Mitochondria structurally remodel near synapses to fuel the sustained energy demands of plasticity

Mitochondria structurally remodel near synapses to fuel the sustained energy demands of plasticity | 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 Mitochondria structurally remodel near synapses to fuel the sustained energy demands of plasticity View ORCID Profile Monil Shah , View ORCID Profile Ilika Ghosh , Luca Pishos , Valentina Villani , Tristano Pancani , View ORCID Profile Ryohei Yasuda , Chao Sun , Naomi Kamasawa , View ORCID Profile Vidhya Rangaraju doi: https://doi.org/10.1101/2025.08.27.672715 Monil Shah 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA 2 International Max Planck Research School for Synapses and Circuits , Jupiter, FL 33458, USA 3 Florida Atlantic University , Boca Raton, FL 33431, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Monil Shah Ilika Ghosh 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Ilika Ghosh Luca Pishos 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Valentina Villani 4 Danish Research Institute of Translational Neuroscience - DANDRITE, Nordic-EMBL Partnership for Molecular Medicine , Denmark 5 Aarhus University, Department of Molecular Biology and Genetics , Aarhus C, Denmark Find this author on Google Scholar Find this author on PubMed Search for this author on this site Tristano Pancani 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Ryohei Yasuda 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Ryohei Yasuda Chao Sun 4 Danish Research Institute of Translational Neuroscience - DANDRITE, Nordic-EMBL Partnership for Molecular Medicine , Denmark 5 Aarhus University, Department of Molecular Biology and Genetics , Aarhus C, Denmark 6 Center for Proteins in Memory - PROMEMO, Danish National Research Foundation, Dept. of Biomedicine, Aarhus University , Denmark Find this author on Google Scholar Find this author on PubMed Search for this author on this site Naomi Kamasawa 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site Vidhya Rangaraju 1 Max Planck Florida Institute for Neuroscience , Jupiter, FL 33458, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Vidhya Rangaraju For correspondence: vidhya.rangaraju{at}mpfi.org Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract The brain is a metabolically demanding organ as it orchestrates and stabilizes neuronal network activity through plasticity. This mechanism imposes enormous and prolonged energetic demands at synapses, yet it is unclear how these needs are met in a sustained manner. Mitochondria serve as a local energy supply for dendritic spines, providing instant and sustained energy during synaptic plasticity. However, it remains unclear whether dendritic mitochondria restructure their energy production unit to meet the sustained energy demands. We developed a correlative light and electron microscopy pipeline with deep-learning-based segmentations and 3D reconstructions to quantify mitochondrial remodeling at 2 nm pixel resolution during homeostatic plasticity. Using light microscopy, we observe global increases in dendritic mitochondrial length, as well as local increases in mitochondrial area near spines. Examining the mitochondria near spines using electron microscopy, we reveal increases in mitochondrial cristae surface area, cristae curvature, endoplasmic reticulum contacts, and ribosomal cluster recruitment, accompanied by increased ATP synthase clustering within mitochondria using single-molecule localization microscopy. Using mitochondria- and spine-targeted ATP reporters, we demonstrate that the local structural remodeling of mitochondria corresponds to increased mitochondrial ATP production and spine ATP levels. These findings suggest that mitochondrial structural remodeling is a key underlying mechanism for meeting the energy requirements of synaptic and network function. Introduction The brain’s ability to adapt via plasticity also relies on its capacity to remain stable. While the neuronal network is constantly adapting to changing synaptic inputs, it is stabilized by homeostatic setpoints that regulate a neuron’s total synaptic strength 1 . This mechanism of homeostatic scaling, which is crucial for learning and development, causes lasting changes in postsynaptic excitatory properties, calcium influx, and the synaptic proteome, resulting in significant and long-lasting energy demands 1 – 3 . However, it is unclear how the brain adapts to these sustained energy demands and whether it adjusts its energy supplies and energy production capacity locally near synapses. Given the complex morphology of a neuron, where most synapses are located far from the cell body, synapses rely on local energy supplies. Particularly, mitochondria, double-membrane structures with their own genome, proteome, and biochemistry, exhibit a remarkable degree of local autonomy and a capacity to adapt spatially and temporally at the synaptic scale. We recently demonstrated that spatially stable dendritic mitochondria respond rapidly near synapses to meet the energy demands of plasticity, providing both immediate and sustained energy for up to an hour 4 – 6 . However, it is unclear whether this spatiotemporal mitochondrial response lasts longer than an hour. Particularly, the mitochondrial structure, organization of cristae and energy-producing complexes, as well as its contacts with other organelles and protein synthesis machinery, are key to energy production 5 – 7 . It is, however, unclear whether mitochondria restructure themselves to increase their energy production capacity and meet the long-lasting synaptic energy demands, such as in homeostatic scaling. This knowledge gap has profound consequences for human health; the disorganization of mitochondrial inner structure, its interorganellar contacts, and subsequent dysfunction are hallmarks of many brain disorders 8 – 10 . Yet, the link between mitochondrial structural features, synaptic energy provision, cognition, and how this link breaks in brain disorders remains elusive. So far, efforts to study neuronal mitochondrial structural adaptations have been limited to global neuronal averages or presynapses, with little knowledge of postsynapse-specific adaptations 11 – 14 . Mitochondrial ultrastructure studies in neurons have so far focused on axonal mitochondria due to the more complex nature of dendritic mitochondria: (i) dendritic mitochondria are long (∼30 μm) with complex cristae adapting tube-and network-like morphologies compared to axonal mitochondria that are short (<2 μm) 4 , 6 , 15 with cristae arranged in the classical stacked lamellar structure; therefore, visualizing dendritic cristae membrane requires advanced high-resolution (∼2 nm) imaging by electron tomography 11 , (ii) lack of automated image and data analysis algorithms to streamline reproducible segmentation and 3D reconstruction, as well as quantification of the complex dendritic mitochondrial inner structure and its interorganellar interactions from the electron microscopy images and tomograms with speed and accuracy, (iii) the lack of tools to make mitochondrial measurements (area, proximity to spines, and functional measurements such as ATP imaging 5 ) by light microscopy and to map the same mitochondria to study its structure using correlative light and electron microscopy (CLEM). To overcome the limitations, we have developed an approach that incorporates a CLEM pipeline to measure mitochondrial area and proximity to dendritic spines using light microscopy, followed by ultrastructural measurements using electron microscopy, deep-learning-based sub-organellar segmentations, and 3D reconstructions for quantification. We observe global increases in mitochondrial length, as well as local increases in mitochondrial area near spines, following homeostatic scaling for 24 hours in hippocampal neuronal cultures. Our CLEM pipeline reveals that the increased mitochondrial area near spines upon homeostatic scaling corresponds to complex, cristae network-like structures, characterized by increases in mitochondrial cristae surface area, cristae junction density, and cristae curvature. In addition, using single-molecule localization microscopy, DNA-PAINT 16 , we demonstrate increased ATP synthase clustering within these mitochondria, suggesting a reorganization of the mitochondrial energy-synthesis machinery to enhance its energy production capacity. We also find increased endoplasmic reticulum contacts and ribosomal clusters near mitochondria following homeostatic scaling, potentially to meet the new lipid and protein demands of local mitochondrial remodeling. This spine-specific structural remodeling of mitochondria corresponds to a local increase in mitochondrial ATP synthesis and spine ATP levels upon homeostatic scaling. We corroborated some of our results in an intact brain using the MICrONS Phase 1 functional connectomics dataset from the visual cortex 17 , and we observe similar local increases in mitochondrial volume near spines. These findings reveal that dendritic mitochondria remodel their inner structure and interorganellar interactions near spines to meet the heightened and sustained synaptic energy demands associated with plasticity. Results Global and local changes in mitochondrial structure near spines upon homeostatic scaling To induce prolonged neuronal and synaptic energy demands, we induced homeostatic plasticity (otherwise called homeostatic upscaling) by treating primary hippocampal neuronal cultures with 2 μM tetrodotoxin (TTX) for 24 hours and compared them to Control neurons ( Fig. 1A ) 1 , 2 . We confirmed the successful induction of homeostatic upscaling (simplified as homeostatic scaling in the rest of the paper) by the increase in the peak amplitude of miniature EPSC events and spine area, compared to the Control, as previously reported 2 , 3 (Fig. S1A-C). Download figure Open in new tab Figure 1. Global and local increase in mitochondrial structure near spines upon homeostatic scaling A Experimental flow for visualizing mitochondrial structural changes near spines by fluorescence and electron microscopy upon homeostatic scaling in primary hippocampal neuronal cultures. B Representative image (of C - E ) showing mitochondria (green, Mito-EGFP) and spines (white, PSD95-mCh) in a dendritic segment. Scale bar 10 μm. C The average mitochondrial length is increased in homeostatic scaling-induced neurons (green) compared to the Control (gray). n in neurons, animals: 41, 3 (Control), 45, 3 (homeostatic scaling). Mann–Whitney U test, p-value: 0.0007 (Control vs. homeostatic scaling). D The average percentage of spines with mitochondria within 1 μm of the spine base increased in homeostatic scaling-induced neurons (green) compared to the Control (gray). n in neurons, animals: 41, 3 (Control), 45, 3 (homeostatic scaling). Mann–Whitney U test, p-value: 0.0364 (Control vs. homeostatic scaling). E The average mitochondrial area within 1 μm radius of the spine base increased in homeostatic scaling-induced neurons (green) compared to the Control (gray). n in neurons, animals: 41, 3 (Control), 45, 3 (homeostatic scaling). Student’s t-test, p-value: 0.0054 (Control vs. homeostatic scaling). See also Figure S1. To measure mitochondrial structural changes near spines, we labeled neuronal mitochondria using EGFP targeted to the mitochondrial matrix (Mito-EGFP) and dendritic spines using mCherry targeted to the spine scaffolding protein, PSD95 (PSD95-mCh) ( Fig. 1A, B ). We observed a global increase in average mitochondrial length in homeostatic scaling compared to the Control ( Fig. 1C ). Additionally, the percentage of spines with mitochondria located within 1 μm of the spine base increased following homeostatic scaling relative to the Control ( Fig. 1D , see Methods). To determine whether there are spine-specific changes in mitochondrial structure, we measured mitochondrial area within 1 μm of each spine base (see Methods). The average mitochondrial area at each spine base increased following homeostatic scaling compared to the Control ( Fig. 1E ). We confirmed that these results are consistent when the imaging was done in the presence of TTX for homeostatic scaling-induced neurons (Fig. S1D, E, see Methods), and in the absence of PSD95 overexpression (using PSD95FingR-EGFP 18 instead of PSD95-mCh) (Fig. S1F, G). These results indicate that mitochondria undergo global and local structural changes near spines in response to the sustained energy demands of homeostatic scaling. Mitochondrial cristae remodel near spines upon homeostatic scaling The spine-specific increase in mitochondrial area near spines in response to prolonged energy demands ( Fig. 1D, E, S1E, G ) suggests that the mitochondrial inner structure may also undergo changes near spines. The mitochondrial inner structure, including the cristae membrane, crista junctions, and their energy-synthesis machinery, plays a crucial role in energy production 7 . For example, increased cristae membrane density indicates an increased energy production capacity of mitochondria 11 , 19 . Increased cristae junction density indicates an increased association of the cristae with the inner mitochondrial membrane, which is critical for metabolite transfer, such as ATP, from the mitochondrial matrix to the intermembrane space and then to the cytosol 20 . Cristae junctions are also essential for stabilization of the cristae folds and are coupled with mitochondrial protein import machinery 21 . In addition, a high cristae curvature indicates local ATP synthase dimerization or multimerization, which is required for cristae formation and increased ATP production efficiency in single-celled organisms 22 – 24 . To visualize mitochondrial cristae, we established a correlative light and electron microscopy pipeline to examine the ultrastructure of dendritic mitochondria near spines as identified under light microscopy ( Fig. 2A , see Methods). We conducted all our measurements in dendrites 50-150 μm from the cell body, as we were interested in spine-specific changes in distal dendritic segments. The target mitochondria and spines of interest, identified under light microscopy, were further resolved under transmission electron microscopy and subsequently examined using dual-axis tilt series to obtain 3D reconstructed tomograms (8 tomograms for Control and 15 tomograms for homeostatic scaling, Fig. 2A , 3A , B). We built custom Python-based pipelines to streamline deep learning-based segmentation and the generation of 3D surfaces for mitochondrial membranes to quantify the surface area of inner mitochondrial membrane, cristae membrane, cristae junctions, and cristae curvature from the tomograms ( Fig. 2A , S2A, 4A , S3A, B, see Methods). Download figure Open in new tab Figure 2. The correlative light and electron microscopy pipeline used to identify and analyze mitochondria near spines A Experimental flow for the correlative light and electron microscopy pipeline to target dendritic spines (white arrowheads) and mitochondria of interest (red arrowheads) using fluorescence microscopy and differential interference contrast imaging (DIC image) at 20X and 63X magnifications, followed by their identification (white and red arrowheads) in transmission electron microscopy images (TEM correlation) for tomogram acquisition, reconstruction, deep-learning-based segmentations and proofreading in primary hippocampal neuronal cultures. Scale bars: 100 μm (20X confocal), 20 μm (63X confocal), 2 μm (63X confocal, inset), 5 μm (low magnification TEM), 2 μm (medium magnification TEM), 0.5 μm (high magnification TEM), 0.5 μm (tomogram). See also Figure S2. Download figure Open in new tab Figure 3. Reconstructed meshes from electron microscopy tomograms. All the reconstructed meshes from electron microscopy tomograms of the mitochondrial regions (of Fig. 4D -H ) showing cristae membrane (yellow) and outer mitochondrial membrane (gray) in Control ( A ) and homeostatic scaling-induced neurons ( B ). mitochondria 4 (control) and 8 (homeostatic scaling) are the same as in Fig. 4C . Scale bar 0.5 μm. Download figure Open in new tab Figure 4. Mitochondrial cristae remodel near spines upon homeostatic scaling A Illustration showing the definitions of outer mitochondrial membrane (OMM), inner boundary membrane (IBM), cristae membrane (CM), inner mitochondrial membrane (IMM), high first principal curvature (k1) CM, and cristae junctions used for segmentations and analysis. B Representative confocal images (of D - H ) showing spines (white, PSD95- mCh) and mitochondria (white, Mito-EGFP) in a dendritic segment, which were identified and imaged using transmission electron microscopy and tomography (tomogram slice) in Control ( top ) and homeostatic scaling-induced neurons ( bottom ). Scale bars 2 μm (confocal images), 0.5 μm (EM images). C Representative 3D reconstructed meshes from electron microscopy tomograms (of D - F, Fig. 3A, B ) of the mitochondrial regions in B (orange box) showing inner membrane (yellow), cristae membrane (yellow), and cristae junctions (purple) in Control ( top ) and homeostatic scaling-induced neurons ( bottom ). Scale bar 0.5 μm. The insets show zoomed-in subregions of the 3D reconstructed meshes for better visualization. Scale bar 0.1 μm. D The average fraction of the inner mitochondrial membrane (IMM) density increased in homeostatic scaling-induced neurons (yellow) compared to the Control (gray). n in 3D mitochondrial reconstructions, animals: 8, 2 (Control), 15, 2 (homeostatic scaling). Student’s t-test, p-value: <0.0001 (Control vs. homeostatic scaling). E The average cristae membrane (CM) density increased in homeostatic scaling-induced neurons (yellow) compared to the Control (gray). n in 3D mitochondrial reconstructions, animals: 8, 3 (Control), 15, 2 (homeostatic scaling). Student’s t-test, p-value: 0.0241 (Control vs. homeostatic scaling). F The average cristae junction density increased in homeostatic scaling-induced neurons (purple) compared to the Control (gray). n in 3D mitochondrial reconstructions, animals: 8, 2 (Control), 15, 2 (homeostatic scaling). Student’s t-test, p-value: <0.0001 (Control vs. homeostatic scaling). G Representative 3D reconstructed meshes from electron microscopy tomograms (of H ) of the mitochondrial regions in B (orange box) showing cristae membrane (yellow) and high first principal curvature (k1) cristae membrane (red) in Control ( top ) and homeostatic scaling-induced neurons ( bottom ). Scale bar 0.5 μm. The insets show zoomed-in subregions of the 3D reconstructed meshes for better visualization. The color scale ranges from low k1 (0 μm -1 ) to high k1 (> 80 μm -1 ) first principal curvature (k1) cristae membrane. Scale bar 0.1 μm. H The average high k1 cristae membrane (CM) density increased in homeostatic scaling-induced neurons (red) compared to the Control (gray). n in 3D mitochondrial reconstructions, animals: 8, 2 (Control), 15, 2 (homeostatic scaling). Student’s t-test, p-value: 0.0391. See also Figure S3. Our analysis revealed a significant increase in the inner mitochondrial membrane area per mitochondrial volume (IMM density) in homeostatic scaling-induced neurons compared to the Control ( Fig. 4A-D ). To further resolve the inner mitochondrial membrane, we separated and quantified the cristae membrane from the inner boundary membrane and determined the cristae membrane density and cristae junction density ( Fig. 4A , see Methods). Mitochondrial cristae in homeostatic scaling-induced neurons exhibited an interconnected and branched network, compared to the Control, with an increased cristae membrane density ( Fig. 4A-C , E). In addition, cristae junction density increased upon homeostatic scaling compared to the Control ( Fig. 4A-C , F). We next determined the first principal curvature of the cristae membrane surface that would indicate local dimerization or multimerization of ATP synthases 14 , 25 ( Fig. 4A , see Methods). Homeostatic scaling resulted in an increase in the surface area of high-curvature cristae membrane ( Fig. 4A, G, H, S3A , B, see Methods) compared to the Control. This result indicates that the spine-specific increase in local mitochondrial area ( Fig. 1D, E, S1E, G ) corresponds to local increases in mitochondrial cristae complexity, characterized by high cristae density and curvature. These results demonstrate that the sustained energy demands of homeostatic scaling drive local structural remodeling of mitochondrial cristae, potentially increasing the density of ATP synthases and their ATP production capacity. ATP synthase clustering increases within mitochondria upon homeostatic scaling We next directly examined ATP synthases within mitochondria upon homeostatic scaling. ATP synthases form differently shaped arrays of dimers and multimers (linear, helical, hexamer, pentagonal pyramids) that generate high membrane curvatures, resulting in cristae reorganization, which potentially promotes ATP production capacity 23 , 26 , 27 . ATP synthase studies have so far relied on single-celled organisms using Cryo-ET 7 . To examine whether the increase in mitochondrial cristae complexity and curvature ( Fig. 4C-H , S3A, B) results in ATP synthase dimerization or multimerization near dendritic spines, we used DNA-PAINT-based single-molecule localization microscopy 16 ( Fig. 5A, B ). While we initially wanted to resolve ATP synthase dimers and linear multimeric arrangements of dimers, 2D DNA-PAINT microscopy only allowed for resolving some of them (see Discussion). Therefore, we focused on clusters containing multiple ATP synthase copies (assumed as multimers) ( Fig. 5B ). We quantified the ATP synthase copies within 2 μm mitochondrial segments to match the high curvature cristae measurements done within ∼2 μm mitochondrial segments in our tomograms ( Fig. 4B-G , see Methods). Because ATP synthases are important for inducing and maintaining cristae curvature, we expected to detect tightly packed, high-copy-number clusters of ATP synthases as multimeric assemblies upon homeostatic scaling 28 . Download figure Open in new tab Figure 5. ATP synthase clustering increases within mitochondria upon homeostatic scaling A Experimental flow for visualizing mitochondrial ATP synthases using DNA-PAINT super-resolution microscopy upon homeostatic scaling in primary hippocampal neuronal cultures. B Representative image (of D ) showing the outer mitochondrial membrane (white, OMM-EGFP using DNA-PAINT) and ATP synthase copy clusters (blue, Anti-OSCP using DNA-PAINT). The insets show zoomed-in subregions of the mitochondria for better visualization of the ATP synthase clusters (white) of different copy numbers, 1-4 (white), 5-8 (yellow), 9-12 (magenta), and 13-24 (red). Scale bars 2 μm, 0.5 μm (insets). C Representative images (of D ) showing 4 (blue, Anti-OSCP using DNA-PAINT) ATP synthase copy clusters within mitochondria (dotted white line, OMM-EGFP using DNA-PAINT) in Control ( top ) and homeostatic scaling ( bottom ). Scale bar 0.2 μm. D The average ATP synthase copy density decreases within copy clusters of 1-4, equalizes with copy clusters of 5-8, and increases with copy clusters of 9-12 and 13-24 in homeostatic scaling-induced neurons (blue) compared to the Control (gray). n in clusters, animals: 1420, 2 (Control), 1392, 2 (homeostatic scaling). Mann–Whitney U test, Control vs. homeostatic scaling, p-values, (1-4): <0.0001, (5-8): 0.3011, (9-12): 0.0198, (13-24): 0.0015. Upon homeostatic scaling, the DNA-PAINT clusters of lower-order ATP synthase copies decreased (1 to 4 ATP synthase copy clusters). This trend equalized (5 to 8 ATP synthase copy clusters) and then increased with higher order ATP synthase copies (9 to 12 and 13 to 24 ATP synthase copy clusters), compared to the Control ( Fig. 5B-D ), consistent with the expectation to find tighter packing of ATP synthases within mitochondria upon homeostatic scaling. This reorganization of ATP synthases within the mitochondria into higher-order ATP synthase multimers, in response to homeostatic scaling, indicates an increased capacity for mitochondrial ATP production in response to sustained energy demands. Enhanced ER contacts and ribosomal cluster recruitment to mitochondria near spines Mitochondrial structural remodeling requires substantial amounts of protein and lipid to meet (i) the increased protein content of electron transport chain complexes and (ii) the increased lipid content of mitochondrial cristae membrane, respectively. The primary source of mitochondrial lipids is the endoplasmic reticulum (ER), which transfers lipids through the ER-mitochondria contact sites (ERMCS) 29 – 32 . The primary source of mitochondrial proteins is the ribosomes (both cytosolic and mitochondrial), which can locally synthesize new proteins on the mitochondrial surface or within the matrix, in addition to transporting newly synthesized proteins from the cytosol into the mitochondria 33 , 34 . The whole neuronal nascent proteome, upon homeostatic scaling and environmental enrichment, reveals several nuclear- and mitochondria-encoded mitochondrial proteins, suggesting that sustained energy demands drive mitochondrial protein synthesis 2 , 35 ; however, this has not been observed within distal dendritic compartments and near spines. We therefore first examined the presence of ER-mitochondria contact sites (ERMCS) near the mitochondrial surface in Control and homeostatic scaling-induced neurons in our tomograms (5 tomograms for Control and 9 tomograms for homeostatic scaling, Fig. 6A-D ). We found a significant increase in ERMCS in homeostatic scaling-induced neurons compared to the Control ( Fig. 6A-D ). Meanwhile, the overall ER density per tomogram (∼2 μm dendritic length) remained unchanged ( Fig. 6A-D ). Download figure Open in new tab Figure 6. Enhanced ER contacts and ribosomal cluster recruitment to mitochondria near spines A All the reconstructed electron microscopy tomograms of the mitochondrial regions (of B-F ) showing ERMCS (dark blue), cristae membrane (yellow), ribosomal clusters (pink), and total ER (light blue) in Control ( top ) and homeostatic scaling-induced neurons ( bottom ). Mitochondria 4 (control) and 2 (homeostatic scaling) are the same as in C . Scale bar 0.5 μm. B Representative tomogram slices (of A , D - F ) identified by the CLEM pipeline in Fig. 2 in Control ( top ) and homeostatic scaling-induced neurons ( bottom ). Scale bar 0.5 μm. The insets show zoomed-in subregions of the tomogram slices for better visualization. Scale bar 0.1 μm. Blue arrowheads indicate ER-mitochondria contact sites (ERMCS) and pink arrowheads indicate ribosomal clusters. C Representative reconstructed electron microscopy tomograms (of A , D - F ) of the mitochondrial regions in B showing ERMCS (dark blue), cristae membrane (yellow), ribosomal clusters (pink), and total ER (light blue) in Control ( top ) and homeostatic scaling-induced neurons ( bottom ). Scale bar 0.5 μm. The insets show zoomed-in subregions of the reconstructed electron microscopy tomograms for better visualization. Scale bar 0.1 μm. D ( top ) The average fraction of ERMCS increased in homeostatic scaling-induced neurons (dark blue) compared to the Control (gray), while ( bottom ) the average ER density remained unchanged. Top: n in 3D ER reconstructions around mitochondria, animals: 5, 2 (Control), 9, 2 (homeostatic scaling). Student’s t-test, p-value: 0.0457 (Control vs. homeostatic scaling). Bottom: n in 3D ER reconstructions around mitochondria, animals: 5, 2 (Control), 9, 2 (homeostatic scaling). Student’s t-test, p-value: 0.6072 (Control vs. homeostatic scaling). E The average ribosome density increased in homeostatic scaling-induced neurons (pink) compared to the Control (gray). n in 3D ribosomes reconstructions around mitochondria: 3, 2 (Control), 6, 2 (homeostatic scaling). Welch’s t-test, p-value: 0.0402 (Control vs. homeostatic scaling). F The average ribosome cluster to outer mitochondrial membrane (OMM) distance decreased in homeostatic scaling-induced neurons (pink) compared to the Control (gray). n in 3D ribosome reconstructions around mitochondria: 3, 2 (Control), 6, 2 (homeostatic scaling). Mann–Whitney U test, p-value: 0.0325 (Control vs. homeostatic scaling). Moreover, homeostatic scaling increased the density of ribosomes per tomogram (∼2 μm dendritic length) compared to the Control ( Fig. 6A-C , E). In addition, ribosomal clusters were recruited closer to the outer mitochondrial membrane (OMM) following homeostatic scaling compared to the Control ( Fig. 6A-C , F). As these measurements were taken in neuronal dendritic segments 50-150 μm from the cell body, this suggests that the ribosomal clusters might be locally synthesizing new proteins near distal dendritic spines during homeostatic scaling, confirming previous observations 3 and providing additional evidence that this process can occur close to the mitochondrial surface. These findings indicate that the machinery for lipid transfer and local protein synthesis is remodeled to potentially meet the lipid and protein demands of mitochondrial structural remodeling upon homeostatic scaling. Mitochondrial structural remodeling corresponds to increased mitochondrial ATP synthesis and spine ATP levels The global and local increases in mitochondrial area near spines ( Fig. 1D, E, S1E, G ), as well as their machinery for energy synthesis, lipid transfer, and protein synthesis ( Fig. 3A , B, 4B -H, S3A, B, 5C, D, 6A-F), suggest an increase in the ATP production capacity of mitochondria upon homeostatic scaling. To test whether the ATP production capacity has increased upon homeostatic scaling, we quantified ATP concentrations in mitochondria and dendritic spines of these neurons, using recently developed ATP reporters, Mito-ATP and Spn-ATP, respectively 5 , 36 ( Fig. 7A-C ). The reporters are based on the engineered firefly luciferase that produces luminescence upon catalyzing the oxidation of luciferin in the presence of ATP and Mg 2+ . Mito-ATP is fused to four repeats of the COX8 signal peptide to target it to the mitochondrial matrix ( Fig. 7B ). Spn-ATP is fused to Homer2, a postsynaptic density scaffolding protein, to target it to dendritic spines ( Fig. 7C ). To calibrate for reporter expression levels and comparison between experimental conditions across neurons and animals, both reporters are fused to a fluorescent protein (mCherry2 for Mito-ATP and mOrange2 for Spn-ATP), allowing for ratiometric luminescence to fluorescence readouts (L/F) using a custom imaging approach, as previously reported (see Methods) 5 , 36 . Download figure Open in new tab Figure 7. Local mitochondrial structural remodeling corresponds to increased ATP mito and ATP spine A Experimental flow for imaging ATP mito and ATP spine upon homeostatic scaling in primary hippocampal neuronal cultures. Illustrations showing the mitochondrial ATP sensor, Mito-ATP ( B ), and spine ATP sensor, Spn-ATP ( C ). D Representative images (of E ) showing Mito-ATP fluorescence (white) and luminescence (orange) within a dendritic segment in Control and homeostatic scaling-induced neurons. Scale bar 10 μm. E The average ATP mito increased in homeostatic scaling-induced neurons (purple) compared to the Control (gray). n in mitochondria, neurons, animals: 111, 18, 3 (Control), 87, 15, 3 (homeostatic scaling). Mann–Whitney U test. p-value: <0.0001 (Control vs. homeostatic scaling). F Representative images (of G, H ) showing Spn-ATP fluorescence (white), luminescence (orange), and mitochondria (white) within a dendritic segment in Control and homeostatic scaling-induced neurons. Scale bar 5 μm. G The average ATP spine increased in homeostatic scaling-induced neurons (purple) compared to the Control (gray). n in spines, neurons, animals: 361, 38, 3 (Control), 271, 18, 3 (homeostatic scaling). Mann–Whitney U test, p-value: <0.0001 (Control vs. homeostatic scaling). H Linear fit between mitochondrial area within 1 μm of spine base and ATP spine shows a significant correlation in homeostatic scaling-induced neurons (purple) compared to the Control (gray). n in spines, neurons, animals: 343, 38, 3 (Control), 248, 18, 3 (homeostatic scaling). Control, Linear fit, Slope: 0.0002 ± 0.0043, p-value (slope): 0.9542; homeostatic scaling, Linear fit, Slope: 0.0168 ±0.0041, p-value (slope): <0.0001. I Illustration showing the local structural remodeling of mitochondria near synapses measured by the increase in cristae membrane (black line), cristae junctions (purple dots), high cristae curvature (red), ATP synthase clusters (blue), endoplasmic reticulum-mitochondria contacts (ERMCS, dark blue; ER in light blue), and ribosomal clusters near mitochondria (pink), corresponding to increased mitochondrial and spine ATP levels (yellow) upon homeostatic scaling compared to Control using the CLEM pipeline for ultrastructural measurements combined with deep-learning-based sub-organellar segmentations and 3D reconstructions for quantification. See also Figure S4. Upon homeostatic scaling, mitochondrial ATP levels increased compared to the Control, suggesting an increased ATP production capacity of mitochondria in these neurons ( Fig. 7B, D, E, S4A ). The increase in mitochondrial ATP levels resulted in increased spine ATP levels compared to the Control ( Fig. 7C, F, G, S4B , C). Interestingly, the increase in spine ATP levels upon homeostatic scaling corresponded to local increases in mitochondrial area ( Fig. 7H , see also 1E). These findings provide functional evidence to support our structural observations that remodeled mitochondrial regions near dendritic spines are capable of local ATP production, which is subsequently distributed to spines to meet their energy needs ( Fig. 7I ). Signatures of mitochondrial structural remodeling near spines in an intact brain To determine whether the mitochondrial structural remodeling near spines observed in dissociated cultures is consistent in an intact brain, we used the volumetric electron microscopy dataset from MICrONS Phase 1 37 (Fig. S5). While this dataset does not measure homeostatic scaling of neurons, as shown in Figs. 1 - 7 , we utilized it because it contains mitochondrial segmentations from the largest EM volume reconstructions compared to any other publicly available dataset from an intact brain 37 . It comprises neurons imaged by serial section electron microscopy, in which neuronal types, their dendrites, spines, and mitochondria are segmented, among others 37 . We used their 3D meshes to render, visualize, and analyze mitochondria 37 (Fig. S5A, B, see Methods). We found long mitochondria (ranging from 6 to 180 μm) within the dendrites, as has been reported previously in the cortex and hippocampus 6 , 15 , 37 , 38 . We developed algorithms to automatically identify spine bases and center them within the dendritic meshes of 107 pyramidal neurons and measured the mitochondrial volume within a 1 μm sphere radius of each spine base (Fig. S5C, see Methods). We found that spine clusters with larger spine size exhibited a strong linear correlation with higher local mitochondrial volume within a 1 μm radius of the spine bases (Fig. S5C, D; see Methods). If we consider the local spine size increase as a proxy for increased synaptic strength and therefore increased energy demands, this analysis suggests that the local energy reserve, the local mitochondrial volume, increases with increasing energy needs. While the MICrONS Phase 1 data analysis revealed a strong global correlation between mitochondrial volume and synapse density 37 , our analysis indicates that this relationship holds even within a 1 μm radius of synapses and is correlated with spine size. Furthermore, we find that local structural features of mitochondria can be utilized as lower-resolution correlates of local changes in synaptic densities. While it was not possible to observe similar high-resolution cristae remodeling (∼2 x 2 x 2 nm, Figs. 2 - 4 , S3) in the relatively low-resolution connectomics data (4 x 4 x 40 nm), we looked for lower-resolution correlates of cristae remodeling within a mitochondrial compartment in the Phase 1 dataset (S5C, see Methods). We detected mitochondrial regions with a high width surrounded by a larger number of spines than mitochondrial regions with no detectable increase in width (Fig. S5C, E). Extending this analysis to our confocal data ( Fig. 1 ), we found similar signatures (Fig. S5F, G). Such observations in low-resolution EM and confocal images could be used in the future as proxies for local mitochondrial structural remodeling, which may indicate local synaptic energy demands and mitochondrial energy production. In summary, our results indicate that mitochondria undergo global and local remodeling of their structure, leading to a robust increase in mitochondrial ATP production, which further enhances spine ATP levels over 24 hours to support the sustained energy demands of neuronal plasticity ( Fig. 7I ). Discussion Using state-of-the-art technology, we investigate whether mitochondria restructure their energy synthesis machinery to increase their energy production capacity near spines in response to energy needs, and whether this spatiotemporal response persists for longer than an hour. We incorporate a CLEM pipeline to measure mitochondrial structural changes using light microscopy, followed by ultrastructural measurements using electron microscopy, deep-learning-based sub-organellar segmentations, 3D reconstructions, and custom algorithms to streamline semi-automatic quantification of mitochondrial length, mitochondrial area near spines, inner mitochondrial membrane density, cristae membrane density, high cristae curvature, cristae junction density, ribosome clusters-to-mitochondria distance, and ERMCS density. We probe the local adaptations of mitochondrial structure-function that drive ATP synthesis and its distribution to spines in response to the sustained energy needs of neuronal plasticity as follows: (1) homeostatic scaling for 24 hours results in global increases in mitochondrial length and local increases in mitochondrial area near spines; (2) increase in mitochondrial area near spines upon homeostatic scaling corresponds to increases in mitochondrial cristae membrane density, cristae junction density, high cristae curvature, mitochondria-endoplasmic reticulum contacts and ribosomal clusters near mitochondria; (3) super-resolution microscopy reveals increased ATP synthase clustering within mitochondria upon homeostatic scaling; (4) the multifarious mitochondrial structural adaptations near spines upon homeostatic scaling corresponds to local increases in mitochondrial ATP synthesis and spine ATP levels; (5) local signatures of mitochondrial remodeling near spines are also observed in the intact brain. The structural remodeling of mitochondria is observed for up to 24 hours ( Fig. 1 - 7 ). Future experiments should investigate whether mitochondria can undergo structural remodeling in response to acute energy demands (within seconds to minutes, as observed in non-neuronal cells 39 ) and whether the remodeling persists for longer than 24 hours. While previous efforts have examined the global relationship between mitochondrial volume, dendritic length, and synapse number 37 , our results emphasize the importance of local modulation of mitochondrial area and volume within the synaptic environment for a more accurate measure of spine-specific, energy-driven changes ( Fig. 1 , 3-6, S3, S5). Future experiments with better spatiotemporal resolution for ATP imaging should reveal whether submitochondrial remodeling results in submitochondrial ATP synthesis hotspots, thereby controlling the spatial scale of ATP availability to a cluster of synapses. Our findings also indicate that future measurements of submitochondrial structural remodeling, even at low resolution in fluorescence and electron microscopy datasets (Fig. S5), could be used as a proxy for hotspots of remodeling in the mitochondrial energy synthesis machinery (e.g., cristae, ATP synthase) and as signatures of energy demand and production in response to the history of local activity or plasticity in a neuronal network. Several molecular players have been studied in remodeling the mitochondrial ultrastructure, particularly in the reorganization of cristae, including Opa1, Tafazzin, the MICOS complex proteins (Mic10, Mic19, Mic60), Prohibitins, stomatin-like protein-2, and DNAJC19 40 – 44 . Most of these studies have been conducted in isolated mitochondria, yeast, and other single-celled organisms or non-neuronal cells. It is unclear whether similar molecular players drive mitochondrial structural remodeling in response to local synaptic energy demands during neuronal plasticity. Moreover, the signaling mechanism that activates these molecular players, particularly near spines, is unknown. For instance, Opa1 homo-oligomerization, which is required for cristae reorganization, is activated by starvation conditions in a Slc25A-dependent manner in non-neuronal mammalian cells 45 . Slc25A is a mitochondrial solute carrier protein that can sense mitochondrial substrates, such as malate and succinate from the Krebs cycle, and respond based on energy demands 46 . Similarly, PINK1-dependent phosphorylation of Mic60 in response to cellular energy demands can drive Mic60 oligomerization and cristae junction formation in fly and human neurons 9 . PKA, a cAMP-dependent protein kinase A, is also known to phosphorylate Mic60 and Mic19 in non-neuronal mammalian cells 47 , 48 . As cAMP is an intracellular second messenger of ATP levels, it can therefore drive cristae remodeling in response to energy demands. Future experiments should investigate these and other potential molecular mechanisms that drive mitochondrial remodeling near synapses in response to spine-specific energy demands. High-curvature cristae are associated with increased ATP synthase enrichment 22 , implying enhanced bioenergetic capacity of mitochondria during neuronal plasticity. The increased cristae curvature and ATP synthase clustering observed within mitochondria ( Fig. 4 , 5 , S3 ) and the subsequent increase in mitochondrial ATP synthesis ( Fig. 7 ) upon homeostatic scaling further support this result. Increases in ATP synthase density, particularly in the form of ATP synthase dimers, are observed in non-neuronal cells and single-celled organisms in response to energy demands 24 , 49 . While we were unable to resolve the ATP synthase dimers due to the resolution limits of our 2D DNA-PAINT measurements (∼14.2 nm in X and Y and ∼ 900 nm in Z, while the ATP synthase dimer-dimer distance is 12-13 nm 50 , 51 ), we observed ATP synthase clustering that may correspond to multimers that are higher-order dimers. Future experiments using cryogenic electron tomography (cryo-ET) should address these resolution limits. Our findings, showing increased mitochondria-ER contacts and ribosome proximity upon homeostatic scaling, align with growing evidence on mitochondria-ER contacts facilitating lipid transfer and membrane biogenesis, as well as the importance of local protein synthesis for organelle maintenance and remodeling 6 , 52 – 54 . It would therefore be helpful to investigate the lipidomic signatures of ER-mitochondria transfer and the proteomic signatures of the newly synthesized proteins that facilitate local mitochondrial restructuring, as well as their roles in plasticity and disease. Neuronal plasticity and network activity are essential for driving various brain-wide mechanisms, which are crucial during learning, memory, development, sleep, and aging. Dysfunctions in these mechanisms are often linked to neurodevelopmental and neurodegenerative disorders as well as memory and sleep disorders 55 , 56 . Specifically, abnormalities in mitochondrial ultrastructure and their structural machinery are observed in multiple neurodegenerative disorders, including Alzheimer’s and Parkinson’s, and cellular aging 8 – 10 , 57 . Notably, the emerging evidence highlighting the significance of mitochondria in sleep and aging 58 , 59 underscores the importance of our findings, which demonstrate that dendritic mitochondria dynamically couple their structure and energy output to local synaptic demands to sustain synaptic stability and neuronal plasticity. Data availability Plasmids generated in this study will be deposited in Addgene. Custom-written scripts used for data analysis, quantification, and statistics with demo images and user instructions will be available at https://github.com/Rangaraju-Lab/Shah-et-al-2025 . Requests for further information and resources should be directed to and will be fulfilled by Vidhya Rangaraju ( vidhya.rangaraju{at}mpfi.org ). Use of AI tools ChatGPT-4o was used to assist with the programming of custom-built algorithms. Competing interests The authors declare no competing interests. Author contributions VR and MS designed experiments. Experiments were carried out by MS, IG, VV, and TP. Data were analyzed by MS, IG, and LP. IG performed the ATP imaging experiments. LP performed the MICrONS data analysis. VV performed the DNA-PAINT experiments, and CS supervised the DNA-PAINT experiments and analysis. TP performed the electrophysiology experiments, and RY supervised them. NK supervised the electron microscopy sample preparation and imaging. All other experiments were conducted and analyzed by MS. VR supervised all experiments and analysis. VR and MS wrote the manuscript, and all authors reviewed and edited it. Methods Animals All experiments were performed in accordance with the Max Planck Florida Institute for Neuroscience IACUC regulations (protocol number 22-005). The DNA-PAINT experiments were conducted in accordance with the National Animal Care guidelines and those issued by Aarhus University and were approved by the local authorities. Plasmid constructs The PSD95FingR-mTagBFP2 plasmid was made by replacing EGFP with mTagBFP2 in the PSD95FingR-EGFP plasmid 18 , purchased from Addgene 46295. The OMM-EGFP plasmid was made by replacing GCaMP6f in the OMM-GCaMP6f plasmid 60 . Cell culture preparation and transfection Unless specified otherwise, all reagents were purchased from Sigma, and all stock solutions were stored at -20 °C. Timed-pregnant Sprague-Dawley rats were obtained from Charles River Laboratories and housed at the animal core facility of the Max Planck Florida Institute for Neuroscience to obtain the postnatal P0 pups subsequently. Hippocampal regions were dissected in ACSF containing (in mM) 124 NaCl, 5 KCl, 1.3 MgSO4:7H2O, 1.25 NaH2PO4:H2O, 2 CaCl2, 26 NaHCO3, and 11 Glucose (stored at 4 °C) and stored in hibernate E buffer (BrainBits LLC, stored at 4°C). Dissected hippocampi were dissociated using the Papain Dissociation System (Worthington Biochemical Corporation, stored at 4 °C) with a modified manufacturer’s protocol. Briefly, hippocampi were digested in papain solution (20 units of papain per ml in 1 mM L-cysteine with 0.5 mM EDTA) supplemented with DNase I (final concentration 95 units per ml) and placed in a shaking heat block for 30 min at 37 °C, 900 rpm. Digested tissue was triturated and set for 5 min, and the supernatant devoid of tissue chunks was collected. The supernatant was centrifuged at 300 g (0.3 rcf) for 5 min, and the pellet was resuspended in resuspension buffer (1 mg of ovomucoid inhibitor, 1 mg of albumin, and 95 units of DNase I per ml in EBSS). The cells were forced to pass through a discontinuous density gradient formed by the resuspension buffer and the Ovomucoid protease inhibitor (10 mg per ml) with bovine serum albumin (10 mg per ml) by centrifuging at 70 g (0.6 rpm) for 10 min at room temperature. The final cell pellet devoid of membrane fragments was resuspended in Neurobasal-A medium (Gibco, stored at 4°C) supplemented with Glutamax (Gibco, stored at -20 °C) and B27 supplement (Gibco, stored at -20°C). Cells were plated on poly-D-lysine-coated coverslips mounted on MatTek dishes (with alphanumeric labels for CLEM experiments) at a density of 70000-80000 cells/cm2. Cultures were maintained at 37 °C and 5% CO2 by feeding them with the same medium every 3-4 days until transfection. Transfections were performed 12-18 days after plating by magnetofection using Combimag (OZ biosciences, stored at 4°C) and Lipofectamine 2000 (Invitrogen, stored at 4°C) according to manufacturer’s instructions. Imaging and optical measurements Live cell imaging was conducted between 17-20 days after plating. All experiments were performed at 37 °C in modified E4 imaging buffer containing (in mM) 120 NaCl, 3 KCl, 10 HEPES (buffered to pH 7.4), 2 CaCl 2 , 0.81 MgCl 2, and 10 Glucose, in the absence of TTX, even for homeostatic scaling-induced neurons, unless specified otherwise. Imaging was performed using a custom-built inverted spinning disk confocal microscope (3i imaging systems; model CSU-W1) with two cameras: an Andor iXon Life 888 electron multiplying charge-coupled device (EMCCD) camera for confocal fluorescence imaging and an Andor iXon3 888 camera for luminescence imaging. The Andor iXon3 888 EMCCD camera was selected for ultralow dark noise, further reduced by cooling to −100 °C. The speed of the Andor iXon3 888 camera used for luminescence measurements was 1 MHz with 1.00 gain and 1000 intensification. Image acquisition was controlled by SlideBook 6 software. Images were acquired with a Plan-Apochromat 63X/1.4 NA. Oil objective, M27 with DIC III prism, using a CSU-W1 Dichroic for 488/561 nm excitation with Quad emitter and individual emitters, at laser powers 50 ms exposure, 4.3 mW (488 nm) and 100 ms exposure, 5.5 mW (561 nm) for confocal fluorescence, and a 720 nm multiphoton short-pass emission filter for luminescence. During imaging, the temperature was maintained at 37 °C using an Okolab stage top incubator with temperature control. Mitochondria and spine imaging, and analysis Confocal and differential interference contrast (DIC) imaging was performed to acquire 20X images of the mitochondria (Mito-EGFP, 488 nm channel) and spines (PSD95-mCherry, 561 nm channel). These images along with additional orientation markings made on the coverslip served as correlative maps for downstream correlative light and electron microscopy (CLEM) analysis. For the same samples, high-resolution 63X confocal Z-stacks were acquired for mitochondria (Mito-EGFP, 488 nm channel) and spines (PSD95-mCherry, 561 nm channel), covering a 200 X 200 X 6 μm 3 imaging volume at a voxel resolution of 0.2 X 0.2 X 0.1 μm 3 . These 63X Z-stacks were used for detailed structural analysis of mitochondria and spine organization. For each neuron, fields of view were selected to maximize dendritic coverage while excluding the soma and ensuring capture of distal dendritic segments within 0-300 μm from the cell body. To improve segmentation quality and signal smoothness, the 3D stacks were upsampled (using a Python library) in the XY dimensions, resulting in an isotropic voxel size of 0.1 μm. The upsampled stacks were then z-score normalized and saved as 32-bit volumes to accommodate varying signal intensities across fields of view. Deconvolution was performed using Huygens (Scientific Volume Imaging, SVI) on the respective 488 nm and 561 nm channels, employing theoretical point spread functions (PSFs) generated within Huygens. The ‘acquity’ parameter was set to −100 to prioritize signal smoothness while preserving original resolution. The deconvolved Mito-EGFP and PSD95-mCherry stacks were subsequently maximum-projected to create 2D representations of mitochondrial morphology and spine distribution using a custom Python script. These were then combined into two-color composite stacks in Fiji, with each composite corresponding to a single neuron containing one or more of its dendrites. Dendrites were manually traced in Fiji using the segmented line tool, with the Mito-EGFP and PSD95-mCherry composite stack serving as a reference. Tracing was performed from the soma toward distal dendritic segments, and each trace was added to the ROI Manager. All dendritic branches were included, with each branch intersecting its parent at the branch point to ensure continuity across the dendritic tree. The segmented line tool generates traces as a series of straight-line segments connected by vertices. However, when drawn manually, these vertices are not uniformly distributed along the dendrite, resulting in segments of varying lengths. Uniform vertex spacing was required for downstream analyses. To address this, we used a custom ImageJ Macro to smooth and interpolate each trace using Fiji’s built-in Interpolate function, redistributing vertices at regular 1 μm intervals between the start and end points of the original trace. The same script was used to export a CSV file containing the unique dendritic branch ID for each ROI, along with the corresponding X and Y coordinates of the vertices in that branch. Using a custom Python script, we imported the maximum-projected mitochondrial and spine images and reconstructed the dendritic traces from the CSV files. These dendritic traces were then used as spatial guides for a custom segmentation pipeline. For mitochondrial segmentation, a sliding bin measuring 10 μm in length and 5 μm in width was moved along each dendritic trace at a step size of 1 μm. At each position, Otsu thresholding was applied to the histogram of pixel intensities within the bin, classifying each pixel as foreground or background. Due to the 90% overlap between consecutive bins (from a 1 μm step size relative to the 10 μm bin length), many pixels were included in multiple thresholding windows. This intentional redundancy allowed each pixel to be evaluated under varying local intensity contexts. The multiple foreground/background assignments for each pixel were aggregated, and the majority vote was used to assign the final label, increasing robustness and confidence in the segmentation. This localized, trace-guided thresholding strategy enabled precise binary segmentation of dendritic mitochondria by accounting for spatially varying intensity and reducing the risk of falsely including axonal mitochondria. For spine segmentation, a custom Python function identified puncta using a Difference-of-Gaussian (DoG) filter applied within a 5 μm width surrounding each dendritic trace along the dendrite in 1 μm steps. The intensity threshold for puncta detection was defined as one standard deviation above the mean intensity of all pixels within the 5 μm width surrounding each dendritic trace, allowing the detection to adapt to local signal variability and exclude background puncta unassociated with dendrites. While the DoG filter does not inherently produce segmentations, it provides the X and Y coordinates of detected puncta. For each detected coordinate, the custom function generated a circular region of interest (ROI) with a defined radius (0.6 μm) and applied Otsu thresholding to the histogram of pixel intensities within that ROI. This pipeline produced a precise binary segmentation of spine puncta, adapting to local intensity variations within each image and across different fields of view. To measure the length of individual mitochondria, a custom Python function skeletonized each segmented mitochondrial compartment. It then identified all endpoints of the skeleton and computed the longest possible (geodesic) path between any two endpoints, summing the actual interpixel distances along the path. This geodesic approach provided a more accurate measurement of mitochondrial length than simply counting the number of pixels. For spine size, we calculated the area of each segmented PSD95 punctum individually. To compute spine area density, we applied a similar mapping strategy as above: each PSD95 pixel was assigned to its closest point on the dendritic trace, and the total number of PSD95 pixels was summed within 10 μm bins. This approach yielded a measure of PSD95 area density along the dendritic length. For spine density, we counted the number of distinct PSD95 puncta falling within a 10 μm bin along the dendritic trace. Electrophysiology measurements and analysis Neurons cultured on 12-mm glass coverslips were transferred to the stage of an Olympus BX51WI microscope and maintained at room temperature in a recording buffer (identical to the E4 imaging buffer, osmolarity: 305–310 mOsm). Neurons expressing Mito-EGFP (mitochondria) and PSD95-mCh (spines) were identified via epifluorescence using a 470 nm and 561 nm LED light source and an X-Cite® 120Q illumination system (Excelitas, PA). Whole-cell voltage-clamp recordings were performed using patch pipettes (3–5 MΩ) pulled from borosilicate glass (Sutter Instruments) and filled with an internal solution containing (in mM): 145 K-gluconate, 14 phosphocreatine, 4 NaCl, 0.3 NaGTP, 4 MgATP, 3 L-ascorbic acid, and 10 HEPES (pH 7.3, 294 mOsm). Seal resistances ranged from 2–6 GΩ. Recordings were excluded if the series resistance exceeded 25 MΩ; series resistance was left uncompensated. After establishing whole-cell configuration, neurons were held at −70 mV, and 3– 5-minute-long recordings were acquired beginning 3–5 minutes post-break-in. Data were recorded using a MultiClamp 700B amplifier (Molecular Devices) and digitized at 10 kHz with custom C# software (FLIMage; https://github.com/ryoheiyasuda/FLIMage_public ) on a Windows PC. Neurons with resting membrane potentials more positive than −50 mV were excluded from analysis. Recordings were low-pass filtered at 1 kHz using a Bessel filter in ClampFit (version 11.4.2.04, Molecular Devices). mEPSC events were detected using the template search function, and only events with amplitudes ≥ 5 pA were included. All mEPSC events from each neuron were pooled by condition (e.g., Control or homeostatic scaling) and analyzed as cumulative distributions representative of the condition. Electron microscopy sample preparation Neuronal cultures were fixed following live-cell imaging for 2 hours at room temperature in a 2% paraformaldehyde (Electron Microscopy Sciences) and 2.5% glutaraldehyde (Electron Microscopy Sciences) solution, prepared in cacodylate buffer (pH 7.4 and osmolarity-adjusted, concentration adjusted to match the osmolarity of the E4 imaging buffer). After fixation, the dishes were washed in 0.14 M or 0.17 M osmolarity-adjusted cacodylate buffer for 2 hours at room temperature. Samples were post-fixed by incubation with 1 ml of 2% osmium tetroxide (OsO₄) (Electron Microscopy Sciences) in osmolarity-adjusted cacodylate buffer for 1 hour on ice. After the initial incubation, 1 ml of 5% potassium ferrocyanide (Sigma Aldrich) was added directly to the OsO₄ solution in the dish, yielding a final working concentration of 1% OsO₄ and 2.5% potassium ferrocyanide. Neuronal cultures were incubated with this mixture for an additional 1 hour on ice. The solution was then removed and replaced with 1 ml of fresh 2% OsO₄, and samples were incubated for another 1 hour on ice. Following fixation, neuronal cultures were washed three times in ultrapure water (Sigma Aldrich) for 1 hour at room temperature. Neuronal cultures were then stained with 1% aqueous uranyl acetate (Electron Microscopy Sciences) for 1 hour at room temperature, followed by a 1-hour wash in ultrapure water. Neuronal cultures were dehydrated through a graded ethanol series (30%, 50%, 70%, 90%, 99.5%, and 100%), with each step carried out for 30 min at room temperature with gentle shaking (30 rpm). Neuronal cultures were infiltrated with Durcupan ACM resin (Sigma Aldrich) without the accelerator component C, through sequential ethanol: resin mixtures (1:3 overnight, 1:1 for 6 hours, and 3:1 overnight), followed by incubation in 100% Durcupan overnight with the accelerator component C under vacuum in a desiccator. To flatten the resin surface, a small piece of ACLAR (Electron Microscopy Sciences) film was placed on top of each well. Dishes were polymerized at 60°C for 72 hours. After polymerization, dishes were left at room temperature overnight. Orientation markings made on the coverslips during live-cell imaging were copied onto the ACLAR sheet. Coverslips were carefully removed using a feather blade (Electron Microscopy Sciences), and the resin disk (in the shape of the MatTek well) containing the embedded neuronal cultures was pushed out and separated from the dish along with the marked ACLAR sheet. Ultrathin sectioning of the target neuron Alphanumeric labels from the coverslip were imprinted onto the surface of the resin disk together with the cultured neurons. X and Y coordinates from live neuronal confocal imaging were extracted to generate a map of relative neuron positions. ACLAR film markings were used to orient the disk, and neurons of interest were located under a stereomicroscope. For each identified neuron, the corresponding alphanumeric position was recorded onto the map to design a cutting strategy. Boxes were drawn around neurons of interest, allowing sufficient margins to account for extraction variability. Each boxed region was carefully excised from the resin disk using a razor blade. Excised resin pieces were mounted onto separately prepared cylindrical resin blocks using cyanoacrylate (Gorilla Glue). For each neuron of interest, a trapezoidal label was placed on the confocal image to define the target region. A matching trapezoid was scratched around the neuron on the resin block and trimmed precisely using a Diatome trimming knife. Serial sections were cut at a thickness of 120 nm or 150 nm using a Diatome Ultra knife. Sections were collected onto formvar-or pioloform-coated (Electron Microscopy Sciences, Ted Pella, Inc.) 3 mm copper grids (Electron Microscopy Sciences) with 2 mm apertures. Electron tomography of target mitochondria Transmission Electron Microscopy (TEM) correlation was performed on an FEI Tecnai G2 BioTwin. All serial sections for a target region were checked, and dendrites from a neuron of interest were visually identified using the live neuron DIC images as correlative guides. PSD95-positive spines and mitochondria at their base were visually identified using 63X live neuron confocal images. TEM reference images were taken at 120 kV using a Veleta EMCCD camera. Given that mitochondria frequently span multiple serial sections and are often absent from the section containing the spine, the section displaying the most prominent mitochondrial cross-section was selected and imaged at a ∼2 × 2 μm field of view and ∼0.48 nm pixel resolution, ensuring optimal capture of ultrastructural features during tilt-series acquisition. Dual-axis tilt series (ranging from -60° to +60° or -68° to +68°) were acquired at 120 kV using SerialEM 61 , 62 . 90° rotations between the two axes were manually performed. 6 out of 24 dual-axis tilt series acquisitions were made at Thermo Fisher Scientific’s Hillsboro NanoPort on a Talos L120C TEM at 120 kV using Thermo Scientific Tomo software. Tilt-series preprocessing and tomogram reconstruction Tilt-series preprocessing and tomographic reconstruction were performed using IMOD 63 . All tilt series were denoised using IMOD’s mtffilter function with a low-pass cutoff of 0.166 and a Gaussian roll-off of 0.08 to suppress high-frequency noise. The tilt series were then downsampled to a 2 nm pixel resolution using IMOD’s squeezevol function to improve contrast. Using Etomo in IMOD, both axes were independently coarse-aligned using cross-correlation and fine-aligned using patch tracking or 10 nm gold fiducials, whenever fiducials were used. Tomograms for each axis were reconstructed using the simultaneous iterative reconstruction technique (SIRT) algorithm within Etomo for 20 iterations, with a cutoff of 0.4 and a Gaussian falloff of 0.08 for high-frequency noise suppression. Dual-axis tomograms were then combined using automatic patch fitting in Etomo’s tomogram combination panel. Combined dual-axis tomograms were flattened within Etomo’s post-processing panel to correct for warping caused by prolonged electron beam exposure during dual-axis tilt-series acquisitions. The resulting tomograms measured approximately 982 × 982 × 50 voxels (XYZ) at an isotropic voxel size of 2 nm, yielding ∼1.96 × 1.96 × 0.1 μm³ volumes that captured partial 3D cross-sections of mitochondria. We tested the feasibility of increasing volumetric coverage and acquired two serial tomograms for one mitochondrion and a two-tile montage tomogram for another. Although it was technically feasible to acquire larger volumes of individual mitochondrial regions, this approach introduced additional challenges during post-processing. We therefore limited subsequent acquisitions to single-section, single-tile volumes, prioritizing the sampling of a broader population of mitochondrial regions to better capture biological variability rather than extending volume coverage of the same mitochondrion. Deep learning-based segmentation of mitochondrial inner structure The inner mitochondrial membrane (IMM) and the outer mitochondrial membrane (OMM) define the boundaries enclosing the intermembrane space (IMS), including the cristae volume. To segment the IMS, we reasoned that outlining this compartment would inherently delineate both the IMM and OMM simultaneously. Reconstructed tomograms were imported into ORS Dragonfly, where we selected sub-volumes of varying sizes that captured the diversity of IMS orientations across different tomograms. Within these sub-volumes, the IMS was manually segmented to generate ground truth training data. From Dragonfly, we exported three aligned stacks: the original electron microscopy (EM) data, a binary mask defining sub-volume boundaries, and a binary segmentation of the IMS within those boundaries. A custom-written Python script used the boundary masks to extract the corresponding EM data and IMS-labeled sub-volumes and saved them as NumPy arrays for downstream analysis (see below). We built a custom deep learning network based on the 3D residual U-Net architecture with anisotropic convolutions 64 , integrated with the three-axis prediction strategy 65 , along with additional modifications (concatenation instead of summation for skip function; median, average, and max pooling between encoder blocks). Accurate IMS segmentation required exposure to the range of orientations present in tomographic data. Therefore, we used the NumPy arrays of manually segmented sub-volumes to generate three sets of training data: EM input patches and corresponding IMS ground truth masks from the XY plane (viewed along the Z-axis), the XZ plane (viewed along the Y-axis), and the YZ plane (viewed along the X-axis). XY-plane patches measured 128 × 128 × 32 voxels, while XZ-and YZ-plane patches were 128 × 32 × 32 voxels. All patches were extracted with 50% overlap to effectively double the training data. Because the Z-dimension of our tomograms was ∼50 voxels, patches extracted in the XZ and YZ planes had reduced dimensions along their second axis. To match the input shape required by the network (128 × 128 × 32), XZ and YZ patches were zero-padded along this axis. In addition to the EM input and IMS mask patches, the patch generation script also created corresponding loss masks for the XZ and YZ views. These loss masks marked the zero-padded regions, instructing the network to ignore these areas during training and preventing it from learning spurious features from blank input regions. To improve the network’s robustness in learning IMS features, we implemented on-the-fly data augmentation. To enable the network to recognize how the same IMS structure appears from different viewing angles, we generated 16 possible patch configurations, derived from all permutations of two rotation states (0° or 90°) and three mirroring operations (horizontal flip, vertical flip, and z-axis inversion) 66 . During training, each patch was passed through an augmentation script that randomly selected one of the 16 configurations and reoriented the entire input patch set (EM, IMS mask, and loss mask). To further improve generalization across varying image qualities, we introduced simulated imaging artifacts. The augmentation script applied Gaussian noise or Gaussian blur with a 10% probability and randomly removed 1–3 consecutive slices (with a 5% chance) from each input patch to mimic missing slices. Training was performed in batches of 32 patches. For each patch, the loss was computed as the masked binary cross-entropy between the predicted IMS and the ground-truth mask. The network weights were updated after each batch using the average batch loss. An iteration consisted of one full pass through all patches along a given axis. Training cycled through the three orthogonal axes (XY, XZ, YZ) and continued for multiple such three-axis iterations until the model stopped improving. Loss was monitored manually and training was stopped after the loss plateaued. The trained model was applied to all tomograms to infer the location of the intermembrane space (IMS). The data augmentation and three-axis strategy used during training were retained during inference. From each tomogram, patches of 128 × 128 × 32 voxels for the XY plane and 128 × 32 × 32 voxels for the XZ and YZ planes were extracted with 50% overlap, following the same procedure as in training. For each axis, every patch was passed through the network in all 16 possible orientations, producing 16 predictions that were combined using a median projection. The resulting median-projected patches were stitched together using a custom normalized 3D Gaussian-like bump function, which assigned higher weights to the center of each patch to minimize edge artifacts. The stitched predictions from the three orthogonal axes were then combined by averaging across axes. This inference strategy yielded highly accurate and smooth IMS segmentations, eliminating the jagged appearance typically introduced by manual tracing between slices, enabling the downstream generation of smooth 3D meshes. To distinguish between the inner boundary membrane (IBM, the portion of the IMM parallel to the OMM) and the cristae membrane (CM, which comprises the invaginations of the IMM), it was necessary to create IBM segmentations that exclude cristae folds. We define IBM segmentation as one that encompasses the mitochondrial matrix and the cristae lumen, while excluding the space between the OMM and IBM, referred to as IBM volume. Using the IMS segmentations generated by deep learning as a guide, we manually segmented the IBM volume in ORS Dragonfly for a subset of tomograms and trained a separate 3D residual U-Net (in the XY plane only) for IBM segmentation. Instead of using the original EM images as input, we used the IMS predictions to simplify the input space and reduce the complexity of features present in raw EM data. This approach enabled the network to efficiently learn to reconstruct the IBM volume based on IMS geometry, effectively learning to fill in the IBM region within the predicted IMS. A voxel-wise union between the IBM and IMS segmentations from each tomogram produced the OMM segmentation, representing the total mitochondrial volume. All deep learning predictions were manually proofread in ORS Dragonfly to correct any missing IMS or IBM labels. These omissions typically occurred in regions where mitochondrial membranes were oriented parallel to the tomogram plane (e.g., the top or bottom surfaces of mitochondria or flat cristae lamellae facing the XY plane). Such membrane orientations are poorly contrasted in tilt-series tomograms and pose challenges even for manual interpretation. We manually filled in these regions to the best of our ability. Additionally, the deep learning model occasionally segmented IMS within axonal mitochondria that appeared only partially in the field of view; these erroneous segmentations were manually removed. Endoplasmic reticulum and ribosome segmentation Endoplasmic reticulum (ER) and ribosomal volumes were manually segmented in ORS Dragonfly and exported as stacks. ER was identified as long tubules of membranes present in the dendrite 67 , 68 . Ribosomes could only be confidently identified when present in or near a cluster. Thus, we only segmented ribosomes in or near clusters. 3D mesh processing and morphometric analysis of cristae, endoplasmic reticulum, and ribosomes Meshes for the IMS, IBM, OMM, endoplasmic reticulum (ER), ribosomes, and a bounding box representing the full tomogram volume were generated using the marching cubes algorithm in the scikit-image Python package 69 , 70 applied to their respective segmentations via a custom Python script and saved as STL files. All meshes were then imported into the open-source platform Blender using the GAMer2 plugin. Within Blender, we applied the built-in ‘remesh’ and ‘triangulate’ functions to all meshes. Voxel-based remeshing was performed at a resolution of 2 nm for ribosomes and 4 nm for the IMS, IBM, OMM, and ER. Subsequent triangulation yielded lighter (i.e., reduced vertex count) and smoother meshes without compromising biological fidelity. All processed meshes were exported from Blender and saved as STL files. At this stage, the IMS mesh includes both the outer (OMM) and inner (IMM) mitochondrial membrane surfaces. In a fully reconstructed mitochondrion, these surfaces would naturally separate; however, since our reconstructions are partial, we implemented an alternative strategy to isolate the IMM. Using a custom Python script, we performed a Boolean subtraction of the IMS mesh from the IBM mesh. This operation produced the IMM mesh by incorporating cristae invaginations into the IBM surface while excluding the OMM. The resulting IMM mesh was imported back into Blender, where the GAMer2 plugin was used to calculate the first principal curvature (k₁). The curvature values were exported from Blender as .pkl files, each containing a k₁ value for every vertex in the IMM mesh. The k₁ values were imported into a custom Python script and assigned to the IMM mesh structure within the Python environment, where they were carefully preserved throughout all downstream mesh operations. At this stage, the IMM mesh included the IBM surface (from the original IBM mesh before Boolean subtraction), the CM surfaces, and the top and bottom capping surfaces corresponding to the start and end of the tomogram. These capping surfaces were required to maintain a closed, watertight mesh for volumetric Boolean operations. However, for surface area measurements, the capping surfaces needed to be removed. To achieve this removal, we used the bounding box mesh of the tomogram to identify and delete any IMM mesh vertices located within 5 nm of the bounding box surface, resulting in an open IMM surface mesh. To isolate the CM surface from the IBM, we used a custom Python function that referenced the original IBM mesh (before Boolean subtraction). Any region of the IMM mesh located more than 5 nm away from IBM mesh vertices was classified as CM, resulting in a CM surface mesh. Cristae junctions were determined automatically using the edges at the interface of CM and the rest of the IMM through a custom script. If these edges formed a ring, the centroid of the ring was marked as a point. Junction density was calculated as the number of points per total mitochondrial volume within the OMM mesh. Using a Python script, we measured the CM surface area from the CM mesh, the IMM surface area from the IMM mesh, and the OMM volume (i.e., the total mitochondrial volume within the tomogram) from the OMM mesh. To quantify the surface area of highly curved cristae, we further extracted regions of the CM mesh with k₁ values greater than 80 cm -1 . Using a Python script, we quantified ERMCS by separating the ER mesh within 20 nm of the OMM and measuring the surface area of the separated mesh, normalized to the total surface area of the ER. For quantifying ribosome density, we utilized the volume of the smallest single 3D mesh component of ribosomal clusters as unit ribosome volume and divided all cluster volumes by the unit volume and normalized per dendritic length, determined as the convex hull length of the mitochondria. The ribosome cluster to OMM distance was measured as the shortest distance between the closest vertices of the ribosomal mesh to the OMM mesh. DNA-PAINT imaging and analysis Dissociated rat primary hippocampal neuron cultures were prepared and maintained as previously described 3 . In brief, hippocampi from postnatal day 1 rat pups of either sex (RRID:RGD_734476; strain Sprague-Dawley) were dissected and dissociated by incubating with L-cysteine-papain solution at 37 °C before being plated onto 35-mm MatTek dishes (MATTEK, A Bico Company) previously coated with poly-d-lysine. Cultured neurons were incubated in Neurobasal A medium (Gibco) supplemented with B-27 (Invitrogen) and Glutamax (Gibco) at 37 °C and 5% CO 2 for 18 days before imaging. Transfections were carried out in neuron cultures at 15 DIV with Lipofectamine 2000 (Invitrogen) diluted in Neurobasal A medium and Glutamax (Gibco) according to the manufacturer’s instructions. Briefly, neuronal cultures were transfected with 1 μg total DNA using PSD95FingR-BFP (to detect spines) and OMM-EGFP (to detect mitochondrial outer membrane). On DIV 18, homeostatic scaling was induced by incubating the neuronal cultures with 2 μM TTX (1 mg/ml stock made in water) for 24 hours. On DIV 19, the neuronal cultures were fixed in paraformaldehyde-sucrose (4% paraformaldehyde; Thermo Scientific, 4% sucrose in phosphate-buffered saline containing MgCl 2 and CaCl 2 ) at room temperature for 20 min, washed three times in PBS. Fixed neuronal cultures were permeabilized with 0.5% Triton X-100 in 1X PBS, pH 7.4, for 15 min and blocked in PBS containing 4% goat serum (Gibco) for 1 hour. For DNA PAINT-based single-molecule localization microscopy, primary antibody staining was performed overnight in PBS containing 4% goat serum (Gibco), followed by three 5-minute washes with PBS. Primary antibodies used were rabbit anti-eGFP (1:500) (Abcam, Ab72226) to detect the mitochondrial outer membrane and mouse anti-OSCP (A8) (1:100) (Santa Cruz, sc-365162) to detect ATP synthases. The respective secondary antibodies were used: anti-rabbit conjugated to a P2-docking oligonucleotide (P2, 1:1000) (Massive Photonics) and anti-mouse conjugated to a single-stranded P1 docking oligonucleotide (P1). After three 5-minute washes in PBS, neurons were briefly fixed for 5 min. Gold fiducial markers were sparsely plated by a 20-minute incubation with 200 μl PBS buffer containing 20 μl a gold nanoparticle stock solution (90 nm, A1190, Nanoparz). For DNA-PAINT imaging, 1000 μl of imaging buffer containing 4 μl of P1 and 0.2 μl of P2 imager oligonucleotide conjugated with Atto655 and Cy3B (Massive Photonics), respectively, was used. DNA-PAINT imaging was performed using an Olympus IX83 with Abbelight SAFe Nexus, and a dual-CMOS, Hamamatsu C15440-S0UP camera (Abbelight) with the Abbelight Neo software package. A 100X oil-immersion objective (Olympus Immoil-F30CC; Numerical aperture 1.518) was used in combination with a motorized TIRF illuminator. Wide-field images of PSD95FingR-BFP were obtained with epifluorescence illumination at 15 to 20 mW (405 nm) laser power at the back aperture. OMM-EGFP was imaged using TIRF illumination with 10 to 15 mW (488 nm) laser power at the back aperture. Dual-camera DNA PAINT acquisition was performed with HILO illumination and laser power of 20 to 30 mW at the back aperture for 640 nm and 561 nm channels. Time-lapse datasets with 60,000 frames and 1x1 binning were acquired for each localization with a 5-Hz frame rate and 10-MHz camera readout bandwidth. DNA-PAINT data were processed using a modified version of a previously published protocol 71 . Acquisitions were first reconstructed in Picasso:Localize using maximum likelihood estimation (MLE) with integrated Gaussian PSF fitting. Localization filtering criteria included localization precision < 19.4 nm (0.2 pixels), PSF width < 2.2 standard deviation, ellipticity 1000, and net gradient > 8000 for the 640 nm channel and > 3000 for the 561 nm channel. These filters retained only high-confidence localizations. Drift correction was performed in Picasso:Render in two stages, an initial coarse correction using redundant cross-correlation (RCC), followed by manual refinement using selected gold fiducials. The 561 nm channel was subsequently aligned to the 640 nm channel using RCC-based registration. A custom Python script was used to render the 561 nm mitochondrial channel into a TIFF image. This TIFF file was opened in Fiji, where mitochondria were manually segmented using the area selection tool and converted into binary masks. The binary masks were used to retain only the 640 nm ATP synthase localizations within the mitochondria using a custom masking script. Because individual hybridization events (i.e., binding of imager strands to the docker strands on antibodies) can persist over multiple frames, localizations were linked into hybridization events in Picasso:Render. Linking parameters included a maximum spatial distance of 19.4 nm (∼1-2 X NeNA: nearest neighborhood analysis values estimated from the localizations, to avoid overlinking tightly packed ATP synthases, which are typically ∼12-20 nm apart) and allowed for up to 20 dark frames. The resulting linked localizations represented discrete hybridization events. To identify ATP synthase-associated hybridization events and suppress noise, we applied HDBSCAN clustering (via a custom Python script) using a minimum cluster size of 5 linked localizations and the ‘leaf’ method to promote homogeneity. The smallest cluster population peaked around 6 ± 1 localizations per cluster and was interpreted as representing single ATP synthase complexes, given that multiple temporally separated localizations correspond to repeated imager binding to the same docker strand 3 . Clusters containing 10 or more localizations were classified as multimeric ATP synthase clusters, representing higher-order ATP synthase assemblies. The estimated copy number for each cluster was calculated by dividing its number of localizations by 5 (i.e., the minimum in single-copy clusters). To assess local copy number density, we calculated the number of ATP synthase copies per 2 μm mitochondrial segment using a custom Python script. For each segment, the number of ATP synthase copies in single-copy and multimeric clusters was measured. The mitochondrial area within the same segment was calculated using the binary mask, and the copy number density (including multimer-to-single-copy ratio) was normalized to mitochondrial area. ATP mito and ATP spine imaging and analysis Neurons were transfected with Mito-ATP (Mito-mCh-Luc) and PSD95FingR-EGFP for imaging mitochondrial ATP and spines, respectively. Neurons were transfected with Spn-ATP (Homer2-mOrg-Luc) and Mito-EGFP for imaging spine ATP and mitochondria, respectively. Transfected neurons were identified by mCh fluorescence of the Mito-ATP or mOrg fluorescence of the Spn-ATP. Test images were acquired simultaneously in the 561 nm (mOrg/mCh) and 488 nm (PSD95FingR-EGFP/Mito-EGFP) channels at exposure times of 100 ms for the 561 nm and 50 ms for the 488 nm channels, and bioluminescence images were simultaneously acquired over 60 s for Spn-ATP and 20 s for Mito-ATP. Only neurons exhibiting both mature spines and healthy dendritic mitochondria were selected for imaging. For each condition, a 3-minute 2D time series of alternating luminescence and fluorescence frames was acquired. To characterize the presence of mitochondria at the base of the spines, a 3D Z-stack of the same neurons was acquired in the 561 nm and 488 nm channels, with a step size of 0.1 μm spanning a range of 6 μm, following the 2D ATP imaging. All image analysis was performed using custom batch-processing macros in Fiji. Each time series was average-projected and background-subtracted. To correct for misalignment between confocal fluorescence and luminescence images, luminescence images were aligned to fluorescence images using the Multichannel SIFT alignment plugin. For Mito-ATP, mitochondria were marked in the 561 nm (mCh) channel using 2 μm diameter circular regions of interest (ROIs). Each circle was then Otsu-thresholded to generate a mitochondrion-shaped region of interest, spanning a 2 μm segment of the mitochondrion. The mean mCh fluorescence and corresponding luminescence intensity were then measured for each mitochondrial ROI to calculate the luminescence-to-fluorescence ratio (L/F). For Spn-ATP, individual spines were manually marked in the mOrg channel using 1 µm diameter circular ROIs. The mean mOrg fluorescence and corresponding luminescence intensity were measured for each marked spine to calculate the luminescence-to-fluorescence ratio (L/F). pH mito and pH spine measurements Neurons were transfected with Mito-pHluorin and PSD95FingR-BFP for imaging mitochondrial pH and spines, respectively. Neurons were transfected with cyto-pHluorin and Mito-DsRed for imaging spine pH and mitochondria, respectively. Transfected neurons were identified by pHluorin fluorescence in the 488 nm channel. Only neurons exhibiting both mature spines and healthy dendritic mitochondria were selected for imaging. Time-lapse pHluorin imaging was performed at 3 frames per second. After ∼50 frames of imaging, the E4 imaging buffer was replaced with an NH₄Cl solution (in mM): 50 NH₄Cl, 70 NaCl, 3 KCl, 2 CaCl₂, 0.81 MgCl₂, 10 Glucose, 10 HEPES; pH adjusted to 7.4 at 37 °C. Time series frames were aligned using the Multichannel SIFT alignment plugin in Fiji, background-subtracted, and averaged projected. For mitochondrial pH measurements, mitochondria were marked in the pHluorin channel using 2 μm diameter circles. Each circle was Otsu-thresholded to define a mitochondrion-shaped region of interest spanning a 2 μm mitochondrial segment. Mean pHluorin intensity was measured across frames for each mitochondrial ROI. For spine pH measurements, individual spines were manually marked in the pHluorin channel using 1 μm diameter circles. Mean pHluorin intensity was measured across frames for each marked spine. The mean intensity of the first ten frames was used as the steady-state fluorescence intensity corresponding to the steady-state pH. The peak intensity upon NH₄Cl application was defined as the fluorescence corresponding to a pH of 7.4. The steady-state spine or mitochondrial pH values were calculated using a modified Henderson-Hasselbalch equation 5 , 36 as follows: where, pKa = 7.1 for pHluorin, F pH is the fluorescence intensity of pHluorin at the measured pH, and FpH 7.4 is the peak fluorescence intensity of pHluorin at pH 7.4 upon adding NH 4 Cl. pH correction for ATP spine measurements Since the pH of the spines upon homeostatic scaling was lower, pH 6.75, than the pH of the control spines, pH 6.95, the luminescence and fluorescence intensities for homeostatic scaling were pH-corrected as follows: where 6.75 is the baseline pH of spines upon homeostatic scaling, L pH6.75 is the luciferase luminescence intensity at pH 6.75, and pKa is 7.03 for luciferase. where 6.75 is the baseline pH of spines upon homeostatic scaling, F pH6.75 is the mOrg fluorescence intensity at pH 6.75, and pKa is 6.5 for mOrg. L pH6.95 /F pH6.95 was calculated to obtain the L/F ratio proportional to ATP spine . Since the mitochondrial pH did not differ significantly between control and homeostatic scaling, we did not pH correct the L/F ratio proportional to ATP mito . Local mitochondrial volume quantification from MICrONS phase 1 meshes Three-dimensional reconstructions of neuronal and mitochondrial meshes were obtained from the MICrONS Phase 1 dataset, along with annotation tables containing mitochondrial IDs, neuronal IDs, synapse IDs, spine size (cleft_vx x 4 x 4 x 40 nm). Of the 112 functionally characterized neurons, 107 had mesh data suitable for analysis; the remaining five were excluded due to mesh processing errors. To quantify local mitochondrial volume at the base of dendritic spines, we developed a custom pipeline. This process required 3D neuronal skeletons that extended along the dendritic arbor, along with fine branches that extended into the spine shafts. Although the MICrONS dataset provides skeletons generated using the TEASAR algorithm 72 , these lie along the mesh surface rather than its center. Therefore, we reskeletonized each neuron using the wavefront propagation algorithm in the Skeletor version 1.3.0 Python package 73 . Prior to skeletonization, meshes were subdivided to increase resolution. The resulting skeletons consist of nodes and edges traversing the center of the neuron, with a local radius value assigned to each node. Manual inspection revealed that terminal skeleton branches between 1–6 μm in length reliably corresponded to spines. However, branch nodes where the spine skeletons emerged from dendrites often showed a small offset toward the spine shaft. To correct this offset, we wrote a custom Python script that separated all spine skeletons from their parent dendritic skeletons and smoothed the dendritic skeletons to remove offsets. We then verified spine identity by cross-referencing each spine skeleton with the MICrONS postsynaptic site coordinates, retaining only spine skeletons located within 1.5 μm of a MICrONS-annotated postsynapse and assigning their corresponding synapse ID. For each validated spine skeleton, we estimated the point where the spine mesh vertices meet the dendritic shaft mesh vertices. This point lies one dendritic radius (mean radius at the branch point ± one vertex) from the last node of the spine’s skeleton, towards the second-last node. This estimated point was then refined by computing the centroid of the 20 nearest vertices on the spine surface mesh. We defined the spine base coordinate as the closest node on the smoothed dendritic skeleton to this centroid. This approach yielded spatially accurate spine base coordinates embedded within the dendrite, which were used for all subsequent measurements of mitochondrial volume. To identify wider mitochondrial regions, we defined them as areas where the local width was substantially greater than the average width of the entire mitochondrial compartment. Specifically, we used a threshold of one standard deviation above the mean width. Using a custom script, each mitochondrial mesh was skeletonized, and the local radius (distance from mesh surface to closest skeleton node) was computed. All mesh vertices with radii exceeding one standard deviation above the mean were retained. These high-radius-value vertices marked high-width regions (and the rest were marked as low-width regions), but existed as disconnected surface fragments. To convert them into volumetric meshes, we grouped spatially contiguous or nearby fragments (within 200 nm) and encapsulated each group with a bounding sphere. Final high-width mitochondrial region volumes were generated via Boolean intersection between these spheres and the original mitochondrial mesh. To measure high-width mitochondrial or low-width mitochondrial region volume near spine bases, using a custom script, we generated 1 μm-radius spherical meshes centered at each spine base coordinate. For computational efficiency, we used spatial proximity to pre-filter mitochondria: any mitochondrial mesh with at least one vertex within 1 μm of a spine base was assigned to that spine’s synapse ID. Boolean intersection was then performed between each 1 μm sphere and its associated low-width mitochondria or high-width mitochondrial region, and the volume of the intersected region was computed. Since multiple spines can originate from proximity to each other and a 1μm sphere at the base of each spine would essentially measure the same mitochondrial volume, we grouped spine bases using DBSCAN clustering. Spine sizes were taken directly from the MICrONS data. The local spine size was calculated as the sum of spine sizes from all spines in a cluster. The number of spines in the cluster was considered the number of spines around the high-width mitochondrial region if a wider region was within 1 μm of the spine base at the center of the cluster. Local mitochondrial volume was determined as the volume of mitochondria within 1 μm of the spine base at the center of the cluster. To detect high-width mitochondrial regions in confocal data, we developed a custom Python function that first skeletonized each segmented mitochondrion, then assigned every mitochondrial pixel to its nearest point on the skeleton. For each skeleton point, the function measured the distance to the five nearest border pixels and computed the mean of those distances. This mean distance was then assigned to all mitochondrial pixels associated with that skeleton point, generating a local width map along the length of the mitochondrion. The Python function compared each pixel’s assigned width value to the average width of that mitochondrial compartment. Any pixel with a width value greater than one standard deviation above the mean was classified as part of a high-width mitochondrial region, and the rest as low-width mitochondrial regions. The number of spines was calculated using the spine segmentations generated in the confocal analysis for Fig. 1 . This approach allowed us to isolate and segment portions of mitochondria that were substantially wider than the rest of the structure, providing a quantitative measure of local signatures of potential mitochondrial structural remodeling. Quantification and statistical analysis All statistical analyses were performed using R (v4.3.0). Measurements made per spine or mitochondria were averaged per neuron to account for cell-to-cell variations unless mentioned otherwise. Mann–Whitney U test was used when data did not pass the Shapiro–Wilk normality test. For normally distributed data, Welch’s t-test was applied when data showed unequal variance, as determined by Levene’s test. For normally distributed data with equal variance, a student’s t-test was used. Linear model or linear mixed-effects models (LMMs) were fitted with the lmer() function from the lme4 package (v1.1-34). Random effects were specified based on the structure of the data, typically including random intercepts for date of the biological replicates, nested neuronal identity, and, where applicable, nested dendritic identity. Fixed effects captured the primary experimental variable condition, Control or homeostatic scaling. Interaction terms between condition and continuous variables were included to test whether the slope of the relationship differed across conditions. Model significance was assessed using Satterthwaite’s approximation for degrees of freedom, as implemented in the lmerTest package (v3.1-3). Estimated marginal means and pairwise contrasts were obtained using emmeans (v1.10.0), with Tukey correction for multiple comparisons. Statistical details of experiments, including statistical tests and n and p values, are mentioned in the figure legends. Each animal corresponds to one weekly batch of neuronal culture preparation. Data visualizations were generated using ggplot2 (v3.5.0) and ggbeeswarm (v0.7.2). In all the figures, the box whisker plots represent the median (line), mean (point), 25th-75th percentile (box), 10th-90th percentile (whisker), 1st–99th percentile (X), min-max (_) ranges, and individual data points. For regression models, fixed-effect prediction lines and confidence ribbons were computed manually from model coefficients and variance-covariance matrices. A p-value of less than or equal to 0.05 was considered significant for all statistical tests. No statistical method was used to predetermine the sample size. Sample sizes were similar to or larger than those reported in the previous publications in the field and sufficient for our claims based on statistical significance. Acknowledgments We thank C. Hanus for the PSD95-mCherry plasmid; J. de Juan-Sanz for Homer2-mOrange2-luc plasmid; and G. Ashrafi for mito-pHluorin plasmid. PSD95FingR-EGFP plasmid was a gift from D. Arnold (Addgene 46295). We are thankful to S. Perez, D. Bhuyan, J. Patarroyo, O. Bapat, and R. Fan for preparing cultured hippocampal neurons and for technical assistance; D. Guerrero-Given and C. I. Thomas at the MPFI Imaging Center Core for help with the transmission electron microscope and the Huygens license; C. Bouchet-Marquis from ThermoFisher for the initial dual-axis tilt series acquisitions; J. Yu at the MPFI Molecular Virology Core for assistance with molecular cloning; and K. Harris, D. Fitzpatrick, H. Inagaki, and all members of the Rangaraju Lab for critical input on the manuscript. CS is supported by the Lundbeck Foundation (Lundbeckfonden grant no. R361-2020-2654), the Novo Nordisk Foundation (NNF23OC0085864), the Danish National Research Foundation (DNRF133), and Independent Research Fund Denmark (Danmarks Frie Forskningsfond; 4283-00041B and 5253-00010B). VR is supported by the Max Planck Society, Louis D. Srybnik Foundation, F.O.R.E Foundation, Chan Zuckerberg Initiative DAF an advised fund of the Silicon Valley Community Foundation grant numbers 2023-331775 and 2024-349543, and the NIH Director’s New Innovator Award (DP2 MH140148-01). Funder Information Declared National Institutes of Health, https://ror.org/01cwqze88 , DP2 MH140148-01 Chan Zuckerberg Initiative (United States), https://ror.org/02qenvm24 , 2023-331775 , 2024-349543 Max Planck Society, https://ror.org/01hhn8329 Footnotes ↵ ‡ Lead contact. References 1. ↵ Turrigiano , G. G. , Leslie , K. R. , Desai , N. S. , Rutherford , L. C. & Nelson , S. B . Activity-dependent scaling of quantal amplitude in neocortical neurons . Nature 391 , 892 – 896 , doi: 10.1038/36103 ( 1998 ). OpenUrl CrossRef PubMed Web of Science 2. ↵ Schanzenbacher , C. T. , Sambandan , S. , Langer , J. D. & Schuman , E. M . Nascent Proteome Remodeling following Homeostatic Scaling at Hippocampal Synapses . Neuron 92 , 358 – 371 , doi: 10.1016/j.neuron.2016.09.058 ( 2016 ). OpenUrl CrossRef PubMed 3. ↵ Sun , C. et al. The prevalence and specificity of local protein synthesis during neuronal synaptic plasticity . Sci Adv 7 , eabj0790 , doi: 10.1126/sciadv.abj0790 ( 2021 ). OpenUrl CrossRef PubMed 4. ↵ Bapat , O. et al. VAP spatially stabilizes dendritic mitochondria to locally support synaptic plasticity . Nat Commun 15 , 205 , doi: 10.1038/s41467-023-44233-8 ( 2024 ). OpenUrl CrossRef PubMed 5. ↵ Ghosh , I. F. , R. ; Shah , M. ; Bapat , O. ; Rangaraju , V. Synapses drive local mitochondrial ATP synthesis to fuel plasticity . bioRxiv , doi: 10.1101/2025.04.09.648032 ( 2025 ). OpenUrl Abstract / FREE Full Text 6. ↵ Rangaraju , V. , Lauterbach , M. & Schuman , E. M . Spatially Stable Mitochondrial Compartments Fuel Local Translation during Plasticity . Cell 176 , 73 – 84 e15 , doi: 10.1016/j.cell.2018.12.013 ( 2019 ). OpenUrl CrossRef PubMed 7. ↵ Kuhlbrandt , W . Structure and function of mitochondrial membrane protein complexes . BMC Biol 13 , 89 , doi: 10.1186/s12915-015-0201-x ( 2015 ). OpenUrl CrossRef PubMed 8. ↵ Choi , K. J. et al. Three-dimensional analysis of abnormal ultrastructural alteration in mitochondria of hippocampus of APP/PSEN1 transgenic mouse . J Biosci 39 , 97 – 105 , doi: 10.1007/s12038-013-9406-8 ( 2014 ). OpenUrl CrossRef PubMed 9. ↵ Tsai , P. I. et al. PINK1 Phosphorylates MIC60/Mitofilin to Control Structural Plasticity of Mitochondrial Crista Junctions . Mol Cell 69 , 744 – 756 e746 , doi: 10.1016/j.molcel.2018.01.026 ( 2018 ). OpenUrl CrossRef PubMed 10. ↵ Bustamante-Barrientos , F. A. et al. Mitochondrial dysfunction in neurodegenerative disorders: Potential therapeutic application of mitochondrial transfer to central nervous system-residing cells . J Transl Med 21 , 613 , doi: 10.1186/s12967-023-04493-w ( 2023 ). OpenUrl CrossRef PubMed 11. ↵ Cserep , C. , Posfai , B. , Schwarcz , A. D. & Denes , A . Mitochondrial Ultrastructure Is Coupled to Synaptic Performance at Axonal Release Sites . eNeuro 5 , doi: 10.1523/ENEURO.0390-17.2018 ( 2018 ). OpenUrl Abstract / FREE Full Text 12. Perkins , G. A. et al. The micro-architecture of mitochondria at active zones: electron tomography reveals novel anchoring scaffolds and cristae structured for high-rate metabolism . The Journal of neuroscience : the official journal of the Society for Neuroscience 30 , 1015 – 1026 , doi: 10.1523/JNEUROSCI.1517-09.2010 ( 2010 ). OpenUrl Abstract / FREE Full Text 13. Smith , H. L. et al. Mitochondrial support of persistent presynaptic vesicle mobilization with age-dependent synaptic growth after LTP . Elife 5 , doi: 10.7554/eLife.15275 ( 2016 ). OpenUrl CrossRef PubMed 14. ↵ Mendelsohn , R. et al. Morphological principles of neuronal mitochondria . The Journal of comparative neurology 530 , 886 – 902 , doi: 10.1002/cne.25254 ( 2022 ). OpenUrl CrossRef PubMed 15. ↵ Lewis , T. L. , Jr. , Kwon , S. K. , Lee , A. , Shaw , R. & Polleux , F . MFF-dependent mitochondrial fission regulates presynaptic release and axon branching by limiting axonal mitochondria size . Nat Commun 9 , 5008 , doi: 10.1038/s41467-018-07416-2 ( 2018 ). OpenUrl CrossRef PubMed 16. ↵ Schnitzbauer , J. , Strauss , M. T. , Schlichthaerle , T. , Schueder , F. & Jungmann , R . Super-resolution microscopy with DNA-PAINT . Nat Protoc 12 , 1198 – 1228 , doi: 10.1038/nprot.2017.024 ( 2017 ). OpenUrl CrossRef PubMed 17. ↵ The MICrONS Consortium ; Bae , J. A. B. , M. ; Bishop , C. A. ; Bodor , A. L. ; Brittain , D. ; Buchanan , J. ; Bumbarger , D. J. ; Castro , M. A. ; Celii , B. ; Cobos , E. ; Collman , F. ; Costa , N. M. ; Dorkenwald , S. ; Elabbady , L. ; Fahey , P. G. ; Fliss , T. ; Froudarakis , E. ; Gager , J. ; Gamlin , C. ; Gray-Roncal , W. ; Halageri , A. ; Hebditch , J. ; Jia , Z. ; Joyce , E. ; Joyce , J. ; Jordan , C. ; Kapner , D. ; Kemnitz , N. ; Kinn , S. ; Kitchell , L. M. ; Koolman , S. ; Kuehner , K. ; Lee , K. ; Li , K. ; Lu , R. ; Macrina , T. ; Mahalingam , G. ; Matelsky , J. ; McReynolds , S. ; Miranda , E. ; Mitchell , E. ; Mondal , S. S. ; Moore , M. ; Mu , S. ; Muhammad , T. ; Nehoran , B. ; Ogedengbe , O. ; Papadopoulos , C. ; Papadopoulos , S. ; Patel , S. ; Pitkow , X. ; Popovych , S. ; Ramos , A. ; Reid , R. C. ; Reimer , J. ; Rivlin , P. K. ; Rose , V. ; Schneider-Mizell , C. M. ; Seung , H. S. ; Silverman , B. Silversmith , W. ; Sterling , A. ; Sinz , F. H. ; Smith , C. L. ; Suckow , S. ; Takeno , M. ; Tan , Z. H. ; Tolias , A. S. ; Torres , R. ; Turner , N. L. ; Walker , E. Y. ; Wang , T. ; Wanner , A. ; Wester , B. A. ; Williams , G. ; Williams , S. ; Willie , K. ; Willie , R. ; Wong , W. ; Wu , J. ; Xu , C. ; Yang , R. ; Yatsenko , D. ; Ye , F. ; Yin , W. ; Young , R. ; Yu , S-c.; Xenes , D. ; Zhang , C. Functional connectomics spanning multiple areas of mouse visual cortex . bioRxiv , doi: 10.1101/2021.07.28.454025 ( 2021 ). OpenUrl Abstract / FREE Full Text 18. ↵ Gross , G. G. et al. Recombinant probes for visualizing endogenous synaptic proteins in living neurons . Neuron 78 , 971 – 985 , doi: 10.1016/j.neuron.2013.04.017 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 19. ↵ Perkins , G. A. & Ellisman , M. H . Mitochondrial configurations in peripheral nerve suggest differential ATP production . J Struct Biol 173 , 117 – 127 , doi: 10.1016/j.jsb.2010.06.017 ( 2011 ). OpenUrl CrossRef PubMed Web of Science 20. ↵ Mannella , C. A . Structure and dynamics of the mitochondrial inner membrane cristae . Biochimica et biophysica acta 1763 , 542 – 548 , doi: 10.1016/j.bbamcr.2006.04.006 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 21. ↵ Gold , V. A. , Chroscicki , P. , Bragoszewski , P. & Chacinska , A . Visualization of cytosolic ribosomes on the surface of mitochondria by electron cryo-tomography . EMBO Rep 18 , 1786 – 1800 , doi: 10.15252/embr.201744261 ( 2017 ). OpenUrl Abstract / FREE Full Text 22. ↵ Blum , T. B. , Hahn , A. , Meier , T. , Davies , K. M. & Kuhlbrandt , W . Dimers of mitochondrial ATP synthase induce membrane curvature and self-assemble into rows . Proceedings of the National Academy of Sciences of the United States of America 116 , 4250 – 4255 , doi: 10.1073/pnas.1816556116 ( 2019 ). OpenUrl Abstract / FREE Full Text 23. ↵ Muhleip , A. W. et al. Helical arrays of U-shaped ATP synthase dimers form tubular cristae in ciliate mitochondria . Proceedings of the National Academy of Sciences of the United States of America 113 , 8442 – 8447 , doi: 10.1073/pnas.1525430113 ( 2016 ). OpenUrl Abstract / FREE Full Text 24. ↵ Davies , K. M. et al. Macromolecular organization of ATP synthase and complex I in whole mitochondria . Proceedings of the National Academy of Sciences of the United States of America 108 , 14121 – 14126 , doi: 10.1073/pnas.1103621108 ( 2011 ). OpenUrl Abstract / FREE Full Text 25. ↵ Venkatraman , K. et al. Cristae formation is a mechanical buckling event controlled by the inner mitochondrial membrane lipidome . The EMBO journal 42 , e114054 , doi: 10.15252/embj.2023114054 ( 2023 ). OpenUrl CrossRef 26. ↵ Davies , K. M. , Anselmi , C. , Wittig , I. , Faraldo-Gomez , J. D. & Kuhlbrandt , W . Structure of the yeast F1Fo-ATP synthase dimer and its role in shaping the mitochondrial cristae . Proceedings of the National Academy of Sciences of the United States of America 109 , 13602 – 13607 , doi: 10.1073/pnas.1204593109 ( 2012 ). OpenUrl Abstract / FREE Full Text 27. ↵ Waltz , F. et al. In-cell architecture of the mitochondrial respiratory chain . Science 387 , 1296 – 1301 , doi: 10.1126/science.ads8738 ( 2025 ). OpenUrl CrossRef PubMed 28. ↵ (!!! INVALID CITATION !!! 22,28). 29. ↵ Janer , A. et al. SLC25A46 is required for mitochondrial lipid homeostasis and cristae maintenance and is responsible for Leigh syndrome . EMBO Mol Med 8 , 1019 – 1038 , doi: 10.15252/emmm.201506159 ( 2016 ). OpenUrl Abstract / FREE Full Text 30. Kornmann , B. et al. An ER-mitochondria tethering complex revealed by a synthetic biology screen . Science 325 , 477 – 481 , doi: 10.1126/science.1175088 ( 2009 ). OpenUrl Abstract / FREE Full Text 31. AhYoung , A. P. et al. Conserved SMP domains of the ERMES complex bind phospholipids and mediate tether assembly . Proceedings of the National Academy of Sciences of the United States of America 112 , E3179 – 3188 , doi: 10.1073/pnas.1422363112 ( 2015 ). OpenUrl Abstract / FREE Full Text 32. ↵ Decker , S. T. & Funai , K . Mitochondrial membrane lipids in the regulation of bioenergetic flux . Cell Metab 36 , 1963 – 1978 , doi: 10.1016/j.cmet.2024.07.024 ( 2024 ). OpenUrl CrossRef PubMed 33. ↵ Rackham , O. , Saurer , M. , Ban , N. & Filipovska , A . Unique architectural features of mammalian mitochondrial protein synthesis . Trends Cell Biol 35 , 11 – 23 , doi: 10.1016/j.tcb.2024.05.001 ( 2025 ). OpenUrl CrossRef PubMed 34. ↵ Richter-Dennerlein , R. , Dennerlein , S. & Rehling , P . Integrating mitochondrial translation into the cellular context . Nature reviews. Molecular cell biology 16 , 586 – 592 , doi: 10.1038/nrm4051 ( 2015 ). OpenUrl CrossRef PubMed 35. ↵ Alvarez-Castelao , B. et al. Cell-type-specific metabolic labeling of nascent proteomes in vivo . Nat Biotechnol 35 , 1196 – 1201 , doi: 10.1038/nbt.4016 ( 2017 ). OpenUrl CrossRef PubMed 36. ↵ Rangaraju , V. , Calloway , N. & Ryan , T. A . Activity-driven local ATP synthesis is required for synaptic function . Cell 156 , 825 – 835 , doi: 10.1016/j.cell.2013.12.042 ( 2014 ). OpenUrl CrossRef PubMed Web of Science 37. ↵ Turner , N. L. et al. Reconstruction of neocortex: Organelles, compartments, cells, circuits, and activity . Cell 185 , 1082 – 1100 e1024 , doi: 10.1016/j.cell.2022.01.023 ( 2022 ). OpenUrl CrossRef 38. ↵ Popov , V. , Medvedev , N. I. , Davies , H. A. & Stewart , M. G . Mitochondria form a filamentous reticular network in hippocampal dendrites but are present as discrete bodies in axons: a three-dimensional ultrastructural study . The Journal of comparative neurology 492 , 50 – 65 , doi: 10.1002/cne.20682 ( 2005 ). OpenUrl CrossRef PubMed Web of Science 39. ↵ Dikov , D. & Bereiter-Hahn , J . Inner membrane dynamics in mitochondria . J Struct Biol 183 , 455 – 466 , doi: 10.1016/j.jsb.2013.06.003 ( 2013 ). OpenUrl CrossRef PubMed 40. ↵ Bohnert , M. et al. Central role of Mic10 in the mitochondrial contact site and cristae organizing system . Cell Metab 21 , 747 – 755 , doi: 10.1016/j.cmet.2015.04.007 ( 2015 ). OpenUrl CrossRef PubMed 41. Barbot , M. et al. Mic10 oligomerizes to bend mitochondrial inner membranes at cristae junctions . Cell Metab 21 , 756 – 763 , doi: 10.1016/j.cmet.2015.04.006 ( 2015 ). OpenUrl CrossRef PubMed 42. Ikon , N. & Ryan , R. O . Cardiolipin and mitochondrial cristae organization . Biochim Biophys Acta Biomembr 1859 , 1156 – 1163 , doi: 10.1016/j.bbamem.2017.03.013 ( 2017 ). OpenUrl CrossRef PubMed 43. Cogliati , S. et al. Mitochondrial cristae shape determines respiratory chain supercomplexes assembly and respiratory efficiency . Cell 155 , 160 – 171 , doi: 10.1016/j.cell.2013.08.032 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 44. ↵ Frezza , C. et al. OPA1 controls apoptotic cristae remodeling independently from mitochondrial fusion . Cell 126 , 177 – 189 , doi: 10.1016/j.cell.2006.06.025 ( 2006 ). OpenUrl CrossRef PubMed Web of Science 45. ↵ Patten , D. A. et al. OPA1-dependent cristae modulation is essential for cellular adaptation to metabolic demand . The EMBO journal 33 , 2676 – 2691 , doi: 10.15252/embj.201488349 ( 2014 ). OpenUrl Abstract / FREE Full Text 46. ↵ Palmieri , F . The mitochondrial transporter family SLC25: identification, properties and physiopathology . Mol Aspects Med 34 , 465 – 484 , doi: 10.1016/j.mam.2012.05.005 ( 2013 ). OpenUrl CrossRef PubMed Web of Science 47. ↵ Akabane , S. et al. PKA Regulates PINK1 Stability and Parkin Recruitment to Damaged Mitochondria through Phosphorylation of MIC60 . Mol Cell 62 , 371 – 384 , doi: 10.1016/j.molcel.2016.03.037 ( 2016 ). OpenUrl CrossRef PubMed 48. ↵ Ould Amer , Y. & Hebert-Chatelain , E. Insight into the Interactome of Intramitochondrial PKA Using Biotinylation-Proximity Labeling . International journal of molecular sciences 21 , doi: 10.3390/ijms21218283 ( 2020 ). OpenUrl CrossRef 49. ↵ Jimenez , L. , Laporte , D. , Duvezin-Caubet , S. , Courtout , F. & Sagot , I . Mitochondrial ATP synthases cluster as discrete domains that reorganize with the cellular demand for oxidative phosphorylation . Journal of cell science 127 , 719 – 726 , doi: 10.1242/jcs.137141 ( 2014 ). OpenUrl Abstract / FREE Full Text 50. ↵ Anselmi , C. , Davies , K. M. & Faraldo-Gomez , J. D . Mitochondrial ATP synthase dimers spontaneously associate due to a long-range membrane-induced force . J Gen Physiol 150 , 763 – 770 , doi: 10.1085/jgp.201812033 ( 2018 ). OpenUrl Abstract / FREE Full Text 51. ↵ Dietrich , L. , Agip , A. A. , Kunz , C. , Schwarz , A. & Kuhlbrandt , W . In situ structure and rotary states of mitochondrial ATP synthase in whole Polytomella cells . Science 385 , 1086 – 1090 , doi: 10.1126/science.adp4640 ( 2024 ). OpenUrl CrossRef PubMed 52. ↵ Chu , Q. J. , W-K. Lipid transfer at mitochondrial membrane contact sites . Mitochondrial Communications 2 , 123 – 128 ( 2024 ). OpenUrl 53. Fusco , C. M. et al. Neuronal processes contain the essential components for the late steps of ribosome biogenesis . Proceedings of the National Academy of Sciences of the United States of America 122 , e2502424122 , doi: 10.1073/pnas.2502424122 ( 2025 ). OpenUrl CrossRef PubMed 54. ↵ Biever , A. et al. Monosomes actively translate synaptic mRNAs in neuronal processes . Science 367 , doi: 10.1126/science.aay4991 ( 2020 ). OpenUrl Abstract / FREE Full Text 55. ↵ Weiss , J. T. & Donlea , J. M . Roles for Sleep in Neural and Behavioral Plasticity: Reviewing Variation in the Consequences of Sleep Loss . Front Behav Neurosci 15 , 777799 , doi: 10.3389/fnbeh.2021.777799 ( 2021 ). OpenUrl CrossRef 56. ↵ Dorszewska , J. , Kozubski , W. , Waleszczyk , W. , Zabel , M. & Ong , K . Neuroplasticity in the Pathology of Neurodegenerative Diseases . Neural Plast 2020, 4245821 , doi: 10.1155/2020/4245821 ( 2020 ). OpenUrl CrossRef 57. ↵ Daum , B. , Walter , A. , Horst , A. , Osiewacz , H. D. & Kuhlbrandt , W . Age-dependent dissociation of ATP synthase dimers and loss of inner-membrane cristae in mitochondria . Proceedings of the National Academy of Sciences of the United States of America 110 , 15301 – 15306 , doi: 10.1073/pnas.1305462110 ( 2013 ). OpenUrl Abstract / FREE Full Text 58. ↵ Sarnataro , R. , Velasco , C. D. , Monaco , N. , Kempf , A. & Miesenbock , G . Mitochondrial origins of the pressure to sleep . Nature , doi: 10.1038/s41586-025-09261-y ( 2025 ). OpenUrl CrossRef 59. ↵ Li , W. et al. Boosting neuronal activity-driven mitochondrial DNA transcription improves cognition in aged mice . Science 386 , eadp6547 , doi: 10.1126/science.adp6547 ( 2024 ). OpenUrl CrossRef PubMed 60. ↵ Ashrafi , G. , de Juan-Sanz , J. , Farrell , R. J. & Ryan , T. A . Molecular Tuning of the Axonal Mitochondrial Ca(2+) Uniporter Ensures Metabolic Flexibility of Neurotransmission . Neuron 105 , 678 – 687 e675 , doi: 10.1016/j.neuron.2019.11.020 ( 2020 ). OpenUrl CrossRef PubMed 61. ↵ Mastronarde , D . N . SerialEM: A Program for Automated Tilt Series Acquisition on Tecnai Microscopes Using Prediction of Specimen Position Microscopy and Microanalysis 9 , 1182 – 1183 , doi: 10.1017/S1431927603445911 ( 2003 ). OpenUrl CrossRef 62. ↵ Mastronarde , D. N . Automated electron microscope tomography using robust prediction of specimen movements . J Struct Biol 152 , 36 – 51 , doi: 10.1016/j.jsb.2005.07.007 ( 2005 ). OpenUrl CrossRef PubMed Web of Science 63. ↵ Kremer , J. R. , Mastronarde , D. N. & McIntosh , J. R . Computer visualization of three-dimensional image data using IMOD . J Struct Biol 116 , 71 – 76 , doi: 10.1006/jsbi.1996.0013 ( 1996 ). OpenUrl CrossRef PubMed Web of Science 64. ↵ Lee , K. Z. , J.; Li , P. ; Jain , V. ; Seung , H. S. Superhuman Accuracy on the SNEMI3D Connectomics Challenge . arXiv:1706.00120 , doi: 10.48550/arXiv.1706.00120 ( 2017 ). OpenUrl CrossRef 65. ↵ Suga , S. et al. An interactive deep learning-based approach reveals mitochondrial cristae topologies . PLoS Biol 21 , e3002246 , doi: 10.1371/journal.pbio.3002246 ( 2023 ). OpenUrl CrossRef PubMed 66. ↵ Zeng , T. , Wu , B. & Ji , S . DeepEM3D: approaching human-level performance on 3D anisotropic EM image segmentation . Bioinformatics 33 , 2555 – 2562 , doi: 10.1093/bioinformatics/btx188 ( 2017 ). OpenUrl CrossRef PubMed 67. ↵ Harris , K. M. et al. A resource from 3D electron microscopy of hippocampal neuropil for user training and tool development . Sci Data 2 , 150046 , doi: 10.1038/sdata.2015.46 ( 2015 ). OpenUrl CrossRef PubMed 68. ↵ Spacek , J. & Harris , K. M . Three-dimensional organization of smooth endoplasmic reticulum in hippocampal CA1 dendrites and dendritic spines of the immature and mature rat . The Journal of neuroscience : the official journal of the Society for Neuroscience 17 , 190 – 203 , doi: 10.1523/JNEUROSCI.17-01-00190.1997 ( 1997 ). OpenUrl Abstract / FREE Full Text 69. ↵ Van der Walt , S. , Schoenberger , J. L. , Nunez-Iglesias , J. , Boulogne , F. , Warner , J. D. , Yager , N. , Gouillart , E. , Yu , T . , and the scikit-image contributors. scikit-image: image processing in Python . PeerJ 2 , doi: 10.7717/peerj.453 ( 2014 ). OpenUrl CrossRef PubMed 70. ↵ Lorensen , W. E. , Cline , H. E . Marching cubes:: A high resolution 3D surface construction algorithm Computer Graphics 21 , 163 – 169 ( 1987 ). OpenUrl CrossRef 71. ↵ Sun , C. et al. An abundance of free regulatory (19S) proteasome particles regulates neuronal synapses . Science 380 , eadf2018 , doi: 10.1126/science.adf2018 ( 2023 ). OpenUrl CrossRef PubMed 72. ↵ Sato , M. , Bitter , I. , Bender , M.A. , Kaufman , A.E. , and Nakajima , M . TEASAR: tree-structure extraction algorithm for accurate and robust skeletons . Proceedings the Eighth Pacific Conference on Computer Graphics and Applications , doi: 10.1109/pccga.2000.883951 ( 2000 ). OpenUrl CrossRef 73. ↵ Schlegel , P. et al. Whole-brain annotation and multi-connectome cell typing of Drosophila . Nature 634 , 139 – 152 , doi: 10.1038/s41586-024-07686-5 ( 2024 ). OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted August 27, 2025. Download PDF Supplementary Material 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 Mitochondria structurally remodel near synapses to fuel the sustained energy demands of plasticity 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 Mitochondria structurally remodel near synapses to fuel the sustained energy demands of plasticity Monil Shah , Ilika Ghosh , Luca Pishos , Valentina Villani , Tristano Pancani , Ryohei Yasuda , Chao Sun , Naomi Kamasawa , Vidhya Rangaraju bioRxiv 2025.08.27.672715; doi: https://doi.org/10.1101/2025.08.27.672715 Share This Article: Copy Citation Tools Mitochondria structurally remodel near synapses to fuel the sustained energy demands of plasticity Monil Shah , Ilika Ghosh , Luca Pishos , Valentina Villani , Tristano Pancani , Ryohei Yasuda , Chao Sun , Naomi Kamasawa , Vidhya Rangaraju bioRxiv 2025.08.27.672715; doi: https://doi.org/10.1101/2025.08.27.672715 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 Neuroscience Subject Areas All Articles Animal Behavior and Cognition (7636) Biochemistry (17705) Bioengineering (13899) Bioinformatics (41967) Biophysics (21460) Cancer Biology (18600) Cell Biology (25526) Clinical Trials (138) Developmental Biology (13384) Ecology (19909) Epidemiology (2067) Evolutionary Biology (24326) Genetics (15613) Genomics (22512) Immunology (17740) Microbiology (40423) Molecular Biology (17193) Neuroscience (88645) Paleontology (667) Pathology (2835) Pharmacology and Toxicology (4825) Physiology (7647) Plant Biology (15159) Scientific Communication and Education (2046) Synthetic Biology (4302) Systems Biology (9825) Zoology (2271)