Asthma affects over 260 million people worldwide, yet diagnosis remains dependent on spirometry and specialist assessment, limiting accessibility in primary care and low-resource settings. Vocal biomarkers offer a promising non-invasive alternative, but prior studies have largely focused on acoustic features without integrating clinical context. We present a multimodal Mixture-of-Experts framework for asthma detection that adaptively combines acoustic embeddings from sustained vowel phonation and reading passage tasks with structured clinical and demographic data. The model was evaluated on a matched cohort of 1,218 asthma cases and healthy controls from the Colive Voice study. The multimodal model achieved an AUROC of 0.85 and Brier score of 0.17, outperforming unimodal and bimodal approaches. Adaptive gating analysis revealed increased reliance on audio features in participants with greater respiratory symptom burden, whereas clinical features contributed more strongly in less symptomatic individuals. These findings support scalable and explainable asthma screening using smartphone-collected voice recordings.

Background Human brain organoids (BOs) are important models for studying early brain development and neurological disorders. While techniques for creating BOs are advancing, they remain developmental structures. Therefore, when human BOs are used to studying glioma-host interactions, the tumor behavior may be influenced by the BO-developmental microenvironment. Here, we describe the maturation of rat brain organoids (rBOs) into fully differentiated BOs and demonstrate their value as a model for studying glioblastoma (GB)-host interactions and their use in testing therapeutic interventions.Materials and Methods rBOs were obtained from fetal cortical brains on the 18th day of gestation. Transcriptomic, proteomic, and metabolomic analyses determined their differentiation into maturity. Their developmental trajectory was compared to human BOs derived from induced pluripotent stem cells as well as to rat brain development. Tumor-rBO interactions, including invasion parameters and therapeutic interactions, were studied using 5 human GB models.Results The rBOs develop into organized structures with myelinated neurons, oligodendrocytes, synapses, and glial cells, mirroring the rat brain development. GB invasion in rBOs matched those observed in orthotopic xenografts, enabling real-time assessment of invasion metrics: cellular heterogeneity, single-cell invasion speed, and tumor progression. The BOs had a strong impact on GB transcriptional activity and can be used to study therapeutic interventions. The rBO differentiation status influenced GB invasion capacity.Conclusions The rBOs serve as an effective target brain structure for studying GB invasion parameters and for evaluating therapeutic interventions. Their rapid development into mature brain tissue makes rBOs a valuable brain avatar system for studying tumor-host interactions.

Glioblastoma (GBM) is an aggressive and lethal brain tumor marked by profound local and systemic immune dysfunction. Despite evidence of peripheral immune impairment, the clinical relevance of these alterations for diagnostic or therapeutic purposes remains poorly defined. We performed multimodal single-cell profiling of peripheral blood mononuclear cells from a single-center cohort of treatment-naïve GBM patients and healthy donors, integrating mass and flow cytometry with single-cell RNA-sequencing. Unsupervised clustering, pseudo-temporal trajectory analyses and cell–cell communication inference were applied to map immune states and their interactions. GBM blood profiles were characterized by heterogeneous changes in classical monocytes, encompassing expanded, reduced and unchanged subsets with distinct functional states, including antigen-presenting, interferon and metabolic programs. Additional myeloid adaptations included myeloid-derived suppressor cell (MDSC) expansion and loss of non-classical monocytes. Trajectory analyses identified a differentiation continuum, evolving from antigen-presenting to metabolic monocyte subsets, and positioning MDSCs as an intermediate state. Antigen‑presenting monocytes displayed tumor‑migratory, precursor‑like profiles that corresponded to tumor‑associated macrophages in public GBM datasets. Across subsets, circulating monocytes shared a “GBM-classical monocytic signature” characterized by low MHC class II expression, altered cell–cell communication and upregulation of anti-inflammatory mediators, including IL1R2 and CD163. Notably, complementary myeloid expression signatures were identified across patients, indicating distinct systemic immune phenotypes. In parallel, lymphocyte alterations included decreased proportions of CD4+ T, natural killer (NK) and CD56+ T cells, retaining relatively conserved activation profiles, exemplified by up-regulation of alarmins S100A8/S100A9. These findings delineate systemic immune reprogramming in primary GBM, characterized by coordinated myeloid and lymphocyte alterations. The identification of circulating monocyte states with transcriptional continuity to the tumour microenvironment, alongside distinct patient-level systemic myeloid signatures, provides a framework for exploring peripheral blood as a source of immune biomarkers in GBM.

Drug resistance of metastatic colorectal cancer (mCRC) remains a major therapeutic challenge. Screening patient-derived tumor cells with diverse compounds in 3D models may overcome the limitations of genomics-based drug response predictions. We describe a personalized functional profiling (PFP) approach in mCRC using patient-derived spheroids (PDS) and assess its utility in predicting drug responses. PDS were established from twelve patients’ tumors and validated by immunohisto(cyto)chemistry and genomic sequencing. Forty-two small molecule anti-cancer drugs, along with five standard-of-care (SOC) drugs in CRC were screened as single agents or in combination, and cell viability was measured using calcein staining or ATP-based assay. Ex vivo results were compared with clinical treatment responses. PDS closely mirrored histopathological and genetic features of the original tumors, supporting their use in PFP. Sensitivity to anti-EGFR drugs distinguished responsive from resistant patients and revealed candidates for anti-ERBB2 therapy, whereas anti-VEGFR screening failed to recapitulate clinical outcomes. SOC drug screening results correlated with clinical outcomes or tumor genetic features in a subset of PDS. This work underscores the predictive value of PFP, its complementarity with genomic sequencing, and the need for refinement to enhance its clinical applicability.

Neuroendocrine neoplasms (NEN) are thought to originate from diffuse neuroendocrine networks and therefore most frequently arise in the gastrointestinal tract and lungs. The liver is a frequent site of metastasis of NEN but also the existence of primary hepatic NEN has been proposed. Due to the impact on disease management, it is urgently required to discriminate the origin of hepatic NEN metastases and to identify clinically relevant subgroups. Using a comprehensive set of NEN (N = 212) from two independent cohorts, we show that the DNA methylation profiles of NEN of distinct anatomical localizations differ significantly and primary tumor-metastasis pairs cluster together, enabling the identification of the tumor origin. Furthermore, the subgroup of hepatic NEN without clinically detectable primary tumor, thus classified as primary hepatic NEN, does not form a distinct cluster by DNA methylation analysis but colocalizes with various subgroups of extrahepatic NEN. Organ-specific subtyping of NEN delineates a foregut-like epigenetic profile for hepatic NEN with unknown primary. We propose a classifier with high prediction accuracy for each of the different organ sites. In conclusion, our results demonstrate that DNA methylation profiling enables precise prediction of NEN origin and suggests that a substantial proportion of presumed primary hepatic NEN may in fact represent misclassified secondary hepatic NEN of unknown primary.

Expression of sEV-related genes correlates with disease progression and poor survival in CLL patients. Gene-expression analysis was performed by RT-qPCR for 7 genes involved in sEV biogenesis and secretion in a cohort of 144 CLL patients. The correlation between gene expression and survival was evaluated by Cox univariate regression analysis. Gene expression in clinical groups was evaluated by differential expression analysis for single genes or by logistic regression (LR) analysis for multiple genes. A, Calculated hazard ratios >1 (red dots, P < 0.05) indicate an increased risk for patients with high single-gene expression in terms of OS. B, Correlation between high or low gene expression and OS. Low and high groups are of identical size (n = 72) Median OS is indicated in months (mo). C, Correlation between high or low combined 7-gene expression and OS. D, Calculated hazard ratios >1 (red dots, P < 0.05) indicate an increased risk for patients with high multiple gene expression in terms of OS. E, Calculated hazard ratios >1 (red dots, P < 0.05) indicate an increased risk for patients with high single-gene expression in terms of TFS. F, Correlation between high or low gene expression and TFS. G and H, Standardized expression of single genes (G) or LR scores for multiple genes (H) in groups of patients according to prognostic markers (CytoG, cytogenetics, group size indicated in each panel). *, P < 0.05; **, P < 0.01. Data are mean with 95% confidence intervals.

LME-sEVs decrease CD8+ T-cell functions. A and B, Percentages (A) and numbers (B) of CD62L−KLRG1+ CD8+ T cells after 48 hours of treatment with LME- and HCME-sEVs assessed by FC. C, Expression of ICP on CD62L−KLRG1+ CD8+ T cells from B. HSNE clustering depicting treatments, cluster identity, and marker expression. D, Hierarchical clustering based on ICP expression. E, Percentages of PD1+TIM3+ICOS+ CD8+ T cells from cluster C1 (top) and of PD1+LAG3+TIM3+TIGIT+ICOS+ CD8+ T cells from cluster C8 (bottom). F–I, CD8+ T cells were isolated from C57BL/6 and CLL cells from TCL1 mice. F, Percentage of T cell–mediated killing of TCL1 cells (cytotoxic assay) in the presence of HCME-sEVs or LME-sEVs (N = 6). G, Quantification of CD8+ T-cell:TCL1 cell conjugates upon treatment with LME- or HCME-sEVs (N = 3–4) and representative images (scale bar, 10 μm). H, Quantification of immune synapse formation (F-actin area in μm2, HCME-, n = 31 and LME-sEVs, n = 39, dashed line representing median) and representative medial optical sections (scale bar, 5 μm) with arrows indicating the synapse. I, Mean Fluorescence intensity (MFI) of GzmB at the synapse between CD8+ T and CLL cells (HCME-, n = 31 and LME-sEVs, n = 51) and representative 3D volume-rendered images. J, miRNA levels quantified by RT-qPCR in CD8+ T cells treated with HCME- or LME-sEVs for 24 hours. K, Protein levels of miRNA targets determined by FC in CD8+ T cells treated for 48 hours with HCME-sEVs or LME-sEVs transfected with scramble or antagomiRs (miR-150, -155, and -378a). Preincubation of LME-sEVs with heparin was used as an inhibitor of sEV internalization. L, ICP levels determined by FC in CD8+ T cells treated for 48 hours with HCME-sEVs or LME-sEVs preincubated with blocking Abs (PD-L1, GAL9, VISTA, and MHC-II) or corresponding isotypes. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001 (unpaired Student t test). Data are mean.

LME-sEVs enter different lymphocyte subsets and modify CD8+ T cells in the microenvironment. A, Percentage of splenocytes internalizing MB488+ sEVs. Splenocytes from C57BL/6 mice were incubated for increasing periods of time with MB488+ LME-sEVs and analyzed by FC. sEV preincubation with heparin sulfate (Hep) was performed for 4 hours. B, Representative confocal microscopy pictures of total splenocytes after 4 hours of treatment with LME-sEVs, in the absence or presence of heparin (scale bar, 5 μm). C, Splenocytes from C57BL/6 mice were incubated for 24 hours with MB488+ LME-sEVs and then analyzed by FC with lymphocyte-lineage markers. D, FACS-sorted CD4+ and CD8+ T cells were incubated for increasing periods of time with MB488+ LME-sEVs and analyzed by FC. E, Representative confocal microscopy pictures of FACS-sorted CD4+ Tconv cells, CD8+ T cells, and Tregs after treatment with LME-sEVs (24 hours; scale bar, 5 μm). F–H, MB570+-LME-sEVs were i.v. injected in C57BL/6 mice. Total splenocytes were harvested 24 hours later and analyzed by FC directly (F) or after staining for specific immune subsets (CD19+ B cells, CD4+ and CD8+ T cells, G–H). I, Volcano plot showing differential expression of genes (DEG) with FDR <0.05 and log2FC >1 in CD8+ T cells isolated from spleens of mice treated with LME- or HCME-sEVs for 1 week. J, Hierarchical clustering of DEG from I. K, t-distributed stochastic neighbor embedding (t-SNE) of samples from I. L, Hierarchical clustering of selected genes from J, grouped by enriched gene ontologies. M and N, mRNA (M) or protein levels (N) of 3 selected DEG from I, quantified by RT-qPCR or FC, in CD8+ T cells treated in vitro for 48 hours with HCME- or LME-sEVs. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001 (unpaired Student t test). Data are mean.

LME-sEVs impact CD8+ T-cell transcriptome, proteome, and metabolome. A, Volcano plot showing DEG identified by RNA-seq from CD8+ T cells treated for 48 hours with LME- (n = 4) or HCME-sEVs (n = 3) with FDR <0.05 and log2FC >1. B and C, Hierarchical clustering of all DEG (B) and of selected genes from relevant cell functions (C). D, Volcano plot showing differentially expressed proteins (DEP) identified by mass spectrometry from CD8+ T cells treated for 96 hours with LME- (n = 3) and HCME-sEVs (n = 3) with FDR <0.05 and log2FC >1. E and F, GzmB mRNA expression and GzmB and perforin levels in CD8+ T cells treated for 48 hours with HCME- or LME-sEVs. G and H, Ontology analysis of enriched (G) or diminished (H) DEP in CD8+ T cells treated with LME- or HCME-sEVs (from D). I, Levels of glucose measured by mass spectrometry in culture medium in CD8+ T-cell treated with LME- or HCME-sEVs for 96 hours. Negative value represents consumption. J, Immunoblot analysis of glycolysis-related proteins from CD8+ T cells treated for 96 hours with LME- or HCME-sEVs. K, Levels of ADP and ATP generated from 13C-glucose measured by mass spectrometry in CD8+ T cells treated with LME- (n = 5) or HCME-sEVs (n = 6) for 96 hours. L, Oxygen consumption measured by SeaHorse assay from CD8+ T cells treated with LME- or HCME-sEVs for 96h. *, P < 0.05; **, P < 0.01; ****, P < 0.0001 (unpaired Student t test). Data are mean and SEM.

LME-sEVs present a specific proteome and miRNA fingerprint. A, Hierarchical clustering of sEVs differentially expressed proteins (DEP with q < 0.05) identified by mass spectrometry between HCME-sEVs (n = 3, isolated from independent pools of 5 C57BL/6 spleens) and LME-sEVs (n = 14). B, Volcano plot showing DEP between LME- and HCME-sEVs with FDR <0.05 and log2FC >1. C, PCA based on DEP. D and E, Ontology analysis of DEP between LME-sEVs and HCME-sEVs. F, Expression of ICP ligands on HCME- or LME-sEVs quantified by bead-based FC. G, Representative pictures of ICP ligand expression on single MB488+CD20+ LME-sEVs visualized by imaging FC. H and I, HSNE clustering of MB488+ LME-sEVs based on CD20, PD-L1, GAL9, and MHC-II expression analyzed by FC and related combinations of markers on CD20+ LME-sEVs. J, miRNA levels measured using RT-qPCR from HCME- (isolated from a pool of 5 C57BL/6 mice spleens) and LME-sEVs (n = 8). Data are mean.

Supplemental Table S3 - Patient clinical information

sEV are enriched in the human and murine leukemic microenvironments. A, Relative mRNA expression of selected genes involved in sEV biogenesis and secretion in B cells from PB of healthy donors (HC, n = 9) and CLL patients (CLL, n = 15; from GSE67640). B, Score based on sEV-related mRNA levels from A. C, mRNA levels of selected genes extracted from A. D, mRNA expression of selected genes according to IGHV mutational status. E, Score combining the expression of Rab10, Rab35, and Rab40C according to IGHV mutational status. F, Relative mRNA expression of selected genes involved in sEV biogenesis and secretion in B cells from C57BL/6 (WT) and Eμ-TCL1 mice (TCL1; from GSE175564). G, Score based on sEV-related RNA levels from F. H,Rab3b mRNA level extracted from F. I, Detailed protocol to isolate and purify sEVs from the murine spleen. J, Amount of proteins (in mg) recovered from LME- (n = 18) or HCME-sEVs (n = 10), normalized per gram of spleen. K, Representative TRPS analysis of ME-sEVs for size and concentration. L, Electron microscopy images of ME-sEVs. M, Western blot analysis of ME-sEVs. N, HSNE clustering analysis of MB488+ LME-sEVs based on CD63, CD81, and CD9 expression measured by bead-free FC (left) and relative percentages of combined expression (right). *, P < 0.05; **, P < 0.01; ****, P < 0.0001 (unpaired Student t test). Data are mean.

Supplemental Table S1 - Proteins detected in sEV preparations or cells treated with sEV

Supplemental Table S2 - Differential expression analysis of genes in leukemic or immune cells treated with sEV

sEVs are crucial for CLL development by impairing the antitumor immune response in vivo.A, Generation of a new TCL1-RAB27DKO mouse model. B, Detection of the human TCL1 transgene and Rab27b excision in gDNA of C57BL/6, TCL1, and TCL1-RAB27DKO mice. C, Immunoblot analysis of RAB27A and RAB27B proteins in the same mice. D, Percentage of CD5+CD19+ CLL cells in the PB of TCL1 (n = 35) or TCL1-RAB27DKO (n = 12) mice over time. E, Survival of mice from D and RAB27DKO mice (n = 10). F, Quantity of proteins recovered from LME-sEVs (n = 18) or LME-sEVsTCL1-RAB27DKO (n = 11) normalized per gram of spleen. G, PCA based on differentially expressed proteins (DEP) between LME-sEVsTCL1-RAB27DKO and LME-sEVs with FDR <0.05 and log2FC >1. H, Volcano plot showing DEP. I, Injection scheme of CLL cells competent (TCL1, red arrows) or deficient in sEV release (TCL1-RAB27DKO, green arrows) into C57BL/6 mice, with or without LME-sEVs (violet arrows). J, Percentage of CD5+CD19+ CLL cells in the blood of C57BL/6 mice injected according to I (n = 16 per condition). Four different clones for each genotype were injected into 4 mice each. K, Injection scheme of CLL cells deficient in sEV release (TCL1-RAB27DKO, green arrows) into C57BL/6 mice, treated with α-CD8 blocking or isotype-control Abs (violet arrows). L, Percentage of CD5+CD19+ CLL cells in the PB of mice injected according to K (n = 6 per group) at days 14 and 21 (left) and in the spleen of the same mice at day 21. M, Injection scheme of CLL cells deficient in sEV release (TCL1-RAB27DKO, green arrows) into C57BL/6 mice, together with α-CD8 blocking Ab (violet arrows) and followed by injection of activated CD8+ T cells treated ex vivo with HCME- (blue arrows) or LME-sEVs (red arrows). N, Percentage of CD5+CD19+ CLL cells at day 10 in the PB of mice injected according to panel M (n = 4 per group). O, Survival of mice from M (n = 4 per group). *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001 (unpaired Student t test for F, L, and N two-way ANOVA followed by the Bonferroni multiple comparison test for D and J, log-rank test for E and O). Data are mean with SEM.

Background Precision medicine has transformed cancer treatment by tailoring therapies to specific molecular aberrations. Integrating high-resolution multi-omics with high-throughput functional profiling in patient-derived organoids of-fers a powerful strategy to further refine patient stratification. While (epi)genetic profiling has drastically improved the classification in diffuse adult gliomas, these advances have not yet translated into effective therapeutic interventions and precision medicine approaches remain to be established. Material and Methods We investigated a panel of 48 patient-derived organoid and orthotopic xenograft models of adult high-grade gliomas, comprehensively characterized at genomic, epigenomic and transcriptomic levels. A functional drug screen was performed on 27 organoid models using a 202-compound library targeting cancer-related pathways and epigenetic regulators. Unsupervised multi-omics factor analysis was employed to identify patient-specific therapeutic vulnerabilities. Validation included dose-dependent drug efficacy assessments, as well as biomarker assessment in patient tumors across molecular subgroups. Results Multi-omics analysis revealed a broad spectrum of molecular profiles capturing the genetic, epigenetic, and transcriptomic diversity of high-grade gliomas. Multi-omics factor analysis, integrating multi-omics and drug response profiles, identified distinct subgroups associated with IDH1 mutation and MYCN amplification. IDH1 mutant grade 4 astrocytomas showed selective sensitivity to histone deacetylase 3 inhibitors, while a MYCN -amplified glioblastoma responded preferentially to histone methyltransferase inhibitors. The differential drug responses were linked to specific (epi)genetic and transcriptomic biomarkers. While other glioblastomas exhibited heterogeneous treatment responses, no robust biomarker-defined responder subgroups were identified. Conclusion Our findings highlight the value of integrating multi-omics and functional profiling to inform precision medicine strategies. This approach enables the stratification of distinct patient subgroups in preclinical models, paving the way for tailored therapeutic interventions. While we observed distinct pharmacogenomic profiles in IDH1 mutant grade 4 astrocytomas and a MYCN -amplified glioblastoma, implementing precision medicine in other glioblastoma subtypes remains a substantial challenge. Key points Study importance To date, attempts to develop effective precision medicine in adult high-grade gliomas failed. Here, we provide a preclinical framework for identifying personalized therapeutic by integrating multi-omics profiling with functional drug screening in patient-derived organoids. We show that IDH1 mutant high-grade astrocytomas present distinct therapeutic vulnerabilities compared to glioblastomas, linked to sensitivity to histone deacetylase 3 inhibitors. Within glioblastomas, we identified a distinct MYCN -amplified tumor, sensitive to histone methyltransferase inhibitors. Applying pharmacogenomic approaches using novel drug libraries holds promise for uncovering additional clinically relevant patient subgroups in the future. ![Figure][1] ### Competing Interest Statement The authors have declared no competing interest. Fonds National de la Recherche, C20/BM/14646004/GLASSLUX, C21/BM/15739125/DIOMEDES, INTER/GACR/23/18089030 MITOFIT, FNR PEARL P16/BM/11192868 FNRS-Télévie, TETHER 7.4632.17, 7.4615.18 [1]: pending:yes

Supplemental Table S4 - Flow cytometry antibody list

Motivation:Deciphering molecular signals from omics data helps understanding cellular processes and disease progression. Effective algorithms for extracting these signals are essential, with a strong emphasis on robustness and reproducibility. Results:R/Bioconductor package consICA implements consensus independent component analysis (ICA)-a data-driven deconvolution method to decompose heterogeneous omics data and extract features suitable for patient stratification and multimodal data integration. The method separates biologically relevant molecular signals from technical effects and provides information about the cellular composition and biological processes. Build-in annotation, survival analysis, and report generation provide useful tools for the interpretation of extracted signals. The implementation of parallel computing in the package ensures efficient analysis using modern multicore systems. The package offers a reproducible and efficient data-driven solution for the analysis of complex molecular profiles, with significant implications for cancer research. Availability and implementation:The package is implemented in R and available under MIT license at Bioconductor (https://bioconductor.org/packages/consICA) or at GitHub (https://github.com/biomod-lih/consICA).

Vocal biomarkers are measurable characteristics of person's voice that provide valuable insights into various aspects of their physiological and psychological state, or health status. The use of standardized voice tasks, such as reading, counting, or sustained vowel phonation are common in vocal biomarker research, but semi-spontaneous tasks where the person is instructed to talk about a particular topic, or spontaneous speech are also increasingly used. However, limited efforts were made to combine multiple voice modalities. In this paper, we propose a simple, yet efficient approach of fusing multiple standardized voice tasks based on vector cross-attention, showing improved predictive capacity for derived vocal biomarkers in comparison to single modalities. The multimodal approach is tested on the assessment of respiratory quality of life from reading and sustained vowel phonation recordings, outperforming single modalities up to 4.2% in terms of accuracy (relative increase of 7%).