Single-cell and multi-layer analysis identifies state-restricted stromal remodeling in steroid-associated osteonecrosis of the femoral head

Technical Summary: State-Restricted Stromal Remodeling in Steroid-Associated Osteonecrosis

Problem Statement
Steroid-associated osteonecrosis of the femoral head (SONFH) is characterized by marrow necrosis and structural failure, driven significantly by stromal remodeling and adipogenic imbalance within the bone marrow mesenchymal stem/stromal cell (BMSC) compartment. While the IRE1α–XBP1 axis is a known stress-adaptation module linked to lipid metabolism and cell survival, its specific contextual behavior in SONFH remains unclear. Specifically, it is unknown whether this stress-adaptation program is uniformly elevated across the stromal compartment, whether it is strictly coupled to terminal adipogenic differentiation, and whether these processes are restricted to specific cellular states or anatomical niches.

Methodology
The study employs a bounded, multi-layer analytical framework using public datasets to construct a hypothesis-generating model rather than a mechanistically closed disease model. The approach integrates four distinct analytical layers:

1. Discovery Single-Cell Analysis: Using dataset GSE316957, the authors defined a purified stromal/BMSC-like compartment (1,922 cells) from 40,722 total cells. This involved stringent quality control, anchor-guided extraction, and decontamination to remove immune, endothelial, and erythroid signals.
2. State-Axis Modeling: Instead of relying on discrete clusters, a continuous computational state axis was defined as rescaled(ADIPO − OSTEO). This allowed for the quantification of stromal dynamics along an osteogenic-to-adipogenic spectrum. The study tested a "reinforcement model" by comparing a predefined "Terminal25" (highly adipogenic) window against a "Middle50" window to see if IRE1α–XBP1-associated stress signals (a "hijack-like" program) increased monotonically with adipogenesis.
3. Spatial and Lesion-Context Mapping:
   • Spatial: Discovery-defined programs were projected onto an independent osteoarthritis (OA)-derived femoral head spatial transcriptomic reference (GSE284089) to assess anatomical contextualization, specifically looking for perivascular preferences.
   • Lesion-Context: Discovery stromal classes were transferred to an independent SONFH-related lesion dataset (GSE290411) to test for class correspondence, explicitly avoiding claims of strict quantitative replication due to expected inter-individual heterogeneity.
4. Structural Hypothesis Assembly: Pre-existing molecular dynamics (MD) readouts were utilized to assemble a structural hypothesis regarding the accommodation of dexamethasone (Dex) within a KIRA6-relevant ATP-binding pocket of IRE1α. No de novo simulations were performed; the analysis focused on local retention metrics (minimum distance, contact occupancy) to propose a testable competition framework.

Key Results

• Non-Uniform Remodeling: Single-cell analysis revealed that SONFH-associated stromal remodeling is not a uniformly elevated whole-compartment phenotype. Instead, it is a structured, state-restricted process. While adipogenic signals (ADIPO) were concentrated in specific subclusters (e.g., C10, C2, C5), the stress-adaptation "hijack-like" score was enriched in distinct niches (e.g., C6, C8). Donor-level averaging failed to detect significant differences between SONFH and OA controls, masking this internal polarization.
• Partial, Non-Monotonic Coupling: Contrary to the hypothesis that stress adaptation reinforces terminal adipogenesis, state-axis modeling showed that the IRE1α–XBP1-associated program did not strengthen along the terminal adipogenic branch. In fact, the relative reinforcement metric declined in the terminal window. Statistical analysis (mixed-effects modeling) confirmed a significant negative main effect of the terminal zone on hijack-like activity, supporting a model of partial and non-monotonic coupling between adipogenic rewiring and stress adaptation.
• Spatial Contextualization: Spatial projection onto the OA-derived femoral head scaffold indicated that transitional stromal states exhibit a local preference for perivascular neighborhoods. While global linear coupling between coupled-program scores and vascularity was weak, neighborhood-distance analysis confirmed that high-coupling spots were significantly closer to vascular regions.
• Lesion-Context Correspondence: Transfer to the independent SONFH lesion dataset confirmed that discovery-defined stromal classes could be recovered (high mapping confidence), but the specific quantitative directional elevation of scores was not uniformly replicated. This supports the view that the remodeling scaffold is structurally compatible but exhibits substantial inter-individual heterogeneity in intensity.
• Structural Hypothesis: MD readouts supported a structurally plausible hypothesis where dexamethasone can be accommodated within a KIRA6-relevant pocket of IRE1α. Dex showed high local retention occupancy, comparable to a rescue system, suggesting a potential competition framework without establishing direct target engagement.

Significance and Claims
The paper positions itself as a "bounded, hypothesis-generating framework" rather than a definitive mechanistic proof. Its primary contribution is reframing SONFH-associated stromal remodeling as a state-restricted and context-sensitive process.

The authors claim that:

1. Adipogenic rewiring in SONFH is only partially coupled to the IRE1α–XBP1 stress-adaptation program, rather than being a linear, monotonic progression.
2. The remodeling is state-restricted, occurring in specific cellular branches and anatomical niches (perivascular) rather than as a global compartmental shift.
3. The study provides a testable structural hypothesis regarding Dex-IRE1α interaction that requires future experimental validation (e.g., competition assays).
4. The findings offer a foundation for future tissue-matched validation and niche-sensitive mechanistic studies, moving away from models that assume uniform disease progression toward models that account for cellular state heterogeneity and spatial context.

The authors explicitly note limitations, including the use of OA as a reference rather than a healthy baseline, the lack of de novo structural simulations, and the cross-sectional nature of the data, which precludes definitive causal inference.