Ethics statement

This study was performed in accordance with the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health (NIH). All experiments using animals followed protocol approved by the Institutional Animal Care and Use Committee at the University of Pennsylvania.

Plasmids

The following plasmids were used: mito-DsRed2 (a gift from T. Schwarz, Harvard Medical School), EGFP–FMRP (gift from G. Bassell, Emory University), GFP-POLG2 (gift from W. Copeland, NIH), Halo–FMRP (subcloned from EGFP–FMRP into pFN21A-HaloTag-CMV vector from Promega), COX8A–BFP (Goldsmith et al.49), LAMP1–Halo (Gallagher and Holzbaur50), GFP–MFF (Addgene, 49153), GFP-RAB7 (Addgene, 12605), GFP–RAB7 T22N (Addgene, 12660), RFP–RAB5 (Addgene, 14437), EGFP–DRP1 (subcloned from pcDNA 3.1-Drp1 (Addgene, 34706) into pEGFP-C1 vector), pCRISPRia-vs2 (Addgene, 84832), PGK 4xMito-mEmerald (Addgene, 200430), pUbC-OsTIR1-myc-IRES-scFv-sfGFP (Addgene, 84563), AID–SunTag–MFF (subcloned from GFP–MFF, MFF cDNA clone (Transomic BC000797) and pUbC-FLAG-24xSuntagV4-oxEBFP-AID-baUTR1-24xMS2V5-Wpre (Addgene, 84561) into pEGFP-N1 vector with EGFP removed), EGFP–TWINKLE (subcloned from TWINKLE–APEX2-V5 (Addgene, 129705) into pEGFP-N1 vector), EGFP–TFAM (subcloned from pCellFree_G03 TFAM into pEGFP-N1 vector) and Halo–Stop (Cason et al. 51). Halo–FMRP ΔRGG (deletion of amino acids 526-552), Halo–FMRP G266E, Halo–FMRP I304N, Halo–FMRP S499A and Halo–FMRP S499D were generated by mutagenesis and subcloning of Halo–FMRP.

Antibodies

The following antibodies were used for ICC, western blot (WB), PLA and DNA-PAINT: anti-FMRP (Millipore, MAB2160) at 1:500 dilution for ICC and WB, anti-HSP60 (Sigma-Aldrich, SAB4501464) at 1:100 dilution for ICC, anti-puromycin (Millipore, MABE343) at 1:250 dilution for PLA, anti-puromycin Alexa Fluor 647 conjugate (Millipore, MABE343-AF647) at 1:250 dilution for ICC, anti-MFF (Proteintech, 17090-1-AP) at 1:1,000 dilution for ICC and PLA, anti-RPS3A (Thermo Fisher, 14123-1-AP) at 1:500 dilution for ICC, anti-Neurofilament Heavy Chain (Aves Labs, NHF) at 1:2,000 dilution, anti-TOM20 (Proteintech, 11802-1-AP) at 1:100 dilution for DNA-PAINT. The following secondary antibodies were used: Alexa Fluor 488 goat anti-rabbit IgG (H+L) (Invitrogen, A11034) at 1:1,000 dilution for ICC, Alexa Fluor 594 goat anti-mouse IgG (H+L) (Invitrogen, A11032) at 1:1,000 dilution for ICC, Alexa Fluor Plus 647 goat anti-mouse IgG (H+L) (Invitrogen, A32728), at 1:1,000 dilution for ICC, IRDye 680RD donkey anti-rabbit IgG (LI-COR, 926-68073) at 1:20,000 dilution for WB, IRDye 800CW donkey anti-mouse IgG (LI-COR, 926-32212) at 1:20,000 dilution for WB. Docking-strand-2-conjugated anti-rabbit secondary antibody and docking-strand-3-conjugated anti-GFP nanobody were custom-made by Massive Photonics and used at 1:100 dilution for DNA-PAINT.

Primary neuron culture

Embryonic day 18 Sprague–Dawley rat hippocampal neurons were obtained from the Neurons R Us Culture Service Center at the University of Pennsylvania. Cells (live imaging, 220,000 cells on 20-mm glass; fixed imaging, 150,000 cells on 20-mm glass; PLA, 110,000 cells on 7-mm glass) were plated in 35-mm glass-bottom dishes (MatTek, P35G-1.5-20-C or P35G-1.5-7-C) that were precoated with 0.5 mg ml−1 poly-l-lysine (Sigma-Aldrich, P1274). Cells were initially plated in attachment medium (MEM supplemented with 10% horse serum, 33 mM d-glucose and 1 mM sodium pyruvate). After 5 h, attachment medium was replaced with maintenance medium (Neurobasal (Gibco) supplemented with 33 mM d-glucose, 2 mM GlutaMAX (Invitrogen), 100 U ml−1 penicillin, 100 mg ml−1 streptomycin and 2% B27 (Thermo Fisher)). On the following day, cytosine arabinoside (1 µM) was added to the cultures to prevent proliferation of non-neuronal cells. Neurons were maintained at 37 °C in a 5% CO2 incubator for 6–12 days before transfection or fixation.

Fmr1 KO (Strain 003025) and C57BL/6J (Strain 000664) mice were obtained from The Jackson Laboratory and housed on a reversed 12-h light–dark cycle at 22 ± 2 °C with 50 ± 20% humidity. Mouse cortex from C57BL/6J (WT) embryos of either sex or a mixture of hemizygous male Fmr1 KO and homozygous female Fmr1 KO embryos was dissected in 1× HBSS (Gibco) at day 15.5. Cortical neurons were isolated by digestion with 0.25% trypsin and trituration through a pipette tip. Cells were plated and cultured as described for rat hippocampal neurons, except that the attachment medium and maintenance medium were supplemented with 37.5 mM NaCl.

Human iPS cell and i3Neuron culture

Human WTC11 iPS cells that harbour a doxycycline-inducible NGN2 transgene at the AAVS1 locus and stably express dCas9-BFP-KRAB were a gift from M. Ward at the NIH and have been previously described24. iPS cells were cultured on dishes coated with human embryonic stem cell-qualified Matrigel (Corning) and fed daily with Essential 8 medium (Thermo Fisher). iPS cells were differentiated into i3Neurons following a previously described protocol24. Following differentiation, i3Neurons were cryo-preserved in i3Neuron medium (BrainPhys Neuronal Medium (StemCell) supplemented with 2% B27 (Gibco), 10 ng ml−1 NT-3 (PeproTech), 10 ng ml−1BDNF (PreproTech) and 1 µg ml−1 laminin (Corning)) with 10% DMSO added. i3Neurons (180,000–200,000 cells) were plated on 35-mm glass-bottom imaging dishes (MatTek, P35G-1.5-20-C) that were precoated with poly-l-ornithine (Sigma-Aldrich, P3655) overnight at 37 °C. i3Neurons were cultured for 21–22 days in 5% CO2 at 37 °C before fixation or transfection. Every 3–4 days, 40% of the medium was replaced with fresh culture medium.

Generation of FMR1 CRISPRi iPS cells

Human iPS cells and i3Neurons with knockdown of FMRP via CRISPR interference (CRISPRi) were made using previously described methods for gene knockdown in this system52,53. In brief, the FMR1-targeting guide RNA sequence GGCGGGCCGACGGCGAGCGC was flanked with BlpI and BstXI cut sites and ligated into BlpI/BstXI-digested pCRISPRia-vs2 using NEB Quick Ligase (New England Biolabs; M2200S). Guide incorporation was verified by sequencing. The sgRNA guide plasmid was then packaged into lentivirus and transduced into iPS cells. HEK293T cells were transfected with sgRNA plasmid, psPAX2 (HIV pol + gag), and pCMVVsv-g. Eight hours after transfection, the medium was replaced with fresh medium supplemented with ViralBoost (Alstem; VB100). Forty hours later, HEK293T medium was collected, filtered and centrifuged with Lentivirus Precipitation Solution (Alstem; VC100) to isolate viral pellet. The virus-containing pellet was resuspended in E8 medium with ROCK inhibitor, aliquoted and frozen at −80 °C. Resuspended virus was then added to iPS cells and incubated together for 2 days. Virus-infected iPS cells were then selected for with puromycin (Takara; 631305) for 4 days, with fresh medium added each day. iPS cells were then collected, cryo-preserved and differentiated into i3Neurons as described above.

HeLa cell culture

HeLa-M (A. Peden, Cambridge Institute for Medical Research) cells were maintained in DMEM (Corning, 10-017-CM) supplemented with 1% GlutaMAX (Thermo Fisher, 35050061) and 10% fetal bovine serum. Cells were maintained at 37 °C in a 5% CO2 incubator. Cells were routinely tested for mycoplasma contamination with a MycoAlert detection kit (Lonza, LT07). Cells were authenticated by STR profiling at the DNA Sequencing Facility at the University of Pennsylvania.

Transfection and nucleofection

Rat hippocampal neurons and mouse cortical neurons were transfected after 6–12 DIV. Neurons were transfected with 0.5–1.5 µg total of plasmid DNA using 4 µl Lipofectamine 2000 Transfection Reagent (Thermo Fisher). Neurons were incubated with lipid–DNA complexes for 45 min before replacing with conditioned medium. Primary rodent neurons were incubated for 18–24 h before fixation or live imaging. i3Neurons (DIV18) were transfected with 1 µg of plasmid DNA using 4 µl Lipofectamine Stem (Thermo Fisher). Neurons were then incubated with lipid–DNA complexes for 90 min before replacing with conditioned medium. i3Neurons were incubated for 72 h before fixation or live imaging. HeLa cells were plated on uncoated 35-mm glass-bottom dishes (MatTek, P35G-1.5-20-C), transfected with 1.5 µg total of plasmid DNA using FuGene 6 (Promega) and incubated for 24 h before live imaging. Mouse cortical neurons were nucleofected before plating with an Amaxa Nucleofector II (Lonza Bioscience) following the manufacturer’s instructions. For each nucleofection, 1 µg total of plasmid DNA was added to 3,000,000 neurons.

Live imaging

For experiments using live-cell imaging of rat and mouse neurons, cells were imaged at 37 °C in Hibernate E medium (BrainBits) supplemented with 2% B27 and 33 mM d-glucose. Neurons expressing Halo-tagged constructs were labelled with 100 nM Janelia Fluor 646-Halo ligand (Promega, GA1121) for 15 min followed by a 30 min washout before imaging. For experiments involving translational inhibition with cycloheximide or puromycin, neurons were treated with 50 µM cycloheximide (Sigma-Aldrich, 01810), 50 µM puromycin or an equivalent volume of DMSO for 30 min before imaging and during image acquisition. For experiments involving microtubule depolymerization with nocodazole, neurons were treated with 2 µM nocodazole (Sigma-Aldrich, M1404-10MG) or an equivalent volume of DMSO for 30 min before imaging and during image acquisition. For experiments examining mitochondrial DNA, neurons were treated with 1:10,000 dilution MitoTracker Deep Red FM (Thermo Fisher Scientific, M22426) and 1:200,000 dilution SYBR Green I (Thermo Fisher, S7563) for 30 min before imaging. For experiments assessing mitochondrial membrane potential, neurons were treated with 2.5 nM TMRE for 30 min before imaging, with 2.5 nM TMRE added to the imaging medium. For live-cell imaging of HeLa cells, the culture medium was replaced with Leibovitz’s L-15 medium (Gibco, 11415064) supplemented with 1% GlutaMAX and 10% fetal bovine serum. For live-cell imaging of i3Neurons, cells were imaged in Hibernate A medium (BrainBits) supplemented with 2% B27, 10 ng ml−1 BDNF and 10 ng ml−1 NT-3. For FRAP experiments, single EGFP–FMRP granules were photobleached in hippocampal neurons using the 488-nm laser line for eight photobleaching cycles with 9 ms per pixel. One pre-FRAP time point was collected and images were captured every second after bleaching for 240 s. For SunTag experiments, hippocampal neurons were transfected with AID–SunTag–MFF, pUbC-OsTIR1-myc-IRES-scFv-sfGFP, COX8A–BFP and Halo–FMRP following the above protocol. Then, 3-indoleacetic acid (Sigma, I2886) was diluted in ethanol and added to neuron cultures overnight at a final concentration of 500 µg ml−1 to degrade pre-existing proteins containing auxin-induced-degron.

All live-cell imaging experiments were performed using a PerkinElmer UltraView Vox spinning disk confocal on a Nikon Eclipse Ti Microscope, which is surrounded by a 37 °C imaging chamber. All samples were given several minutes to equilibrate before imaging, with each dish imaged for a maximum of 1 h. Videos were acquired at one frame per second for 5 min, one frame per 3 s for 15 min or one frame per 5 s for 30 min. Experiments were imaged on either a Hamamatsu EMCCD C9100-50 camera or a Hamamatsu CMOS ORCA-Fusion (C11440-20UP). The EMCCD camera was used with Volocity Software (Quorom Technologies/PerkinElmer) and the CMOS camera was used with VisiView (Visitron).

DNA-PAINT super-resolution microscopy

Hippocampal neurons were plated in Lab-Tek II eight-well chambered coverglass (Thermo Fisher, 155409) at a density of 10,000-15,000 cells per well. Chambers were precoated with poly-l-lysine and neurons were cultured as described above. Neurons in each well were transfected with 125 ng EGFP–FMRP following the above protocol and incubated for 24 h. Neurons were fixed for 10 min using warm PBS with 4% paraformaldehyde and 4% sucrose. Blocking buffer (PBS containing 10% donkey serum, 0.1% saponin and 0.05 mg ml−1 sonicated salmon sperm single-stranded DNA (Stratagene)) was then added for 1 h at 25 °C before incubation with rabbit anti-Tom20 (see above) in blocking buffer for 1 h at 25 °C. Neurons were then washed once with PBS and three times for 5 min in 1× Wash Buffer (Massive Photonics) before incubating with docking-strand-2-conjugated secondary anti-rabbit antibody and docking-strand-3-conjugated GFP nanobody. Neurons were then washed three times for 7 min with 1× Wash Buffer and twice with PBS.

Hippocampal neurons were imaged on a Nanoimager microscope (ONI) equipped with a ×100 oil immersion objective (NA 1.45), 405, 488, 561 and 640-nm lasers, 498–551 and 576–620-nm band-pass filters in channel 1, 666–705–839 nm band-pass filters in channel 2 and an 840 Hamamatsu Flash 4 V3 sCMOS camera. Imager strands 2 and 3 conjugated to either Cy3B or ATTO655 were added at a concentration of 0.25 nM in Imaging Buffer (Massive Photonics). Images were collected at HiLo illumination angle with 100 ms exposure using a laser programme alternating between 100 frames collected of the TOM20 channel and 200 frames for the EGFP–FMRP channel, repeated for a total of 25,000 frames. Localizations were generated and drift corrected by Nanoimager operating and analysis software (ONI) and exported as csv files for downstream processing with the following localization filter parameters: 300 photon minimum, 30 nm maximum localization precision in x and y directions and 10–250 nm sigma in x and y directions.

Correlative light and electron microscopy and cryo-electron tomography

Cryo-electron microscopy (EM) grids (Quantifoil 2/2, LF, 200 mesh) were glow-discharged, placed in 35-mm glass-bottom dishes (MatTek, P35G-1.5-20-C) and coated with poly-l-lysine as described above. Rat hippocampal neurons were plated at a density of 70,000–150,000 cells across 3–4 grids and cultured as described above. Cells were transfected with EGFP–FMRP and/or mito-DsRed2 at DIV5. Twenty-four hours after transfection, grids were supplemented with 10-nm colloid gold beads and blotted from the back side before plunge-freezing using a Leica EM GP2 robot. Cryo-grids were imaged with a Leica EM Cryo-CLEM Cryo Light Microscope or Zeiss LSM900 Airyscan2 cryo-confocal microscope. Z-stack images were collected for individual cells with both bright-field and fluorescence channels, from which maximum-intensity projection images were produced for subsequent correlative imaging.

Grids were transferred to a Titan Krios transmission electron microscope (Thermo Fisher) for tomography, which was operated at 300 kV acceleration voltage, equipped with a post-column BioQuantum energy filter and a K3 direct electron detector (Gatan). Full-grid montages were captured at ×82 magnification to identify cells of interest. Each cell was further imaged at ×470 magnification to obtain maps covering a grid square, which were used for correlation with fluorescence images using CorRelator54 program. The regions of interest were located by the EGFP fluorescence signal overlaid on cryo-EM maps, and the candidate targets were further imaged at ×4,800 and ×19,500 magnifications, at which features of mitochondria could be clearly recognized. The targets for data collection were selected at the interface between the EGFP fluorescence spot and nearby mitochondria. Tilt series were collected at ×33,000 magnification, with a calibrated pixel size of 2.65 Å, following the dose-symmetric scheme, spanning from −60° to 60° with 3° increments. Images were recorded with an electron dose rate of 40 e− per pixel per s. Each tilt was exposed for 0.45 s, which was fractionated into five movie frames, resulting in an accumulative dose of 105 e−/Å2 for each tilt series. To boost the contrast of cellular features, the Volta phase plate was used during data collection. The defocus range of the images was −1 to −3 μm.

Immunocytochemistry

Hippocampal neurons and i3Neurons were fixed for 10 min using warm PBS with 4% paraformaldehyde and 4% sucrose. Cells were washed three times with PBS, permeabilized with 0.2% Triton X-100 in PBS for 15 min and washed three times with PBS. Cells were blocked for 1 h in blocking solution (5% goat serum and 1% BSA in PBS) and incubated overnight at 4 °C with primary antibodies diluted in blocking solution. Cells were then washed three times with PBS, incubated with secondary antibodies diluted in blocking solution for 1 h and washed three more times. Neurons were imaged using the spinning disk confocal setup described above, with Z-stacks collected at 200-nm step-size.

Puromycylation assay

Hippocampal neurons were transfected with mito-DsRed2 and EGFP–FMRP following the above protocol. After 24 h, neurons were treated with 2 µM puromycin for 10 min then fixed and permeabilized according to the ICC protocol. Neurons were incubated with anti-puromycin Alexa Fluor 647 conjugate (see above) for 16 h at 4 °C, washed three times with PBS, and imaged using a PerkinElmer UltraView Vox spinning disk confocal, as described above, with Z-stacks collected at 200-nm step-size. Puromycin enrichment for EGFP–FMRP was defined as overlap with a puromycin-positive puncta. For experiments involving translational inhibition with cycloheximide, neurons were treated with 50 µM cycloheximide or equivalent volume DMSO for 30 min before the addition of puromycin.

Single-molecule RNA-FISH

Custom single-molecule RNA-FISH probes with Quasar 670 dye were designed against the mRNA coding sequences for each gene using the Stellaris Probe Designer (Biosearch Technologies; Supplementary Table 1). The predesigned human GAPDH with Quasar 670 Dye (Biosearch Technologies, SMF-2019-1) was used as a control. DIV21 i3Neurons or DIV10 mouse neurons were fixed with 4% paraformaldehyde in PBS for 10 min at room temperature and washed three times with PBS. Neurons were stored in 70% ethanol at 4 °C for up to 2 weeks. Neurons were then washed with Wash Buffer (2× SSC (20× SSC (Corning 46-020-CM)) and 10% formamide diluted in nuclease-free water (Thermo Fisher, BP561-1)) for 5 min. Pools of Quasar 670-conjugated FISH probes at 250 nM were combined with anti-FMRP and anti-HSP60 antibodies (see above) in Hybridization Buffer (2× SSC solution containing 5% dextran sulfate (Sigma-Aldrich, D8906) and 10% formamide diluted in nuclease-free water) and hybridized to cells for 16 h at 37 °C. Neurons were then washed in Wash Buffer for 30 min at 37 °C to remove unbound probes and incubated in Wash Buffer with secondary antibodies (see above) for 30 min at 37 °C. Neurons were then washed three times in 2× SSC buffer for 5 min and mounted in GLOX Buffer (2× SSC solution containing 10 mM Tris-HCl, pH 8.0, 0.4% glucose, 0.5 mg ml−1 glucose oxidase (Sigma-Aldrich, G0543) and 470 U ml−1 catalase (Sigma-Aldrich, C3155)). Neurons were imaged using a PerkinElmer UltraView Vox spinning disk confocal microscope, as described above, with Z-stacks collected at 200-nm step-size.

Proximity ligation assay

Hippocampal neurons were transfected with COX8A–BFP and EGFP–FMRP following the above protocol. After 24 h, neurons were treated with 2 µM puromycin for 10 min, fixed in PBS containing 4% paraformaldehyde and 4% sucrose for 10 min at room temperature, washed three times in PBS, permeabilized for 15 min in PBS with 0.2% Triton X-100, and washed with PBS. Newly synthesized proteins were detected using Duolink In Situ PLA Mouse/Rabbit kit with red detection reagents (Sigma-Aldrich, DUO92101-1KT) according to the manufacturer’s protocol with a mouse anti-puromycin antibody in combination with a rabbit anti-MFF antibody (see above for details). Neurons were imaged using a PerkinElmer UltraView Vox spinning disk confocal, as described above. Z-stacks were collected at 200-nm step-size. PLA puncta were then manually counted and scored for overlap with EGFP–FMRP or COX8A–BFP.

Immunoblotting

Neurons were washed twice with ice-cold PBS and lysed with ice-cold RIPA buffer (50 mM Tris-HCl, 150 mM NaCl, 0.1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS and 2× Halt Protease and Phosphatase inhibitor). Cells were snap frozen in liquid nitrogen and incubated in RIPA buffer for 30 min on ice. Samples were then centrifuged for 10 min at 17,000g. The protein concentration of the supernatant was determined by BCA assay, then samples were denatured in sample buffer containing SDS at 95 °C. Samples were analysed by SDS–PAGE and transferred onto PDVF Immobilon FL membranes (Millipore). After drying for 1 h, membranes were rehydrated in methanol and stained for total protein (LI-COR REVERT Total Protein Stain). Following imaging of total protein, membranes were de-stained with 0.1 M NaOH supplemented with 30% methanol, blocked for 1 h in EveryBlot Blocking Buffer (Bio-Rad), and incubated overnight at 4 °C with primary antibodies diluted in Blocking Buffer with 0.2% Tween-20. Membranes were washed three times for 5 min in 1× TBS Washing Solution (50 mM Tris-HCl, pH 7.4, 274 mM NaCl, 9 mM KCl and 0.1% Tween-20), incubated in secondary antibodies diluted in Blocking Buffer with 0.01% SDS for 1 h and washed three times for 5 min in Washing Solution. Membranes were imaged using an Odyssey CLx Infrared Imaging System (LI-COR). Band intensities were measured in Image Studio (LI-COR).

Image analysisFraction of FMRP on mitochondria

To quantify the number of endogenous FMRP granules contacting mitochondria in i3Neurons and rat hippocampal neurons, we generated max projections of the mitochondria and FMRP channels in ImageJ. Regions containing only neurites were selected from these max projections. We then converted the area for FMRP and mitochondria to binary masks using the pixel classification module of Ilastik, a machine-learning based image segmentation program55. The masked images of FMRP and mitochondria were then overlayed and scored for overlap in ImageJ. To quantify the number of EGFP–FMRP granules contacting mitochondria, granules were manually counted and scored for overlap with mitochondria. For presentation purposes, sections of individual neurons were ‘straightened’ using the Straighten command in ImageJ.

DNA-PAINT

Mitochondria localizations were Voronoi-segmented and clustered based on a minimum Voronoi area of 5,476 nm2 and minimum of 15 localizations using a custom-made MATLAB code. Mitochondria clusters with area less than 0.110 µm2 were considered nonspecific background clusters and discarded. FMRP localizations were Voronoi-segmented and clustered based on a minimum Voronoi area of 548 nm2 and minimum of ten localizations. FMRP clusters with area less than 0.018 µm2 were considered nonspecific background clusters and discarded. Images surrounding mitochondria in neuronal processes were cropped and used for downstream analysis.

A custom colocalization analysis was performed using MATLAB (https://github.com/LakGroup/Data_Analysis_Software_Fenton_etal) on the Voronoi-segmented data. The mitochondria channel was used as ‘reference’ channel, and the FMRP channel as the ‘colocalization’ channel. The colocalization analysis determined the overlap between the reference and colocalization clusters, and they were considered to colocalize when there is overlap between them. This analysis split the data of the colocalization (FMRP) channel into two subsets: FMRP that is localized within a mitochondrion and FMRP that is localized outside mitochondria. In practice, an initial search was performed to find reference clusters with colocalization clusters in their vicinity, without limiting the number of colocalization clusters associated with a reference cluster, and without limiting the number of reference clusters a colocalization cluster was associated with. This was performed to speed up calculations, and achieved by making a low-resolution polygon out of the reference clusters, and then isometrically enlarging them with five pixels (585 nm). Any colocalization cluster found within this region was then associated to a reference cluster and considered in further steps of the colocalization analysis. The second step in this colocalization analysis described the reference clusters that have colocalization clusters associated to them as high definition alphaShape objects (a polytope that is a subset of the Delaunay triangulation of the clusters and includes all localizations in the cluster, but also describe holes present in the clusters). The individual localizations of any colocalization cluster that were associated to a reference cluster were then evaluated to be inside or outside that reference cluster. Colocalization clusters with localizations inside a reference cluster were considered colocalized, whereas colocalization clusters without localizations inside a reference cluster were not. This separated the colocalization clusters in two groups on which the different metrics (area, distance to mitochondrion and circularity) were calculated.

The area of the individual FMRP clusters was calculated from their alphaShape rendering, as well as their perimeter. The circularity of the clusters was then obtained by applying the following formula: \(\mathrm }=\,\frac}}}}^}\), which is 1 for a perfect circle and near 0 for highly noncircular shapes. To measure the distance from FMRP to mitochondria for FMRP clusters that do not make mitochondrial contact, we determined which ten mitochondrial clusters were closest to the FMRP cluster and then calculated the border-to-border distance (FMRP to mitochondrion) between these. The lowest distance between any of these results is reported.

FMRP–mitochondria contact analysis over time

FMRP–mitochondria contacts were determined by manually drawing a line along the length of an axon or dendrite from a confocal video. A kymograph was then generated using the Multiple Kymograph plugin for ImageJ. Tracks of individual FMRP granules in contact with mitochondria were manually traced in ImageJ to determine contact duration and displacement. Motile FMRP granules (anterograde or retrograde) were defined by a net displacement ≥5 µm, whereas nonmotile granules were defined by a net displacement ˂5 µm.

Normalized position along mitochondria

The positioning of FMRP granules or MFF puncta along mitochondria was measured by manually drawing a line along the length of an axon or dendrite from a confocal video. A kymograph was then generated in ImageJ and used to determine the ends of each mitochondrion. The location of each FMRP granule or MFF puncta in contact or within 1 μm distance from each mitochondrion was recorded and used to determine the distance to both mitochondrial ends. The distance from the proximal mitochondrial end was then divided by the mitochondrial length to determine the relative position along the mitochondrion, with ends normalized to 0 and 1.

Determination of fission events

Mitochondrial fission was defined as any event that showed clear division of a single mitochondrion into two distinct mitochondria that display uncoupled movement following division. A fission event was considered positive for a marker if the relevant marker was within 500 nm distance of the fission location in the frame before fission occurs. To determine when a marker appeared at a fission site before fission, a circular region of interest (ROI) with 1 µm radius was drawn around the fission site. The number of seconds preceding fission while each marker remained within the ROI was reported. Fission events in which both FMRP and DRP1 or MFF puncta are present in the ROI at the beginning of the recording were excluded from analysis. The position of mitochondrial fission was measured by manually drawing a line along the length of the mitochondrion in the frame before fission.

To measure fission rates, regions of neurons were selected in which individual mitochondria were easily resolvable and did not leave the focal plane. The mitochondrial network was then segmented using Ilastik to determine total mitochondrial area. To account for changes in mitochondrial density over time, the final mitochondrial area was determined by averaging the areas in the ROI for the first and last frames of the video. Fission rates are indicated as the number of fission events per square micron of mitochondria.

Colocalization analysis

To quantify the number of mRNA puncta in contact with FMRP granules and mitochondria we generated max projections of each channel in ImageJ. Regions containing only neurites were selected from these max projections. We then converted each channel to a binary mask using Ilastik, as described above. The masked images for mRNA puncta and FMRP were then overlayed and scored for overlap in ImageJ. For mRNA puncta that overlap with FMRP, the segmented FMRP granule was scored for overlap with mitochondria. For presentation purposes, sections of individual neurons were ‘straightened’ using the Straighten command in ImageJ. Manders’ overlap coefficients for FMRP-RPS3A overlap were determined using the JACoP plugin in ImageJ56.

FRAP

The mean fluorescence intensity was measured for each ROI containing an EGFP–FMRP granule. The ROI was moved if and when each granule moved out of the initially determined ROI. The intensities of the ROIs were normalized in Microsoft Excel such that the prebleached intensity is 1 and the first photobleached frame is 0.

TMRE quantification

Max projections were generated for each neuron and selected a region containing a single axon or dendrite in ImageJ. We then converted the mitochondria channel to a binary mask using Ilastik, as described above. The mean grey value of the TMRE channel was then measured within the mitochondrial regions in ImageJ. The TMRE intensity was normalized in Microsoft Excel so the average intensity across biological replicates of the WT condition is 1.

Tomogram reconstruction and subtomogram averaging

The movie frames of each tilt angle were aligned to correct the beam-induced motion and assembled into tilt-series stacks in Warp57. Initial tilt-series alignment was performed with EMAN2 (ref. 58) software and the tomograms were reconstructed with a binning factor of 4. The ribosome particles were manually picked with e2spt_boxer.py and an initial model was calculated using e2spt_sgd.py which clearly resolved the features of large and small subunits of the ribosome. The particle orientations were refined stepwise using e2spt_refine.py with particles at 4× and 2× binning levels. To achieve the best possible resolution, we made use of the multiparticle refinement approach of Warp-Relion-M pipeline59. The tilt series were re-aligned with autoalign_dynamo using the 10-nm gold beads as fiducials. Tomograms were reconstructed using Warp with deconvolution filter applied to facilitate visualization and particle picking in IMOD60. A total of 378 particles were picked from three tomograms. Subtomograms were first reconstructed with a binning factor of 4. A round of three-dimensional refinement in Relion-3.1 (ref. 61) produced a reconstruction at 30 Å resolution, estimated by the Fourier shell correlation cutoff of 0.143. The unbinned subtomograms were then reconstructed and fed to M for multiparticle refinement57, which resulted in a density map at 27 Å resolution (Fig. 4b,f).

Tomogram postprocessing, segmentation and annotation

Due to multiple steps of transfer, the sample got some surface contamination by crystalline ice. These dense features produced some smeared artifacts in multiple slices of the tomogram. To remove these artifacts, a mask covering the ice density was created in IMOD, which was used to reproject the density to tilt-series images for signal subtraction using the Masktorec program62. Clean tomograms were reconstructed with the signal-subtracted tilt images and contrast-enhanced with the nonlinear anisotropic diffusion filter in IMOD. The tomogram was segmented manually using IMOD and the density for each feature was cropped out using the imodmop function to facilitate three-dimensional visualization and analysis. The subtomogram average map of the ribosome was placed back into the tomogram according to the original particle position and angular information using ArtiaX63. The figures were rendered with either IMOD or Chimerax64.

Postprocessing of correlation

The bright-field light microscopy images were first correlated with the ×470 cryo-EM map using the hole centroid as fiducials. The same transformation matrix was applied to the fluorescence channel images. For registration of other cryo-EM images at higher magnifications, the correlation was performed solely between EM images at adjacent magnifications, from which the transformation matrix was derived and applied to fluorescence images for stepwise correlation. The ice contamination served as fiducials for correlating EM images between ×470 and ×4,800 magnifications and the 10-nm gold colloid beads were used for higher magnifications. The tomogram slices were extracted and correlated with the ×19,500 EM images using the structural features as fiducials, which allowed the annotation of tomogram features with fluorescence signals.

Simulating FMRP distributions

Random FMRP distributions were generated in ImageJ. For each neuron, a mask was created to outline the boundaries of each neurite. The number of FMRP granules within each mask was manually counted. A custom macro was used to generate points with random xy coordinates within the masked region. The same number of points as FMRP granules were generated for each mask. The distance from each mtDNA marker to the nearest generated point was then measured for all mtDNA puncta within each mask.

Statistics and reproducibility

Statistical analyses and graphing were performed using GraphPad Prism v.10.1.0. When appropriate, we performed statistical tests only on data from biologically independent replicates. A D’Agostino–Pearson test was performed to determine whether to use parametric or nonparametric statistical tests. Two-tailed unpaired Student’s t-tests were used to compare two normally distributed datasets. A two-tailed paired t-test was used to compare the timing of puncta appearance at fission sites. An ordinary one-way ANOVA with Tukey’s multiple comparisons test or Dunnett’s multiple comparisons test was used to compare normally distributed datasets in experiments with more than two conditions. *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001 were considered significant. All statistical analyses were performed on data from 3–5 independent experiments. All representative images and kymographs were selected from datasets with at least three independent experiment showing similar results. No statistical method was used to predetermine sample size. For experiments with multiple conditions, neuron cultures from the same dissection batch were randomly assigned to each experimental condition. Data collection and analysis were not performed blind to the conditions of the experiments. For imaging experiments, cells were excluded from analysis if the cell displayed signs of stress, such as excessive rounding of mitochondria or aggregation of FMRP or if image quality was poor.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.