Oncogenic drivers shape the tumor microenvironment in human gliomas

Oncogenic drivers shape the tumor microenvironment in human gliomas | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Oncogenic drivers shape the tumor microenvironment in human gliomas Bo Zhao , Chae Yun Cho , Lingqun Ye , Tula Keal , Thomas Mitchell , Ines Martin-Barrio , Christina Abi Faraj , Mesut Unal , Jessica Alsing , Nora M. Lawson , Ellie Rahm Kim , Francisco Chavez , Kristy Mendoza Rangel , Sanjay K. Singh , Joseph H. McCarthy , David M. Ashley , W. K. Alfred Yung , Vinay Puduvalli , Guy Nir , Jeffrey S. Weinberg , Linghua Wang , Keith L. Ligon , Giannicola Genovese , Peter Van Loo , P. Andrew Futreal , Jason T. Huse , Jesse R. Dixon , Frederick F. Lang , Kadir C. Akdemir doi: https://doi.org/10.1101/2025.05.16.654515 Bo Zhao 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Chae Yun Cho 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Lingqun Ye 2 Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Tula Keal 3 Gene Expression Laboratory, Salk Institute for Biological Studies , La Jolla, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Thomas Mitchell 4 Department of Genetics, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ines Martin-Barrio 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Christina Abi Faraj 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Mesut Unal 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jessica Alsing 5 Department of Biochemistry and Molecular Biology, University of Texas Medical Branch , Galveston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Nora M. Lawson 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ellie Rahm Kim 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Francisco Chavez 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kristy Mendoza Rangel 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Sanjay K. Singh 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Joseph H. McCarthy 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site David M. Ashley 6 Department of Neurosurgery, Duke University School of Medicine , Durham, North Carolina, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site W. K. Alfred Yung 7 Department of Neuro-Oncology, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Vinay Puduvalli 7 Department of Neuro-Oncology, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Guy Nir 5 Department of Biochemistry and Molecular Biology, University of Texas Medical Branch , Galveston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jeffrey S. Weinberg 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Linghua Wang 2 Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Keith L. Ligon 8 Department of Pathology, Dana-Farber Cancer Institute, and Harvard Medical School , Boston, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Giannicola Genovese 2 Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center , Houston, TX, USA 9 Department of Genitourinary Medical Oncology, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Peter Van Loo 2 Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center , Houston, TX, USA 4 Department of Genetics, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site P. Andrew Futreal 2 Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jason T. Huse 10 Department of Anatomical Pathology, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jesse R. Dixon 3 Gene Expression Laboratory, Salk Institute for Biological Studies , La Jolla, CA, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Frederick F. Lang 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Kadir C. Akdemir 1 Department of Neurosurgery, The University of Texas MD Anderson Cancer Center , Houston, TX, USA 2 Department of Genomic Medicine, The University of Texas MD Anderson Cancer Center , Houston, TX, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: kcakedemir{at}mdanderson.org Abstract Full Text Info/History Metrics Preview PDF Abstract Gliomas are aggressive and heterogeneous brain tumors with limited treatment options. While oncogenic mutations in gliomas have been well-characterized, their impact on the tumor microenvironment remains poorly understood. To investigate how genomic alterations may influence the glioma microenvironment, we performed an integrative multiomic and spatial transcriptomic analysis of 93 glioma samples (46 IDH-mutant gliomas and 47 IDH-wildtype glioblastoma) from 69 patients representing both primary and recurrent stages. Using whole-genome sequencing, chromatin conformation capture (Hi-C), RNA-seq, and single-cell spatial transcriptomics (Xenium), we defined how major driver mutations influence spatial tumor organization. We found that IDH-mutant gliomas frequently harbored inflammatory microglia expressing CX3CR1 specifically within their astrocyte-like malignant neighborhoods. In contrast, glioblastomas demonstrated relatively higher T-cell infiltration and enrichment of immunosuppressive myeloid cell populations. We further compared glioblastomas harboring EGFR amplifications due to extrachromosomal DNA (ecDNA) amplifications versus linear chromosomal 7 gains. Tumors with EGFR ecDNA displayed increased presence of mesenchymal-like malignant cells, and higher interactions between pericytes and mesenchymal-like malignant cells, likely driven by hypoxia-associated vascular proliferation. Our findings reveal that the mode of oncogene amplification—linear versus ecDNA—shapes distinct tumor architectures and transcriptional dynamics. Taken together, our study highlights the role of oncogenic drivers in shaping the glioma microenvironment, revealing subtype-specific cellular ecosystems that could inform targeted therapeutic strategies. Introduction Gliomas, the most prevalent primary brain tumors in adults, exhibit significant cellular and molecular heterogeneity and clinical outcomes 1 . Current clinical classification of gliomas is based on mutations in the isocitrate dehydrogenase genes (IDH1 and IDH2) 2 . IDH-mutant gliomas, including astrocytomas (IDH-mutant) and oligodendrogliomas (IDH-mut-codel, harboring chromosome 1p/19q deletions), have distinct molecular features and clinical outcomes when compared with IDH-wildtype gliomas, which are predominantly glioblastomas (IDH-wildtype) 3 , emphasizing the importance of incorporating somatic alterations in clinical classification of gliomas. A hallmark of IDH-mutant gliomas is the accumulation of the oncometabolite D-2-hydroxyglutarate (D-2HG), which leads to widespread DNA hypermethylation phenotype in glioma cells 4 and consequentially may disrupt molecular processes including chromatin organization around key oncogenes 5 – 10 and influence the tumor microenvironment. In contrast, IDH-wildtype gliomas are characterized by higher genomic instability 11 , notably the high frequency of extrachromosomal DNA (ecDNA) amplifications 12 , 13 . These ecDNA elements drive oncogene heterogeneity and are associated with poorer patient outcomes 14 , 16 . Recently, several multi-omic analyses of gliomas have deepened our understanding of glioma tumors 17 – 24 . For example, single-cell RNA sequencing-based studies revealed four primary cellular states: neural progenitor-like (NPC), oligodendrocyte progenitor-like (OPC), astrocyte-like (AC), and mesenchymal-like (MES) enriched in glioma tumors 20 . IDH-mutant gliomas are enriched in OPC-like malignant cell states, whereas IDH-wildtype gliomas predominantly exhibit AC-like and MES-like malignant cell states. The distribution of these cell states influences tumor morphology; for instance, spatial transcriptomics studies of gliomas have demonstrated that tumors with hypoxia-enriched MES-like cell states display more organized compositions compared with their less hypoxic counterparts 21 , 22 . Despite these insights, the impact of somatic mutations on tumor composition and spatial architecture remains incompletely understood, particularly in gliomas harboring ecDNAs amplifications which can only be reliably detected through whole-genome sequencing approaches. There is a pressing need for comprehensive multi-omic studies that integrate somatic genome alterations in gliomas with spatial characterization of the tumor microenvironment to elucidate how driver mutations influence the glioma tumor microenvironment. To address this need, we performed a multi-omic analysis of 93 gliomas from 69 patients, including longitudinal samples from 19 patients with multiple time points. Of these 93 gliomas, 47 were primary gliomas, and 46 were recurrent gliomas. Disease recurrence times in this cohort spanned from 2 months to 22 years (Supplementary Table 1). In total, we profiled 449 unique specimens from these 69 patients across our cohort using different molecular assays (Supplementary Table 2). To better characterize structural variants, ecDNA, and complex alterations, half of the samples (n=39) were also analyzed using Oxford Nanopore long-read whole-genome sequencing. To associate somatic genome alterations with in vivo tumor architecture, the majority of patients were also profiled with single-cell spatial transcriptomics via the 10x Genomics Xenium platform (n=62 unique specimen from 53 distinct tumors). To further investigate our findings, select tumors were subjected to additional analyses: single-cell analysis (whole-genome sequencing=5; ATAC-seq and/or RNA-seq n=16), bulk chromatin conformation Hi-C (n=20). We used the adjacent tumor sections for the various profiling methods to reduce the effects of tumor heterogeneity in interpretation of our integrative analysis. Tumor purities ranged from 21% to 89% with a mean of 51%, and tumors with more than 25% purities were considered for our analyses. Throughout the manuscript, we integrated these diverse datasets to assess tumor heterogeneity and clonal architecture, providing a comprehensive understanding of the molecular underpinnings of glioma subtypes on tumor microenvironment. Characterization of somatic alterations in glioma samples To investigate the somatic alterations in our glioma samples, we analyzed bulk whole-genome sequencing data from tumor and matched normal (blood) samples. We built a robust data processing pipeline for identifying somatic single-nucleotide variations (SNV), structural variants (SV), copy number alterations (CNA), and extrachromosomal DNA (ecDNA) amplifications (Methods, Supplementary Figure 1 ). Overall, we observed somatic alterations that are known to be present in specific glioma subtypes 25 . Specifically, chromosome 1p/19q co-deletions were prevalent in IDH-mutant-codel gliomas, while chromosome 7 amplifications and chromosome 10 deletions were common in both IDH-mutant (astrocytoma) and IDH-wildtype gliomas ( Figure 1a , Supplementary Table 3). Somatic single-nucleotide hypermutation (tumor mutation burden > 10/mb) was observed in certain temozolomide-treated gliomas, overlapping with earlier findings 26 . Clinically, the recurrence timelines aligned with established patterns 1 , where patients with IDH-wildtype gliomas progressed more rapidly (median=1.1 years), followed by those with IDH-mutant non-codel gliomas (median =3.4), and patients with IDH-mutant-codel gliomas had the longest time to recurrence (median 4.2 years, see Supplementary Figure 2 ). Download figure Open in new tab Figure 1. Characterization of somatic mutation landscape in profiled gliomas. A) Summary of genomic alterations across glioma samples, stratified by molecular subtype (IDH-mutant-codel, IDH-mutant, IDH-wildtype). Shown from top to bottom are total somatic single nucleotide variants (SNVs), temozolomide (TMZ) treatment status, mutational signature contributions (Signature 1, 5, and 11), structural variant (SV) burden, extrachromosomal DNA (ecDNA) amplification presence, chromosomal-level copy number alterations (chr1p/19q, chr7, chr10), complex SV types, and gene-level mutations and CNAs (e.g., IDH1, EGFR, CDKN2A, TERT). Clinical features including ploidy, purity, and tumor stage (primary, recurrence 1–3) are shown in the bottom heatmaps. B) Oncogene amplifications detected by AmpliconArchitect and AmpliconClassifier across glioma samples. Orange circles indicate ecDNA classifications; purple circles denote complex, non-cyclic amplifications lacking evidence for ecDNA topology. * denotes ecDNAs missed by the AmpliconArchitect algorithm but identified using long-read whole-genome and chromatin conformation datasets. C) Fluorescence in situ hybridization (FISH) in metaphase spreads of glioma stem cell lines derived from profiled tumors confirms extrachromosomal amplification. We next examined structural variations (SVs) across our glioma samples and found comparable SV levels among the different subtypes ( Figure 1a ). Notably, IDH-wildtype gliomas exhibited more complex SVs, characterized by multiple breakpoints within confined genomic regions (Methods). Extra-chromosomal DNAs are frequently observed in glioblastoma genomes but given the complexity of these alterations, whole-genome sequencing approaches are necessary to reconstruct the architecture of such alterations. To characterize complex rearrangements and ecDNAs, we used two alternative methods based on short read sequencing. First, using the AmpliconArchitect (AA) algorithm followed by AmpliconClassifier workflow 27 , 28 , we identified focal gene amplifications and detected the presence of extrachromosomal DNA (ecDNA) in 15 tumors—14 from IDH-wildtype gliomas and one was a recurrent IDH-mutant tumor—consistent with previous reports on ecDNA prevelance 14 – 16 ( Figure 1b ). Additionally, several tumors exhibited high-level focal amplifications, but AmpliconArchitect did not find sufficient evidence to classify them as circular ecDNA. Second, we also used a graph-based structural variation algorithm, JabBA 29 , to identify potentially missed ecDNA samples by the AmpliconArchtitect algorithm. Three of such tumors were labeled as ecDNA* given the evidence provided by the JabBA algorithm and long-read whole-genome sequencing datasets. Three additional tumors were labeled as complex non-cyclic amplifications ( Figure 1b , purple). As a result, 18 tumors were identified with putative ecDNAs in our cohort where the predicted ecDNAs predominantly harbored the EGFR oncogene (n=15). In four of these EGFR-amplified tumors, MDM2 and/or CDK4 genes were predicted to be co-amplified, while CDK6 was co-amplified in three tumors. Other amplified oncogenes - CCNE1 , MET , MDM4 , MYCN , PDGFRA - were found independent of EGFR . The single IDH-mutant glioma (p30R1) with ecDNA harbored MYCN amplification. This tumor, which had recurred and had been treated with temozolomide, also exhibited SNV hypermutation (226,689 somatic mutations, 75 mutations/megabase), suggesting multiple evolutionary bottlenecks triggered by treatment during disease progression, an observation consistent with previous studies of recurrent IDH-mutant gliomas 15 . To validate our ecDNA predictions, we utilized patient-derived glioma stem cell (GSC) lines, successfully generated from the some of our profiled tumors. In matched GSCs, we used fluorescence in situ hybridization (FISH) to confirm 13 the high copy number and extrachromosomal nature of detected oncogene amplifications in metaphase spreads ( Figure 1c , Supplementary Figure 3 ). Integrative analysis enabled more accurate ecDNA reconstruction Accurate reconstruction of ecDNA remains challenging in certain tumors, even when using whole-genome sequencing approaches. Given the discrepancies in ecDNA predictions based on short read algorithms, we complemented short-read whole-genome sequencing with high-coverage (>30×) Oxford Nanopore long-read sequencing and bulk Hi-C datasets from ecDNA-predicted tumors. This multimodal dataset enabled a comparative analysis of ecDNA reconstruction algorithms using both short- and long-read whole-genome sequencing in matched tumor samples (Supplementary Table 4). In tumors with relatively simple ecDNA configurations (e.g., patient P-46p), we observed high concordance between short-read and long-read sequencing reconstructions ( Figure 2a ). However, tumors with more complex ecDNA structures including those from patients P-59 and P-68 were more difficult to resolve using a single approach. AmpliconArchitect predicted a single ecDNA containing the oncogenes EGFR (from chromosome 7) and CDK4 (from chromosome 12) in both cases. For the P-68 tumor, we observed high concordance between short-read and long-read reconstructions ( Figure 2b , Supplementary Figure 4a ). In contrast, for P-59, we detected discrepancies between the reconstructions generated by AmpliconArchitect and those from long-read-based algorithms (Decoil 30 and Coral 31 , Supplementary Figure 4b ). While AmpliconArchitect predicted a single circular amplification, long-read sequencing revealed two distinct ecDNA circles—one containing EGFR and the other containing CDK4 ( Figure 2c-d ). This two-circle model was more consistent with the observed differences in copy number between EGFR (average CN: 91) and CDK4 (average CN: 50) in bulk whole-genome sequencing data. To further evaluate the long-read findings, generated Hi-C data for these samples to characterize their architecture and genomic contiguity. We applied the ec3d algorithm, which integrates Hi-C data for ecDNA reconstruction 32 , to compare the single versus multiple ecDNA configuration predictions. In patient P-59, ec3D analysis showed an increased interaction frequency signal at the corners of the maps reminiscent of circular DNA organization 33 , corresponding to distinct ecDNA circles derived from chromosomes 7 and 12 ( Figure 2e-f ). In contrast, the AmpliconArchitect prediction of a single circle resulted in an abnormal Hi-C pattern, suggesting that the single-circle reconstruction does not accurately reflect the predominant ecDNA structure in this tumor ( Supplementary Figure 4c ). Accurate reconstruction of ecDNA is critical for understanding the expression and regulation of ecDNA-residing genes. To investigate enhancer relocation within ecDNAs, we ran neo-loop analysis 34 using Hi-C data from patient P-59. This showed an enhancer-hijacking event within the chr12-derived ecDNA circle, where HOXC11 and HOXC13 genes (chr12: 53.9Mb) were brought into close proximity with a distal enhancer (chr12:20.1-20.2Mb). Both the expression of HOXC11 , HOXC13 genes, as well as the accessibility of distal enhancer element, were elevated only in P-59 tumor, compared to other IDH-wildtype gliomas lacking this enhancer-hijacking event ( Figure 2g-h ). Download figure Open in new tab Figure 2. Integrative analysis reveals complex ecDNA architectures and enhancer hijacking event inside an ecDNA in gliomas. A) Short-read (top) and long-read (bottom) ecDNA reconstructions for tumor P-46p demonstrate high concordance in a simple EGFR amplification event on ecDNA. B) Short-read reconstruction of P-68p reveals a complex ecDNA containing EGFR (chr7), CDK4, and MDM2 (chr12). C) Short-read-based prediction for tumor P-59p suggests a single ecDNA circle spanning EGFR (chr7), CDK4, and MDM2 (chr12), but lacks consistency with observed copy number differences. D) Long-read sequencing reconstruction of P-59p reveals two distinct ecDNA circles derived from chromosomes 7 and 12, consistent with copy number differences and structural variation. E–F) Hi-C interaction maps for P-59p show elevated corner signals (black arrows) indicative of circular architecture in reconstructions of the chr12-derived ecDNA circle (E) and chr7-derived circle (F), supporting the long-read-based ecDNA topology over single-circle prediction. G) P-59p Hi-C contact matrix shows neo-loop formation between the HOXC11/13 gene cluster (chr12: ∼54Mb) and a distal enhancer (chr12: ∼20.2Mb), consistent with enhancer hijacking within the chr12 ecDNA. RNA-seq and ATAC-seq tracks show high expression and accessibility at both loci. H) A control IDH-wildtype tumor (P-19p) without ecDNA shows no enhancer interaction or activation at these loci. ecDNAs contribute to temporal heterogeneity during IDH-wildtype glioma progression Access to longitudinal samples from the same patients enabled us to study ecDNA dynamics during disease progression. Previous studies have reported that ecDNAs often preserve their structures during progression 16 . Consistent with this, we found that in two patients (P-27 and P-68), both primary and recurrent glioma tumors harbored identical EGFR ecDNAs ( Figure 3a , Supplementary Figure 5a -b). Despite radiation and chemotherapy after initial surgery ( Supplementary Figure 5c ), ecDNA configurations remained unchanged in the recurrent tumors, suggesting limited therapeutic impact on ecDNA-containing clones. Download figure Open in new tab Figure 3. Longitudinal tracking of EGFR ecDNA reveals divergent evolutionary trajectories during glioma progression. A) Genome-wide copy number profiles for patient P-68 showing persistence of EGFR ecDNA amplification in both the primary (P-68p) and recurrent (P-68r1) tumors. B) Copy number analysis of patient P-12 across three timepoints (P-12p, P-12r1, P-12r2) reveals loss of EGFR ecDNA in recurrent tumors, despite retention of other clonal CNAs. C) Bulk RNA-seq analysis shows a marked decrease in EGFR expression in recurrent tumors from P-12, while CDK6 (residing on chromosome 7) expression remains relatively stable, indicating selective loss of the EGFR ecDNA clone. D) Hi-C contact matrix from P-12p and P-12r1 shows absence of ecDNA-associated stripe signals in recurrent tumor (upper half matrix) indicated by black arrow. Shared SVs (circles) and chromosome 3 shattering (blue arrows) are indicated for both samples. E) Single-nucleus whole-genome sequencing confirms loss of EGFR ecDNA-amplified clones in P-12r1 and P-12r2, with preservation of other chromosomal aberrations (e.g., chr9 and chr10 deletions). F) Longitudinal MRIs from P-12 reveal multi-centric disease evolution; recurrent lesions (P-12r1 and P-12r2) emerged outside the initial resection zone (P-12p blue arrow), suggesting spatial heterogeneity and ecDNA clone exclusion from the dominant progression path. Notably, our longitudinal ecDNA analysis also revealed unusual patterns in two cases. For patients P-12 and P-29, EGFR ecDNAs were present in the primary tumors, but absent in the subsequent recurrent tumors, based on copy number profiling, even though other clonal variations were retained ( Figure 3b , Supplementary Figure 6a ). Since previous studies 16 reported the emergence of ecDNAs as subclonal events during disease progression or ecDNA persistence in recurrent tumors (as in patients P-27 and P-68), this loss of EGFR ecDNA was unexpected. Consistent with the loss of the ecDNA, we observed a marked decrease in EGFR gene expression in recurrent tumors from RNA-seq data, while expression of other oncogenes such as CDK6 also residing on chromosome 7 remained stable ( Figure 2c , Supplementary Figure 6b ). This suggests that the loss of the ecDNA clone was specific to EGFR . To confirm that the lack of ecDNA in the recurrent sample was not due to insufficient sampling of the recurrent samples, we performed multiple additional genomic assays to analyze the presence of ecDNAs in these samples. Hi-C data from these tumors also supported the loss of the ecDNA in the recurrent samples, where the presence of a genome-wide stripe pattern in the primary tumors, a hallmark of ecDNA 35 – 37 , was absent in recurrent tumors despite sharing other SVs ( Figure 3d , Supplementary Figure 6c ). Finally, we performed single-nucleus whole-genome sequencing 38 on primary and recurrent tumor samples from P-12 and P-29, which confirmed the absence of ecDNA-amplified clones in recurrent tumors ( Figure 3e , Supplementary Figure 7a ,b). However, several other clonal events, such as chromosome 9, 10 deletions in P-12 and chromosomes 6, 10 deletions in P-29 were maintained, indicating that primary and recurrent tumors originated from the same ancestor but diverged following ecDNA loss. Given the unusual pattern of ecDNA subclone loss in the recurrent tumors, we reviewed the clinical information from these patients. Patient P-12’s first and second recurrences likely originated from an area that was enhancing outside of the primary resection area ( Figure 3f , blue arrow). Patient P-29 underwent a partial resection, and recurrences developed in the unresected regions ( Supplementary Figure 7c ). These findings suggest that the original tumor exhibited spatial and temporal heterogeneity and that ecDNA-containing clones were not the dominant clones leading to disease progression in these cases. Glioma genetic drivers are associated with distinct spatial tumor organizations The observed differences in spatial and temporal ecDNA composition during the evolution of the tumors within our samples highlighted the importance of examining spatial organization of gliomas with different oncogenic driver events. To address this, we performed single-cell spatial transcriptomics (10X Xenium) on matched samples for which we also had whole-genome profiling data. We developed a panel of probes targeting 366 unique genes, enabling the identification of distinct malignant cell states and major cell types within the glioma microenvironment (Supplementary Table 5). In total, we generated over 775 million transcripts from 4.66 million cells across 62 different tumors (Supplementary Table 6). First, we evaluated different cell segmentation methods, including a 5-micron nuclear expansion approach, Baysor 39 , and Proseg 40 algorithms. Notably, Proseg produced distinct cell state clusters compared with the other methods, offering unique insights into our dataset ( Supplementary Figure 8a ). Using our probe set, we were able to characterize major glioma cellular states, including oligodendrocyte progenitor-like (OPC-like), astrocyte-like (AC-like), and mesenchymal-like (MES-like) malignant cell populations. Additionally, non-malignant microenvironmental components such as pericytes, myeloid-derived cells, and T cells were mapped across different glioma subtypes by utilizing a consensus non-negative matrix factorization method ( Figure 4a-b ). To quantify the spatial organization of these cellular states in different glioma subtypes, we annotated each cell based on the highest program usage (Methods). This analysis showed that IDH-mutant gliomas are enriched in OPC-like cell states compared with IDH-wildtype gliomas, which contain significantly higher MES-like malignant cells, in agreement with previous publications 21 – 22 . Additionally, we found that myeloid, pericyte and T-cell cell states are significantly higher in IDH-wildtype gliomas compared to IDH-mutant gliomas ( Figure 4c ). T-cell infiltrations in gliomas has been noted to be lower compared to other solid tumor types 41 , therefore previous publications focusing on immune microenvironment of gliomas generally require a sorting step for enriching immune cell types. Our data suggests that single-cell spatial transcriptomics, compared with spot-based methods, offer higher-resolution characterization of the glioma microenvironment ( Figure 4d-e ). Relative enrichment of T-cell infiltration in IDH-wildtype gliomas compared to IDH-mutant gliomas can be partially explained by the excessive D2HG levels in IDH-mutant glioma microenvironment which is reported as toxic for T-cells 42 . We also observed significant differences in the myeloid cell populations between IDH-wildtype and IDH-mutant gliomas. Using a classification scheme based on a recent study 43 , we identified two major distinct microglial subtypes in our cells: inflammatory microglia expressing CX3CR1 and immunosuppressive scavenger cells expressing CD163 ( Supplementary Figure 8b -c). Immunosuppressive microglia were enriched in both IDH-mutant and IDH-mutant codel samples compared to IDH-wildtype gliomas ( Figure 4f ). In contrast, CD163 -positive scavenger cells were more abundant in IDH-wildtype tumors, consistent with previous reports ( Figure 4f ). Interestingly, although IDH-mutant gliomas had a higher proportion of OPC-like malignant cells, microglia cells, particularly the inflammatory microglia, were preferentially enriched within AC-like malignant cell neighborhoods and were depleted in OPC-like neighborhoods within the same tumor specimen, suggesting that the microglial response is shaped by the malignant cell state within the tumor microenvironment ( Figure 4g , Supplementary Figure 8d –h). Lastly, to understand the spatial organization of these tumors, we computed a coherence score (Methods), which showed that IDH-wildtype gliomas, harboring higher MES-like cell states, also exhibit more structured tissue organization compared with IDH-mutant, overlapping with previous findings 22 ( Figure 4h-i ). Download figure Open in new tab Figure 4. Glioma subtypes have distinct spatial tumor organizations. A) Correlation matrix of cell state programs derived from consensus non-negative matrix factorization (cNMF) across spatial transcriptomic profiles highlights distinct malignant and non-malignant states, including OPC-like, AC-like, MES-like, immune, and vascular compartments. B) UMAP embedding of over 4 million spatially resolved cells from 62 glioma tumors colored by annotated cell type identities. C) Comparison of cell type ratios across IDH-mutant-codel, IDH-mutant non-codel, and IDH-wildtype gliomas shows higher OPC-like states in IDH-mutant and increased MES-like, myeloid, pericyte, and T cell populations in IDH-wildtype tumors. P-values were calculated using the two-sided Wilcoxon rank-sum test (* P < 0.05, ** P < 0.01, *** P < 0.001). D) Representative H&E image (left) and Xenium spatial transcriptomic map (right) from an IDH-wildtype (P-65r1) glioma tumor section annotated with major cell types. E) High-resolution visualization of T cells (left) and spatial heatmap of T cell marker gene (CCL5, CD2, CD3G, CYTIP, GZMA, PTPRC, TRAC) expression (right) within an IDH-wildype glioma (P-65r1) section. F) Myeloid cell population is compared between IDH-wildtype and IDH-mutant (codel and non-codel combined) samples. P-values were calculated using the two-sided Wilcoxon rank-sum test (*** P < 0.001). G) Representative high-magnification spatial maps from P-43p (IDH-mutant-codel tumor) highlighting the enrichment of myeloid cells within AC-like malignant cell neighborhoods compared to OPC-like malignant cell neighborhood. H) Spatial coherence scores quantifying tissue organization show significantly higher values in IDH-wildtype gliomas compared to IDH-mutant subtypes. P-values were calculated using the two-sided Wilcoxon rank-sum test (* P < 0.05). J) Spatial cell type annotation maps from representative IDH-mutant (P-02p) and IDH-wildtype (P-44p) tumors, illustrating structural differences in tumor organization. Extrachromosomal EGFR amplifications lead to distinct spatial tumor organization compared to gliomas with linear EGFR amplifications Although ecDNAs are frequently observed in IDH-wildtype gliomas, their contribution to spatial tumor organization remains poorly understood. Leveraging the high prevalence of EGFR amplification in our cohort, we utilized our multi-omic framework to stratify IDH-wildtype tumors by the structural context of their EGFR amplifications. For this analysis, we focused on the primary tumor’s treatment effects as a confounder. As a result, we identified four tumors with chromosome 7 gains but without ecDNA amplifications of any oncogenes, hereafter referred to as linear amplifications, (e.g., P-67) and five tumors with EGFR ecDNAs (e.g., P-46; Figure 5a ). Next, we analyzed the cell composition of the microenvironment in these tumors, which showed that tumors with EGFR amplifications on ecDNA exhibit significantly higher proportions of MES-like and pericyte cells compared with tumors with linear EGFR amplifications ( Figure 5b ). To validate these results, we also analyzed the bulk RNA-seq data available from the same tumors, which showed that ecDNA-containing tumors have significantly higher hypoxic and metabolic activity compared to linear EGFR -amplified tumors ( Supplementary Figure 9a ). This result differs from the current classification of EGFR -amplified gliomas 20 , 24 , which suggests that these tumors are generally enriched in AC-like cells. While the linear EGFR amplification cases do show enrichment of the AC-like cell state, we find that EGFR amplifications driven by ecDNA are associated with a high fraction of MES-like cell states. This highlights that our classification of IDH-wildtype gliomas based on whole-genome sequencing provides a more accurate resolution of EGFR -amplified subtypes. One key reason for this improvement is that single-cell RNA-seq-based copy number inference methods or exome-sequencing techniques often fail to reliably detect focal, ecDNA-like amplifications. We next examined whether the enriched MES-like and pericyte cells show close spatial proximity in ecDNA-containing tumors ( Figure 5c–d , Supplementary Figure 9b ). Indeed, in tumors with EGFR ecDNA amplifications, MES-like malignant cells and pericyte cells were significantly closer in spatial proximity compared to tumors with linear EGFR amplification. The enrichment of pericyte and MES-like malignant cells within shared spatial neighborhoods potentially reflects hypoxia-associated perivascular infiltration. Overall, our results show that ecDNA-containing tumors possess a distinct tumor microenvironment compared to those with linear amplification. Download figure Open in new tab Figure 5. EGFR ecDNA amplifications drives distinct spatial microenvironment and transcriptional clustering in IDH-wildtype gliomas. A) Genome-wide copy number profiles of two representative IDH-wildtype gliomas: P-67p with linear amplification of chromosome 7 (top), and P-46p with focal EGFR amplification on ecDNA (bottom). B) Cell type composition of tumors with linear versus ecDNA-based EGFR amplification shows significant enrichment of MES-like and pericyte cells in ecDNA cases, while AC-like cells predominate in linear cases. P-values were calculated using the two-sided Wilcoxon rank-sum test (* P < 0.05). C) Spatial proximity between MES-like and pericyte cells is significantly closer in EGFR ecDNA-containing tumors than in linear amplification tumors. P-values were calculated using the two-sided Wilcoxon rank-sum test (*** P < 0.001). D) Post-Xenium H&E (top) and spatial transcriptomic cell-type map (bottom) from the same region of P-46p shows MES-like (orange) and pericyte (purple) cells in spatial proximity (highlighted). E) EGFR expression distribution is higher in ecDNA (P-46p) versus linear amplified (P-67p) tumors (left). Subcellular mapping (right) reveals significantly higher EGFR transcript levels (yellow dots) in P-46p. F) EGFR expression stratified by cell state (AC-like vs. MES-like) shows significantly higher levels in ecDNA tumors compared to linear amplified cases. P-values were calculated using the two-sided Wilcoxon rank-sum test (*** P < 0.001). G) Distribution of EGFR transcript cluster sizes across cells reveals larger and more frequent transcriptional clusters in ecDNA tumors (P-46p) than in linear amplified tumors (P-67p). H) Xenium-detected EGFR transcript clusters (yellow) co-localize with high-copy EGFR regions visualized by post-Xenium FISH from the same cells (magenta) in P-46p, confirming spatial overlap and transcriptional hubs. Dashed white lines shows identified EGFR transcript clusters. I) Scatterplots of EGFR and CDK4 expression in tumors P-59p (EGFR and CDK4 on separate ecDNAs) and P-68p (co-amplified on same ecDNA) show stronger expression correlation in P-68p. J) Spatial maps of co-localized EGFR (yellow) and CDK4 (darkpink) clustered transcripts polygon in P-59p (top) and P-68p (bottom) show similar levels of transcript clustering despite architectural differences, suggesting hub formation between distinct ecDNA species. Given the high copy number heterogeneity introduced by ecDNAs, we examined EGFR expression dynamics in tumors with either linear or ecDNA-based amplifications. This analysis revealed that EGFR expression was significantly higher in ecDNA-containing tumors across both AC-like and MES-like malignant cell populations, compared to tumors with linear EGFR amplification ( Figure 5e,f ). Next, we leveraged the subcellular resolution of the Xenium platform to identify EGFR transcriptional clusters within individual cells. We defined clusters as the presence of at least three EGFR transcripts within a 1-micron radius. EGFR ecDNA-containing malignant cells exhibited significantly more transcriptional clusters than cells with linear EGFR amplification ( Figure 5g ). As a control for high transcript counts of EGFR in ecDNA-containing cells, we identified FABP7 gene in our panel which showed similarly high expression levels as EGFR in P-46 ( Supplementary Figure 9c ). Performing the same transcriptional cluster analysis showed that the number of FABP7 transcriptional clusters was lower than that of EGFR in the same tumor ( Supplementary Figure 9d ), suggesting that the observed clustering is not solely due to high expression levels. In addition, the EGFR clusters within ecDNA containing tumors showed substantial nuclear overlap, with a median of 70%, suggesting that transcriptional bursts occur within nuclear regions. To further validate this observation, we performed EGFR FISH on the same post-Xenium slides. FISH confirmed the presence of high EGFR copy number clusters that spatially overlapped with transcriptional clusters identified by Xenium ( Figure 5h ), supporting the hypothesis that nuclear ecDNA hubs drive localized transcriptional activity 37 , 44 . Finally, we examined how differences in ecDNA architecture influence oncogene co-regulation. Our integrative analysis using long-read and Hi-C data revealed the complex ecDNA structures in tumors P-59 and P-68, both of which harbored amplifications of EGFR and CDK4. In P-59, EGFR and CDK4 were located on separate ecDNA circles, whereas in P-68, both oncogenes were co-amplified on the same ecDNA molecule ( Figure 2 ). As expected, EGFR and CDK4 expression showed a stronger correlation in P-68 (r 2 = 0.61) than in P-59 (r 2 = 0.34) ( Figure 5i,j ). MDM2 was amplified only in the ecDNA of P-68, and we observed a strong correlation between EGFR and MDM2 expression in P-68 (r² = 0.81), compared to a much weaker correlation in P-59 (r² = 0.07) ( Supplementary Figure 9e ). We also checked the clustered transcripts of EGFR and CDK4 in both cases. This data showed despite being on two different circles in P-59, clustered transcripts of EGFR and CDK4 genes are overlapping in similar levels observed in P-68 cells ( Figure 5k , Supplementary Figure 9f ), suggesting these distinct ecDNAs can form hubs across tumor microenvironment as reported in cancer cells containing different ecDNA species 44 . These results support the accuracy of our ecDNA reconstruction and highlight the importance of ecDNA architecture in modulating oncogene co-expression within the tumor microenvironment. Discussion Our integrative multi-omic and spatial analyses reveal that oncogenic drivers in gliomas exert profound effects on both tumor cell-intrinsic programs and the composition and organization of the tumor microenvironment. In this work, we particularly focused on the effects of IDH mutations and extrachromosomal DNA amplifications in the glioma microenvironment. IDH mutations lead to excessive accumulation of oncometabolite, D2HG, in the tumor microenvironment. Our analysis showed that IDH -mutant gliomas harbor significantly reduced T-cell infiltration compared with IDH -wildtype gliomas, although the level of T-cell infiltration is low for wildtype gliomas compared with other solid tumors. Recent work 42 showed that D2HG can alter T-cell metabolism and reduced T-cell proliferation. Similarly, D2HG has been reported to affect NF-KB signaling and subsequently elevate CX3CR1 expression in IDH-mutant glioma cells 45 . This may partially explain the increased presence of inflammatory microglia in IDH-mutant glioma microenvironment. Although IDH-mutant tumors were enriched in OPC-like malignant states, inflammatory microglia expressing CX3CR1 were preferentially localized within AC-like regions and were significantly depleted from OPC-like territories within the same tumor sections. This suggests that microglial interactions are not solely dictated by tumor genotype but are also shaped by the local malignant cell state. We demonstrate that the structural context of oncogene amplification, particularly the amplification of EGFR on extrachromosomal DNA, plays a pivotal role in shaping transcriptional dynamics, spatial architecture, and cell–cell interactions within gliomas. ecDNAs have been reported to emerge during disease progression and exhibit spatial heterogeneity in gliomas 46 – 48 . Our longitudinal analyses further show that ecDNA can be selectively lost during disease progression likely due to spatial heterogeneity, where the ecDNA-containing subclone is resected in the primary tumor, while the remaining non-ecDNA clones drive subsequent recurrences. These divergent trajectories highlight the dynamic nature of ecDNA and underscore the importance of sampling spatially distinct tumor regions to capture ecDNA-related phenotypes. We confirm a high prevalence of EGFR -containing ecDNA in these tumors and show that ecDNA amplifications are not only associated with increased copy number heterogeneity but also with spatially organized transcriptional hubs. Using subcellular spatial transcriptomics and combined DNA FISH, we observe that ecDNA-amplified EGFR loci form discrete nuclear transcriptional clusters, likely contributing to enhanced gene expression. Importantly, ecDNA-containing tumors are characterized by a microenvironment enriched in MES-like malignant cells and pericytes, as well as increased hypoxic and metabolic activity—features linked to aggressive tumor behavior. Indeed, a recent study which utilized spatial chromatin accessibility technology in a glioblastoma sample also demonstrated that malignant cells with EGFR ecDNA molecules are enriched in hypoxic neighborhoods 49 . These findings also overlap with other recent work suggestive of hypoxic features in ecDNA containing pancreatic cancer cells 50 , 51 . These observations raise a critical question: does hypoxia-induced genomic instability facilitate ecDNA biogenesis, or do ecDNA-containing cells actively remodel the tumor microenvironment to promote hypoxia? Similarly, ecDNA-containing cells may be better adapted to the harsher conditions of hypoxia due to their fitness advantage over cancer cells lacking ecDNA. Understanding the spatio-temporal dynamics of this bidirectional relationship will be essential for understanding ecDNA dynamics and their role in tumor progression. By integrating short- and long-read sequencing, chromatin conformation capture, and spatial transcriptomic technologies, we provide a comprehensive view of how genomic alterations shape tumor architecture. Our findings highlight that accurate reconstruction 52 of ecDNA molecules in tumor samples generally requires the combined use of multiple technologies, as each platform, and the algorithms built upon them, carry inherent biases. This integrative approach will be particularly important for identifying and stratifying patients for ecDNA-targeted clinical trials 53 . Together, our study advances the understanding of glioma biology by revealing how genetic drivers orchestrate distinct cellular ecosystems. Data and Code Availability The raw bulk data reported in this study will be deposited in European Genome Archive. Any additional information is available from the lead contact upon request. Analysis scripts are deposited in the Akdemir Lab GitHub: https://github.com/akdemirlab/glioma-multiome . Author contributions K.C.A., J.R.D., J.T.H. and, F.F.L. conceived and supervised the study. B.Z. performed the bulk RNA-Seq and spatial transcriptomics analysis and visualization. C.Y.C, T.K, E.R.K, L.Y, and F.C performed the Hi-C data analysis. L.Y. built the somatic mutation analysis pipeline and performed the short-read whole-genome sequencing analysis. T.M. performed the short-read and long-read ecDNA reconstruction analysis. I.M-B. performed the cell line and post-Xenium FISH analysis. K.M.R, C.A.F. conducted clinical annotation. All authors supported clinical contextualization and tumor information integration. J.H.M., D.M.A, K.A.Y, V.P., J.S.W, K.L.L, J.T.H., and F.F.L. provided expert guidance on pathology, and glioma biology. L.W, G.G, P.V.L., and P.A.F. provided critical insights on genomic analysis and interpretation. K.C.A. and J.R.D. wrote the manuscript with input from all authors. Competing Interests The authors declare no competing interests. Supplementary Information Supplementary Tables Supplementary Table 1 Clinical characteristics of 69 glioma patients analyzed in this study. Supplementary Table 2 The list of 93 tumor samples collected from 69 glioma patients, comprising 47 primary tumors and 46 recurrent tumors (29 first recurrence, 14 second recurrence, and 3 third recurrence). Samples were classified by IDH status as IDH-wildtype (n = 47), IDH-mutant (n = 25), or IDH-mutant-codel (n = 21). Different data types generated per sample is denoted. Supplementary Table 3 Summary of somatic mutation and structural variant analyses from short-read whole-genome sequencing of glioma samples. Supplementary Table 4 This table compares ecDNA reconstruction results obtained using short-read whole-genome sequencing (AmpliconArchitect) and long-read sequencing analyzed via the Coral algorithm with two support thresholds (≥5 and ≥10 supporting reads). Supplementary Table 5 The list of Xenium gene probe set used in this study. Supplementary Table 6 Overview of single-cell spatial transcriptomic data generated using the 10X Genomics Xenium platform across the glioma cohort. Download figure Open in new tab Supplementary Figure 1 Overview of the somatic variant and ecDNA analysis pipeline. This schematic outlines the computational workflow used for identifying high-confidence somatic single-nucleotide variants (SNVs), structural variants (SVs), and extrachromosomal DNA (ecDNA) in glioma whole-genome sequencing data. SNVs were detected using both Mutect2 and Strelka2 within the GATK (Sarek) pipeline, with only concordant variants retained. Structural variants were identified using JaBbA, SvABA, Smoove, and Purple, and filtered based on multi-tool support (≥3 tools for translocations; ≥2 tools for other SV types). ecDNA was reconstructed from high-confidence copy number and breakpoint data using AmpliconArchitect and classified with AmpliconClassifier. This integrative workflow ensured robust detection of somatic alterations and ecDNA across glioma samples. Download figure Open in new tab Supplementary Figure 2 Survival of patients according to their IDH-mutation classification. Download figure Open in new tab Supplementary Figure 3 Fluorescence in situ hybridization (FISH) in interphase cells of glioma stem cell lines derived from profiled tumors showed high oncogene levels reminiscent of extrachromosomal amplification. Download figure Open in new tab Supplementary Figure 4 A) Long-read ecDNA reconstructions for tumor P-68 primary (top) and recurrent (bottom) demonstrate high concordance in a complex EGFR/CDK4/MDM2 amplification event on ecDNA. B) Long-read sequencing reconstruction of P-59p reveals two distinct ecDNA circles derived from chromosomes 7 and 12, consistent with copy number differences and structural variation. C) Hi-C interaction maps for the P-59p amplification reconstructed using AmpliconArchitect’s single-circle model showed high variability in off-diagonal signals, suggesting that regions from chromosomes 7 and 12 are unlikely to reside within the same circular structure. Download figure Open in new tab Supplementary Figure 5 A) Genome-wide copy number profiles for patient P-27 showing persistence of EGFR ecDNA amplification in both the primary (P-27p) and recurrent (P-27r1) tumors. B) Short-read ecDNA reconstructions for tumor P-27 primary (top) and recurrent (bottom) demonstrate high concordance in a complex EGFR/CDK4/MDM2 amplification event on ecDNA. C) Clinical timeline and treatment information for patients 27 and 68. Download figure Open in new tab Supplementary Figure 6 A) Copy number analysis of patient P-29 across three timepoints (P-29p, P-29r1, P-29r2) reveals loss of EGFR ecDNA in recurrent tumors, despite retention of other clonal CNAs. Bottom) Long-read sequencing reconstruction of P-29p reveals circular ecDNA structure. B) Bulk RNA-seq analysis shows a marked decrease in EGFR expression in recurrent tumors from P-29. C) Hi-C contact matrix from P-29p and P-29r1 shows absence of ecDNA-associated stripe signals in recurrent tumor (upper half matrix) indicated by black arrow. Download figure Open in new tab Supplementary Figure 7 A) Single-nucleus whole-genome sequencing confirms loss of EGFR ecDNA-amplified clones in P-12r1 and P-12r2, with preservation of other chromosomal aberrations (e.g., chr9 and chr10 deletions). B) Single-nucleus whole-genome sequencing confirms loss of EGFR ecDNA-amplified clones in P-29r1, with preservation of other chromosomal aberrations (e.g., chr6 and chr10 deletions). C) Longitudinal MRIs from P-29 reveal multi-centric disease evolution; recurrent lesions (P-29r1 and P-29r2) emerged from the residual tumor area after P-29p surgery, suggesting spatial heterogeneity and ecDNA clone exclusion from the dominant progression path. Download figure Open in new tab Supplementary Figure 8 A) Comparison of cell segmentation strategies using nuclear expansion, Baysor, and Proseg algorithms. B-C) Classification of microglial subtypes based on dominant marker gene expression. D-E) Neighborhood analysis to investigate microglia-enriched spatial neighborhoods. F-G) Analysis of microglial subtype enrichment within malignant cell neighborhoods. Scavenger immunosuppressive microglia were enriched in MES-like neighborhoods in IDH-wildtype gliomas (F), whereas inflammatory microglia were enriched in AC-like neighborhoods and depleted in OPC-like neighborhoods in IDH-mutant gliomas (G). H) Representative spatial map from P-43p (an IDH-mutant-codel tumor) highlighting the enrichment of myeloid cells within AC-like malignant neighborhoods (R1) compared to OPC-like malignant neighborhoods (R2). Download figure Open in new tab Supplementary Figure 9 A) Bulk RNA-seq data analysis showed that ecDNA-containing tumors have significantly higher hypoxic and metabolic activity compared to linear EGFR-amplified tumors B) Cell type composition of tumors with linear versus ecDNA-based EGFR amplification shows significant enrichment of MES-like and pericyte cells in ecDNA cases, while AC-like cells predominate in linear cases. P-values were calculated using the two-sided Wilcoxon rank-sum test (* P < 0.05). C) FABP7 and EGFR expression levels are comparable in P-46p tumor Xenium data. D) Distribution of EGFR transcript cluster sizes across cells reveals larger and more frequent transcriptional clusters than FABP7 transcription clusters in P-46p tumor Xenium data. E) Scatterplots of EGFR and MDM2 expression in tumors P-59p ( EGFR amplified, MDM2 not amplified) and P-68p ( EGFR and MDM2 co-amplified on same ecDNA) show stronger expression correlation in P-68p. F) Distribution of polygon distances between EGFR and CDK4 clusters in P-59p (amplified within separate ecDNAs) and P-68p (amplified within the same ecDNA). Methods Human Tissue Collection All patients provided written informed consent under protocols approved by the Institutional Review Board of MD Anderson, in accordance with ethical standards for genomic research. The pathological evaluation confirmed that these tissues were malignant glioma specimen. DNA and RNA extraction was performed using standard methods for fresh frozen tissues. Illumina whole genome sequencing (WGS) (tumor: 60x, normal: 30x), bulk RNA-sequencing was generated by UT MD Anderson Moonshot Program Cancer Genomics Laboratory using standard Illumina library preparations and sequenced on Illumina NovaSeq platform. Oxford Nanopore Technology (ONT) long read sequencing was performed on the PromethION platform by MD Anderson Advanced Genome Sequencing Core. Single nuclei DNA sequencing for 2 pairs of primary and brain metastasis samples was performed by the CPRIT single cell genomic center (SCGC) at MD Anderson Cancer Center. 10x Genomics Xenium spatial transcriptomics were performed with our inhouse machine as described below. Short-read WGS Data Processing and Alignment WGS data pre-processing was conducted in accordance with the GATK Best Practices using Sarek (nf-core-sarek-3.1.2 54 ). BWA 55 (v0.7.17) was used to align the data to the genome (GRCh38/hg38) using bwa mem. For single nucleotide variant (SNV) calling, GATK 56 MarkDuplicates was used to identify duplicate reads, followed by Quality Score Recalibration (BQSR) using GATK’s BaseRecalibrator and ApplyBQSR to improve base quality scores. Fingerprint analysis Given the multi-omic nature of our study, we implemented a single-nucleotide polymorphism matching algorithm to ensure accurate correspondence between tumor sections and their respective tumors. We performed DNA fingerprint analysis using Picard (v2.27.4) CrosscheckFingerprints 57 . The hg38 haplotype map, a collection of SNP “blocks,” was obtained from the pre-computed fingerprint map repository ( https://github.com/naumanjaved/fingerprint_maps ). Initially, we conducted cross-checks across all randomly paired WGS samples to identify and correct mismatched samples. Subsequently, we compared fingerprints between WGS, Hi-C and RNA-seq datasets in a pair-wise manner, to confirm that all datasets originated from the same patient. Purity and Ploidy Estimation Purity and ploidy of tumor samples were estimated with ascat 58 (v4.5.0) using matched tumor and normal samples. An SNP GC correction file was downloaded from the ascat GitHub page. For samples with abnormal ASCAT sunset figures (e.g., p18P, total of three tumors), purity and ploidy were estimated using the mean values from Hatchet 59 , FACETS 60 (v0.6.2), and Purple 61 (v3.7.2). Somatic Variant Calling Single nucleotide variants (SNVs) and small insertions/deletions (indels) were identified using GATK Mutect2 56 (GATK v4.3.0.0) and Strelka2 62 (v2.9.10) within the Sarek pipeline. To obtain high-confidence SNVs, results from Mutect2 and Strelka2 were merged using pysam. Only SNVs that passed all quality metrics and showed consistent ALT type between the two tools were retained. Variants were annotated with VEP 63 (v110.0) and SnpEff 64 (v5.1). Structural Variant Calling Structural variants (SVs) were detected using JaBbA (v1.1), SvABA 65 (v1.2.0), Purple (v3.7.2), and Smoove (v0.2.5, https://github.com/brentp/smoove ). In JaBbA, the coverage ratio was calculated using FragCounter ( https://github.com/mskilab/fragCounter ), while ploidy and purity were derived from ascatNgs or mean values from ascatNgs, FACETS (v0.6.2) and Purple (v3.7.2) as mentioned in the Purity and Ploidy Estimation section. SV input of JaBbA was from unfiltered results from SvABA. To get high confidence SVs, SVs from these four tools were merged. SVs with the same type and within a 1kb distance at both ends were considered as the same SV. Translocation SVs were only retained if supported by at least three tools, whereas other SVs required support from at least two. Driver Mutation Identification Putative driver mutations in a selected gene list ’IDH1’, ‘IDH2’, ‘EGFR’, ‘CDKN2A’, ‘ATRX’, ‘TERT’, ‘CIC’, ‘FUBP1 ’ were identified using GATK Mutect2 (GATK v4.3.0.0) or Purple (v3.7.2). Mutations from Purple with a likelihood greater than 0.75 were retained. Mutect2 results were annotated with VEP, with a maximum distance set to 500 bp. ecDNA detection from bulk WGS To detect ecDNA, all samples in the WGS cohort were analyzed using the AmpliconSuite pipeline (v1.3.4), which integrates AmpliconArchitect (v1.3.r6) and AmpliconClassifier (v1.2.1). The pipeline ( AmpliconSuite-pipeline.py ) was launched using BAM files with default settings. DNA copy number alterations were identified using CNVkit 66 (v0.9.12), and the resulting CNV calls were used to generate “seed regions.” Seed regions were defined as regions with a copy number ≥4.5, a length >50 kb, and at least 2 copies above the median chromosome arm ploidy, while masking repetitive and poorly mappable sequences. AmpliconArchitect used these seed regions to construct a local genome breakpoint graph, representing genomic copy number and structural variants (SVs). The ultimate amplicon graph is formed by merging regions from other locations joined by additional SVs to this graph’s regions. Paths and cycles were then extracted from the genome graph and analyzed with AmpliconClassifier to predict the presence of ecDNA. Bulk RNA-seq analysis Total RNA was extracted from approximately 20 mg of flash-frozen brain tumor tissue. mRNA was purified from ∼100 µg of DNase-digested total RNA and used to generate sequencing libraries, which were run on the Illumina NextSeq 6000 platform. Processed reads were aligned to the GRCh38 reference genome using STAR (v2.7.9a). Gene-level read counts were calculated using featureCounts (subread v2.0.6). Gene expression levels were quantified as transcripts per million (TPM) based on the read counts. Gene enrichment analysis Differential gene expression analysis was performed using PyDESeq2 67 (v0.5.0). Genes with |Log2FoldChange| > 1 and padj < 0.05 were considered significantly differentially expressed (DEGs). Enrichment analysis of DEGs was conducted with GSEApy.enrichr 68 (v1.1.7) using the Hallmark pathways database from MSigDB. Additionally, ssGSEA was performed with GSEApy.ssgsea, and the NES scores for the top Hallmark pathways in each sample were obtained. Hi-C Data Processing and Alignment Hi-C reads were aligned to the human reference genome (GRCh38/hg38) using BWA (v0.7.17) with bwa mem with the -SP5M flag. To generate Hi-C .cool files, the BAM files were first converted to a binned contact matrx files using Pairtools. These matrix files were then converted to.mcool files using Cooler 69 . Resolutions of 5k, 10k, 50k, and 100k were retained. Two different normalizations were applied: the first using the ICE (Iterative Correction and Eigenvector) algorithm with cooler balance, and the second using CNV balance by NeoLoopFinder’s correct-cnv. Hi-C Structural Variant Calling and Neoloop Identification SVs were identified from the Hi-C ICE-balanced .cool files using Eagle-C 70 by analyzing changes in contact patterns between genomic regions. The parameters for Eagle-C were set to --prob-cutoff-5k 0.8, --prob-cutoff-10k 0.8, and --prob-cutoff-50k 0.99999. Neoloops were identified using neoloop-caller with --prob set to 0.99 and --no-clustering. Single-nucleus whole-genome sequencing Single nucleus whole genome sequencing (snWGS) was generated from frozen tissue samples with the ARC-well protocol. Data was aligned to hg38 with BWA mem 2. Samtools sorting, picard duplicate marking, and samtools indexing were performed. ASCAT.sc (v0.1) was used to call copy number alterations with the multiPCF setting for consistent segmentation between cells. Ploidies were fit accordingly to their DAPI FACS peak. Peaks that resembled diploid or G2/M were fit to a ploidy range of 1.5-2.5. Peaks that resembled a whole genome duplication were fit to a ploidy range of 3.5-6. Peaks that weren’t clearly distinguishable between G2/M and whole genome duplications were fitted to a broader range of 1.5-6 to account for the uncertainty. The focal amplification region was specified with the svinput argument, which contained the largest interval from AmpliconArchitect cycles file. A modified version of the ASCAT.sc quality control function was used to select high quality cells for further analysis with the following settings: at least 2^16 reads per cell, the upper 90% of cell qualities, and within 0.03 distance of a line fitting the cells between number of reads and logR noise. Copy number segments were adjusted to be roughly proportional to their genomic length for plotting, which was performed with ComplexHeatmap 71 . In order simplify the main figure for visualization, further filtering was performed to remove cells that were nearly diploid (healthy cells) or highly aneuploid (like whole genome duplicated cells). To do this, cells with less than 25% or more than 75% of their genomic segments equal to 2 were filtered out. All cells that passed quality control are visualized in the supplementary figures. Single-nucleus Multiome Single-nucleus Multiome (ATAC + Expression) analysis was performed on tumor samples. Nuclei were isolated from approximately 30 mg of freshly frozen tissue using the 10x Genomics Nuclei Preparation Kit, with slight modifications. Briefly, tissue samples were thawed in 200 µL of the nuclei lysis buffer, cut into small pieces, and homogenized with a pestle until no visible tissue fragments remained. The samples were then incubated on ice for 5–10 minutes with an additional 300 µL of nuclei lysis buffer. After centrifugation, nuclei were washed twice with wash buffer, resuspended in diluted nuclei buffer, and counted. Library preparation followed the 10x Genomics user guide (CG000338), and sequencing was performed on the Illumina NovaSeq 6000 platform. Reads were aligned to the human genome (GRCh38) using cellranger-arc (v2.0.0, 10x Genomics) to generate UMI counts and ATAC-seq fragments. snRNA cells were filtered as: 30000 ≥ UMI counts ≥ 500, 6500 ≥detected gens ≥ 200, 10 > Mitochondrial count percentage, doublet removal (predicted by Scrublet). Genes detected in less than 3 nuclei were also removed. snATAC cells were filtered as: 100000 ≥ counts ≥ 1000, TSSe ≥ 5 and doublet removal (predicted by Scrublet). Features in the blacklist (hg38-blacklist.v2) were excluded. Finally, only cells present in both the snRNA and snATAC datasets were retained for downstream analysis. snRNA data were analyzed using Scanpy 72 (v1.10.2), while snATAC data were processed with SnapATAC2 73 (v2.7.0). For snRNA data, UMI counts were normalized to a target sum of 10,000 and log transformed. The 2,000 most variable genes were selected for principal component analysis (PCA). Data integration was performed using harmonypy (v0.0.10), followed by clustering with the Leiden algorithm. Marker genes for each cluster were identified using scanpy.tl.rank_genes_groups with the default Wilcoxon method. Cells were manually annotated based on marker gene expression, and these annotations were also mapped to the snATAC dataset. For snATAC data, the cell × feature count matrix was reduced in dimensionality using snapatac2.tl.spectral , integrated with snapatac2.pp.harmony , and clustered using snapatac2.tl.leiden . Copy number variations (CNVs) were inferred from snRNA data using infercnvpy (v0.5.0) and from snATAC data using epiAneufinder 74 (v1.0.1). Oxford NanoPore Long-read Sequencing Oxford Nanopore Technology (ONT) sequencing was performed with the R10.4 flowcell. Basecalling and alignment to the hg38 genome was performed using Dorado (v0.9). The aligned bam was sorted, MD tags were added, and indexed with Samtools. Bams were merged into a single bam file for each sample. Nanoplot 75 was used to obtain sequencing quality information. Coral was used to reconstruct ecDNA amplifications with default settings with two adjusted parameters: coral seed --gain 10, coral reconstruct --min-bp-support 5, 10. Both 5 and 10 were used for --min-bp-support because they resulted in some samples with distinct AmpliconClassifier results. Coral reconstruct --min-bp-support 10 results were used further analysis. AmpliconClassifier were then used to classify focal amplifications. Spatial transcriptomics with 10x Xenium Analyzer Fresh Frozen (FF) and Formalin-Fixed Paraffin-Embedded (FFPE) tissues were used in this study. FF tissues embedded in OCT were sectioned to a thickness of 10 µm onto Xenium slides and processed according to the 10x Genomics user guide (CG000581) for fixation and permeabilization. FFPE blocks were sectioned to a thickness of 5 µm, placed on Xenium slides, and processed following the user guide (CG000580) for deparaffinization and decrosslinking. Xenium in situ gene expression analysis was performed according to the appropriate 10x Genomics user guides, either without cell segmentation staining (CG000582) or with cell segmentation staining (CG000749). A custom gene panel of 100 genes (Supplemental Table 5) was added to the pre-designed human brain gene panel. A mix of pre-designed and custom gene expression probe hybridization was conducted at 50 °C overnight. Unhybridized probes were removed through multiple washing steps, after which the probes were ligated and Rolling circle amplification (RCA) primers were annealed at 37 °C for 2 hours to generate circularized padlock probes, which were subsequently amplified at 30 °C for 2 hours. If cell segmentation staining was used, it was performed at 4 °C overnight according to the CG000749 user guide. Lastly, background fluorescence was quenched, nuclei were stained, and the prepared slides were loaded onto the Xenium Analyzer. The Xenium Analyzer performed multiple rounds of fluorescence probe hybridization, imaging, and probe removal automatically. Z-stacked fluorescence images were acquired and analyzed using Xenium Onboard Analysis (v3.0.0), which included decoding, cell segmentation, analysis, and quality control (QC). After the Xenium run, slides were stored hydrated in PBS-T (0.05% Tween-20 in PBS) at 4 °C. For H&E staining, slides were processed according to the user guide (CG000613), mounted with coverslips, dried, and scanned with a slide scanner (PE Vectra Polaris) at 20x magnification. DNA-FISH on post-Xenium slides Post-Xenium FFPE slides were treated with Quencher Removal solution as suggested by 10x Genomics prior to performing Fluorescence in situ hybridization (FISH). Slides were processed according to the Agilent FISH Protocol on FFPE Samples using the DAKO Histology FISH kit. A set of commercially available probes from MetaSystems and Leica Biosystems were used to target and hybridize with specific genes to confirm abnormalities observed previously by Xenium and molecular analysis. Centromere probes, labeled in green, were used as controls, while locus-specific probes targeting genes were labeled in red. Cell and signal imaging were conducted using an ECHO REVOLVE R4 fluorescent microscope. Xenium data processing We performed Xenium spatial transcriptomics on 62 samples using either probes combined with the human brain base probe panel. Xenium Analyzer output data (transcript locations, gene assignments, cell x gene tables, cell/nucleus boundaries, and stacked nuclei images) were processed using the SpatialData framework 76 (v0.2.2). Individual samples within each Xenium dataset were separated using spatialdata.polygon_query and stored in Zarr format. For datasets using P1 probes without cell segmentation staining, the cell segmentation was initially performed by expanding nuclear boundaries by 5 μm in XOA. To optimize cell segmentation, we compared two re-segmentation methods: Baysor (v0.7.0) and Proseg (v2.0.2). Proseg produced superior clustering results, so Proseg generated cell x gene tables were used for downstream analysis. Cells with ≤20 or ≥800 transcripts, or ≤4 detected genes were filtered out. Leiden clustering was then applied to each sample using Scanpy (v1.10.2) after normalizing to a target sum of 10,000 and log transformation. Clusters with median transcript counts ≤50 were excluded. The cell x gene tables from all samples were merged into an AnnData object (v0.10.8), normalized (target sum = 10,000) and log transformed. Gene Expression Program (GEP) analysis was conducted using cNMF 77 (v1.5.4) with Harmony batch correction. cNMF was run for 500 iterations with k values from 5 to 15, with k = 11 selected as the optimal GEP clustering number. The top 15 marker genes from each GEP were selected and duplicates removed, resulting in 155 unique marker genes. A GEP was assigned to a cell if it had minimum 10% usage and higher usage than any other program. These GEP assignments were used for cell type annotation. For FFPE tissue p46P, which used P1 probe with cell segmentation staining, GEP usage derived from cNMF analysis was transferred using starCAT 78 (v1.0.9). The GEP with the highest program usage determined the cell type annotation. Main marker gene’s used for cell annotations include: AC-like (EGFR, FABP7, MT3, GFAP, AQP4, TTYH1); OPC-like (BCAN, OLIG1, OLIG2, PDGFRA, SOX8, SOX4, SOX11); MES-like (IGFBP5, HILPDA, VEGFA, NDRG1); Oligodendrocyte (MAG, CNDP1, SOX10, MOG, CNTN2, ERMN, MOBP, MAL, CLDN1); Neuron (SNAP25, ANO3, SLC17A7, C1QL3, GAD1, GAD2, ANK1); Myeloid (TMEM119, P2RY12, CX3CR1, CD163, MSR1, MRC1, C1QA, C1QB, IL1B); Cycling (CCNB2, CDK1, CENPF, MKI67, TOP2A); Endothelial (PECAM1, FLT1, PHLDB2, CLDN5); Pericyte (COL12A1, COL3A1, PDGFRB, DCN, FBLN1); T cell (CD3D, CD8A, CD8B, FOXP3, CCL5, CD2, CD3G, CXCR4, CYTIP, CD48, GZMA, TRAC ). Neighborhood analysis Spatial niche analysis was conducted using CellCharter 79 (v0.3.1) and Banksy 80 algorithms. To identify the optimal niche number for each sample, CellCharter encoded cells into a network where nodes represented cells and edges represented cell connectivity. Neighborhoods for each cell were aggregated into a single feature to the layer of 4 using the gr.aggregate_neighbors function. A Gaussian Mixture Model was created and fitted to predict niche labels for all cells, with the optimal niche number determined by tl.ClusterAutoK. Banksy was then run with the identified niche number and a lambda value of 0.8 to define spatial niches. To identify consistent niches across all sample cohorts, we computed cell type composition ratio within each niche, performed Pearson correlation analysis between niches, and applied hierarchical clustering. This analysis revealed eight distinct meta-neighborhoods characterized by high intra-group correlation. We assessed spatial proximity between cell types on the connectivity graph using CellCharter’s gr.nhood_enrichment function. The analysis focused on malignant (AC-like, OPC-like, and MES-like) cells and two subtypes of myeloid cells (inflammatory microglia and immunosuppressive scavenger). The normalized enrichment score for each cell type pair was calculated within individual samples, and the scores were averaged across samples for final evaluation. Spatial coherence analysis The spatial coherence score, a measure of the structural organization or disorganization within a tumor, was defined previously 22 . To calculate the score, the tissue sample with cell annotations was evenly divided into 200 x 200 bins, and the number of cells from each cell type was determined for each bin. The dominant cell type within each bin, defined as the one with the highest cell count, was assigned as the program for that bin. The coherence score was calculated by examining each bin and its surrounding 3 x 3 square neighbors. If the neighboring bins shared the same program as the central bin, the bin was assigned a score of 1; otherwise, it was assigned a score of 0. This process was repeated for all bins and programs. The total scores for each program were summed and divided by the total number of bins associated with that program. The resulting ratios were scaled between 0 and 1, and the average of these scaled ratios was used as the overall coherence score for the sample. Neighboring cells analysis In each sample, the 8 nearest neighboring cells of each pericyte cell were identified using KDTree (scipy, v1.14.1). Neighboring cells located more than 200 µm away were excluded. The remaining neighboring cells were aggregated to calculate the ratio of different cell types. The nearest AC-like or MES-like cell to each pericyte was identified within a 200 µm radius, and their Euclidean distance was measured. Gene transcripts clustering analysis The X and Y coordinates (2D) were used to analyze gene transcript clustering within each cell. For each transcript, all other transcripts of the same gene within 1 µm were identified and counted. A transcript was classified as part of a cluster if it had at least 3 neighboring transcripts. The frequency of clusters of all sizes was then recorded. A polygon representing each cluster was generated using Shapely (v2.0.4), and the distance between EGFR and CDK4 polygons, or between EGFR and nucleus polygons, was calculated for each cell. Patient-derived glioma stem cell lines Following thawing at 37°C, the cells were resuspended and cultured in DMEM/F12 media supplemented with B-27, hEGF, hbFGF, and antibiotics (McDonald, M. F. et al., 2024). The cultures were maintained in an incubator at 37°C with 5% CO2. Once the cells reached approximately 80% confluency, they were passaged into new flasks. For harvesting, the cells were treated with colcemid and exposed to a hypotonic KCl solution. Carnoy’s fixative solution (3:1 methanol:acetic acid) was added, and the cell pellets were stored at 4°C until slide preparation. Fluorescence in situ Hybridization (FISH) Slides were labeled with appropriate identification details, and two drops of the cell suspension were applied. The slides were then dried on a slide warmer before being aged overnight at 60°C. Following this, the slides were processed for FISH. A set of commercially available probes including EGFR (7q11.2)/SE7 (D7Z1-centromeric region), RUO- MDM2 (12q15)/SE12(D12Z3), RUO-MYCN/SE2 from Leica Biosystems were used to study the mutations in each cell line. Centromere probes, labeled in green, were used as controls, while locus-specific probes were labeled in red, except for PAX6, which was also labeled in green. The signal analysis was performed using an ECHO REVOLVE R4 fluorescent microscope and 100 interphase cells were analyzed for each FISH probe. Acknowledgements We thank the patients and their families for their selfless contribution to this study. We especially dedicate this work to M. Sachs, whose generosity empowered and further strengthened our commitment to this study. We thank Drs. Sangeeta Goshwami and Preeti Ramadoss for her critical reading of this manuscript. We would also like to thank MD Anderson CATALYST Program and the Brain Tumor Center members, including Lisa Norberg, Chetna Wathoo, Truc Kuo, Stephanie Jenkins, Brittany Parker Kerrigan, Douglas Nielsen, and Jennifer Ritchie, for their essential contributions to patient sample collection and preparation. This work was supported by Brain Cancer SPORE (P50CA127001 to K.C.A., J.T.H., F.F.L), Kleberg Innovative Investigator Award (K.C.A), The Broach Foundation for Brain Cancer Research (to F.F.L), The Elias Family Fund for Brain Tumor Research (to F.F.L), The Priscilla Hiley Cancer Research Fund (to F.F.L), The Sweet Family Brain Cancer Research Fund (to F.F.L), The Jim & Pam Harris Fund, the TLC Foundation From the Heart (to F.F.L), and The Mary Harris Pappas Endowed Fund for Glioblastoma Research (to F.F.L), the Pew Biomedical Scholars Program (to J.R.D.), the Rita Allen Foundation Scholars Program (to J.R.D.), and the 4D Nucleome consortium (U01CA260700 to J.R.D.). This work was also supported by the generous philanthropic contributions to The University of Texas MD Anderson Cancer Center Glioblastoma Moon Shots Program™. We also acknowledge this work utilized services from CPRIT Single-cell CORE Facilities Grant (RP180684) and ATGC core grant (CA016672). Footnotes ↵ * Co-first authors ↵ # Co-senior authors: jhuse{at}mdanderson.org , jedixon{at}salk.edu , flang{at}mdanderson.org , kcakedemir{at}mdanderson.org References 1. ↵ Ostrom , Q. T. et al. The epidemiology of glioma in adults: a ‘state of the science’ review . Neuro. Oncol . 16 , 896 – 913 ( 2014 ). OpenUrl CrossRef PubMed 2. ↵ Louis , D. N. , et al. The 2021 WHO classification of tumors of the Central Nervous System: A summary . Neuro. Oncol . 23 , 1231 – 1251 ( 2021 ). OpenUrl CrossRef PubMed 3. ↵ Weller , M. , et al. Glioma . Nat. Rev. Dis. Primers 10 , 33 ( 2024 ). OpenUrl CrossRef PubMed 4. ↵ Dang , L. et al. Cancer-associated IDH1 mutations produce 2-hydroxyglutarate . Nature 465 , 966 ( 2010 ). 5. ↵ Noushmehr , H. et al. Identification of a CpG island methylator phenotype that defines a distinct subgroup of glioma . Cancer Cell 17 , 510 – 522 ( 2010 ). OpenUrl CrossRef PubMed Web of Science 6. Malta , T. M. et al. The epigenetic evolution of glioma is determined by the IDH1 mutation status and treatment regimen . Cancer Res . 84 , 741 – 756 ( 2024 ). OpenUrl CrossRef PubMed 7. Flavahan , W. A. et al. Insulator dysfunction and oncogene activation in IDH mutant gliomas . Nature 529 , 110 – 114 ( 2016 ). OpenUrl CrossRef PubMed 8. Rahme , G. J. et al. Modeling epigenetic lesions that cause gliomas . Cell 186 , 3674 – 3685 .e14 ( 2023 ). OpenUrl CrossRef PubMed 9. Turcan , S. et al. IDH1 mutation is sufficient to establish the glioma hypermethylator phenotype . Nature 483 , 479 – 483 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 10. ↵ Lu , C. et al. IDH mutation impairs histone demethylation and results in a block to cell differentiation . Nature 483 , 474 – 478 ( 2012 ). OpenUrl CrossRef PubMed Web of Science 11. ↵ Xie , T. et al. Pervasive structural heterogeneity rewires glioblastoma chromosomes to sustain patient-specific transcriptional programs . Nat. Commun . 15 , 3905 ( 2024 ). OpenUrl CrossRef PubMed 12. ↵ Turner , K. M. et al. Extrachromosomal oncogene amplification drives tumour evolution and genetic heterogeneity . Nature 543 , 122 – 125 ( 2017 ). OpenUrl CrossRef PubMed 13. ↵ deCarvalho, A. C., et al. Discordant inheritance of chromosomal and extrachromosomal DNA elements contributes to dynamic disease evolution in glioblastoma . Nat. Genet . 50 , 708 – 717 ( 2018 ). OpenUrl CrossRef PubMed 14. ↵ Kim , H. et al. Extrachromosomal DNA is associated with oncogene amplification and poor outcome across multiple cancers . Nat. Genet . 52 , 891 – 897 ( 2020 ). OpenUrl CrossRef PubMed 15. ↵ Bailey , C. et al. Origins and impact of extrachromosomal DNA . Nature 635 , 193 – 200 ( 2024 ). OpenUrl CrossRef PubMed 16. ↵ Kim , H. et al. Mapping extrachromosomal DNA amplifications during cancer progression . Nat. Genet . 56 , 2447 – 2454 ( 2024 ). OpenUrl CrossRef PubMed 17. ↵ 17. Cancer Genome Atlas Research Network . Comprehensive genomic characterization defines human glioblastoma genes and core pathways . Nature 455 , 1061 – 1068 ( 2008 ). OpenUrl CrossRef PubMed Web of Science 18. 18. Cancer Genome Atlas Research Network, et al. Comprehensive, integrative genomic analysis of diffuse lower-grade gliomas . N. Engl. J. Med . 372 , 2481 – 2498 ( 2015 ). OpenUrl CrossRef PubMed 19. Barthel , F. P. et al. Longitudinal molecular trajectories of diffuse glioma in adults . Nature 576 , 112 – 120 ( 2019 ). OpenUrl CrossRef PubMed 20. ↵ Neftel , C. et al. An Integrative Model of Cellular States, Plasticity, and Genetics for Glioblastoma . Cell 178 , 835 – 849 .e21 ( 2019 ). OpenUrl CrossRef PubMed 21. ↵ Ravi , V. M. et al. Spatially resolved multi-omics deciphers bidirectional tumor-host interdependence in glioblastoma . Cancer Cell 40 , 639 – 655 .e13 ( 2022 ) OpenUrl CrossRef PubMed 22. ↵ Greenwald , A. C. et al. Integrative spatial analysis reveals a multi-layered organization of glioblastoma . Cell 187 , 2485 – 2501 .e26 ( 2024 ). OpenUrl CrossRef PubMed 23. Spitzer , A. et al. Deciphering the longitudinal trajectories of glioblastoma ecosystems by integrative single-cell genomics . Nat. Genet . 1 – 11 ( 2025 ). 24. ↵ Nomura , M. et al. The multilayered transcriptional architecture of glioblastoma ecosystems . Nat. Genet . 1 – 13 ( 2025 ). 25. ↵ Brennan , C. W. et al. The somatic genomic landscape of glioblastoma . Cell 155 , 462 – 477 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 26. ↵ Touat , M. et al. Mechanisms and therapeutic implications of hypermutation in gliomas . Nature 580 , 517 – 523 ( 2020 ). OpenUrl CrossRef PubMed 27. ↵ Deshpande , V. et al. Exploring the landscape of focal amplifications in cancer using AmpliconArchitect . Nat. Commun . 10 , 392 ( 2019 ). 28. ↵ Luebeck , J. et al. Extrachromosomal DNA in the cancerous transformation of Barrett’s oesophagus . Nature 616 , 798 – 805 ( 2023 ). OpenUrl CrossRef PubMed 29. ↵ Hadi , K. et al. Distinct Classes of Complex Structural Variation Uncovered across Thousands of Cancer Genome Graphs . Cell 183 , 197 – 210 .e32 ( 2020 ). OpenUrl CrossRef PubMed 30. ↵ Giurgiu , M. et al. Reconstructing extrachromosomal DNA structural heterogeneity from long-read sequencing data using Decoil . Genome Res . 34 , 1355 – 1364 ( 2024 ). OpenUrl Abstract / FREE Full Text 31. ↵ Zhu , K. et al. CoRAL accurately resolves extrachromosomal DNA genome structures with long-read sequencing . Genome Res . 34 , 1344 – 1354 ( 2024 ). OpenUrl Abstract / FREE Full Text 32. ↵ 32. Chowdhury , B. , et al. Reconstructing the three-dimensional architecture of extrachro-mosomal DNA with ec3D . bioRxiv ( 2025 ) doi: 10.1101/2025.02.01.636064 . OpenUrl Abstract / FREE Full Text 33. ↵ Helmsauer , K. et al. Enhancer hijacking determines extrachromosomal circular MYCN amplicon architecture in neuroblastoma . Nat. Commun . 11 , 5823 ( 2020 ). OpenUrl CrossRef PubMed 34. ↵ Wang , X. et al. Genome-wide detection of enhancer-hijacking events from chromatin interaction data in rearranged genomes . Nat. Methods 18 , 661 – 668 ( 2021 ). OpenUrl CrossRef PubMed 35. ↵ Zhu , Y. et al. Oncogenic extrachromosomal DNA functions as mobile enhancers to globally amplify chromosomal transcription . Cancer Cell 39 , 694 – 707 .e7 ( 2021 ). OpenUrl CrossRef PubMed 36. Chang , L. et al. Droplet Hi-C enables scalable, single-cell profiling of chromatin architecture in heterogeneous tissues . Nat. Biotechnol . 1 – 14 ( 2024 ). 37. ↵ Hung , K. L. et al. ecDNA hubs drive cooperative intermolecular oncogene expression . Nature 600 , 731 – 736 ( 2021 ). OpenUrl CrossRef PubMed 38. ↵ Wang , K. et al. Archival single-cell genomics reveals persistent subclones during DCIS progression . Cell 186 , 3968 – 3982 .e15 ( 2023 ). OpenUrl CrossRef PubMed 39. ↵ Petukhov , V. et al. Cell segmentation in imaging-based spatial transcriptomics . Nat. Biotechnol . 40 , 345 – 354 ( 2022 ). OpenUrl CrossRef PubMed 40. ↵ Jones , D. C. et al. Cell Simulation as Cell Segmentation . bioRxiv ( 2024 ) doi: 10.1101/2024.04.25.591218 . OpenUrl Abstract / FREE Full Text 41. ↵ Yeo , A. T. et al. Single-cell RNA sequencing reveals evolution of immune landscape during glioblastoma progression . Nat. Immunol . 23 , 971 – 984 ( 2022 ). OpenUrl CrossRef PubMed 42. ↵ Notarangelo , G. et al. Oncometabolite d-2HG alters T cell metabolism to impair CD8+ T cell function . Science 377 , 1519 – 1529 ( 2022 ). OpenUrl CrossRef PubMed 43. ↵ Miller , T. E. et al. Programs, origins and immunomodulatory functions of myeloid cells in glioma . Nature 640 , 1072 – 1082 ( 2025 ). OpenUrl CrossRef PubMed 44. ↵ Hung , K. L. et al. Coordinated inheritance of extrachromosomal DNAs in cancer cells . Nature 635 , 201 – 209 ( 2024 ). OpenUrl CrossRef PubMed 45. ↵ Han , C.-J. et al. The oncometabolite 2-hydroxyglutarate inhibits microglial activation via the AMPK/mTOR/NF-κB pathway . Acta Pharmacol. Sin . 40 , 1292 – 1302 ( 2019 ). OpenUrl CrossRef PubMed 46. ↵ Favero , F. et al. Glioblastoma adaptation traced through decline of an IDH1 clonal driver and macro-evolution of a double-minute chromosome . Ann. Oncol . 26 , 880 – 887 ( 2015 ). OpenUrl CrossRef PubMed 47. Wang , Q. et al. Spatial 3D genome organization reveals intratumor heterogeneity in primary glioblastoma samples . Sci. Adv . 11 , eadn2830 ( 2025 ). 48. ↵ Mathur , R. et al. Glioblastoma evolution and heterogeneity from a 3D whole-tumor perspective . Cell 187 , 446 – 463 .e16 ( 2024 ). OpenUrl CrossRef PubMed 49. ↵ Huang , Y.-H. et al. Unified molecular approach for spatial epigenome, transcriptome, and cell lineages . Proc. Natl. Acad. Sci. U. S. A . 122 , e2424070122 ( 2025 ). OpenUrl CrossRef PubMed 50. ↵ Fiorini , E. et al. MYC ecDNA promotes intratumour heterogeneity and plasticity in PDAC . Nature 1 – 10 ( 2025 ). 51. ↵ Perelli , L. et al. Evolutionary fingerprints of epithelial-to-mesenchymal transition . Nature 1 – 10 ( 2025 ). 52. ↵ Bafna , V. & Mischel , P. S . Extrachromosomal DNA in cancer . Annu. Rev. Genomics Hum. Genet . 23 , 29 – 52 ( 2022 ). OpenUrl CrossRef PubMed 53. ↵ Noorani , I. , Mischel , P. S. & Swanton , C . Leveraging extrachromosomal DNA to fine-tune trials of targeted therapy for glioblastoma: opportunities and challenges . Nat. Rev. Clin. Oncol . 19 , 733 – 743 ( 2022 ). OpenUrl CrossRef PubMed 54. ↵ Garcia , M. et al. Sarek: A portable workflow for whole-genome sequencing analysis of germline and somatic variants . F1000Res . 9 , 63 ( 2020 ). OpenUrl 55. ↵ Li , H. & Durbin , R . Fast and accurate short read alignment with Burrows-Wheeler transform . Bioinformatics 25 , 1754 – 1760 ( 2009 ). OpenUrl CrossRef PubMed Web of Science 56. ↵ McKenna , A. et al. The Genome Analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data . Genome Res . 20 , 1297 – 1303 ( 2010 ). OpenUrl Abstract / FREE Full Text 57. ↵ Javed , N. et al. Detecting sample swaps in diverse NGS data types using linkage disequilibrium . Nat. Commun . 11 , 3697 ( 2020 ). OpenUrl CrossRef PubMed 58. ↵ Raine , K. M. et al. ascatNgs: Identifying Somatically Acquired Copy-Number Alterations from Whole-Genome Sequencing Data . Curr. Protoc. Bioinformatics 56 , 15.9.1 – 15.9.17 ( 2016 ). OpenUrl CrossRef PubMed 59. ↵ Zaccaria , S. & Raphael , B. J . Accurate quantification of copy-number aberrations and whole-genome duplications in multi-sample tumor sequencing data . Nat. Commun . 11 , 4301 ( 2020 ). OpenUrl CrossRef PubMed 60. ↵ Shen , R. & Seshan , V. E . FACETS: allele-specific copy number and clonal heterogeneity analysis tool for high-throughput DNA sequencing . Nucleic Acids Res . 44 , e131 ( 2016 ). OpenUrl CrossRef PubMed 61. ↵ Shale , C. et al. Unscrambling cancer genomes via integrated analysis of structural variation and copy number . Cell Genom 2 , 100112 ( 2022 ). 62. ↵ Kim , S. et al. Strelka2: fast and accurate calling of germline and somatic variants . Nat. Methods 15 , 591 – 594 ( 2018 ). OpenUrl CrossRef PubMed 63. ↵ McLaren , W. et al. The Ensembl Variant Effect Predictor . Genome Biol . 17 , 122 ( 2016 ). 64. ↵ Cingolani , P. et al. A program for annotating and predicting the effects of single nucleotide polymorphisms, SnpEff: SNPs in the genome of Drosophila melanogaster strain w1118; iso-2; iso-3 . Fly (Austin) 6 , 80 – 92 ( 2012 ). OpenUrl PubMed 65. ↵ Wala , J. A. et al. SvABA: genome-wide detection of structural variants and indels by local assembly . Genome Res . 28 , 581 – 591 ( 2018 ). OpenUrl Abstract / FREE Full Text 66. ↵ Talevich , E. , Shain , A. H. , Botton , T. & Bastian , B. C . CNVkit: Genome-Wide Copy Number Detection and Visualization from Targeted DNA Sequencing . PLoS Comput. Biol . 12 , e1004873 ( 2016 ). OpenUrl CrossRef PubMed 67. ↵ Love , M. I. , Huber , W. & Anders , S . Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2 . Genome Biol . 15 , 550 ( 2014 ). 68. ↵ Fang , Z. , Liu , X. & Peltz , G . GSEApy: a comprehensive package for performing gene set enrichment analysis in Python . Bioinformatics 39 , ( 2023 ). 69. ↵ Abdennur , N. & Mirny , L. A . Cooler: scalable storage for Hi-C data and other genomically labeled arrays . Bioinformatics 36 , 311 – 316 ( 2020 ). OpenUrl CrossRef PubMed 70. ↵ Wang , X. , Luan , Y. & Yue , F . EagleC: A deep-learning framework for detecting a full range of structural variations from bulk and single-cell contact maps . Sci. Adv . 8 , eabn9215 ( 2022 ). 71. ↵ Gu , Z. , Eils , R. & Schlesner , M . Complex heatmaps reveal patterns and correlations in multidimensional genomic data . Bioinformatics 32 , 2847 – 2849 ( 2016 ). OpenUrl CrossRef PubMed 72. ↵ Wolf , F. A. , Angerer , P. & Theis , F. J . SCANPY: large-scale single-cell gene expression data analysis . Genome Biol . 19 , 15 ( 2018 ). 73. ↵ Zhang , K. , Zemke , N. R. , Armand , E. J. & Ren , B . A fast, scalable and versatile tool for analysis of single-cell omics data . Nat. Methods 21 , 217 – 227 ( 2024 ). OpenUrl CrossRef PubMed 74. ↵ Ramakrishnan , A. et al. epiAneufinder identifies copy number alterations from single-cell ATAC-seq data . Nat. Commun . 14 , 5846 ( 2023 ). OpenUrl CrossRef PubMed 75. ↵ De Coster , W. & Rademakers , R . NanoPack2: population-scale evaluation of long-read sequencing data . Bioinformatics 39 , ( 2023 ). 76. ↵ Marconato , L. et al. SpatialData: an open and universal data framework for spatial omics . Nat. Methods 22 , 58 – 62 ( 2025 ). OpenUrl CrossRef PubMed 77. ↵ Kotliar , D. et al. Identifying gene expression programs of cell-type identity and cellular activity with single-cell RNA-Seq . Elife 8 , ( 2019 ). 78. ↵ Kotliar , D. , et al. Reproducible single cell annotation of programs underlying T-cell subsets, activation states, and functions . bioRxivorg ( 2024 ) doi: 10.1101/2024.05.03.592310 . OpenUrl Abstract / FREE Full Text 79. ↵ Varrone , M. , Tavernari , D. , Santamaria-Martínez , A. , Walsh , L. A. & Ciriello , G . CellCharter reveals spatial cell niches associated with tissue remodeling and cell plasticity . Nat. Genet . 56 , 74 – 84 ( 2024 ). OpenUrl CrossRef PubMed 80. ↵ Singhal , V. et al. BANKSY unifies cell typing and tissue domain segmentation for scalable spatial omics data analysis . Nat Genet 56 , 431 – 441 ( 2024 ). OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted May 21, 2025. Download PDF Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Oncogenic drivers shape the tumor microenvironment in human gliomas Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Oncogenic drivers shape the tumor microenvironment in human gliomas Bo Zhao , Chae Yun Cho , Lingqun Ye , Tula Keal , Thomas Mitchell , Ines Martin-Barrio , Christina Abi Faraj , Mesut Unal , Jessica Alsing , Nora M. Lawson , Ellie Rahm Kim , Francisco Chavez , Kristy Mendoza Rangel , Sanjay K. Singh , Joseph H. McCarthy , David M. Ashley , W. K. Alfred Yung , Vinay Puduvalli , Guy Nir , Jeffrey S. Weinberg , Linghua Wang , Keith L. Ligon , Giannicola Genovese , Peter Van Loo , P. Andrew Futreal , Jason T. Huse , Jesse R. Dixon , Frederick F. Lang , Kadir C. Akdemir bioRxiv 2025.05.16.654515; doi: https://doi.org/10.1101/2025.05.16.654515 Share This Article: Copy Citation Tools Oncogenic drivers shape the tumor microenvironment in human gliomas Bo Zhao , Chae Yun Cho , Lingqun Ye , Tula Keal , Thomas Mitchell , Ines Martin-Barrio , Christina Abi Faraj , Mesut Unal , Jessica Alsing , Nora M. Lawson , Ellie Rahm Kim , Francisco Chavez , Kristy Mendoza Rangel , Sanjay K. Singh , Joseph H. McCarthy , David M. Ashley , W. K. Alfred Yung , Vinay Puduvalli , Guy Nir , Jeffrey S. Weinberg , Linghua Wang , Keith L. Ligon , Giannicola Genovese , Peter Van Loo , P. Andrew Futreal , Jason T. Huse , Jesse R. Dixon , Frederick F. Lang , Kadir C. Akdemir bioRxiv 2025.05.16.654515; doi: https://doi.org/10.1101/2025.05.16.654515 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Genomics Subject Areas All Articles Animal Behavior and Cognition (7622) Biochemistry (17645) Bioengineering (13867) Bioinformatics (41873) Biophysics (21420) Cancer Biology (18550) Cell Biology (25447) Clinical Trials (138) Developmental Biology (13361) Ecology (19866) Epidemiology (2067) Evolutionary Biology (24289) Genetics (15587) Genomics (22473) Immunology (17707) Microbiology (40322) Molecular Biology (17144) Neuroscience (88457) Paleontology (666) Pathology (2826) Pharmacology and Toxicology (4815) Physiology (7634) Plant Biology (15111) Scientific Communication and Education (2042) Synthetic Biology (4285) Systems Biology (9813) Zoology (2268)