Introduction

Memory engram cells are crucial for the formation and storage of long-term memories, serving as the cellular substrate underlying the brain’s cognitive functions1,2,3,4,5,6. These specific neuronal populations undergo significant alterations during learning, forming tightly interconnected networks that mediate memory consolidation and retrieval7,8,9. Synaptic plasticity is fundamental to the stability of memory engram networks. It enhances the connectivity between engram cells by modifying the structural and electrophysiological properties of synapses10,11,12. These adaptations are crucial for stabilizing memory traces and ensuring effective reactivation of the memory network upon exposure to reminder cues13.

Mitochondria play a critical role in supporting synaptic plasticity by providing the energy necessary for synaptic transmission and neuronal function14,15,16,17,18. As the primary site of ATP production, mitochondria ensure a consistent supply, especially during periods of high energy demand such as synaptic transmission and plasticity19. The energy consumption at the synapse is substantial, and mitochondria, strategically located near synapses18,20,21, are essential for maintaining synaptic function by rapidly supplying ATP. Recent studies using electron microscopy (EM) have revealed the presence of mitochondria within dendritic spines and highlighted their involvement in processes related to learning and neurological disorders17,22,23. Despite their crucial role, current research on mitochondria has primarily focused on overall mitochondrial function rather than detailed protein-level analysis within specific cellular contexts.

Recent advancements in single-molecule localization microscopy techniques, such as MINFLUX, offer high-resolution insights into cellular structures and molecular dynamics24,25. MINFLUX excels in providing nanometer-scale resolution and microsecond-range single-molecule tracking, making it particularly valuable for studying mitochondrial membrane proteins26. However, most studies using MINFLUX have been conducted in controlled environments like cultured cells, leaving a knowledge gap regarding mitochondrial protein distribution in native biological tissues.

This study aims to address this gap by utilizing 3D MINFLUX to investigate the spatial distribution of mitochondrial proteins at the synaptic level in identified memory engram cell populations within brain tissue. Our research discovered that the presence of mitochondria within dendritic spines is significantly increased in memory engram cells, and that the mitochondrial inner membrane protein α-F1-ATP synthase (ATP5a) is redistributed within these spines, concentrating towards synaptic contact sites. Additionally, using in vitro neuron culture combined with chemical LTP (cLTP) induction to simulate the learning process, we found that the redistribution of ATP5a occurs at both spine and shaft synapses. In contrast, translocase of outer mitochondrial membrane 20 (TOMM20) did not exhibit such a polarized redistribution. Notably, pharmacological inhibition of ATP5a by oligomycin-A revealed a correlated disruption: the cLTP-induced spatial reorganization of ATP5a failed to occur, concurrently with impaired structural plasticity. Collectively, these observations point to an activity-dependent repositioning of the inner mitochondrial membrane component ATP5a. This dynamic nanoscale targeting mechanism thereby advances our understanding of mitochondrial contributions to cognitive processes.

Results

Nanoscale distribution of mitochondrial ATP5a in dendrites of memory engram cells revealed by MINFLUX nanoscopy

To investigate the function of mitochondria in the synaptic plasticity of memory engram cells, activity-dependent labeling was performed in the hippocampal region of cFos-CreER mice. The TRAP (Targeted Recombination in Active Populations) system27,28 was employed for activity-dependent labeling. A mixture of two AAV vectors was injected unilaterally into the dentate gyrus (DG) of the hippocampus: one carried a Cre-dependent (floxed) mCherry reporter, and the other expressed EGFP under the control of the CaMKIIα promoter. This strategy enabled activity-dependent labeling of memory-encoding neurons (engram cells) while simultaneously labeling nearby non-activated neurons (non-engram cells) (Fig. 1a). Following viral infection and during learning-induced neuronal excitation, the expression of the cFos gene drove the transcription of CreER. Tamoxifen, administered 24 hours prior, bound to CreER, removing its spatial hindrance and allowing its nuclear entry. This facilitated recombination at loxP sites, leading to the expression of mCherry and marking the memory engram cells (Fig. 1b). A timeline of the experimental procedure is illustrated in Fig. 1c. After a 2-week recovery period post-injection, the mice underwent intraperitoneal injection of Tamoxifen (150 mg/kg), followed by contextual fear conditioning (CFC) training 24 hours later. During the training process, activity-dependent genetic recombination was triggered in activated neurons, establishing permanent molecular labeling of engram cells. Three days post-training, the mice were sacrificed, and brain tissue was collected and processed for sectioning and subsequent analysis. Confocal imaging revealed the presence of mCherry-labeled memory engram cells in DG, while EGFP marked the surrounding non-engaged neurons (Fig. 1d; Supplementary Fig. 1a). Activity-dependence of labeling was further confirmed by controls of tamoxifen-only, TRAP in homecage, and EE conditions (Supplementary Fig. 1b–d). Consistent with previous findings7, structural analysis revealed that memory engram cells exhibited significantly more dendritic spines compared to non-engaged cells, and dendritic spine width was statistically greater in the memory engram cell populations (Fig. 1e, f), indicating significant structural plasticity and enhanced synaptic connectivity.

Fig. 1: 3D MINFLUX super-resolution imaging of protein distribution in dendritic mitochondria of hippocampal memory engram cells.

a–c Labeling of memory engram and non-engram cells in the hippocampal DG region by AAV viral vectors. a Expression of EGFP protein driven by the CaMKII promoter for labeling excitatory neurons in the hippocampal DG region. mCherry protein expression driven by the Syn promoter, with loxP sequences, labels activity-dependent memory engram cells. b Overview of the TRAP technique. By combining TRAP with tamoxifen (TM) injection, active neurons within a specific time window can be fluorescently labeled. c Schematic representation of the behavioral protocol for labeling in cFos-CreER mice. d Representative confocal images showing memory engram and non-engram cells. Blue: DAPI, Red: mCherry, Green: EGFP. Scale bar: 200 μm. e, f Observation and statistical analysis of dendritic spine density in memory engram and non-engram cells. e Cells with co-localized red and green signals, along with dendrites, are classified as memory engram cells, while cells with only green fluorescence are categorized as non-engram cells. Scale bar (cell images):10 μm, Scale bar (dendrite images): 2 μm. f Quantification of spine density and spine width (n-engram cells: 21 dendritic segments, n-non-engram cells: 36 dendritic segments, from three mice). Significance: **P < 0.01, ***P < 0.001 (spine density: Unpaired t test, spine width: Mann–Whitney U test), P = 0.007 for spine density quantification and P = 0.0026 for spine width quantification. g 3D volume view of dendrites and mitochondria in each group. Mitochondria labeled with anti-TOMM20-AF647 (validated in Supplementary Fig. 2). h Quantification of mitochondrial localization within dendritic spines of memory engram and non-engram cells (n-engram: 27 dendritic segments; n-non-engram: 35 dendritic segments; from 11 brain sections/4 mice). Significance: **P < 0.01 (Mann–Whitney U test), P = 0.0015. i Representative 3D volume renderings of dendritic spines (with/without mitochondria) in engram cells. j 3D MINFLUX imaging of ATP5a protein in mitochondria located in mCherry-positive dendrite (left) and EGFP-positive dendrite (right). The z-axis is color-coded. ATP5a was first labeled using a primary antibody, followed by secondary antibody conjugation with AF647. Scale bar (confocal images): 2 μm, Scale bar (3D MINFLUX images): 500 nm. In the box-and-whisker plot, the box represents the interquartile range (IQR, 25th–75th percentiles), divided by the median. The whiskers span the full data range, terminating at the minimum and maximum observed values, individual data points are displayed in dot plots while violin plots depict probability density distributions.

Having successfully targeted the memory engram cell populations, we then examined the distribution of mitochondria in the dendritic spines of both engram and non-engram cell populations using immunohistochemistry. Mitochondria were visualized via microscopy (Fig. 1g; Supplementary Fig. 1e–g) after labeling with an anti-TOMM20 antibody conjugated to AF647 dye (validated in Supplementary Fig. 2). The results showed a significant increase in the presence of mitochondria in dendritic spines of memory engram cells in the DG compared to non-engram cells (Fig. 1h, 1.66 ± 0.35% in engram cells vs. 0.49 ± 0.19% in non-engram cells, Mann–Whitney U test, p = 0.0015). Next, we explored the potential significance of the mitochondria’s distribution in dendritic spines (Fig. 1i). Mitochondria possess a wide variety of membrane proteins embedded in both the inner and outer membranes, which are essential for their metabolic and signaling functions29,30. However, traditional confocal microscopy, due to its resolution limitations, is unable to resolve the distribution of these proteins. To capture the finer details of mitochondrial changes within synapses, we turned to 3D MINFLUX imaging. Before conducting MINFLUX imaging on brain sections, several considerations were made to optimize tissue preparation. Due to the thickness of brain tissue compared to cultured cells, sectioning parameters were adjusted by increasing the oscillation frequency of the microtome and reducing the blade advancement speed, which allowed for the production of thinner sections of brain tissue (10–15 µm). Additionally, to reduce non-specific antibody binding, which could lead to excessive fluorescence signal, two key adjustments were made: the blocking time with BSA was extended to minimize non-specific binding; and during the immunohistochemistry process, the PBS wash time and volume were optimized to further reduce background fluorescence (Supplementary Fig. 3a, b). Following traditional mounting protocols of MINFLUX imaging in cultured cells26, GLOX buffer that contains 30 mM mercaptoethylamine (MEA) was used for later imaging studies.

As shown in Fig. 1j, 3D MINFLUX imaging revealed the distribution of ATP5a proteins (labeled with primary antibody, followed by secondary antibody conjugated to AF647) within the dendrite and the spines of the marked memory engram cells and non-engram cells (Supplementary Fig. 3c). Additionally, we also achieved 3D dual-color MINFLUX imaging in brain sections, where TOMM20 was labeled with primary antibody followed by secondary antibody conjugated with FL640 dye, and ATP5a was labeled with primary antibody followed by secondary antibody conjugated with FL680 dye31 (Supplementary Fig. 3d, f). Using DBSCAN clustering and spatial analysis32, the distribution pattern of ATP5a molecules was analyzed (Supplementary Fig. 4a, b). The spatial resolution of protein localization was confirmed with XYZ axis precision of 6-7 nm (Supplementary Fig. 4c). The nearest-neighbor distance distribution of ATP5a molecules within memory engram cells revealed several distinct peaks, with a major peak at 21.27 nm, and additional peaks at 52.25 nm, 72.16 nm, 87.65 nm, and 116.4 nm (Supplementary Fig. 4d), validating the high spatial resolution of MINFLUX and confirming its applicability for observing molecular organization at the nanoscale.

Polarized redistribution of mitochondrial inner membrane protein ATP5a after learning-induced synaptic plasticity

To explore the role of mitochondrial reorganization during learning, particularly within synapse-associated dendritic spines, the distribution of ATP5a in spine areas with pre-synaptic contact was specifically examined. Confocal microscopy revealed a tight spatial relation between the localization of mitochondria and pre-synaptic marker synaptophysin (Syn), suggesting that mitochondria play a significant role in synaptic function (Fig. 2a, c). 3D MINFLUX data revealed that mitochondrial ATP5a molecules within engram cell spines exhibit preferential accumulation near the side with high Syn intensity, as indicated by green arrows. These regions displayed a higher density of AF647 molecule nanoclusters identified by MINFLUX. In contrast, ATP5a distribution in non-engram cell spines appeared more random, without a clear spatial correlation with syn signals (Fig. 2b, d). To quantify this phenomenon, DBSCAN-based clustering analysis was performed to locate the labeled molecules and the local density of each ATP5a molecule was then calculated based on the spatial locations (Fig. 2e), revealing a higher local density of ATP5a molecules near directions with high Syn intensity. This was further confirmed by radial measurements of local density using a segmentation approach (Fig. 2f), which indicated a significant increase in ATP5a local density in the direction with high Syn intensity (Syn+) in dendritic spine regions of memory engram cells compared to non-engram cells (Fig. 2g). These results suggest that the molecular reorganization of ATP5a within dendritic spines is closely tied to synaptic plasticity during the learning process (Fig. 2h).

Fig. 2: Polarized redistribution of mitochondrial inner membrane protein ATP5a in dendritic spines of DG memory engram cells.

a, b Representative MINFLUX imaging of ATP5a in dendritic spines of memory engram cells. a Left: mCherry-labeled dendritic spines (Gray). Middle: ATP5a (red) and synaptophysin (green). Right: line intensity profile. b Z-projection of 3D MINFLUX data for mitochondrial ATP5a in the region indicated, green arrows highlight regions of high syn intensity. Confocal scale bar, 500 nm, MINFLUX scale bar, 200 nm. c, d Representative MINFLUX imaging of ATP5a in dendritic spines of non-engram cells. c Left: EGFP-labeled dendritic spines (Gray). Middle: ATP5a (red) and synaptophysin (green). Right: line intensity profile. d Z-projection of 3D MINFLUX data for mitochondrial ATP5a in the region indicated, green arrows highlight regions of high syn intensity. Confocal scale bar, 500 nm, MINFLUX scale bar, 200 nm. e, f Analysis pipeline for correlating ATP5a MINFLUX local density with syn intensity in dendritic spines. e Regions of interest (ROIs) containing mitochondria within dendritic spines of an engram cell were selected for MINFLUX imaging. Clusters of ATP5a molecules were identified using DBSCAN-based clustering, and local densities were calculated to assess ATP5a distribution. Scale bar, 200 nm. f ATP5a local density and syn intensity data were measured using an 8-segment division centered on the mitochondrial staining centroid. Regions with the highest synaptic intensity (syn+) and the lowest synaptic intensity (syn−) were compared to analyze differences in ATP5a molecular density. g Spatial redistribution of mitochondrial ATP5a relative to synaptic contact regions in dendritic spines of memory engram cells versus non-engram cells. Statistical analysis revealed significant differences in the spatial distribution of ATP5a between these two groups. (n-memory engram = 26 spine mitochondria from 12 brain sections from nine mice; n-non-engram = 13 spine mitochondria from 6 brain sections from 4 mice). Significance: **P, ##P < 0.01,***P < 0.001 (Aligned Rank Transformation ANOVA, Group factor: F(1, 77) = 11.3, P = 0.0012; Region factor: F(1, 77) = 6.96, P = 0.0102; Interaction, F(1, 77) = 9.53, P = 0.0028; Bonferroni post-hoc test, Pnon-engram_syn– vs. non-engram_syn+ > 0.9999; Pengram_syn– vs. engram_syn+ < 0.001; Pengram_syn+ vs. non-engram_syn+ = 0.0034). h Proposed molecular model of ATP5a reorganization in dendritic spines of memory engram cells during learning-related processes.

Chemically-induced long-term potentiation triggers polarized reorganization of ATP5a in dendritic spines

To further explore the relationship between synaptic plasticity and mitochondrial inner membrane protein reorganization, we conducted MINFLUX imaging experiments in cultured cortical primary neurons. AAV-mediated expression of membrane GFP was conducted in order to visualize the morphology of the neuron (Fig. 3a). To simulate learning-related processes in vitro, chemically-induced long-term potentiation (cLTP) was used21 (Supplementary Fig. 5a, b). DIV17-21 neurons were exposed to a short 5 min duration of 100 μM glycine stimulation, followed by fixation and immuno-staining for ATP5a and Syn. Confocal imaging showed that mitochondria in dendritic spines preferentially localized near presynaptic contact sites. Rod-shaped mitochondria were also frequently found adjacent to shaft synapses, as highlighted in the magnified insets and line profiles (Fig. 3b, c). Consistent with the findings described above, 3D MINFLUX imaging revealed that ATP5a molecules in dendritic spines of cultured neurons showed a significant redistribution following cLTP induction, with a clear preference towards regions with high Syn intensity. In contrast, such spatial redistribution of ATP5a was not observed in the control condition (Fig. 3d–f). To better understand how long synaptic plasticity-induced polarized redistribution of ATP5a can be maintained in cultured neurons, we performed a time-course analysis. Neurons were fixed immediately after cLTP induction, and at 30 minutes, 12 hours, and 3 days post-stimulation (Fig. 3g). The results showed that ATP5a reorganization persisted for at least 12 hours after cLTP induction and then diminished by 3 days (Fig. 3h), suggesting a short-term but significant role for ATP5a in synaptic plasticity and the difference between neural networks in vitro and memory engram networks. We also investigated whether similar ATP5a redistribution occurs in shaft synapse regions. Specifically, ATP5a MINFLUX localizations were collected from shaft synapses that exhibited strong Syn signals and were in close contact with dendritic mitochondria, as well as from adjacent areas with weaker pre-synaptic Syn signals (Supplementary Fig. 6a). To quantify ATP5a spatial distribution, the local density of each identified ATP5a molecule was calculated (Supplementary Fig. 6b). Further analysis of ATP5a local density maps in shaft synapse regions revealed that ATP5a local density was significantly higher in regions with high Syn intensity compared to nearby low Syn intensity regions in the cLTP group (Supplementary Fig. 6c–e).

Fig. 3: Polarized redistribution of mitochondrial inner membrane protein ATP5a during cLTP.

a Representative images of DIV21 cultured primary cortical neurons expressing mGFP. Scale bar: 50 μm. b Mitochondria were labeled with anti-ATP5a and AF647-conjugated secondary antibodies. Presynaptic sites were labeled with anti-synaptophysin and AF555-conjugated secondary antibodies. Red: ATP5a, Green: Syn, Gray: mGFP. c Representative images showing postsynaptic mitochondria in spine synapses and shaft synapses. Line profiles show ATP5a and Syn confocal signals across the postsynaptic sites in spine and shaft synapses. Scale bar: 1 μm. d–f Reorganization of mitochondrial ATP5a in dendritic spines under Control (d) and cLTP (e) conditions. Confocal images show mitochondrial ATP5a (red) and Syn (green) in dendritic spines of cultured neurons. The white dashed line indicates the mitochondria area. The bottom panel shows z-projection of MINFLUX data for ATP5a localization. Confocal scale bar: 400 nm; MINFLUX scale bar: 200 nm. f Quantification of ATP5a local density enrichment near synaptic contact regions in Control and cLTP conditions. (n-cLTP = 51 spine mitochondria from 34 neurons/4 cultures, n-Control = 47 spine mitochondria from 32 neurons/ 5 cultures). Significance: #P < 0.05,**P < 0.01. (ART ANOVA, Group factor: F(1, 195) = 1.14, P = 0.2873; Region factor: F(1, 195) = 0.55, P = 0.4573; Interaction, F(1, 195) = 6.60, P = 0.0110; Bonferroni post hoc test, PControl_syn– vs. Control_syn+ = 0.5784; PcLTP_syn– vs. cLTP_syn+ = 0.0047; PcLTP_syn+ vs. Control_syn+ = 0.0159). g, h Distribution of mitochondrial ATP5a protein in dendritic spines at different time points after cLTP induction. g Top: confocal images showing mitochondrial ATP5a (red) and adjacent Syn (green) in dendritic spines. Bottom: z projection of MINFLUX data showing ATP5a localization in the same dendritic spines. Confocal scale bar: 400 nm; MINFLUX scale bar: 200 nm. h Quantification of local density enrichment of ATP5a near synaptic contact regions. (n-Control = 40 spine mitochondria from 27 neurons/3 cultures, n-cLTP = 22 spine mitochondria from 15 neurons/3 cultures, n-30 min = 40 spine mitochondria from 20 neurons/4 cultures, n-12 h = 28 spine mitochondria from 19 neurons/3 cultures and n-3days = 38 spine mitochondria from 25 neurons/3 cultures). Significance: *P, #P < 0.05, **P, ##P < 0.01. (ART ANOVA, Group factor: F(4, 339) = 4.08, P = 0.0031; Region factor: F(1, 339) = 0.24, P = 0.6227; interaction, F(4, 339) = 4.57, P = 0.0013; Bonferroni post hoc test: PCtrl_syn–vs. Ctrl_syn+ > 0.9999; PcLTP_syn–vs. cLTP_syn+ = 0.0084; P30min_syn–vs. 30 min_syn+ = 0.0212; P12hrs_syn–vs. 12 h_syn+ = 0.0025; P3days_syn–vs. 3days syn+ > 0.9999; PCtrl_syn+ vs. cLTP_syn+ = 0.0032; PCtrl_syn+ vs. 30 min_syn+ = 0.0426; PCtrl_syn+ vs. 12 h_syn+ = 0.0056; PCtrl_syn+ vs. 3days_syn+ > 0.9999;).

To test the functional link between cLTP-induced ATP5a redistribution and local energy demands during synaptic plasticity, we pharmacologically inhibited ATP synthase using oligomycin-A33. A 30-minute treatment led to a dose-dependent reduction in intracellular ATP levels. At 1 µM (Supplementary Fig. 7a), ATP levels were significantly decreased without inducing detectable cell death, as assessed by propidium iodide (PI) staining (Supplementary Fig. 7b). These results support the use of 1 µM for 30 minutes as an effective, non-toxic condition for further experiments. We then immunostained ATP5a under four experimental conditions—Control, cLTP, oligomycin-A, and oligomycin-A combined with cLTP, with all groups fixed 30 minutes after treatment, followed by 3D MINFLUX imaging (Supplementary Fig. 7c). This analysis revealed that ATP synthase inhibition abolished the cLTP-induced ATP5a enrichment in Syn-high regions within dendritic spines (Supplementary Fig. 7d–f). Under these conditions, morphological analysis showed that the cLTP-induced increases in spine density, spine head width, as well as mitochondrial recruitment to dendritic spines, were abolished (Supplementary Fig. 7g–i), indicating impaired structural plasticity. Together, these results indicated that ATP synthase activity is required for the activity-dependent, polarized redistribution of ATP5a and that this redistribution is functionally linked to synaptic structural remodeling after cLTP induction.

Lack of cLTP-induced polarized reorganization of the outer membrane protein TOMM20

Mitochondria consist of both inner and outer membrane proteins with distinct functional roles. While the reorganization of the inner membrane protein ATP5a was observed during synaptic plasticity, it remained unclear whether outer membrane proteins such as TOMM20 exhibit similar changes. To address this, we performed dual labeling of mitochondrial inner and outer membrane proteins, using immunolabeling of ATP5a and TOMM20 with FL640 and FL680, respectively (Supplementary Fig. 8a, b). The localization precision of both proteins was quantified at ~6–7 nm (Supplementary Fig. 8c), validating the high spatial accuracy of the MINFLUX system. We next applied dual-color 3D MINFLUX nanoscopy to visualize the spatial organization of ATP5a and TOMM20 within spine mitochondria of cultured neurons under control and cLTP conditions (Fig. 4a). Subsequent segment-based analysis of local protein density and its relationship to Syn intensity revealed that, in contrast to ATP5a, TOMM20 did not undergo a polarized redistribution following cLTP induction (Fig. 4b, c; Supplementary Fig. 8d–h). These results suggest that TOMM20, as an outer membrane protein, does not undergo the same dynamic reorganization as ATP5a during synaptic plasticity (Fig. 4d), highlighting the differential roles of mitochondrial membrane proteins in synaptic remodeling.

Fig. 4: Distinct redistribution patterns of inner and outer mitochondrial membrane proteins in dendritic spines.

a Representative 3D dual-color MINFLUX imaging of ATP5a and TOMM20 in dendritic spines under Control and cLTP conditions. Confocal images show the inner mitochondrial membrane protein ATP5a (red), outer membrane protein TOMM20 (green), and Syn (blue). MINFLUX imaging reveals the distribution of TOMM20 and ATP5a. Maximum z-projection (z-proj.) and 3D reconstructions were generated using ImageJ. Pseudo-coloring is consistent across all modalities. b, c Quantification of local density enrichment of TOMM20 (b) and ATP5a (c) near synaptic contact regions in the Control and cLTP groups. Local density calculations for dual-color imaging are provided in the Supplementary Fig. 8. (n-cLTP: 20 spine mitochondria from 14 neurons/3 cultures, n-Control: n = 16 spine mitochondria from 11 neurons/3 cultures) Significance: **P, ##P < 0.01. (TOMM20:ART ANOVA, Group factor: F(1, 71) = 0.7498, P = 0.3896; Region factor: F(1, 71) = 0.9172, P = 0.3416; Interaction (Group×Region): F(1, 71) = 1.3784, P = 0.2445. Bonferroni post hoc test: PControl_syn–vs. Control_syn+ <0.9999; PcLTP_syn– vs. cLTP_syn+ = 0.5494; PcLTP_syn+ vs. Control_syn+ = 0.3472; ATP5a:ART ANOVA, Group factor: F(1, 71) = 10.0817, P = 0.0023; Region factor: F(1, 71) = 1.9934, P = 0.1625; Interaction F(1, 71) = 5.9228, P = 0.0176. Bonferroni post hoc test: PControl_syn-vs. Control_syn+ > 0.9999; PcLTP_syn–vs. cLTP_syn+ = 0.0098; PcLTP_syn+ vs. Control_syn+ = 0.0066). d Proposed model showing that mitochondrial inner membrane protein ATP5a, but not outer membrane protein TOMM20, relocates within synapses in response to cLTP induction.

Discussion

Recent studies have elucidated the cellular mechanisms underlying the formation and maintenance of memory engrams in the brain, primarily using immediate early gene-based labeling systems combined with in vivo imaging, optogenetics, and chemogenetics9,34,35. Through observation of memory engram cells in the hippocampal DG region, researchers have found that the dendritic spine density of memory engram cells is significantly higher than that of non-engram cells36. Furthermore, by expressing light-sensitive channel proteins in the entorhinal cortex (EC) and using light to activate the projections to the DG, researchers observed that memory engram cells in the DG produced higher excitatory postsynaptic currents36. These findings indicate a specific increase in synaptic strength in the memory engram cell population. The plasticity of synapses after learning is dependent on new protein synthesis, and this process at the synaptic structures of neurons is highly energy-demanding, which highlights the critical role of mitochondrial function in synaptic plasticity. In line with this, studies have found that synaptic activities promote the presence of mitochondria in the dendritic protrusions20 and knockout of PINK1, a regulator of mitochondria dynamics, shows a reduction of mitochondria localizations in dendritic spine and creates a defect in learning and memory37. In our study, we further explored the mitochondrial dynamics in memory engram cells and observed an increased presence of mitochondria in dendritic spines compared to non-engram cells. We utilized the 3D MINFLUX nanoscopy technique, which offers single-molecule localization capabilities, to observe that the ATP5a on the inner mitochondrial membrane in the dendritic spines of memory engram cells in the mouse hippocampus undergoes redistribution. This suggests that, after learning, mitochondria actively participate in the plasticity changes occurring at the dendritic spines. The polarized distribution of ATP5a likely facilitates efficient ATP delivery to synaptic sites, promoting synaptic plasticity, including dynamic changes in actin, AMPA receptor trafficking, and local protein translation processes14. The redistribution phenomenon observed in the shaft synapses further supports this conclusion. It is known that in excitatory principal neurons, a certain proportion of glutamatergic synapses remain located in the dendritic shaft38. The maturation process of shaft synapses, which protrude to form more mature spines after learning, depends on F-actin remodeling39 and exocytic trafficking from recycling endosomes40, and optimized ATP5a protein distribution may further promote this process.

Interestingly, in the in vitro cLTP model used to simulate learning, we observed that the ATP5a redistribution phenomenon only lasted for up to 12 hours. However, in memory engram cells, the redistribution was observed on the third day after learning. This discrepancy may reflect the temporal dynamics of memory consolidation. After memory formation, the brain undergoes systems-level consolidation processes, including hippocampal replay during sleep, which serve to stabilize and reorganize memory traces41,42. Increasing evidence suggests that sleep is an active state during which neuronal ensembles engaged during learning are reactivated, supporting the consolidation of recent experiences43. More specifically, studies combining in vivo calcium imaging with the c-fos tTA transgenic mouse system and the photoconvertible protein KikGR have demonstrated that memory engram cells are preferentially reactivated during post-learning sleep44. This selective reactivation suggests that offline engram cell activity plays a key role in maintaining the integrity of memory ensembles across time. Such sustained activity patterns may provide continued plasticity signals, thereby prolonging the polarized redistribution of mitochondrial components in these cells to support memory stabilization. At the same time, we also observed that the outer mitochondrial membrane protein TOMM20 did not undergo specific redistribution during the synaptic plasticity process. This suggests that the distribution of inner and outer mitochondrial membrane proteins is regulated by different mechanisms, as studies reported that Ca2+ can activate aerobic respiration through a direct effect on ATP5a45,46. TOMM20, as a core component of the mitochondrial translocase complex, is responsible for importing proteins into mitochondria. Interestingly, another protein import component, TIM23, shows localization redistribution when protein synthesis and import are pharmacologically inhibited47. Therefore, the redistribution of protein reporters like TOMM20 may be regulated by different cellular signals.

MINFLUX, a localization concept proposed in 201748, has pushed 3D multicolor single-molecule imaging and tracking to unprecedented levels, achieving single-digit nanometer precision and ~100 µs time resolution, with a scalable field of view24. Our study presents the application of 3D MINFLUX imaging in brain tissues, with a focus on memory engram cell populations that are activated during learning and encode specific memories. In vitro cultured neuron systems often differ greatly in network characteristics from those in the biological brain, making observations in brain tissue essential for obtaining more accurate biological insights. Additionally, synaptic plasticity occurs at the nanoscale level of synaptic connections. Previous studies using STED/STORM super-resolution imaging research have provided valuable insights, such as the aligned nanomodules of pre- and postsynaptic proteins that contribute to structural plasticity49 and the existence of trans-synaptic molecular nanocolumns50. Given these findings, we believe MINFLUX can further enhance our understanding of nanoscale biological processes involved in learning, as well as the abnormalities in synaptic regulation seen in neurological diseases, thus providing deeper insights into disease mechanisms.

Materials and methods

Mouse subjects

In this study, male cFos-CreER mice (stock number 021882, The Jackson Laboratory) and wild-type C57BL/6 mice were housed in the Laboratory Animal Facility at ShanghaiTech University, which is accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International, under a 12-hour light/12-hour dark cycle (9 a.m.–9 p.m.) with unrestricted access to food and water. The experimental mice were males, aged 3 months. All mice were allowed to recover for at least 1 week from virus injection before any behavioral tasks. All animal protocols and experiments were reviewed and approved by ShanghaiTech University (license number 20201218002), following the Guide for the Care and Use of Laboratory Animals and in accordance with Chinese law (Laboratory Animal–Guideline for Ethical Review of Animal Welfare, GB/T 35892). This study adheres to all relevant ethical regulations for animal testing and research and received ethical approval from the Scientific Research Ethics Committee at ShanghaiTech University.

Mouse virus Injection

M