Spatial transcriptomics of meningioma refines postoperative progression risk by myeloid cell state
Background: Prognostic information from cell states beyond pathological grade
Myeloid cell states in the meningioma tumor microenvironment may help refine the risk of progression after surgery. Researchers at the University of Toronto and University Health Network examined cell states inside and outside tumors using single-nucleus RNA sequencing and spatial transcriptomics. They reported that adding cell-state information to an existing molecular classification reduced prognostic prediction error.
Meningioma is the most common brain tumor in adults, but its clinical course varies. Molecular classifications have been developed to capture differences that World Health Organization (WHO) pathological grade alone cannot readily explain. This study examined cell states and their spatial organization in greater detail, both within tumors and in adjacent dura. The findings were published in Nature Genetics on June 9, 2026.
Key findings: Differences by myeloid cell state and location
The researchers analyzed 580,185 high-quality nuclei from 102 samples using single-nucleus RNA sequencing. They separately obtained high-resolution spatial transcriptomics data from 43 samples and compiled expression profiles for 7,146,538 cells segmented in images. The figure of approximately 7.15 million refers to cells in the spatial analysis. The study also used 712 bulk DNA methylation and RNA analyses, 88 proteomic analyses, and 59 plasma methylation analyses. Because the analysis groups overlapped, these numbers cannot be added to derive a total patient count.
In the subset used to build the cell atlas, myeloid cells made up 22.3% of all cells and most of the non-tumor cells. Expression programs associated with cell cycle, inflammation, tissue repair, and metabolism, among other functions, were distinguished in these cells. The distribution of these states also varied by tumor molecular group. Here, a cell state describes which gene sets are active in cells of the same type.
Differences also emerged between the tumor and surrounding dura. In three spatial transcriptomics samples with preserved boundaries, myeloid-cell expression programs related to tumor necrosis factor (TNF) showed a declining pattern across the first 2 mm from the dural boundary into the tumor. This result shows a location-associated difference; it does not directly establish that cells change state as they move or identify the cause of the change.
CellChat and CellNEST analyses inferred distinct ligand–receptor interactions for different myeloid cell states. Both analyses linked the interaction between insulin-like growth factor 1 (IGF1) in myeloid cells and its receptor IGF1R in tumor cells to tissue-repair programs. These are inferences based on expression data, so actual signaling and effectiveness as a therapeutic target require functional testing.
Implications and outlook: Turning improved prediction into a clinical tool
The researchers combined seven expression programs associated with prognosis into an integrated cell-state (ICS) score. A model combining these programs with existing molecular-group information had lower Brier error—a measure of prognostic prediction error—than models using WHO grade or molecular group alone. The ICS high-risk group had a higher risk of postoperative progression or recurrence even after adjustment for WHO grade, extent of resection, receipt of adjuvant radiotherapy, and molecular group. The adjusted hazard ratio was 2.23 (95% confidence interval, 1.58–3.16); this indicates a prognostic association, not a treatment effect.
Plasma methylation analysis in 59 people was a proof of concept for noninvasively estimating myeloid cell states. The overall study was a retrospective cohort analysis, and its comparison of multiple regions within tumors was limited to four meningiomas. Further validation is needed before applying the findings to clinical decisions. Translating candidate interactions into treatments also requires testing in preclinical models that preserve the tumor microenvironment.
Source: Original Nature Genetics paper
Nature Genetics, Published online: 09 June 2026. DOI: 10.1038/s41588-026-02615-w
Summary: Bypassing the low predictive velocities and spatial structure stripping errors that historically cloud empirical histological classification in intracranial oncology, this multi-omic translation scales a programmable single-cell mapping infrastructure. Utilizing high-depth single-cell and spatial transcriptomic workflows synchronized across a cohort comprising more than seven million discrete cells, the computing platform systematically charts the tumor-intrinsic heterogeneity driving meningioma progression. The model deciphers the precise mathematical covariance linking malignant compartments to highly variable stromal microenvironments where localized fibroblastic-immune networks dynamically reshape tissue architectures. This molecular calibration delivers a validated, non-invasive computational baseline to isolate raw positional variance, predict early recurrence thresholds, and guide prospective universal patient stratification.
This study shows that immune-cell activity states, as well as immune-cell abundance, may add information to meningioma prognostic assessment. Follow-up studies could test whether cell states estimated from bulk RNA analysis of surgical tissue, when added to existing molecular classifications, distinguish progression risk within the same molecular group. The current results do not establish clinical utility for changing surveillance intervals or radiotherapy decisions.
For drug research, the findings provide a starting point for modulating candidate signals such as IGF1–IGF1R in preclinical models that preserve interactions between myeloid and tumor cells, and assessing tumor responses. Plasma methylation could also be investigated further as a noninvasive marker, but the current data do not establish the performance of a clinical test or companion diagnostic.