Remodeling and Tenacity of Inhibitory Synapses: Relationships with Network Activity and Neighboring Excitatory Synapses

Abstract

Glutamatergic synapse size remodeling is governed not only by specific activity forms but also by apparently stochastic processes with well-defined statistics. These spontaneous remodeling processes can give rise to skewed and stable synaptic size distributions, underlie scaling of these distributions and drive changes in glutamatergic synapse size "configurations". Where inhibitory synapses are concerned, however, little is known on spontaneous remodeling dynamics, their statistics, their activity dependence or their long-term consequences. Here we followed individual inhibitory synapses for days, and analyzed their size remodeling dynamics within the statistical framework previously developed for glutamatergic synapses. Similar to glutamatergic synapses, size distributions of inhibitory synapses were skewed and stable; at the same time, however, sizes of individual synapses changed considerably, leading to gradual changes in synaptic size configurations. The suppression of network activity only transiently affected spontaneous remodeling dynamics, did not affect synaptic size configuration change rates and was not followed by the scaling of inhibitory synapse size distributions. Comparisons with glutamatergic synapses within the same dendrites revealed a degree of coupling between nearby inhibitory and excitatory synapse remodeling, but also revealed that inhibitory synapse size configurations changed at considerably slower rates than those of their glutamatergic neighbors. These findings point to quantitative differences in spontaneous remodeling dynamics of inhibitory and excitatory synapses but also reveal deep qualitative similarities in the processes that control their sizes and govern their remodeling dynamics.

Fig 1. mTurq2:Geph expression in cortical neurons growing on MEA dishes.

Fluorescence images of two neurons expressing mTurq2:Geph (left) and brightfield images of the same regions (right). Note the punctate distribution of mTurq2:Geph along dendritic shafts. The position of these neurons relative to nearby extracellular electrodes (opaque circles) and leads (transparent) is shown in the right-hand panels (yellow arrowheads). Bar, 30μm.

Fig 2. mTurq2:Geph colocalizes with GABA_A receptors and presynaptic boutons of GABAergic neurons.

A,B) GABA_A receptors were labeled in live hippocampal neurons using antibodies against extracellular epitopes of the GABA_A receptor γ₂ subunit (A) or GABA_A receptor β_2,3 subunits (B). Left panels: mTurq2:Geph; Middle panels: antibody labeling; Right panels: combined images. Note the good colocalization of mTurq2:Geph puncta with clusters of labeled receptors (arrowheads point to some examples). Bars, 10μm. C) Functional presynaptic boutons of GABAergic neurons were labeled in live neurons using anti vesicular GABA transporters (VGAT) antibodies. Top panels—mTurq2:Geph (left) and anti-VGAT (right). Middle panels—differential interference contrast (DIC) image of the same field (left) and the combined image (right). Bottom panels—enlarged images of region enclosed in rectangle in right middle panel. Note the good colocalization of mTurq2:Geph puncta with clusters of labeled VGAT (arrowheads). Bars, 20μm (middle), 10μm (bottom). D,E) Correlation between mTurq2:Geph fluorescence and GABA_Aγ₂ labeling (D; 3 Experiments, 18 cells, 1284 synapses) and VGAT labeling (E; 4 Experiments, 17 cells, 985 synapses).

Fig 3. Long term imaging of mTurq2:Geph puncta.

A) A rat cortical neuron expressing mTurq2:Geph within a network of cortical neurons growing on a thin-glass MEA dish (starting at 18 days in vitro). Bar, 20μm. B) Time-lapse imaging of the region enclosed in rectangle in (A). Images were collected at one hour intervals for 161 hours (~7 days). Only a subset of the data is shown here. Bar: 20μm. C) High temporal and spatial resolution images of region enclosed in rectangle in (B). Note the excellent ability to follow the same synapses over time. Fluorescence measurement data for the three synapses enclosed in colored rectangles is provided in Fig 5A. The bottom panel demonstrates the programmatic detection of mTurq2:Geph puncta, used thereafter for counting and fluorescence quantification. Bar, 10μm.

Fig 4. Population measures of inhibitory synapse sizes and numbers are stable and only modestly affected by the suppression of spontaneous activity.

A) Spontaneous activity recorded for 72 hours in one network, starting with the mounting of the preparation on the combined MEA recording/imaging system. Every dot represents the total number of action potentials (AP) recorded from all electrodes during a one minute period, divided by 60. Action potentials were identified on-line as threshold crossing events (with the threshold set at -0.4 mV), in traces such as those shown in the inset (6 action potentials recorded from two electrodes). Bars, 1msec, 1mV. B) Changes in spontaneous activity levels—pooled data from 6 experiments. Data was normalized to activity levels measured at t = 72 hours. C) Changes in mTurq2:Geph puncta numbers pooled from 27 neurons in 4 experiments in which images were obtained at one hour intervals. Counts for each neuron were normalized to initial puncta counts at t = 0. D) Changes in mTurq2:Geph fluorescence intensities pooled for the same 27 neurons as in C). Fluorescence values were normalized to mean fluorescence at t = 0. E) Distributions of mTurq2:Geph puncta fluorescence intensities averaged over three consecutive, 24-hour windows for the same 27 neurons as in C,D). F) Changes in mTurq2:Geph puncta numbers following the abrupt suppression of network activity by exposing the networks to TTX at t = 0 (red line, see also S3 Fig). G) Changes in mean mTurq2:Geph puncta fluorescence intensities following the suppression of network activity at t = 0. H) Distributions of mTurq2:Geph puncta fluorescence intensities during 23-hour consecutive time windows before and after exposure to TTX (4 Experiments, 27 neurons, ~4,000 synapses). All values in B-D,F,G represent means ± standard deviations.

Fig 5. Spontaneous remodeling of individual inhibitory synapses.

A) Changes over time in the fluorescence intensity of three mTurq2:Geph puncta (enclosed in color-coded rectangles in Fig 3C) measured over the course of 161 hours. Raw data is shown as thin lines whereas thick lines show the same data after smoothing with a 3-point low-pass filter. The filtered data was used in all subsequent analyses. B) Distributions of the range of changes exhibited by individual mTurq2:Geph over the course of 24 hours, before and after exposure to TTX (4 Experiments, 27 neurons, 749 synapses). The ranges of change of individual puncta were expressed as “range over mean” calculated as illustrated in inset, that is, by subtracting the minimal intensity from the maximal intensity and dividing this difference (the range) by the mean fluorescence intensity (eq 1 in main text).

Fig 6. Spontaneous remodeling of individual inhibitory synapses analyzed in the framework of a Kesten process.

A) The fluorescence of individual mTurq2:Geph puncta at increasing times (1,8,24 and 42 hours) as a function of their fluorescence at an arbitrary reference time point (t = 0). Linear regression lines are shown in black, whereas the unity line is shown as a dashed red line. Note that with time, the slope grows shallower, the offset increases and the goodness of fit (R²) decreases. 749 puncta from 27 neurons. B) The slopes and offsets of linear regressions such as those shown in (A) for all one-hour intervals from t = 1 to t = 42 hours). C) Obtaining the average value of ε (i.e. <ε>). According to eq 3 (main text) the slope in linear regression plots such as those shown in (A) should be 〈ε〉^k where k is the number of steps (hours, in this case); therefore log(slope) = log〈ε〉 ∙ k. Thus, when log(slope) is plotted as a function of k as shown here, log〈ε〉 can be estimated from a linear regression line in this plot (black), giving log〈ε〉 = -0.00253 and 〈ε〉 = 0.9942. D) Extrapolation of the slopes expected in plots such as those shown in (A) for longer durations. The expected slopes were calculated as (0.9942)^k where k is given in hours from t = 0. The empirical data shown in (B) and (C) is overlaid onto the calculated curve. E) Slopes and <ε> are similar in spontaneously active networks and in networks in which activity was blocked for 24 hours. Analysis similar to that outlined in B-D was performed for two time windows before (-25 to -1 hours) and after (22 to 46 hours) exposure to TTX. The derived values of <ε> were essentially indistinguishable.

Fig 7. Suppression of spontaneous activity leads to transient changes in remodeling dynamics.

A) The slopes and offsets of linear regressions similar to those shown in Fig 6A for three consecutive 24 hour periods beginning 24 hours before exposure to TTX, immediately after exposure to TTX and 22 hours after exposure to TTX. Note that immediately after exposure to TTX, the slopes increase and offsets decrease, but these effects are temporary and mostly gone after 12 hours. B-E) Changes in the fluorescence of individual mTurq2:Geph puncta as a function of their initial fluorescence during 12 hour windows before (B) and after (C-E) exposure to TTX. ΔF was calculated by subtracting the fluorescence at the beginning of each time window (F₀) from the fluorescence at its end. Solid lines are linear regression fits. P denotes the statistical significance of the regression lines. Note the transient change in the trends of the regression line immediately after TTX addition (C) and the recovery in subsequent time windows (D,E).

Fig 8. Concomitant imaging of GABAergic and glutamatergic synapses.

A) A rat cortical neuron expressing both mTurq2:Geph and PSD-95:EGFP. B) Time-lapse imaging of the region enclosed in rectangle in (A). Images were collected at 1 hour intervals for 100 hours (only a subset of the data is shown here). Bars: 20μm (A), 5μm (B). C) Changes over time in the fluorescence intensity of one mTurq2:Geph and three neighboring PSD-95:EGFP puncta shown in (B) over 40 hours. D) Normalized fluorescence intensities of tracked PSD-95:EGFP and mTurq2:Geph puncta, 24 hours before (left) and after (right) exposure to TTX (4 Experiments, 6 neurons, 49 mTurq2:Gephyrin puncta and 163 PSD-95:GFP puncta, mean ± SEM).

Fig 9. Inhibitory synapse configurations deteriorate more slowly than excitatory synapse configurations.

A) Goodness of fit, expressed as the coefficient of determination (R²) of linear regression lines calculated as explained in Fig 6A. B) Same as in (A) but for 24 hour periods in TTX. Note the slower decay rates of slopes and R² values for mTurq2:Geph in both active and silenced networks.

Fig 10. Relationships between the remodeling of nearby excitatory and inhibitory synapses.

A) PSD-95:EGFP fluorescence intensity as a function of distance from nearest inhibitory synapse before (left) and after (right) exposure to TTX. NS = non-significant. B) Covariance (Pearsons’ correlation) of PSD-95:EGFP and mTurq2:Geph fluorescence as a function of distance between the synapses before (left) and after (right) exposure to TTX. C) Average (± SEM) covariance of PSD-95:EGFP and mTurq2:Geph within groups (but irrespective of distance) before and after exposure to TTX. Same synapses as in Fig 9. Open bars (“Shuffled”) correspond to the average (±SEM) covariance calculated for all possible pairs of PSD-95:EGFP and mTurq2:Geph in the entire data set (irrespective of neuron or experiment).