ABSTRACT: Astrocyte reactivity is increasingly recognised as a central contributor to neurological disease, yet astrocyte states do not resolve into discrete categories and their morphological organisation remains poorly understood. While transcriptomic studies have revealed substantial molecular heterogeneity, scalable frameworks for reconstructing the topology and plasticity of astrocyte morphological states are lacking. Here, we generated a multimodal dataset of human iPSC-derived astrocytes from ALS patients carrying VCP mutations and matched controls, profiled by high-content fluorescence imaging and bulk RNA sequencing under basal and inflammatory conditions. We developed SI-SimCLR, a spatially informed self-supervised learning framework that learns biologically meaningful image representations without segmentation or manual annotation. Building on these representations, we introduce an optimal transport (OT)-based framework that reconstructs astrocyte phenotypic landscapes as interconnected morphological state manifolds, enabling inference of attractor-like, transient, and highly plastic intermediary substates. SI-SimCLR outperformed standard self-supervised and pretrained baselines in capturing disease- and inflammation-associated morphology across experimental batches. OT-based landscape analysis revealed that VCP-mutant astrocytes occupy a constrained genotype-specific region under basal conditions, while inflammatory stimulation selectively remodels these states through highly connected intermediary configurations toward control-associated morphologies. Substate-level analysis further uncovered heterogeneous cell-autonomous reactive programmes within untreated VCP-mutant astrocytes that differentially recapitulate inflammation-associated states. Integration with matched transcriptomic profiling revealed a striking dissociation between molecular and morphological organisation: inflammatory stimulation dominated transcriptional variation, whereas ALS-associated mutation was the principal driver of morphological state organisation. Together, these findings establish a general framework for topology-aware reconstruction of cellular phenotypic landscapes and demonstrate that morphology encodes disease-associated cellular organisation beyond transcriptional programmes in ALS astrocytes. Here, we addressed this by generating a large, multimodal dataset of human iPSC-derived astrocytes from ALS patients carrying VCP mutations and matched controls, profiled by high-content fluorescence imaging and bulk RNA sequencing under basal conditions and controlled pro-inflammatory stimulation (IL-1α, TNF, C1q). We developed SI-SimCLR, a spatially informed contrastive learning framework that learns biologically meaningful representations from microscopy images, without segmentation or predefined labels, outperforming standard self-supervised and pretrained baselines in capturing disease- and inflammation-associated morphological variation across experimental batches. Unsupervised analysis of SI-SimCLR embeddings revealed that astrocytes occupied a structured morphological landscape composed of twelve substates with distinct biological composition and transition dynamics. Using Optimal Transport to construct a morphological transition graph, we found that VCP-mutant astrocytes occupied a constrained genotype-specific region under basal conditions, while high-exposure inflammatory stimulation partially shifted these states toward control-like configurations. Substate-resolved analysis further highlighted a subset of untreated VCP-mutant astrocytes occupying cell-autonomous morphological states that overlap with inflammation-induced reactive phenotypes. Integration with bulk RNA sequencing across the same conditions revealed a striking dissociation: while inflammatory stimulation dominates transcriptional variation, ALS mutations were the primary driver of morphological organisation. This indicates that morphological and transcriptional responses to disease and inflammation represent partially independent axes of astrocyte dysfunction. Together, these results establish a scalable, annotation-free framework for high-resolution characterisation of cellular phenotypic heterogeneity, uncover a dissociation between morphological and transcriptional modes of disease-associated variation, and provide a principled foundation for substate-resolved analysis of astrocyte biology in neurodegeneration.
