Tumor cell phenotype and heterogeneity differences in IDH1 mutant vs wild-type gliomas

Glioma is recognized to be a highly heterogeneous CNS malignancy, whose diverse cellular composition and cellular interactions have not been well characterized. To gain new clinical- and biological-insights into the genetically-bifurcated IDH1 mutant (mt) vs wildtype (wt) forms of glioma, we integrated multiplexed immunofluorescence single cell data for 43 protein markers across cancer hallmarks, in addition to cell spatial metrics, genomic sequencing and magnetic resonance imaging (MRI) quantitative features. Molecular and spatial heterogeneity scores for angiogenesis and cell invasion differ between IDHmt and wt gliomas irrespective of prior treatment and tumor grade; these differences also persisted in the MR imaging features of peritumoral edema and contrast enhancement volumes. Longer overall survival for IDH1mt glioma patients may reflect generalized altered cellular, molecular, spatial heterogeneity which manifest in discernable radiological manifestations.

induced angiogenesis and identified that the relative cerebral blood volume (rCBV) MRI was able to 117 predict IDH1 mutations status with an 88% accuracy (32). Unlike the latter studies that predict IDH1 118 mutational status, we seek to reveal correlations between MRI derived quantitative features, cellular 119 composition and spatial cellular heterogeneity to understand the mechanism of disease progression in 120 relation to IDH1 mutational status. Such knowledge could enable creation of predictive models on MRI 121 of disease progression or treatment response without the need for an invasive biopsy.
122 We show lower cell-level protein expression in IDH1mt vs wt cases. Further, IDH1mt gliomas, 123 irrespective of grade, showed greater spatial heterogeneity but lower molecular heterogeneity of 124 biomarkers associated with angiogenesis (VEGR2, CD31, SMA, S100A4) and invasion (n-cadherin, cofilin, 7 135 Cohorts of 20 treatment-naïve gliomas (grades 2, 3, and 4 from the Ohio Brain Tumor Study) and 16 136 post-treatment recurrent glioblastoma (grade 4 from University of California San Francisco(33)) were 137 retrieved based on appropriate patient consent, suitable MR images, FFPE tissue availability, and 138 specimens suitable for next-generation sequencing ( Table 1 for patient summary and S-Tables S-1 and   139 S-2 for additional details).
140 processing, registration and single cell analysis, was used (21) (S1 A-C Fig.). After a two-step antigen 197 retrieval step, the sample underwent repeated cycles of staining, imaging and dye signal quenching (S1-198 A Fig.) for a total of 43 biomarkers (S3 Table) Fig.). Three of these, VEGFR2, Vimentin and HLA1 were also included in the multi-variate model 407 using Random Forest which provided an AUC of 0.87 (error rate 5%) in predicting IDH mutation status 408 (S6-B Fig.).  Clusters 1 and 4 with above average expression of most hallmarks were composed of cells 421 from just two IDH1wt patients (Fig. 2). Clusters 2 and 6 contained the largest numbers of cells (21.0% 422 and 21.9%, respectively, S7 Fig.) from the greatest number of cases (12 and 11 cases, respectively,  Fig.). Overall, similar staining 442 profiles and biomarker patterns in IDHmt vs wt cases were found in the recurrent GBM cohort. 443

Cell cluster alignment with IDH and other glioma related mutations
474 Fig. 4 shows cluster distribution aligned with IDH mutation status and the other most common 475 mutations in treatment-naïve glioma. In concordance with known biology, IDH1 mutations were found 476 to be mutually exclusive of EGFR and PTEN mutations (Fig. 4, panel B). IDH1mt samples appeared to be more homogenous, particularly those with concurrent ATRX mutation, and were mostly dominated by 478 the cluster 6 cell phenotype (lower than average expression of most markers (panel A). Approx. 50% of 479 IDH1wt cases with EGFR amplification had a high proportion of cluster 2 cells (overall, average 480 biomarker expression, and lower DNA damage and stem cell markers, higher iron metabolism markers). Cell cluster alignment with RNA expression and IDH status 486 The degree to which single cell clusters agreed with deconvoluted, transcript-based cell class 488 expression data of all measured genes, we identified three cell classes using CellDistinguisher, each class 489 having 50 or more distinguisher genes (S9 Fig.). Exceeding three classes resulted in a very short list of 490 distinguisher genes for some classes, which diminishes the utility of comparing behavior or functions 491 across the classes. Classes 2 & 3 were qualitatively similar to protein derived cell clusters 6 and 2 492 respectively. Ratios of the average staining intensities for 21 markers in clusters 6 and 2 were calculated 493 (Fig. 5A). The ratios of the expression values for the same 21 genes were compared between RNA 494 classes 2 and 3 (Fig. 5B). Fractional composition of IDHmt and wt cases within cell cluster 2 or 6 ( Fig. 5C) 495 or within RNA class 2 or 3 (Fig. 5D) was determined. Consistent with earlier results, tumors dominated 496 by cluster 2 cells were more likely to be IDHwt, while cases with dominance of cluster 6 were mostly 497 IDH1mt. Similarly, the IDHwt tumors were mainly comprised of RNA class 3 markers while class 2 was 498 more abundant in the IDH1mt (Fig. 5D). were enriched in genes related to cancer hallmarks of "inducing angiogenesis", 516 "enabling replicative immortality" and "evading growth suppression" (see S4 Fig. and S2 517 Table).

519
We have found noteworthy similarity between the cell types and patient compositions identified from 520 the MxIF biomarker intensities and the gene expression data. Except for FASN, GSK3b and NCad, good 521 directional correlation was observed in differential protein and gene expression between cell clusters 522 and RNA classes in the IDH1mt and IDHwt populations (Fig. 5). Lack of concordance between H2AX 523 protein and transcript likely is due to staining intensity by anti-H2AX antibody reporting only the post-  Fig. 6A, which shows the discretized (high 538 (2), medium (1), low (0)) expression values for each marker, and corresponding color-coding for each 539 cell. Heterogeneity calculated from the distribution of these states in different tumors shows an inverse 540 correlation between molecular and spatial heterogeneity in both treatment-naïve glioma and recurrent 541 GBM cohorts. IDHwt tumors had higher molecular heterogeneity while IDHmt tumors were more 542 spatially heterogenous (S10 Fig.). Similar trends were present in both cohorts. Fig. 6B shows a scatter 543 plot of heterogeneity in the inducing angiogenesis hallmark with the range of spatial and molecular 544 heterogeneity metrics for gliomas and recurrent GBM samples, also encoded by IDHmt (red) and wt 545 (blue) status. Trends in heterogeneity of this hallmark were similar to those observed for the 546 proliferation hallmarks as well as activating invasion motility hallmark (S10 Fig.). No other significant 547 differences in heterogeneity were found.

MR feature differences between IDH1 mutant and wildtype patients
568 Simple features derived from the MR images uncovered differences in discernable elements of brain 569 tumor dispersion from IDH1wt and IDH1mt patients. IDH1wt patients had larger enhancing cores 570 (feature "Normalized enhancing core volume"), but less contrast uptake in the peri-tumoral edema 571 regions (feature "Edema T1 post"). On the other hand, the IDH1mt patients lack a clearly defined 572 enhancing core, but have increased contrast uptake on the T1 post contrast MRI protocol in the peri-573 tumoral edema region (Fig. 7). These trends were observed both in the treatment-naïve glioma as well 574 as the recurrent GBM, and are not surprising since the IDH1mt are known to have less contrast 575 enhancement than the IDH1wt (49). i.e. S100A4 that is known to promote angiogenesis and metastasis development (50) , and VGFR2 that 602 plays a fundamental role in neovascularization (51). These found associations were consistent When investigating multimodal associations (Fig. 8), we can also observe a consistent trend across the 605 two cohorts of patients. Not surprisingly, IDH1 mutations are found in lower grade tumors, younger 606 patients and have better overall survival. As also shown in Fig. 7 and S11 Fig., IDH1mt tumors have 607 smaller enhancing cores but more contrast uptake in the edema regions and show reduced expression 608 levels of RNA and protein from the Inducing Angiogenesis hallmark (Fig. 8, highlighted box). Of the five 609 angiogenesis hallmark cell clusters, cluster 4 (above average expression of VEGFR2, SMA and CD31) and 610 cluster 5 (above average expression of VEGFR2 and S100A4), which are characterized by higher well-defined enhancing rim and with higher uptake in the edema region, on account of infiltrating cells.

662
Previous studies have linked poor survival with the peritumoral edema volume (52) and tumor volume Fig.). This is consistent with increased sequestration of iron, making it unavailable for oxidative DNA 676 damage leading to evasion of ferroptosis. A more in-depth analysis of this pathway that includes iron 677 transport, storage and utilization is necessary to determine if evasion of ferroptosis is indeed driving the 678 tumor growth in these patients (53,54). Inter-and intra-tumoral molecular heterogeneity is a well-679 recognized feature of GBM (6,55,56) and is believed to be the main reason behind treatment failure.

680
Emergence of several single cell analysis platforms has fueled the investigations of intra-tumoral

693
The key limitations of this study include small sample size, lack of registration of sample derived for 694 molecular analysis to MR images and a limited number of markers representing different hallmarks. The 695 intent of this study was not to generate a diagnostic signature but to evaluate correlation between 696 imaging and molecular features at the hallmark level and to generate a work flow for integrating 698 treatment-naïve glioma cohort was limited, the fact that similar cell clusters existed in another cohort 699 (recurrent GBM) and the correlations between MR and molecular features of angiogenesis hallmark 700 hold for both cohorts is encouraging. Having developed methods to integrate and evaluate such a 701 complex data set, we are in the process of designing a more focused study to interrogate the biology of 702 a specific molecular subtype of GBM that will consider and address the aforementioned shortcomings.