insufficiently characterized.

We constructed an integrated immune landscape of LUAD by combining bulk transcriptomic data, multi-omics profiling, and a large-scale single-cell atlas of non–small cell lung cancer. Immune subtypes were identified through integrative clustering approaches. A machine learning–derived Differential Phenotype Immune Score (DPIS) was developed to quantify immune-related phenotypic variation. Single-cell mapping, regulatory network inference, pan-cancer analyses, protein-level validation, and functional assays were conducted to interrogate key molecular drivers.

Three recurrent immune states were identified, including the Wound Healing, IFN-γ Dominant, and Inflammatory subtypes, each exhibiting distinct immune compositions, metabolic features, signalling activities, and clinical trajectories. Although tumors classified as IFN-γ Dominant or Inflammatory showed comparable sensitivity to immune checkpoint blockade, their baseline prognoses differed substantially, suggesting that immune activation alone does not fully explain outcome heterogeneity. DPIS consistently stratified overall survival across six independent cohorts and was predominantly localized to highly proliferative malignant cells at single-cell resolution. Regulatory network analysis revealed that DPIS-high tumors were governed by cell cycle–associated transcriptional programs. Among the DPIS components, TPX2 emerged as a central regulator linking proliferative signalling to immune suppression, characterized by impaired antigen presentation, reduced immune cell infiltration, and unfavorable immunotherapy responses. Functional experiments further demonstrated that TPX2 promotes tumor cell proliferation, migration, and resistance to apoptosis.

This study identifies a proliferation-driven immune suppression program in LUAD, establishes DPIS as a robust and clinically applicable framework for immune stratification, and highlights TPX2 as a potential therapeutic target for overcoming immune resistance.