Background <p>Neoadjuvant therapy plays an important role in the treatment of glioblastoma (GBM), but a considerable proportion of patients remain unresponsive to the combination of immune checkpoint blockade (ICB) and antiangiogenic therapy. Understanding the mechanisms underlying resistance to this treatment and developing novel therapeutic strategies are crucial.</p> Methods <p>We integrate extensive single-cell and spatial transcriptomic data to dissect the cellular composition and spatial organization of the GBM tumor microenvironment and validate our findings through experiments such as multiplex immunohistochemistry and atomic force microscopy. We applied 101 machine learning algorithms to evaluate the prognostic and immunological value of COL6A3⁺&#xa0;tumor-associated fibroblasts (TAFs) and GPNMB⁺&#xa0;monocyte-derived macrophages (MDMs) in multiple GBM cohorts and immunotherapy cohorts.</p> Results <p>We constructed a stromal cell atlas in GBM and identified a distinct subset of COL6A3⁺ TAFs with functional characteristics of matrix fibroblasts. We found that COL6A3⁺ TAFs are significantly enriched in non-responders to neoadjuvant combination therapy. These fibroblasts drive the spatial-reprogramming of anti-tumorigenic MDMs into a pro-tumorigenic phenotype. In turn, these reprogrammed immunosuppressive GPNMB⁺ MDMs promote vascular fibrosis mediated by COL6A3⁺ TAFs through the GPNMB-ITGB5 interaction.</p> Conclusions <p>Our findings highlight the critical role of COL6A3⁺ TAFs in regulating MDM function and spatial distribution, as well as their contribution to fibrotic tumor vasculature formation. Additionally, we propose targeting COL6A3⁺ TAFs with cilengitide as a potential therapeutic strategy.</p>