Animals used for neuron cultures
Pregnant female Sprague-Dawley rats were purchased from Charles River Laboratories (Wilmington, MA) and housed in the Center for Comparative Medicine and Surgery at the Mount Sinai School of Medicine which holds an Animal Welfare Assurance Number (A3111-01) signifying that it adheres to all National Institutes of Health guidelines for the care and treatment of laboratory animals. To obtain embryos for neuron culture preparation, pregnant rats were euthanized at E18.5 by CO2 asphyxiation and the embryos removed by caesarean section. All procedures were approved by the Mount Sinai School of Medicine Institutional Animal Care and Use Committee (IACUC) (protocol number 04-0548).
Neuron culture preparation and immunocytochemistry
To prepare dissociated neuron cultures for CLEM analysis of GFP-filled spines, hippocampi were harvested from E18.5 rat embryos and dissociated. 3–4.5×106 neurons were electroporated with 3 µg pEGFP-N1 plasmid DNA using the Rat Neuron Nucleofector Kit (Lonza Group Ltd., Basel, Switzerland) and plated on gridded glass bottom dishes (MatTek Corp., Ashland, MA) at approximately 1.7×105 neurons/dish in MEM (Invitrogen, Carlsbad, CA) supplemented with L-glutamine. Before plating, the glass was treated for 12–24 hours with 1 M hydrochloric acid, washed 3×20 minutes with water, left to dry completely, treated for 8 hours with 1 mg/ml poly-L-lysine (Sigma-Aldrich, St. Louis, MO) in borate buffer, washed 3×10 minutes with water, then left to dry completely. Four hours after plating, the MEM was replaced with Neurobasal (Invitrogen) supplemented with B27 (Invitrogen) and L-glutamine. Neurons remained in this media at 37°C and 5% CO2 until fixation at 21 DIV with 4% glutaraldehyde in 0.1 M cacodylate buffer.
To prepare dissociated neuron cultures for immunocytochemistry, neurons were obtained as above and plated on 18 mm diameter coverglasses (Fisherbrand coverglass for growth, Thermo Fisher Scientific, Waltham, MA) then fixed with 4% paraformaldehyde at 7 to 27 DIV. Prior to plating, these coverglasses were treated with >65% pure nitric acid for 48–72 hours, washed 3×20 minutes with water, then sterilized and dried completely in an oven. Immunolabeling was carried out as described 
using primary antibodies against vGlut1 (Millipore, Billerica, MA), PSD95 (Thermo Fisher Scientific), and SV2 (Developmental Studies Hybridoma Bank, University of Iowa, Iowa City, IA).
To prepare dissociated neuron cultures for long-term live laser scanning confocal microscopy (LSCM), neurons were again obtained as above and divided equally between two tubes, each containing 1.5–4.5×106
neurons. The neurons in one tube were electroporated with 3 µg VAMP2-DsRed plasmid DNA (obtained from Dr. Kimberley McAllister, U.C. Davis) 
and the other with 3 µg pEGFP-N1 as described above then plated in glass bottom dishes at a ratio of 1.5 VAMP2-DsRed to 1 GFP-transfected neuron at a total density of 4×106
neurons/dish. This density and plating ratio maximized the number of axodendritic contacts between GFP-filled spines and VAMP2-DsRed-labeled boutons. The cultures were grown for 14 to 25 DIV before long-term live LSCM.
LSCM was performed using a Zeiss LSM 510 META (Carl Zeiss, Inc., Oberkochen, Germany) with a 63× oil objective (plan apochromat, numerical aperture 1.4). In addition, a 10× air objective was used to obtain low magnification images of the neurons in their locations on the gridded coverslips to facilitate neuron relocation for CLEM. The EGFP was excited by 488 nm wavelength argon laser light and its emission detected through a long pass 505 nm filter through a pinhole set to 1 Airy unit. For the 63× LSCM, z-stacks of 50 to 100 images were taken with an x-y pixel size of 0.14 µm2, an optical slice thickness of 0.7 µm and a z-step of 0.10, 0.14, 0.33, or 0.37 µm. The number of images in each z-stack depended on the thickness of the dendritic segment in the stack. Automated analysis of spine head distance was performed only with z-stacks taken with a 0.10 or 0.14 µm z-step. The scan speed was 3.2 µs/pixel with 4 line averaging. Z-stacks were deconvolved using AutoDeblur (version X1.4.1, MediaCybernetics, Bethesda, MD) prior to automated spine analysis in NeuronStudio.
For long-term live LSCM, the same 63× objective was used, the EGFP was excited by and detected through the same laser and filter, the x-y pixel size was 0.14 µm2
, the optical slice thickness was 0.7 µm, the z-step was always 0.37 µm, and the scan speed was 1.6 µs/pixel with 2 line averaging. DsRed tagging VAMP2 was excited by 543 nm wavelength helium neon laser light and its emission detected through a band pass 585–615 nm filter. The GFP and DsRed were detected independently in separate channels. Z-stacks were taken every hour for 8 to 35 hours with a scan time of 3 to 10 minutes for each image depending on the size of the region in the z-stack. Prior to placing the dish on the microscope, the microscope chamber was equilibrated to 37°C and 5% CO2
for at least 1 hour. After focusing the objective on the dendritic segment to be recorded, the dish was left to equilibrate for at least 1 more hour to prevent vertical drift during the long-term live LSCM. Dishes were covered with a PTFE (polytetrafluoroethylene) membrane (DuPont, Wilmington, DE) sealed around the dish using a rubber band to prevent media evaporation while allowing gas exchange. This covering has been used previously to ensure long-term survival during live microscopy 
Neuron processing for CLEM
After conducting LSCM of GFP-filled dendritic segments, synaptic contacts were verified by CLEM using methods previously described 
. Briefly, the neurons in the dishes were treated with 1% osmium tetroxide plus 1.5% potassium ferricyanide in 0.1 M cacodylate buffer, dehydrated in an ascending ethanol series (50%, 60%, 70%), left in 3% uranyl acetate in 70% ethanol for 12 hours at 4°C, washed in 70% ethanol, then further dehydrated in an ascending ethanol series (80%, 90%, 100%). The neurons were infiltrated with a 1
1 solution of resin (Embed 812 kit, Electron Microscopy Sciences) and 100% ethanol for 24 hours at room temperature. The resin-ethanol solution was then replaced with a thin layer of pure resin and neurons of interest were embedded and sectioned through at 70 nm. The grid on the coverslip transferred to the resin (Fig. S1C
) and was used as a guide during sectioning. Serial sections were collected on Formvar-coated slot grids. Sections were contrasted with lead citrate and uranyl acetate and serial sections of the cell of interest were recorded on an Hitachi H-7000 (Hitachi, Ltd., Tokyo, Japan) transmission electron microscope at 12–40 k magnification and 75 kV voltage. High (63×) and low (10×) magnification LSCM images were used as maps for relocating regions of interest based on dendrite and spine morphology.
Automated dendritic spine detection
Deconvolved LSCM z-stacks of GFP-filled dendritic segments containing both CLEM-verified MSB spine pairs and control pairs of randomly selected adjacent spines were loaded into the NeuronStudio program (available at http://research.mssm.edu/cnic/tools.html
) for automated analysis 
. Only well-isolated, well-developed spiny dendritic segments not in contact with other dendrites, glial cells, or excessive axons were selected for analysis. Dendritic segments of interest were selected manually then traced and reconstructed in 3D for automated spine detection. Tracing errors and non-spinous entities misidentified as spines were eliminated manually. All spines were numbered for identification.
Spine parameters for both MSB and control spine pairs were imported into Excel (Microsoft Corp., Redmond, WA) for analysis. The distance between spine heads was measured as the distance between the centers of mass of the heads using the x-y-z coordinates determined by NeuronStudio. Spine orientation was determined manually by visual inspection of the reconstructed dendrite in 3D. For this analysis, the set of 22 control spines was different than that used for the other analyses because spine orientation had to be manually determined and was not clearly discernable in the original set. Spine with heads nearer than bases were classified as angled towards each other, spines with heads further than bases were classified as angled away from each other, and spines with equidistant heads and bases were classified as in parallel.
Spine stability analysis
Live LSCM z-stack projections containing both MSB and SSB synapses were loaded into ImageJ (version 1.42q, freely downloadable from the ImageJ website, http://rsbweb.nih.gov/ij/
) for processing and analysis. Brightness and contrast were adjusted separately for the VAMP2-DsRed and GFP detection channels and noise was reduced separately for each channel by automatic removal of outlier pixels, defined as single pixels with any brightness having only completely dark adjacent neighbors. To match the duration across each live LSCM series, only the first 15 hours of each series were included for analysis.
Spine stability was evaluated for MSB spine pairs compared to SSB spines paired with their nearest clearly observable neighbors. The volume of each spine head was measured as the integrated density of the GFP fluoresence of each spine head calculated by ImageJ. Integrated density is mean brightness multiplied by area. The area of each spine head was defined manually by tracing the contour of each spine head. To normalize the data, the dominant spine for each pair was preselected as the spine with the higher average integrated density over the 15 hour LSCM series. Then, a difference index was calculated for each time point as the integrated density of the dominant spine minus the nondominant spine divided by the sum of the two according to the formula:
is the difference index, IntDen
is the integrated density, Dom
is the dominant spine, and Nondom
is the nondominant spine.
The difference indices were plotted over time from frame to frame using Excel and fit to linear trend lines. The slopes of the trend lines were taken as a measure of the degree of dominance, or the dominance rate. R2 values were also calculated for the trend lines.
One-tailed t tests were used to compare spine head distances and dominance rates for MSB-contacting versus control spines. A two-tailed Mann-Whitney U test was used for the spine head distance distribution. A χ2 test was used for the spine orientation distribution. Significance was set at an α level of 0.05 for all tests.