SpaNorm, SpaNormSVG, and SpiDE: A Spatial Transcriptomics Toolkit for Atlas-Scale TME Characterisation and Clinical Response Prediction in Non-Small Cell Lung Cancer
Spatial transcriptomics is transforming our ability to interrogate tumour ecosystems in situ, yet integrating high-dimensional molecular data with pathology and clinical outcomes remains a key challenge. Here, we present a single-cell, spatially resolved whole-transcriptome atlas of non-small cell lung cancer (NSCLC), comprising ~750,000 cells from 527 patient biopsies profiled using CosMx Whole Transcriptome imaging. This atlas captures 18,935 genes per cell and is unified across five independent clinical cohorts, enabling robust characterisation of the tumour microenvironment (TME) at scale.
We developed a computational framework incorporating novel quality control, spatial normalisation (SpaNorm), spatially variable gene detection (SpaNormSVG), and a spatial influence differential expression model (SpiDE). Integration with pathology-guided annotations enabled accurate cell typing and the identification of six conserved tissue neighbourhoods, linking spatial architecture to biological function. Across histology, stage, and sex, we observed distinct compositional and transcriptional reprogramming within neighbourhoods, highlighting context-dependent immune dynamics.
Focusing on immunotherapy, we analysed patients treated with immune checkpoint inhibitors (ICI) and identified spatially resolved features associated with response. Responders exhibited enrichment of plasma cells and macrophages with enhanced antigen presentation signatures, particularly within tumour–immune interface regions. In contrast, non-responders demonstrated immunosuppressive architectures, including regulatory T cell enrichment. SpiDE revealed that proximity to tumour cells and fibroblasts drives distinct immune cell transcriptional states, underscoring the importance of spatial context in modulating function.
Using pathway-level feature selection and spatial metrics, we derived predictive models of treatment response and progression-free survival with high accuracy, outperforming conventional biomarkers. Key pathways implicated include antigen presentation, interferon signalling, and immune regulation. Findings were validated in independent cohorts using orthogonal spatial platforms.
This work establishes a clinically annotated spatial atlas and computational toolkit that links tissue architecture to therapeutic response, providing a scalable framework for spatial biomarker discovery and precision oncology.
Bio
Dr Dharmesh D Bhuva is an NHMRC Emerging Leadership Fellow (EL1) at the Frazer Institute, University of Queensland, who is passionate about understanding how complex systems of gene regulation and signalling lead to organisation of healthy and diseased tissues. He completed his PhD in 2020 at the University of Melbourne and WEHI where he developed new systems biology approaches to study molecular function and gene regulation in cancer. He then undertook his post-doctoral studies at the world-renowned WEHI Bioinformatics division, where he embarked on developing novel approaches to study cancer tissues using spatial molecular data. In 2023, he joined the computational systems oncology division at the South Australian Immunogenomics Cancer Institute (SAiGENCI) to continue his cutting-edge research in developing computational approaches to study tissue architecture. Computational methods developed by Dr. Bhuva have been downloaded >240,000 times globally. Dr. Bhuva has been recently awarded a MRFF grant and a NHMRC investigator grant (EL1) to identify spatial biomarkers in cancer systems, particularly biomarkers of response to immunotherapies in NSCLC.
