Long-term potentiation (LTP) is an important model for examining synaptic mechanisms of learning and memory. A key question is whether the enhanced synaptic transmission occurring with LTP involves the addition of new synapses, the enlargement of existing synapses, or a redistribution in synaptic weight among synapses. Two experimental designs were used to address this question. In the first experimental design three conditions were evaluated across hippocampal slices maintained in vitro, including slices with LTP analyzed at 2 hr post-tetanus, slices tetanized in the presence of APV, and control slices receiving test stimulation only. In the second experimental design independent LTP and control (low-frequency stimulation) sites were examined. Synapse density was estimated by an unbiased volume sampling procedure. Synapse size was computed by three-dimensional reconstruction from serial electron microscopy (EM). Serial EM also was used to compute synapse number per unit length of dendrite. In both experimental designs there were no significant effects of LTP on total synapse number, on the distribution of different types of synapses (thin, mushroom, stubby, or branched dendritic spines and macular, perforated, or segmented postsynaptic densities), on the frequency of shaft synapses, nor on the relative proportion of single or multiple synapse axonal boutons. There was also no increase in synapse size. These results suggest that LTP does not cause an overall formation of new synapses nor an enlargement of synapses at 2 hr post-tetanus in hippocampal area CA1, and these results support the hypothesis that LTP could involve a redistribution of synaptic weights among existing synapses.
- long-term potentiation
- area CA1
- serial electron microscopy
- dendritic spines
- postsynaptic density
- axonal boutons
- three-dimensional reconstructions
- hippocampal slice
- in vitro
The cellular mechanisms for storing memory in the brain are not known, although prevailing theories include changes in synapse number or structure (Tanzi, 1893; Ramon y Cajál, 1911; Hebb, 1949; Wallace et al., 1991; Bailey and Kandel, 1993). Long-term potentiation (LTP) is an enduring enhancement of synaptic transmission that results from specific patterns of activation, occurs in many different brain regions, and is widely accepted as a cellular mechanism of learning and memory (Bliss and Lomo, 1973; Bliss and Collingridge, 1993). Hence, a useful step toward testing the structural theory of memory is to establish whether, when, and where LTP involves a change in synapse number or structure.
Considerable research has been devoted to determining whether LTP has a structural correlate (for review, see Wallace et al., 1991; Horner, 1993; Edwards, 1995). Most of the structural studies have been done in the dentate gyrus of the hippocampal formation in vivo and involved repeated stimulation of the perforant path input. In this region some changes in synapse number and/or structure have been reported to occur as early as 2–30 min after induction of LTP (Van Harreveld and Fifkova, 1975; Fifkova and Van Harreveld, 1977; Fifkova et al., 1982; Desmond and Levy, 1986a,b, 1988, 1990; Trommald et al., 1990) and to have lasted for hours to days (Geinisman et al., 1991,1994). Across the many studies, however, there are puzzling inconsistencies as to which changes are specific to LTP. Area CA1 in the hippocampal slice preparation was chosen in the present study because it has become the dominant model system for investigating the cellular and molecular bases of LTP. In previous studies single-section analyses (Lee et al., 1980; Chang and Greenough, 1984; Chang et al., 1991; Buchs and Muller, 1996) or confocal microscopy (Hosokawa et al., 1995) was used to evaluate the number or structure of hippocampal CA1 synapses after electrically or chemically induced LTP. Single-section analyses are now recognized to be inaccurate because individual synapses might be identified incorrectly or missed, and variability in synapse size, shape, or orientation substantially influences the probability of viewing them on a single section (Dubin, 1970; De Groot and Bierman, 1983; Sterio, 1984; Braendgaard and Gundersen, 1986;Coggeshall and Lekan, 1996). In addition, the irregular shapes of synapses and dendritic spines make it impossible to extrapolate from measurements made on single sections to the true three-dimensional values (Harris et al., 1992; Harris and Sultan, 1995; Spacek and Harris, 1997). Confocal microscopy does not distinguish short or curved dendritic spines from the overlapping dendritic shaft nor the occurrence of shaft synapses, and it also has insufficient resolution to measure synaptic dimensions (Harris, 1994; Trommald et al., 1995).
In the present work serial electron microscopy (EM) was used to overcome many of these shortcomings. Unbiased sampling was achieved by a volume disector, and viewing through serial sections was used to distinguish different types of synapses. Spine numbers obtained along reconstructions of individual dendritic segments validated results from the unbiased sampling. Total postsynaptic density (PSD) area was measured as an indicator of overall synapse size because it is well correlated with the dimensions of other components of the synapse (Chicurel and Harris, 1992; Harris et al., 1992; Lisman and Harris, 1993; Sorra and Harris, 1993; Spacek and Harris, 1997). The results provide the first detailed measurements of synapse number and size in hippocampal slices and suggest that LTP does not result in synapse enlargement nor in more synapses at 2 hr post-tetanus in the mature hippocampal area CA1.
MATERIALS AND METHODS
Preparation and maintenance of hippocampal slices in vitro. Hippocampal slices were prepared from male rats of the Long–Evans strain 60–70 d old (weight 236–310 gm), according to methods adapted from (Harris and Teyler, 1984) and in accordance with National Institutes of Health guidelines and approved animal care protocols. After decapitation, the left hippocampus was removed, and four to six slices (400 μm each) were cut from the middle third of the hippocampus into ice-cold media containing (in mm) 116.4 NaCl, 5.4 KCl, 3.2 CaCl2, 1.6 MgSO4, 26.2 NaHCO3, 1.0 NaH2PO4, and 10 d-glucose. Slices were transferred to nets positioned over wells containing media with or without 50 μm dl-amino-5-phosphonovaleric acid (APV; Sigma, St. Louis, MO) at the interface of humidified O2 (95%) and CO2 (5%) maintained at 30–31°C. Slices were equilibrated for at least 1 hr before physiological recordings.
Physiological recordings. Two concentric bipolar stimulating electrodes (Ultrasmall, 50 μm pole separation; Fred Haer, Brunswick, ME) were positioned 600–800 μm apart in the middle of stratum radiatum on either side of a single extracellular recording electrode (glass micropipette filled with 0.12 m NaCl; see Fig.1 a). Test stimuli consisted of alternating stimuli at each electrode delivered at one per 15 sec at an intensity that evoked field EPSPs (fEPSPs) with initial slopes measuring ∼1 mV/msec. Increasing stimulus intensity was used to generate an input–output (I–O) curve, and the half-maximal responses were monitored for 20–40 min for each physiologically independent pathway before induction of LTP (see the protocols below). All responses were digitized and analyzed with Scope software (RC Electronics, Santa Barbara, CA) and converted to Lotus 1-2-3 (Lotus Development, Cambridge, MA) and Microcal Origin (Microcal Software, Northhampton, MA) for subsequent analysis and graphing.
Across-slice experiments. Slices from a total of seven animals were used for these experiments. Four conditions were represented, including (1) LTP, (2) control stimulation in normal media, (3) control stimulation in media containing 50 μmAPV, and (4) slices tetanized in the presence of APV (APVtet). LTP was induced by delivering tetanic stimulation (100 Hz for 1 sec at half-maximal stimulation) at both of the stimulating electrodes, either sequentially or simultaneously. Sequential tetani were separated by 10 min intervals, and pairs of simultaneous tetani were separated by 20 min intervals. Two or three sets of tetanic simulation were delivered to ensure that LTP was saturated. Enduring LTP was defined as an increase in the initial slope of the fEPSP of at least 40% above the pretetanus value monitored for 2 hr post-tetanus (see Fig.2 a). This 2 hr time point was chosen because it is long enough after tetanus for gene induction to have occurred (Ghosh et al., 1994; Impey et al., 1996), it is in a phase of LTP that had been reported to require new protein synthesis (Otani et al., 1989; Huang and Kandel, 1994), and it is consistent with times used in other structural studies of LTP in area CA1 (Lee et al., 1980; Chang and Greenough, 1984; Chang et al., 1991; Hosokawa et al., 1995; Buchs and Muller, 1996). Control slices, with or without APV in the media, were tested with the I–O procedure, and their responses were monitored at the rate of one stimulus per 15 sec for 10 min at half-maximal stimulation intensity alternating between the two stimulating electrodes (Fig. 2 b). Other control slices received the same tetanic stimulation protocols as the LTP slice, but in the presence of APV, which blocked LTP (APVtet, Fig. 2 c). In two experiments all four conditions were obtained from four different hippocampal slices of the same animal. In other experiments an LTP and a control slice or an APVtet and an APV control slice were obtained from the same animal. All of the results from the two control conditions that were given test stimuli only in the presence or absence of APV (conditions 2 and 3) were statistically equivalent, so these two control conditions were combined as one group (“untetanized controls,” e.g., Figs. 4, 5).
Slices from each condition were immersed in mixed aldehydes during 8 sec of microwave irradiation to ensure rapid fixation throughout the slices immediately after the physiological recordings (Jensen and Harris, 1989). Indirect lighting and a dissecting microscope were used to identify the locations of the two stimulating electrodes as depressions in the surface of stratum radiatum in area CA1. These depressions were used to guide manual trimming of the activated region from the rest of the hippocampal slice. Then the trimmed tissue piece was processed for transmission EM, and semithin (∼ 1 μm) sections were cut parallel to the stimulating electrode tracks left on the surface of the slice (Fig. 1 b) down to ∼100–150 μm into the core of the slice. Test thin sections were evaluated by EM. Tissue preservation was judged acceptable if cell bodies, membranes, and cellular organelles (endoplasmic reticulum, mitochondria, microtubules, etc.) were intact at the ultrastructural level, and few or no dark (degenerating) processes were present in the neuropil (e.g., Fig. 3below). A small trapezoid was centered between the two stimulating electrodes where the recording electrode had been located. The tip of the recording electrode was too small to leave a recognizable mark in the tissue, but its position had been measured relative to the two stimulating electrodes during the electrophysiological experimentation. This trapezoid was trimmed to a depth that was within the range of best tissue preservation at 100–180 μm from the air surface of the slices, and serial thin sections were cut for photomicroscopy and quantitative analyses of synapses (Fig.1 a).
Within-slice experiments. A second pair of experiments was done to evaluate synapse number and structure directly beneath the source of electrical stimulation. Two stimulating electrodes were positioned in the middle of s. radiatum, separated by ∼600 μm with a recording electrode centered between them, as in Figure1 a. In these experiments tetanic stimulation was delivered to induce LTP at one electrode, and control low-frequency stimulation was delivered to the other electrode within the same slice. AnI–O curve was obtained, and then the fEPSP was monitored at approximately half-maximal stimulation for at least 20 min at both sites. During the pretetanus monitor a total of three low-frequency stimulations (LFS; 5 Hz stimulation for 20 sec as two trains separated by a 20 sec interval) were delivered to the control pathway, separated by ∼5 min each for a total of 600 pulses. At 5 min after the last LFS, tetanic stimulation (100 Hz for 1 sec as two trains separated by a 20 sec interval) was delivered via the other stimulating electrode to the LTP pathway. Additional pairs of tetanic stimulation were delivered at 10 and 20 min later to ensure that LTP was saturated (Fig. 7, below). Responses at both pathways were monitored at a rate of one stimulation every 30 sec for 2 hr post-tetanus. Input specificity was demonstrated by the occurrence of LTP from tetanic stimulation at one electrode and by no enhancement in response to stimulation at the control electrode.
In another set of experiments (data not shown) the recording electrode was positioned within 50 μm of the stimulating electrode, and the same tetanic stimulation protocol was given. These experiments demonstrated that LTP (148% ± 18%, n = 3) occurs immediately adjacent to the stimulating electrode as well as centered between the two stimulating electrodes.
The exact positions of both stimulating electrodes were indicated, with respect to identifiable cellular land marks in the slice, at the end of each physiology experiment to ensure correct identification of the LTP and control sites for subsequent anatomical analyses. Slices were fixed as described above for the across-slice experiments, and area CA1 containing the stimulating electrodes was dissected free from the surrounding tissue. A single cut was made across the depth of the slice through the middle of the stimulating electrodes, and the tissue beneath each electrode was trimmed for serial EM (Fig.1 b).
Unbiased adjustment of synaptic densities with the volume disector. These methods were extended from earlier studies (Harris et al., 1989, 1992; Harris, 1994). Using a calibrated photographic screen of the electron microscope, we divided a reference section in the middle of each series into equal fields. One of the fields was selected randomly to photograph through serial sections. All series were coded and analyzed blind as to experimental condition. Clear plastic sheets covered each micrograph so that markers could be used to number and map each synaptic complex as it was viewed through serial sections. A rectangular sampling frame was drawn on the middle reference section to contain the sampling area. Two sides of the rectangle were assigned randomly as inclusion or exclusion edges to create a counting frame that minimized potential edge effects across samples (Gundersen, 1978).
To measure the sample areas, we placed the reference section from each series under a video camera and digitized it by a PC-based frame grabber (Vision-8, Insync Technologies, San Leandro, CA). The sample fields were traced and their areas computed with software entitled V8, which was developed in the Image Graphics Laboratory at Children’s Hospital. For the across-slice experiments 27 sample fields were analyzed for a total area of 3314 μm2 (123 ± 20 μm2 per sample); for the within-slice experiments 24 sample fields were analyzed for a total area of 2703 μm2 (113 ± 9 μm2 per sample). The sectioned areas of elements appearing nonuniformly in the sample field (i.e., cell bodies, large dendrites, and myelinated axons) were traced and then subtracted from the sample areas to obtain the “homogeneous neuropil area” (HNA). The total HNA was 2671 μm2 (99 ± 18 μm2 per sample) for the across-slice experiments and 2269 μm2 (95 ± 9 μm2 per sample) for the within-slice experiments.
Any PSDs falling within the sample frame or on the two inclusion lines were counted, and those falling outside of the frame or on the two exclusion lines were not counted. Synapses were always counted if the PSD and synaptic vesicles both occurred in the sample frame. PSDs sectioned en face or obliquely were included if the synaptic vesicles appeared in the next section. Totals of 1790 and 1438 synapses were analyzed for the across-slice and within-slice experiments, respectively. Synaptic and spine morphologies were identified as described in Results by tracing them across serial sections. One-half of the sample field in each coded series was analyzed by each author for all experiments.
PSDs have different shapes and sizes, and the probability of viewing them on the reference section differs in proportion to the number of sections they occupy. Thus, the number of serial sections each PSD occupied was counted. When the synapses were grouped by different PSD morphologies, spine shapes, or bouton types, the mean number of sections for PSDs in each category was computed. The probability of viewing different synaptic elements also varies with section thickness. Every effort was made to obtain uniform section thickness at the time of cutting (platinum-colored sections in the boat of the diamond knife); however, uniform section colors do not necessarily translate into uniform section thickness (Peachey, 1958). To correct for this potential bias, we estimated section thickness for each series by measuring the diameters of longitudinally sectioned mitochondria. The number of serial sections each mitochondrion appeared in was counted, and section thickness was estimated as thickness (μm/section) equaled measured diameter per number of sections (Harris and Stevens, 1988,1989; Harris, 1994). Five to twenty mitochondria were included in each series to obtain this estimate of section thickness. The biased synapse density (BSD) equaled the number of synapses per HNA · 100. Then this value was adjusted to account for the probability of viewing synapses of different shapes and sizes and for differences in section thickness. The adjusted synapse density (ASD) was the number of synapses/100 μm3 = BSD · (1/mean number of sections per PSD) · (1/mean section thickness).
Three-dimensional reconstructions. The surface areas of PSDs were reconstructed and measured through serial sections, using the PC-based reconstruction software (V8) developed in the Image Graphics Laboratory at Children’s Hospital. EM micrographs were digitized, and the images of adjacent sections were microaligned for reconstruction by flickering between the stored image and the live image and moving the live image to minimize motion of the profiles in the field. The area of cross-sectioned PSDs equaled PSD length on adjacent sections multiplied by section thickness and added across sections. For en facePSDs the enclosed areas were measured, and a connector was drawn to estimate where the areas overlapped in adjacent sections; the total area equaled the enclosed area plus the length of each connector multiplied by section thickness.
Statistical analysis. Lotus-1-2-3 (Lotus) was used to organize the database and to compute quantitative analyses on synapse densities. Microcal Origin (Microcal) and Statistica software (StatSoft, Tulsa, OK) were used to graph, to obtain means and SDs, to test for normality, and to perform the tests of significance described in Results. Nonparametric analyses were used for the across-slice experiments, because the data failed to approximate normal distributions as revealed with the Shapiro–Wilks’ W test. The distribution-free Kruskal–Wallis ANOVA by ranks (wherek > 2) and Mann–Whitney U (wherek = 2) tests were used to evaluate the significance of differences between sample populations. A two-factor ANOVA was used for the within-slice experiments. The significance criterion was set atp < 0.05.
Physiological responses from a typical set of across-slices experiments are illustrated in Figure 2. For the LTP slices (e.g., Fig. 2 a), the fEPSPs were potentiated at 2 hr post-tetanus, with the change in slope mean ± SD = 190 ± 40% relative to the pretetanus baseline responses (n = 6 slices from six different animals). Untetanized control slices (n = 6 slices in control media and three slices in APV containing media from seven different animals; Fig. 2 b) achieved maximal fEPSP slopes of 2.9 ± 1.3 mV/msec, which was comparable to the pretetanus maximal fEPSP slopes of 2.9 ± 0.64 mV/msec for the LTP slices. Slices tetanized in the presence of APV (Fig. 2 c) showed no change in response (95 ± 10%, n = 3 slices from three different animals, which were the same as some of the animals used for the LTP and control slices, as discussed in Materials and Methods).
Synapses were evaluated in one to three series from each of the slices for a total of 25 EM series, including 10 series from LTP slices (20–62 sections per series), 10 series (25–58 sections per series) from untetanized control slices maintained without (n = 6) or with APV (n = 3), and 5 series (25–47 sections per series) from the three slices that were tetanized in the presence of APV. Representative examples of the slice neuropil and synapses in each of the three treatment conditions are illustrated in Figure3. Individual synapses were identified as macular if the PSD profiles were continuous or as perforated if electron-lucent regions divided the PSD on adjacent serial sections.
The total synapse densities were not significantly different among the LTP, the untetanized, and the APV conditions (Fig.4 a). Macular synapses predominated across all three conditions (Fig. 4 b) although only ∼10% of the synapses were perforated. When macular and perforated synapses were analyzed separately, no significant differences were found across the treatment conditions for either type of synapse.
The dendritic spines associated with the reference section PSDs were viewed through serial sections and subjectively classified into four shape categories: thin, mushroom, stubby, and branched (Peters and Kaiserman-Abramof, 1970; Harris et al., 1992). Spines were classified as thin if their lengths were greater than their neck and head diameters; mushroom, if the heads were much wider than the necks; stubby, if the neck diameters were similar to total length; and branched, if the spines possessed more than one head. On average, 8% of the spines fell on the borders, especially between thin and mushroom shapes, and these were assigned equally to the two categories. LTP had no significant effect on the distribution of dendritic spines in these different shape categories (Fig.5 a).
Presynaptic boutons were identified as single synapse boutons (SSBs) if they synapsed with only one postsynaptic target and as multiple synapse boutons (MSBs) if they synapsed with two or more targets (Sorra and Harris, 1993). No differences were found across the three conditions in the frequencies of these different bouton types (Fig. 5 b), suggesting that the number of connections between individual boutons and their postsynaptic targets was not altered by the LTP.
Shaft synapses were identified as asymmetric (excitatory) or symmetric (inhibitory) and as occurring on the spiny pyramidal cell dendrites or the nonspiny dendrites of local interneurons (Harris and Landis, 1986;Peters et al., 1991) (Figure 6). In Figure 5 a all types of shaft synapses are combined; on average, their overall frequency is <5% of all synapses in the neuropil. Spiny dendrites predominated in all of the samples; however, the occasional nonspiny dendrite could add a disproportionate number of asymmetric shaft synapses. Because each type of shaft synapse has a distinctly different function and/or location, it was important to analyze them separately with respect to the experimental conditions. No significant differences in the frequencies of the different types of shaft synapses (asymmetric or symmetric) on either dendrite type (spiny or nonspiny) were detected (Table 1).
The results from the across-slice experiments showed no statistically significant differences in a variety of measurements of synapse number among the three treatment conditions. However, there appeared to be a trend toward fewer synapses in the two control conditions. To address whether variability in the data masked an underlying effect of LTP on synapse number, we did a new set of experiments. These within-slice experiments were designed to minimize the variation across slices from the same and from different animals. To achieve this goal, we took advantage of the well characterized phenomenon of input specificity of LTP, wherein LTP can be induced at one stimulating electrode while low-frequency stimulation at a control site in the same slice results either in no change in response or a contrasting depression in response.
In this way, a pair of within-slice experiments (named for rat numbers 68 and 69) was performed to sample a concentrated population of LTP (i.e., potentiated) synapses beneath one stimulating electrode and nonpotentiated synapses beneath another control stimulating electrode. The neuropil was sampled directly beneath the sources of electrical stimulation at the LTP and control sites (see Fig. 1 a,c for electrode positioning). Physiological responses from both experiments are shown in Figure 7, and representative electron micrographs are illustrated in Figure8.
The physiological and anatomical data from each slice were graphed separately to control for across-animal differences. A total of 24 EM series (25–38 sections) were photographed, with six series from each slice at each of the LTP and control sites (i.e., 2 slices, 2 animals, 12 LTP series, 12 control series). This approach controlled for differences between animals and/or slices and for the amount of time a particular slice spent in vitro before fixation. In the within-slice experiments there were no significant differences between the LTP and control sites in the total synapse densities (Fig.9 a). Similarly, there were no differences in the relative frequencies of macular or perforated PSDs (Fig. 9 b) or in the different types of shaft synapses (Table1, within-slice experiments). There was, however, a significant difference between animals: the slice from rat 69 had a higher synapse density than rat 68 at both the LTP and control sites (p < 0.01).
Synapse number along individual dendritic segments
The ASD calculation, like all density estimates, could be influenced by differential growth or proliferation of other cellular elements in the neuropil. For example, glial proliferation with LTP could lead to synapses being pushed apart in a sample volume of tissue. In fact, some evidence in the literature suggests that glial changes occur with LTP (Wenzel et al., 1991). If both synaptogenesis and glial proliferation occurred with LTP, then a change in synapse number might not be detected with the ASD calculations described above. The absolute number of spines per unit length of dendrite is a measure of synapse number that is insensitive to growth of other elements in the neuropil. This measure of synapse number could be affected by dendrites that are elongating or contracting; however, current evidence suggests that total dendritic length remains constant during LTP (Hosokawa et al., 1995; Trommald, 1995).
To compute spine number per unit length of dendrite, we identified and visually reconstructed cross-sectioned dendrites from the LTP and control sites of the within-slice experiments across serial sections. Previous work demonstrated that the lateral thin dendrites have diameters ≤1 μm, and the apical dendrites have diameters >1 μm (Harris et al., 1989, 1992). Preliminary analysis from the within-slice experiments revealed that the thicker apical dendrites also had more spines per unit length (our unpublished observations), although the thinner dendrites constituted the majority. Thus, to ensure that comparable dendritic populations were sampled, only the thinner dendrites (diameters ≤1 μm) were included here. Two hundred dendritic segments were evaluated, 50 from each of the LTP and control sites for both animals. All spines emerging from each cross-sectioned dendritic segment were counted through serial sections, and dendritic segment length was computed by multiplying the number of serial sections by section thickness. This analysis confirmed the ASD results reported above and showed no significant difference between the LTP and control sites in the number of dendritic spines per unit length of dendrite (Fig. 10 a).
Branched spine analysis
Earlier studies of LTP in the hippocampal dentate gyrus provided evidence for an increased frequency of branched dendritic spines along the lengths of dendrites (Trommald et al., 1990; Trommald, 1995). To test whether branched spines were selectively affected by LTP in area CA1, we randomly selected 40 dendritic segments (10 from each of the two LTP and control sites) for a three-dimensional analysis of the occurrence of branched spines along their lengths. Branched spines were rare along dendritic segments, and their frequency was not increased with LTP (Fig. 10 b). There was a trend, however, toward more branched spines along dendrites at the control sites in both slices.
PSD area was shown previously to correlate with the size of the dendritic spine head and the amount and complexity of their subcellular constituents as well as the size of the presynaptic bouton and the number of vesicles it contains (Harris and Stevens, 1989;Harris and Sultan, 1995; Spacek and Harris, 1997). For this reason, PSD area was measured as an overall indicator of whether synapse size differed between the LTP and control sites. Sample fields from the EM series were divided into four quadrants. One or two quadrants from each series were selected randomly, and all of the PSDs contained within these quadrants were reconstructed in three dimensions. More macular than perforated synapses were reconstructed, because more macular synapses occur in area CA1 neuropil [see above and Harris et al. (1992)]. To achieve a sufficiently large population of perforated PSDs, we generated a set of random numbers, and we also reconstructed those perforated PSDs occurring outside the reconstruction quadrant having those numbers. In total, 520 PSDs were reconstructed. These reconstructions, illustrated schematically in Figure11 a, revealed no differences between the LTP and control sites in the sizes of the macular PSDs (Fig. 11 b,c) or the perforated PSDs (Fig. 11 d), except in one slice in which the perforated PSDs were on average smaller at the LTP site (p < 0.05, Fig.11 e).
Segmented PSDs are a subset of perforated PSDs (Fig.11 a). They are characterized by a single presynaptic bouton synapsing with separate PSD areas on a single dendritic spine head (or dendritic shaft) (Geinisman et al., 1993; Harris and Sultan, 1995). In contrast, nonsegmented, perforated PSDs have electron-lucent holes in the middle of an otherwise macular PSD or are sufficiently irregular in shape that they appear perforated on some sections. A coded analysis was done of the 120 perforated PSDs that were reconstructed above to determine the relative frequencies of segmented synapses (Table2). Only 3–27% of the perforated PSDs in these samples were fully segmented. Because the perforated PSDs are only 10% of the total synapse population (see Fig. 4 babove), these findings indicate that only 0.3–2.7% of all synapses have segmented PSDs. There were no consistent differences in the frequency or size of segmented PSDs associated with the LTP versus control sites (Table 2).
These results suggest that LTP does not produce a change in synapse number or size at 2 hr post-tetanus in the mature hippocampal area CA1. A parsimonious interpretation is that nonstructural synaptic mechanisms are sufficient to support LTP during this phase. Alternatively, concurrent synaptogenesis and synapse elimination could result in no net change in synapse number or size.
When these results are compared with those from other laboratories, several factors should be considered, including the sampling and measurement strategies, the time post-tetanus analyzed, and the experimental design. The volume disector used in this study is an unbiased procedure to adjust synaptic densities for known variables that influence sampling probability, including size, shape, and orientation of the synapse, plus the nonuniform occurrence of large and small dendrites, cell bodies, etc. The volume disector allows for inclusion of every synapse in the sampling frame, whereas other unbiased stereological approaches eliminate synapses occurring on a reference section that do not disappear on a parallel “look up” section (Gundersen, 1985; Braendgaard and Gundersen, 1986). Computation of synapse number along dendrites controls for other variables, such as growth of astrocytes, and also provides valid numbers for modeling. Finally, one of the key advantages of serial EM is that serial viewing is needed for unambiguous identification of all features of the synapse.
The main disadvantage is that serial EM is time-consuming, which makes the analysis of large numbers of samples difficult. If LTP were to have multiple forms and the anatomical effects were not always invoked, then detection of a small but significant change could require a large number of samples. Alternatively, LTP might be concentrated in a small number of synapses. We used two different approaches to address these issues. First, many slices from different animals were analyzed in the across-slice experiments. Second, many samples were obtained in the region where the density of potentiated synapses should be highest in the two within-slice experiments. The data from the across-slice experiments were more variable than the data from the within-slice experiments; however, the consistent findings from both experiments strengthen the basic result of no increase in overall synapse number or size at 2 hr after LTP.
Synapse density in the neuropil
Many other studies also have reported no change in total synapse number after LTP (Lee et al., 1980; Chang and Greenough, 1984; Desmond and Levy, 1986a,b, 1988, 1990; Gomez et al., 1990; Schuster et al., 1990; Chang et al., 1991; Geinisman et al., 1991, 1993, 1996; Grabs et al., 1991; Hosokawa et al., 1995; Buchs and Muller, 1996). However, some earlier studies in area CA1 reported a subtle but statistically significant increase in the number of stubby dendritic spines and shaft synapses, although <2/100 μm2 were detected under all conditions (Lee et al., 1980; Chang and Greenough, 1984; Chang et al., 1991). Here we detected a higher absolute density of stubby spines and shaft synapses, but no selective effect of LTP.
Because our absolute values are higher, it seems unlikely that we missed synapses that were counted on single sections in the earlier studies (Lee et al., 1980; Chang and Greenough, 1984; Chang et al., 1991). Instead, discrepancies could result from the ambiguous identity of spine shapes on single EM sections. In the earlier studies identification was limited to spines that were sectioned longitudinally; hence fewer spines could be identified. Our higher values probably result from the reliable identification of all synapses across serial sections. For example, the heads of cross-sectioned stubby spines are often indistinguishable from the heads of mushroom spines on one section; however, the stubby spines are easily distinguished from mushroom spines with constricted necks by tracing them to their origins with the parent dendrites. Unambiguous identification is a prerequisite for unbiased and meaningful statistical comparisons between treatments (Gundersen, 1985;Braendgaard and Gundersen, 1986).
In the earlier studies from area CA1, asymmetric shaft synapses were distinguished only by the diameter of their parent dendrite (Chang and Greenough, 1984). A more explicit distinction between spiny and nonspiny dendrites is crucial, because asymmetric shaft synapses occur more frequently on the dendrites of nonspiny interneurons and both cell types have thick and thin dendrites. Asymmetric and symmetric shaft synapses also must be delineated because they are functionally distinct. The maintenance phase of LTP was analyzed recently in the dentate gyrus at 13 d after four consecutive daily periods of tetanic stimulation to the perforant pathway. The only statistically significant increase was in the asymmetric shaft synapses (Geinisman et al., 1996). Because asymmetric synapses occur more frequently on the shafts of nonspiny interneurons and these were not distinguished from the spiny granule cell dendrites, it is a possibility that the long-term changes in the frequency of shaft synapses occurred on the nonspiny interneurons.
Spine number along individual dendrites
Trommald et al. (1990, 1995) reported increases in spine number along dendritic segments in the dentate gyrus at 30 min after tetanic stimulation of the perforant path input. Their analyses used serial EM reconstructions from two animals, along 34 and 28 dendritic segments in the control and LTP conditions, respectively. On average, there was one spine per micrometer in the control condition and three to four spines per micrometer in the LTP condition. In area CA1 we analyzed 200 dendrites, also from two different animals, and found approximately four spines per micrometer both in the control condition and in the LTP condition at 2 hr post-tetanus. It will be interesting to determine whether LTP induces more spines in area CA1 at 30 min post-tetanus, with a concomitant loss after 2 hr.
In the same analyses Trommald and colleagues detected a selective increase in branched dendritic spines at 30 min post-tetanus. In area CA1 we found no significant differences between the LTP and control conditions for branched spines. Branched spines composed <5% of the total spine population in the mature area CA1 (K. Sorra, J. Fiala, K. Harris, unpublished observations) and ∼2% in the dentate gyrus (Trommald et al., 1990; Trommald, 1995). Thus, it will be valuable to increase the sample size in both regions at 30 min to determine whether spine branching provides a consistent structural basis for the early phase of LTP.
Spine shape and PSD size
Spine shape is highly irregular, and spine heads vary by >100-fold in volume (Harris et al., 1992; Harris and Kater, 1994;Trommald and Hulleberg, 1997). The functional implications of this variation are not known; however, it is thought that the spine heads provide biochemical compartmentalization for synaptic plasticity (for review, see Harris and Kater, 1994; Koch et al., 1995; Denk et al., 1996). A spine head can appear large and indented on one section and small and convex two sections later, thereby making it difficult to interpret measurements of spine perimeters and areas on single sections (Van Harreveld and Fifkova, 1975; Lee et al., 1980; Fifkova and Andersen, 1981; Fifkova et al., 1982; Chang and Greenough, 1984;Desmond and Levy, 1988, 1990; Petit et al., 1989). We used the three-dimensional criteria of Peters and Kaiserman-Abramhoff to analyze spine shape as thin, mushroom, or stubby because they provide a useful convention to estimate whether spine dimensions change with LTP (Harris et al., 1992). An increase in spine volume would be expressed as a shift from the stubby and thin to the mushroom shapes and vice versa. Spine shortening would be detected by a shift from thin to stubby spines, although lengthening probably would have been missed because stubby spines are so rare in the mature hippocampus. Similarly, an enlarging of existing mushroom spines would have been missed. No significant shifts among spine shapes were observed.
To examine this question further, we measured the area of the PSD, because it correlates with total spine volume, the number of docked and nondocked presynaptic vesicles, and intra-spine organelles (r > 0.9) (Harris and Stevens, 1989; Harris et al., 1992; Harris and Sultan, 1995; Spacek and Harris, 1997; Trommald and Hulleberg, 1997). No significant change in PSD size occurred with LTP. It is possible, however, that during the different phases of LTP other indicators of synaptic efficacy, such as vesicle number and clustering (Applegate et al., 1987; Meshul and Hopkins, 1990; Chang et al., 1991;Malgaroli et al., 1995; Ryan and Smith, 1995; Murthy et al., 1997), become disassociated from PSD size, and these deserve examination in future studies.
Our observations suggest that segmented PSDs are not selectively affected by LTP in hippocampal area CA1. Geinisman et al. (1996, 1993,1991) reported no change in the total synapse number or in the incidence of macular synapses after LTP in the hippocampal dentate gyrus, consistent with our findings above; however, the frequency of segmented PSDs was greater after LTP in the dentate gyrus. This discrepancy could reflect differences across brain regions or in sampling strategies.
Other work in area CA1 revealed calcium precipitates in dendritic spines with perforated PSDs at 35 min after LTP, and it was concluded that LTP caused a threefold increase in perforated PSDs (Buchs and Muller, 1996). The Ca2+ precipitates, however, are more likely to be detected in those spines that contain calcium-sequestering tubules of smooth endoplasmic reticulum (SER).Spacek and Harris (1997) have shown recently that only ∼20% of spines with macular PSDs contain SER, whereas 100% of the spines with perforated PSDs contain SER. Thus, the analysis of Buchs and Muller (1996) was restricted mostly to large spines that already had perforated PSDs. An alternative interpretation would be that more bound calcium is sequestered in spines with SER after LTP.
Redistribution of synaptic efficacy?
Our results are consistent with the hypothesis that LTP in the mature brain results in a redistribution of synaptic weight among existing synapses. Alternatively, synapse populations could replace one another after LTP and not be detected as a shift in the overall number or size. To test these hypotheses, specific labels will be needed to distinguish recently potentiated synapses from those that were modified by past experience (Frey and Morris, 1997; Schuman, 1997) or by concurrent heterosynaptic depression (Bear and Abraham, 1996; Coussens and Teyler, 1996; Scanziani et al., 1996). Whether structural correlates of LTP can be detected will depend on whether changes in synaptic weight result in identifiable synapse-specific tags (Frey and Morris, 1997).
This work was supported by National Institutes of Health Grants NS21184 and NS33574 (K.M.H.), MR Center Grant P30-HD18655 (Dr. Joseph Volpe, PI), the Program in Neuroscience and the Division of Medical Sciences at Harvard University, and the Natural Sciences and Engineering Research Council of Canada (K.E.S.). We thank Ms. Marcia Feinberg for expert technical assistance. We also thank Drs. John Davis and John Fiala for developing and assisting us with the reconstruction and image analysis systems in the Image Graphics Laboratory at Children’s Hospital.
Correspondence should be addressed to Dr. Kristen M. Harris, Division of Neuroscience, Enders 260, Children’s Hospital, 300 Longwood Avenue, Boston, MA 02115.