Brain mitochondrial diversity and network organization predict anxiety-like behavior in male mice

The study addresses how mitochondrial biology across the brain relates to behavior under the premise that brain function is energetically constrained by oxidative phosphorylation. Prior work shows mitochondria influence neuronal development, synaptic function, neurogenesis, and inflammation, and stress alters both behavior and mitochondrial function. Yet, systematic, brain-wide biochemical mapping of mitochondria-behavior associations has been lacking. The authors hypothesize that mitochondrial phenotypes in specific brain areas—and potentially distributed mitochondrial networks—are associated with stress-related and anxiety-like behaviors. They aim to quantify multiple mitochondrial features across many brain regions and link them to behavioral variability, testing whether coherent mitochondrial networks exist and whether they are behaviorally relevant and aligned with gene expression and structural connectivity.

The paper situates its work within evidence that: (1) mitochondria support diverse neuronal functions and the brain’s energy demands; (2) behavioral phenotypes (e.g., anxiety, social avoidance) vary naturally and with stressors like chronic social defeat stress (CSDS) and chronic corticosterone (CORT); (3) mitochondrial function varies naturally and recalibrates with stress; (4) causal perturbations of mitochondrial respiratory chain (RC) activities influence behaviors such as working memory, social dominance, and anxiety. Human and animal studies implicate mitochondrial dysfunction in psychiatric and neurodegenerative disorders, with metabolic imaging linking regional energy metabolism (e.g., nucleus accumbens) to cognition and anxiety. Despite this, biochemical differences across brain areas and their relationship to behavior are poorly mapped, motivating a multi-region, multi-feature mitochondrial assessment and network perspective.

Design and cohort: Inbred male mice with natural and induced behavioral variation were studied. Subgroups underwent chronic corticosterone (CORT; ~3 months) or chronic social defeat stress (CSDS; 10 days); a subset of CSDS mice recovered for ~2 months. Behavioral testing included open field test (OFT), elevated plus maze (EPM), novelty suppressed feeding (NSF), and social interaction (SI). Not all animals completed all tests due to pairing with interventions. Total cohort size for key analyses was n≈27 mice (5–6 per group). Sampling: 571 samples were obtained across 17 brain areas (bilateral 1 mm punches, 200 µm thick sections, frozen tissue) at defined stereotaxic coordinates, plus 5 peripheral tissues (adrenal glands, liver, heart, soleus, white gastrocnemius). Tissue processing preserved mitochondrial integrity (rapid decapitation, flash-freeze, controlled cryostat sectioning). Punch mass was estimated (≈0.163 mg per punch). Mitochondrial assays: Miniaturized, high-throughput spectrophotometric assays in 96-well plates quantified enzymatic activities of complex I (CI), complex II (CII/SDH), complex IV (CIV/COX), and citrate synthase (CS; mitochondrial content marker). Assays were validated against traditional cuvette formats (r=0.81–0.96 across assays) and optimized for minimal tissue input; duplicates were used with positive controls per plate and normalization to control for batch effects. Non-specific activity controls were included. mtDNA quantification: From the same homogenates, mtDNA and nuclear DNA were quantified by TaqMan qPCR (COX1, B2M). Two metrics were computed: mtDNA copy number (mtDNAcn, mtDNA:nDNA) and mtDNA density (copies per unit tissue). Given up to ~8.5-fold cell density differences across brain areas, mtDNA density was favored as a more generalizable metric in brain tissue. Composite index: A mitochondrial health index (MHI) was computed as (CI+CII+CIV)/(CS+mtDNA density+1)*100 after mean-centering features, reflecting RC capacity per mitochondrion. Behavioral scoring: Behavioral z-scores were transformed so higher scores indicate higher anxiety/avoidance for all tests (OFT, EPM, NSF, SI), with NSF latency capped at 600 s. SI composite scores averaged multiple measures. Topological and network analyses: TDA-based Mapper analysis characterized stress-induced mitochondrial recalibrations across areas. Correlation-based connectivity matrices assessed inter-regional similarity. Multi-slice community detection (six slices for the six mitochondrial features) identified cohesive mitochondrial networks; stability was assessed via module allegiance and consensus clustering. Cross-modal comparisons: Mitochondria-derived communities were compared against Allen Mouse Brain Atlas datasets: genome-wide gene co-expression (AGEA) and structural connectome (EYFP-labeled projections). Permutation tests (10,000 shuffles) evaluated whether within-network connectivity exceeded chance using strength fraction and modularity (Q_mod). Transcriptomic mitotyping: Using MitoCarta 3.0 gene lists (mapped n=946), mitochondrial gene expression and 149 mitochondrial pathway scores were computed per area (Allen Atlas ISH data, dorsal/ventral DG combined for 16 areas). PCA, hierarchical clustering, pathway scoring, and differential pathway ranking (log2 fold differences) characterized network-specific mitochondrial molecular phenotypes. Statistics: Effect sizes (Hedges g), ANOVAs with Tukey adjustments, binomial tests, t-tests, Spearman correlations for mito-behavior associations, graph-theoretic metrics (participation coefficient). Where power was limited, effect sizes and unadjusted p-values were reported. Plate normalization via z-scored positive controls.

- Assay platform and measurement validity: Miniaturized enzymatic assays achieved sub-milligram sensitivity (<1 mg, ~0.33 mg per two punches) and correlated strongly with standard cuvette assays (r=0.81–0.96). Enzymatic activity did not correlate with RC protein abundance (SDHA) in cerebellar layers (r²=0.02–0.07), indicating protein levels are not surrogate for activity. - Variation across animals and tissues: Brain mitochondrial measures (CI, CII, CIV, CS, mtDNA density) showed substantial inter-animal variability across 17 areas (mean C.V. 36%; up to 2.9-fold differences within a given area). Peripheral tissues varied less (mean C.V. 25%). - Stressor effects: Both CORT and CSDS altered mitochondrial phenotypes area-specifically. Trends: CORT increased activities in ~60% of brain areas, CSDS decreased in ~82% of areas; decreases were significant for CI, CIV, and MHI. Notable effects: Amygdala CII +49% with CORT (p=0.03, unadjusted); PAG CI −42% with CSDS (p=0.02, unadjusted). Mapper analysis showed CSDS induced a more integrated, coherent mitochondrial response across regions (higher participation coefficient) than CORT, which produced more segregated responses (~25% lower PC in CORT vs CSDS; p<0.05). - Mito-behavior associations: Across all animals, MHI correlated significantly with behavior on OFT (p<0.01), EPM (p<0.0001), and SI (p<0.0001), but not NSF. Strong examples: M1 CII vs OFT r=0.51 (p=0.025, unadjusted); NAc MHI vs EPM r=0.92 (p=0.0005, unadjusted); SN MHI vs SI r=−0.78 (p=0.0035, unadjusted). Higher mitochondrial metrics generally associated with higher anxiety in OFT/EPM, and with lower avoidance (greater sociability) in SI. Brain mitochondrial measures related to behavior more robustly than peripheral tissue measures. - Brain-wide mitochondrial connectivity and modularity: Inter-regional mitochondrial features were modestly positively correlated (mean r=0.22), with modular structure indicating within-area mitochondrial features co-regulate more than across areas. Nodal degree varied by area (highest cerebellum r=0.34; lowest vestibular nucleus r=0.05). Brain mitochondria were largely uncorrelated with peripheral tissue mitochondria (brain–periphery mean r≈0.02). - Mitochondria-derived networks: Multi-slice community detection identified three networks: (1) Cortico-striatal network: CPu, V1, M1, mOFC, mPFC, NAc; (2) Salience/Spatial navigation: Cerebellum, VN, VTA, Thalamus, CA3, DGd, DGv; (3) Threat response: Amygdala, Hypothalamus, SN, PAG. - Network-level behavioral relevance: Network 1 mitochondrial features showed the strongest associations with behavior: OFT r=0.40 (p=0.039), EPM r=0.74 (p=0.015; r²≈0.54, adjusted r²≈0.48), and SI r=−0.69 (p=0.003; adjusted r²≈0.43). Network 2 showed weaker associations (r≈0.05–0.35 across behaviors). - Cross-modal overlap: Mitochondria-derived communities exhibited higher-than-chance within-network similarity in gene co-expression (S.F. p=0.020; Q_mod p=0.008) and structural connectivity (S.F. p=0.029; Q_mod p=0.015). - Transcriptional mitochondrial specialization: Network 1 areas were enriched for synaptic signaling, morphogenesis, and phosphorylation-based enzyme regulation, and under-expressed metabolism- and sensing-related processes. Mitochondrial gene PCA and pathway analyses showed cortico-striatal network 1 clustered distinctly from networks 2/3 and was enriched for the glycerol-3-phosphate shuttle, amidoxime and vitamin D metabolism, but under-expressed vitamin B2/B6 metabolism, BH4 synthesis, and MICOS complex components. The vestibular nucleus expressed the highest OxPhos component levels among areas analyzed.

The findings demonstrate that brain mitochondrial properties are diverse across regions, form coherent large-scale networks, and are behaviorally relevant. Direct biochemical measures of RC activity and mtDNA density uncovered brain-wide patterns not captured by protein abundance alone. Stressors differentially recalibrate mitochondrial phenotypes: CORT induced more region-specific changes, while CSDS led to coordinated reductions across regions, suggesting distinct neuroendocrine and temporal dynamics of mitochondrial adaptation. MHI, integrating multiple mitochondrial features, provided the strongest and most consistent associations with behavior, outperforming individual features and peripheral tissue measures, underscoring brain specificity. Mitochondrial connectivity revealed moderate global co-regulation across brain regions but little coherence with peripheral tissues, indicating organ- and region-specific mitochondrial regulation. The identification of three mitochondria-based brain networks aligns broadly with known functional systems (cortico-striatal decision/action, salience/spatial navigation, and threat response) and significantly overlaps with gene expression and structural connectivity, yet remains distinct—suggesting mitochondria capture an independent facet of neural system organization. The cortico-striatal network’s mitochondrial phenotype explained a substantial portion of inter-individual variance in anxiety-like and social behaviors, strengthening the functional significance of mito-based networks. Methodologically, the work highlights the utility of multi-feature indices (MHI) akin to multivariate neuroimaging approaches, and cautions against interpreting mtDNA copy number per cell (mtDNAcn) as a proxy for mitochondrial content in brain due to large cellularity differences. Overall, the study supports a network-based, multimodal view of how mitochondrial function contributes to brain-behavior relationships.

This study introduces a scalable, miniaturized biochemical approach to map mitochondrial phenotypes across the mouse brain and links them to behavior. It reveals that mitochondria form distinct large-scale networks with specific biochemical and transcriptional signatures, and that the cortico-striatal mito-network is especially predictive of anxiety-like and social behaviors. Cross-modal concordance with gene expression and structural connectivity supports the biological validity of these mito-based networks. Future directions include: causal manipulations of mitochondrial function within identified networks to test behavioral effects; longitudinal studies to map the dynamics of mitochondrial recalibrations under different stressors and recovery; higher-resolution, cell-type–specific and subcellular (e.g., synaptic) assessments; translation to other species and to human brain studies; and integration with in vivo metabolic imaging to relate maximal capacity measures to real-time metabolic fluxes.

- Sample size limited (≈27 mice; 5–6 per group) reducing power for multiple comparisons; many reported p-values are unadjusted; effect sizes for some associations may be inflated. - Not all animals underwent all behavioral tests due to experimental design; NSF latency cap (600 s) reduces correlation precision. - Frozen tissue assays measure maximal RC capacity, not in vivo flux influenced by instantaneous neural activity. - Tissue punches may include adjacent regions given 1 mm diameter vs smaller target areas; punches were not weighed directly; mass estimated. - Plate design limited cross-tissue comparisons due to plates accommodating no more than two tissue types; normalization mitigates but does not eliminate batch effects. - Analyses restricted to male mice; sex differences not addressed. - Cell-type heterogeneity across regions may contribute to observed differences; although assessed, full resolution to single-cell level was not performed. - Some mitochondrial measures missing due to technical issues (n=8/579 excluded).