0306-4522/91 $3.00+ 0.00 PergamonPress plc © 1991 IBRO
Neuroscience Vol. 42, No. 2, pp. 351-363, 1991
Printed in Great Britain
SYNAPTIC REORGANIZATION BY MOSSY FIBERS IN H U M A N EPILEPTIC FASCIA D E N T A T A T. L. BABB,*'~ W. R. KUPFER,'~ J. K. PRETORIUS,t P. H. CRANDALL*~§ and M. F. LEVESQUE§ *Department of Neurology, tBrain Research Institute and the §Division of Neurological Surgery, UCLA School of Medicine, University of California, Los Angeles, CA 90024-1769, U.S.A.
Al~traet--This study was designed to identify whether synaptic reorganizations occur in epileptic human .hippocampus which might contribute to feedback excitation. In epileptic hippocampi, (n = 21) reactive synaptogenesis of mossy fibers into the inner molecular layer of the granule cell dendrites was demonstrated at the light microscopic and electron microscopic levels. There was no inner molecular layer staining for mossy fibers in autopsy controls (n = 4) or in controls with neocortex epilepsy having no hippocampal sclerosis (n = 2). Comparing epileptics to controls, there were statistically significant correlations between Timm stain density and hilar cell loss. Since hilar neurons are the origin of ipsilateral projections to the inner molecular layer, this suggests that hilar deafferentation of this dendritic zone precedes mossy fiber reafferentation. Quantitative Timm-stained electron microscopy revealed large, zinc-labelled vesiclesin terminals with asymmetric synapses on dendrites in the inner molecular and granule cell layers. Terminals in the middle and outer molecular layers did not contain zinc, were smaller and had smaller vesicles. These histocbemical and ultrastructural data suggest that in damaged human epileptic hippocampus, mossy fiber reactive synaptogenesis may result in monosynaptic recurrent excitation of granule cells that could contribute to focal seizure onsets.
It has been well established that following deafferentation of mammalian dendrites, "new" synapses are formed within a few weeks. 38-4° This process of axonal growth into vacated dendrites has been well studied in the hippocampus (for review see Ref. 19), and referred to as reactive synaptogenesis because the new synapses form in response to loss of normal afferents. Granule cell dendrites in the fascia dentata (FD) receive afferents to the outer molecular layer (OML) from lateral entorhinal cortex, to the middle molecular layer (MML) from medial entorhinal cortex and to the inner molecular layer (IML) from a variety of regions, most prominently the hilar ipsilateral associational and commissural system. 35 Most of the reports on the hilar cell denervation and IML reactive synaptogenesis used transection or removal44'47'48of only the commissural inputs to the IML layer in adult or neonatal r a t : 7 Electron microscopic studies showed that terminal degeneration was followed by recovery of synapses; however the origin of these fibers was not fully determined. The authors suggested that the ipsilateral associational axons in the IML sprouted and innervated the vacated "com:~To whom correspondence should be addressed. Abbreviations: CA3/CA4, Cornu Ammonis regions; FD,
fascia dentata; GLD, gray level difference; H and E, hematoxylin and eosin; HS, hippocampal sclerosis; IML, inner molecular layer; KA, kainic acid; MF, mossy fiber; MFT, mossy fiber terminal; MML, middle molecular layer; OML, outer molecular layer; Pb, lead salts; PM, polymorph layer; R, correlation; SG, stratum granulosum; Ur, uranyl acetate.
missural" zones."45 The functional significance of this new synaptic rearrangement remained a puzzle, even though this new circuit possibly could increase the hilar ipsilateral projections back to the granule cells and thereby form a stronger disynaptic excitatory feedback system: i.e. granule cell mossy fibers would excite hilar neurons which would then excite IML dendrites of granule cells. Laurberg and Zimmer36used the Timm's technique to demonstrate that following IML denervation, mossy fibers from granule cells sprout back into the IML, apparently forming a monosynaptic feedback circuit. This circuit was confirmed using Golgielectron microscopy)° Using the kainic acid (KA) lesion of CA3/CA4, Tauck and Nadler 6° demonstrated that this mossy fiber reorganization produces calcium-dependent monosynaptic multiple population spikes in granule cells after single activation by antidromic stimulation. They found that the incidence of multiple firing was positively correlated with the greatest loss of hilar neurons and with the highest rating of Timm's staining density in the IML. Other studies using KA hippocampal lesions have shown faster hippocampal kindling27 and occasional spontaneous seizures17 which have been associated with IML sprouting.2° Hence, the available evidence suggests that following denervation of the IML there is a functional reinnervation that results in hyperexcitability. Our interest is in the anatomy of synaptic reorganization in the IML because those dendrites would be selectively denervated in human hippocampal
351
352
T.L. BABBet al.
lepsy by loss o f hilar neurons. Because e n t o r h i n a l cortex n e u r o n s are spared in h u m a n h i p p o c a m p a l epilepsy, neither the M M L n o r O M L is denervated (cf. Ref. 6). The hilar n e u r o n s are considered to be the cells o f origin for ipsilateral associational axons to the IML, 13'35a n d they would be reduced at the time t h a t severe h i p p o c a m p a l sclerosis occurred. We hypothesized t h a t loss o f hilar n e u r o n s occurred at the time of severe h i p p o c a m p a l sclerosis, deafferentation o f the I M L followed, a n d reactive synaptogenesis by mossy fibers would form a b n o r m a l m o n o s y n a p t i c recurrent excitatory synapses o n the I M L o f granule cells, a circuit t h a t could contribute to epileptogenicity. Previous reports have described a n o m a l o u s inn e r v a t i o n of the h u m a n epileptic I M L by mossy fibers 9'10'16'32'5s a n d by peptide-containing axons. 23,32
buffer for 48 h then stored at -80°C. Sections (30 # m thick) were cut at -70°C, returned to buffered ethylene glycol and glycerol for six months at -70°C, then placed in routine Timm solution. Although paraformaldehyde fixation was used prior to the sulphide solution, the low concentration of paraformaldehyde and high concentration of sodium sulphide resulted in a Timm reaction comparable to the routine procedure. Hence, the procedures with this hippocampus differed in many ways; however, the routine highsulphide Timm stain pattern was comparable to other non-epileptic FD. (3) A12189 was a 56-year-old male who died of an adenocarcinoma of the sigmoid colon and was autopsied 20 h post mortem. A 1.0 cm block of unfixed normal hippocampus was placed in neo-Timm, lowsulphide-glutaraldehyde fixative within 30 rain and processed with the neo-Timm procedure. (4) A12430 was a 59-year-old male who died of acute myocardial infarction without p r i o r ischemic changes and no evidence of gross brain damage. The brain was removed 24 h post mortem, and a 1.0 cm block of the hippocampus was processed with the neo-Timm procedure.
EXPERIMENTAL PROCEDURES
Criteria for identified epileptic hippocampus All 21 epileptic hippocampi had sclerosis typical of hippocampal epilepsy (cf. Ref. 5). Following hippocampal resection all patients had either no seizures or rare seizures, indicating that the hippocampus was epileptogenic. Additionally, intrahippocampal recordings from macroelectrodes26 or microelectrodes 8 verified that seizures originated in the hippocampus within 1-2cm of the tissues subsequently processed for the current studies. Electrodes in orbitofrontal gyrus and over neocortical regions approximating the International 10-20 System 3a provided a physiological detection of extrahippocampal seizure activity (for further details see Ref. I 1). In nine of the 21 epileptic patients, temporal lobeetomy was performed without intrahippocampal seizure recordings (for diagnostic criteria see Ref. 26). These nine hippocampi were also sclerotic, and the surgical resection which included the anterior 3.5 cm of the hippocampus resulted in seizure relief. Control comparison hippocampi The density of Timm's sulphide silver stain processed by either routine Timm (n = 9) or neo-Timm (n = 12) in identified epileptic hippocampus was compared to six nonepileptic hippocampi. Two patients (4295, 4406) had seizures arising outside the hippocampus and mild neuron loss not typical of hippocampal epilepsy. Patient 4406 had a grade II astrocytoma in the superior temporal gyrus, and all seizures were recorded from lateral temporal cortex. Patient 4295 also had seizures originating from lateral temporal cortex; however, no specific pathology was found in the resected temporal lobe. These hippocampi were stained by the routine high-sulphide Timm method exactly as the epileptic hippocampi. Four hippocampi removed at autopsy were also studied. The fixation protocols prior to Timm stain differed. (1) A11702 was a 64-year-old female who died from intestinal metastatic adenocarcinoma and was autopsied 4 h post mortem. A 1.0 cm block of the unfixed normal hippocampus was placed into the routine Timm solution within 30 min and processed identically to the epileptic hippocampus. This was a normal brain, and hence the Timm procedure differed from the epileptic's procedure only in the extra 4 h delay. (2) C587-L was a hippocampus kindly provided to us by Drs David Amaral and Ricardo Insausti, The Salk Institute, San Diego, U.S.A. The brain of a 63-year-old male who died of epidermoid carcinoma was removed 90 min after death, flushed through one carotid and the basilar artery with buffered heparinized normal saline followed by the following fixative: 4% paraformaldehyde, 0.002% picric acid and 2.7% NaCI in 0.1 M phosphate buffer. One-centimeter blocks were placed in 20% glycerol in 0.1 M phosphate
Hippocampal fixation and development Two methods were used to visualize zinc and other metals in the hippocampus. The first method (routine Timm) was used for light microscopic analysis, while the second method (neo-Timm) allowed both light and electron microscopic analysis from a single block of tissue. Routine Timm stain. The solution consisted of 1.2% sodium sulphide in 0.I M sodium phosphate buffer at pH 7.4. Tissue blocks were immersed at 4°C for 24-96 h. Blocks were than transferred to a cold 3% glutaraldehyde, 15% sucrose solution pH 7.4 for a maximum of 24 h. Tissue was taken directly from the glutaraldehyde-sucrose solution, quick-frozen and sectioned with a cryostat (-15°C) at 30#m. Sections were mounted on chromiumalum-gelatin-coated slides and allowed to air dry. Mounted sections were immersed in a "physical developer" in darkness at 21°C. The physical developer was prepared as follows: to 180ml gum arabic (500 g/1 water) add 30ml aqueous solution of 7.65 g citric acid and 7.05 g sodium citrate; just prior to use add (in the dark) 5.0 g hydroquinone in 90 ml distilled water and 1.5 ml of 15% silver nitrate solution. Two batches of different staining density were made for each patient based on visual inspection of staining, staining times varied from 30-120 min. After removal from developer, sections were washed first in distilled water for 2 min, in running tap water for 10 min, and finally dehydrated and coverslipped. Routinely, alternate sections were processed for hematoxylin and eosin (H and E) staining to localize anatomical fields and provide cell counts using the Abercrombie I correction, as described previously. 6 Hilar cell counts were limited to a region between stratum granulosum (SG) and the pyramids of CA4 or CA3c3 This region was at all times at least 50/~m from SG and 100 #m from CA4 (see hilus in Fig. 1A). The entire areas of the hilus counted ranged from 0.3 to 1.5 mm 2. The hilar cell counts were used for correlations with routine Timm staining density by quantitative densitometry. The hilar cell losses adjacent to neo-Timm-stained tissues were rated from 0 (no loss) to 3 (greatest loss) for subsequent correlations with neo-Timm staining density which was rated 0-3. Neo-Timm stain. To increase the specific staining of zinc and improve visualization of the mossy fiber system at both light and electron microscopic levels, a second Timm's staining method was used. 2~'5°The fixative consisted of 4% glutaraldehyde, 0.1% sodium sulphide, and 0.002% calcium chloride in 0.12M Millonig's buffer at pH7.3. Tissue blocks, approximately 1.0 cm thick, were immersed in fixative for 24-72 h at room temperature. Blocks were then transferred to cold 0.12 M Millonig's buffer with 0.002% calcium chloride at pH 7.3 for periods ranging from 2 h (room temperature) to two days (6°C). Next, blocks were
Fig. 1. Photomicrographs of human hippocampus to demonstrate patterns of Timm stains in control (B) and epileptic (C-E) FD (A, box). (A) Cresyl Violet stained autopsy normal hippocampus (case C587-1) shows normal cytoarchitecture of coronal section. Dashed line segregates the CA4 pyramidal cells from the hilus of the FD. (B) Timm stain of normal section adjacent to A showing dense stain only in polymorph or hilus (h) but not in SG, IML, MML, or OML. There is a faint irregular trace of Timm stain in parts of SG. (C) Timm stain in epileptic, sclerotic hippocampus showing dense stain in IML. (D) Timm stain puncta in IML from boxed area in C. (E) Timm stain puncta in SG from boxed area in C. Note that there is less Timm stain in PM layer of epileptic (C) than control PM (B) because there is a 55% granule cell loss in the epileptic (C). g, Granule cells; hf, hippocampal fissure; PM, polymorph layer. Scale bars in A = 500 gm; in B, C = 100 gm; in D, E = 20/~m. 353
354
T.L. BABBet aL
embedded in agar and sectioned with a Vibratome at 40 p m for light microscopy or 500#m for electron microscopy, Light microscopic sections were mounted, dried and developed as described previously for the routine Timm procedure. Alternate 40-pm sections were stained with H and E for routine cytology and measures of cell density. Neo- Timm's electron microscopy Electron microscopic sections (500 p m thick) were transferred to fresh buffer for 2 h, then into 15 ml glass vials containing a developing solution. The developer was prepared as follows: to 180 ml gum arabic (500 g/1 water) a 30 ml aqueous solution of 7.65 g citric acid and 7.05 g sodium citrate was added; just prior to use 2.5 g hydroquinone dissolved in 45 ml water and 45 ml 0.73% silver lactate were added in darkness. Development took place in darkness with constant agitation in a 26°C water bath for 60-80min; then sections were washed for 15min, wetmounted and coverslipped. Using low power light microscopy, areas of interest with optimal staining were selected and recorded by drawings. These areas (approximately 4 mm 2) were microdissected from the section and transferred to fresh buffer. Sections were post-fixed for 1 h in a solution of 2.0% osmium tetroxide, 0.12 M Millonig's buffer, 0.002% calcium chloride and 0.54% glucose, at pH 7.3. After two 10-min washes in 2.4% sodium chloride, the tissue was dehydrated through alcohols to propylene oxide then embedded in epoxy. Initial polymerization was in BEEM capsules for three days at 60°C. Tissue was observed by light microscopy in the block, trimmed to the F D and again recorded by drawing. Sections ( 5 ~ p m thick) were cut and dropmounted on glass slides, and those with optimal anatomy and Timm's stain were selected for re-embedding. Optimal Timm's staining was always within 30 # m of the tissue surface due to limits of the penetration of reagents. To obtain a fiat tissue surface for light microscopic observation and thin sectioning, the tissue was re-embedded in a BEEM capsule directly on the glass slide. After one day of polymerization at 60°C the block was removed from the glass slide by sliding a razor blade between the block and the glass. After two or three more days of polymerization, blocks were trimmed to contain the FD from the sub-granular polymorph layer through the granule cell and molecular layers to the hippocampal fissure. Once again, reference drawings were made. Thin sections (60 nm, silver interference color) were cut with a diamond knife and mounted on 200- or 300-mesh copper grids. Tissue contrast was increased by drop staining for 1 h in uranyl acetate (Ur) and 3-5 min in lead salts (Pb). Photographs were taken on a Siemens model 1A at an accelerating voltage of 80 kV. Anatomical distinctions between layers of the FD were made by correlating reference drawings with the thin section in the electron microscope. Data analysis To quantify the density of routine Timm's stain in the FD, low-power light microscopic densitometric measurements were made of the various layers using a Ziess IBAS interactive image analysis system. A video image of a light microscope field including SG, IML, M M L and OML was digitized into the system and each layer outlined using a cursor. The mean densitometric gray level was measured in
each layer, and the measurement repeated at the same location on six tissue slides for each patient; three slides were stained lightly (short development time) and three slides were stained heavily (long development time). Light and dark staining provided a control for the effect of staining density on relative density differences within each tissue section. A length of approximately 1 mm of FD was sampled in each slide. The sample location was selected as being representative of the overall staining pattern in F D prior to any knowledge of the patient's hippocampal epilepsy or pathology. A total of 13 hippocampi with routine Timm staining (nine epileptic, four control) were examined. IML staining positively correlated with hippocampal sclerosis (HS), whereas M M L staining was relatively light, increasing slowly with development time but not correlated with HS. A normalized measure of IML staining density was computed by subtracting the gray level difference (GLD) of the M M L from the IML. To quantify the density of neo-Timm staining, the 0-3 rating procedure of Tauck and Nadler 6° was used on coded slides by three independent scorers (T.L.B., W.R.K. and J.K.P.) and averaged. Similarly, hilar cell loss was scored 0-3 on H and E-stained slides coded without knowledge of the relation to the Timm slides and averaged. Pearson product moment correlations (R) were made of the Timm densitometry or the neo-Timm density scores against variables such as hilar neuron loss, age and background staining density. Student's t-tests were used for statistical significance (Ref. 25, p. 78). Ultrastructural aspects of neo-Timm-stained axon terminals [presumed mossy fiber terminals (MFTs)], and synapses were analysed in each layer of FD for: (i) the number of stained terminals, (ii) mean terminal size, (iii) average vesicle size and (iv) average vesicle density. Stratified electron microscopic sampling from the layers of FD was made by excluding sampling in transition areas between those layers, a procedure similar to that reported by McWilliams and Lynch. 44 Ten photographs ( × 17,000) were taken of axon terminals in each layer of the FD. If Timm's staining was rare or absent in a layer, photographs were taken randomly of well-preserved unstained axon terminals. The criteria used to identify terminals in any F D layer included: (i) a continuous spherically shaped membrane containing vesicles and mitochondria, (ii) a pre-synaptic density adjacent to a synaptic cleft, and (iii) a postsynaptic density. In addition MFTs were categorized by the presence of zincspecific silver in or adjacent to vesicles. The number of terminals used for each layer was SG (n = 17), IML (n = 23), and M M L (n = 13). Terminals were measured by planimetry on a Zeiss IBAS image analyser. Vesicle size and density were measured on photographic prints at a final magnification of × 55,000. Vesicle diameter was measured for the five largest vesicles per terminal because larger diameters more closely approximate a section through the center of a vesicle, i.e. the true diameter of a vesicle. Vesicle density was measured by counting all vesicles contained in a 30 x 30-mm square overlay, or a 30 × 15-mm rectangular overlay in smaller terminals. To avoid sampling errors in vesicle density measures, terminals smaller than either the square or rectangular overlay were not sampled. For statistical comparisons between layers, a full design Fisher least significant difference ANOVA was computed and differences were accepted at the 0.05 level with a one- or two-tail test, depending on the hypothesis being tested.
Fig. 2. Photomicrographs of control (A, B: A12430) and epileptic (C, D: 4855) hippocampus stained with H and E (A, C) and neo-Timm (B, D). In control (B) there are no IML (arrowheads) mossy fiber terminals above the SG (small circles), but in epileptic (D) there is a dense Timm stain around the entire IML (arrowheads). Note the hilus (h) is more densely stained in control (B) than epileptic (D) because there is a 69% loss of granule cells (see C, small circles). Note in B the characteristic infra- and suprapyramidal mossy fiber termination zones (straight arrows) in CA3 (curved arrow). Scale bar = 500 #m.
Fig. 2.
T.L. BABBet al.
356 RESULTS
Light microscopy of normal and epileptic Timmstained hippocampus Control F D processed with the routine (high sulphide) Timm procedure was always darkly stained in the hilus, CA3 dendritic zones, OML near the hippocampal fissure, and in granule cells. The IML and M M L had little or no staining. Control tissue processed with the neo-Timm (low sulphide) procedure was darkly stained in the hilus (Figs 1B, 2B) and CA3 (Fig. 2B) and showed no staining in the IML, M M L (Figs 1B, 2B) or granule cells. Figure 1A shows a coronal section of a control hippocampus (autopsy C587-L) Cresyl Violet-stained to illustrate the anatomy of the F D [hilus, SG, hippocampal fissure (h0] for comparison with CA3 pyramids proper (CA3) and the extension of pyramidal cells into the F D (termed CA3c or CA4 pyramids). The dashed line is included to show where the hilar neurons would be distinct from the CA4 pyramids. The box spans the polymorph zone to the hippocampal fissure, the area magnified 55 times in Fig. 1B (control, A12189) and C (epileptic, 4727). Figure 1B shows the normal pattern of Timm staining of mossy fibers, where only the polymorph-hilar (PM) region is well stained. By contrast, in epileptic hippocampus (Fig. 1C) the inner molecular layer is Timmstained, and these MFTs are shown at higher magnification as dark puncta in Fig. 1D (arrowheads). The 40-/~m-thick section of the SG in Fig. IC shows very little Timm stain compared to the high magnification of Fig. 1E where Timm-stained puncta appear among the granule cells (g). These puncta in SG may be located on cell bodies, but our ultrastructural studies indicate that they are probably terminating on dendrites passing through SG, arising from or actually on the apical shafts of granule cells (see Fig. 4). In all the physiologically and pathologically identified epileptic hippocampi (n = 21) the only reliable difference from controls (n = 6) in the pattern of Timm stain was a dense band in the IML (Figs 1C, D; and 2D). The width of the band was variable, but usually encompassed the proximal third of the molecular layer. For example, Figure 2D shows the dense wide band (arrowheads) in a sclerotic epileptic hippocampus, and the IML staining varies in thickness and density around the dentate gyrus. The H and E stain (Fig. 2C) is shown to indicate that there is severe hippocampal gliosis and cell loss in the hilus (Fig. 2C). Figure 2A is an H and E stain from a control autopsy (A12430) to show the normal cell density in the hilus and CA3. The neo-Timm-stained section (Fig. 2B) is 200#m away and shows the normal pattern of mossy fibers projecting from the granule cells (Fig. 2B, small circles), through the hilus, to terminate above and below (Fig. 2B, straight arrows) the CA3 pyramids (Fig. 2B, curved arrow). These supra- and infrapyramidal terminations are typical of mossy fibers.
Ultrastructure of mossy fiber terminals & epileptic fasc& dentata Electron microscopy of neo-Timm-stained F D in five epileptic hippocampi was studied quantitatively. Figure 3C shows a typical M F T labeled with zincspecific silver granules (arrowheads). This terminal, located in the upper SG, is very large, has large zinc-containing vesicles, and forms an asymmetrictype (possibly excitatory) synapse (arrow) on a dendrite (d). These characteristics are similar to those of MFTs in the hilus and CA3. All mossy fiber synapses in epileptic F D were the assymmetric type. We have not yet found a M F T synapsing on a granule cell body. Such axosomatic synapses may simply be very rare compared to the mossy fiber synapses on dendrites coursing through the SG. Figure 3A shows a large, zinc-labelled (arrowheads) N F T surrounding a dendrite (d) in the IML and forming three asymmetric synapses (arrows) in this ultrathin section. The presence of a mossy fiber synapsing on a dendrite in either SG or IML would not necessarily prove that a "recurrent" circuit exists. These mossy fibers could synapse on dendrites not originating from granule cells. However, it is most likely that granule cell dendrites are the mossy fiber targets, based on the light microscopic evidence of the dense terminals in only SG and IML where there are comparatively few dendrites of non-granule neurons. Figure 4 is an example of electron microscopic evidence of a granule cell whose apical shaft (Fig. 4B) and dendritic processes (Fig. 4A, B) have zinc-labeled MFTs. There are also numerous MFTs in close contact with the apical shaft (Fig. 4B) that may make synapses out of the plane of this section. When visually sampling SG and IML in the electron microscope it appeared that the Timm-positive synaptic density was greater in the IML than in the SG; however, no significant differences could be found when computing the actual sampled regions of Table 1. The average synapse density was 17.09 + 1.88/200/lm 2 in the IML and 13.97 + 1.44/200 # m 2 in the SG. It is important to note that in the SG all the measures of zinc-labeled terminals had to be made in the neuropil between granule cell bodies which would occupy a greater portion of the area sampled than would be the case for dendrites in the IML. In contrast to Fig. 3A and C, Fig. 3B (from the MML) shows a smaller, zinc-free terminal (T) and smaller dendrite (d) with an asymmetric synapse (arrow). There were only a few zinc-labeled terminals found in the M M L of our study, and none were found in the OML. Zinc-label was never observed in dendrites (e.g. Fig. 3) or cell bodies (e.g. Fig. 3C). Table 1 summarizes the quantitative analyses of Timm-positive and -negative terminals in three layers of Timm-stained epileptic FD: (i) SG, (ii) IML, and (iii) MML. These comparisons provide further
Synaptic reorganization in human fascia dentata
Fig. 3. Neo-Timm electron micrographs from epileptic sclerotic FD to show zinc-label (arrowheads) in MFTs with asymmetric synapses (thick arrows) on dendrites (d). Label is present in upper SG (C) and IML (A) but not in terminal (T) in MML (B). cm, cell membrane of granule cell. Ur and Pb stain. Scale bar = 0.5 #m.
357
O0
Synaptic reorganization in human fascia dentata
359
Table 1. Mean terminal densities and terminal areas in neo-Timm-stained electron microscopy of the fascia dentata in five epileptics to document the location and characteristics of Timm-positive and -negative terminals FD layer MML IML SG
Timm-positive Terminals/200/~m 2 Mean terminal area [3.74 + 1.76] 17.09 + 1.88 13.97 __+1.44
1.38 +__0.51 #m 2 [1.29 + 0.30 #m 2] 2.32 -1-0.31 #m 2
evidence for the identification of M F T s innervating the SG and I M L and less so the MML. There were no M F T s detected in the OML. The most important findings between layers or columns are bracketed in Table 1: (i) significantly fewer Timm-positive terminals in M M L compared to both SG (P < 0.005, two-tail test) and to I M L (P < 0.002, two-tail test), (ii) significantly more Timm-negative terminals in M M L compared to both SG (P < 0.002, two-tail test) and to I M L (P < 0.05, two-tail test), and (iii) slightly smaller Timm-positive terminal areas in the IML compared to the SG (P < 0.05, with a one-tail test). It is important to note that the mean area of Timmpositive terminals in the I M L was smaller than the other layers, which was an unexpected finding. There were no significant differences in the Timm-negative terminal areas between any layer. The Timm-negative mean terminal area in SG was larger but much more variable than in the IML and MML. Valid statistical comparisons cannot be made between Timm-positive and -negative ultrastruetures because of a possible sampling bias. Electron micrographs were taken to include Timm-positive terminals in each layer; hence the incidence of positive to negative terminals in the total data set could be biased toward detecting more Timm-positive terminals. Nevertheless, the random sampling of terminals in M M L yielded fewer than one positive to every five negative terminals (Table 1). By contrast, in both the SG and I M L fewer than one negative terminal was detected for every four positive terminals. Note also that the standard errors of the means for each of these conditions were relatively low, indicating that the ratios of positive to negative terminals across each of the three layers were similar between the five patients (Table 1). There was no a priori sampling bias in measuring mean areas between Timm-positive and negative terminals. As predicted, Timm-positive terminals were larger than negative terminals in SG (P < 0.05, using one-tail test), in I M L (P < 0.05, two-tail test) and in M M L (P < 0.05, two-tail test). Further evidence that the Timm-positive terminals in SG and IML are characteristic of MFTs, based on
Timm-negative Terminals/200 #m 2 Mean terminal area [21.7 _ 3.62] 4.42 __+1.10 3.01 -I-0.88
0.41 _ 0.07 #m 2 0.41 + 0.06 #m 2 1.11 + 0.54 gm 2
vesicle sizes being significantly larger than vesicles in terminals of the M M L or O M L is provided by Table 2. Additionally, the larger MFTs (see Table 1) have significantly fewer vesicles per/~m 2 in SG and I M L than in M M L or OML. Inner molecular layer Timm staining and hilar cell loss Correlation studies between routine Timm I M L densitometry and percentage hilar cell loss or between neo-Timm I M L rating and hilar cell loss rating were both statistically significant and supported the hypothesis that hilar cell loss is positively correlated with the density of I M L Timm staining only when both controls and epileptics are included (see Table 3). However, for epileptics alone the positive correlations were not statistically significant (see Table 3); that is, our study did not demonstrate a linear relation between extent of local hilar cell loss and the nearby extent of MFTs in the IML within the epileptic tissue. There was no significant correlation between I M L stain density and age of epileptics. DISCUSSION The present study was designed to determine if there are synaptic reorganizations in damaged epileptic human hippocampus. It was likely that hippocampal sclerosis in the F D and CA45'7 was associated with depopulation of hilar neurons and deafferentation of the IML; however, the cells of origin for possible reactive synaptogenesis into the I M L were not known. In fact, Golgi studies of dendrites of granule cells in hippocampal epilepsy showed degenerated branches and spines s3 and loss of complex spines in both apical and basilar proximal dendrites (see Ref. 4, Figs 3 and 4), which might indicate failure of I M L reinnervation. However, new data strongly indicate that dendrites of the "epileptic" IML are reinnervated with mossy fibers from granule cells, forming putative monosynaptic recurrent excitatory synapses. 6,TAe,32,s8The importance of this type of synaptic reorganization is that monosynaptic reexcitation is probably much more effective than
Fig. 4. Electron micrographs to illustrate zinc-labeled MFTs synapsing on granule cell dendritic processes (A, B) and apical shaft (B). (A) Low magnification of granule cell body (N, nucleus; n, nucleolus), apical shaft (AS) and mossy fiber synapse (thick arrows) on dendrite (d). Inset is magnified 2.6 times to show that the MFT forms an asymmetric synapse. (B) Electron micrograph of zinc-labeled mossy fiber synapses on granule cell apical shaft (AS, large arrow, on dendritic processes (arrowheads), and other MFTs in close contact with the apical shaft (small arrows). Some glial fibrils are marked with white asterisks. Ur and Pb stain. Scale bar in A = 2 # m ; in B = 1.0/~m.
360
T.L. BABBet al.
Table 2. Mean vesicle sizes (nm) and vesicle densities (per pm 2) in four layers of fascia dentata in two epileptic hippocampi processed with neo-Timm electron microscopy simultaneously to control for possible shrinkage differences FD layer
Mean vesicle size (nm)
OML MML IML SG
56 _ 2 51 + 3 68 __+2 76_+4
Vesicle density (#m ~) 190 + 15 196 + 18 141 _ 12 161 _+ 11
OML
MML IML SG Statistically significant differences between layers (P < 0.05 two-tail Fisher ANOVA test) shown by closed arrows; open arrows not significant. disynaptic feedback or feedforward excitation. For example, in vitro studies of CA3 neurons have demonstrated that current-evoked action potentials in one neuron may elicit excitatory postsynaptic potentials and/or discharges in a second (postsynaptic) neuron. 41 Such recurrent or local excitation detected intracellularly in vitro was not c o m m o n (6% of 88 paired neurons); however, Lebovitz et al. 37 had used extracellular techniques in the deafferented fornix in vivo preparation to demonstrate shortlatency recurrent excitation in CA3 pyramids. After a fornix-induced antidromic spike, an orthodromic spike could be detected 1 ms later, just before the massive recurrent inhibition blocked further CA3 firing (Ref. 37, p. 104). Miles and Wong 46 demonstrated that blockade of recurrent inhibition in vitro would lead to even greater numbers of neurons being synchronously discharged by activation of a single neuron. These results suggest that in CA3 monosynaptic recurrent excitatory circuits exist but are normally controlled by the more powerful recurrent inhibition. There are very few papers on the physiology of normal dentate granule cells; however, Fricke and Prince 28 have concluded that there was an " . . . absence of significant recurrent excitatory circuitry within the dentate gyrus itself (p. 195)." Using
the in vitro hippocampal preparation, they were unable to evoke more than one granule cell action potential with supramaximal orthodromic stimulation and did not find burst discharges after blockade of inhibitory postsynaptic potentials. By contrast, granule cells exhibited typical recurrent inhibition in 27 of 34 cells, which is a comparable incidence to that of hippocampal pyramidal cells. These physiological results suggest that recurrent excitatory circuits are not normally present or not very effective in rat dentate gyrus. Golgi-electron microscopic studies have demonstrated in primates that M F T s may normally synapse on dendrites in the FD, suggesting a normal monosynaptic feedback circuit25 Timm studies of primate indicate, however, that there are very few mossy fibers in SG or I M L ) 8 Golgi studies in rats have suggested a disynaptic recurrent excitatory path where mossy fibers excite hilar mossy cells that secondarily re-excite the IML of the FD. zS~ Hence, it appears that normal F D has weak monosynaptic or disynaptic re-excitation and would not be a site for epileptogenesis, as previously demonstrated in vivo 24 and in vitro. ~ However, with a more potent monosynaptic recurrent excitatory synapse, granule cell discharges and re-excitation would more likely contribute to epileptogenicity. Species differences and ultrastructure in fascia dentata Timrn staining
Electron micrographs revealed Timm's silver precipitate only in vesicles and free intervesicular space of apparently typical M F T s in the IML, SG and a few in the MML. With longer development times, silver capsules grow on the zinc, rupture vesicles and eventually fill the entire terminal. Hence, given the longer development allowed for light microscopy, each Timm puncta seen with light microscopy usually represents a terminal seen with electron microscopy. The distribution of silver in h u m a n M F T s (see Fig 3A, C) appears to be similar to that previously described in the rat hippocampus. 33,5° For example, their silver- or zinc-labeled giant boutons had round clear vesicles with only about 10% of them containing the silver label, and all labeled terminals made only asymmetric synapses on dendrites. The two
Table 3. Results of two independent studies correlating extent of hilar cell loss with extent of inner molecular layer Timm staining
Expt 1 Routine Timm IML density Expt 1 Routine Timm IML density
Controls and epileptics (n = 13) Percentage hilar cell loss r = 0.61, P < 0.05 (two-tail) Epileptics only (n = 9) Percentage hilar cell loss r =0.16, P
Neuroscience Vol. 42, No. 2, pp. 351-363, 1991
Printed in Great Britain
SYNAPTIC REORGANIZATION BY MOSSY FIBERS IN H U M A N EPILEPTIC FASCIA D E N T A T A T. L. BABB,*'~ W. R. KUPFER,'~ J. K. PRETORIUS,t P. H. CRANDALL*~§ and M. F. LEVESQUE§ *Department of Neurology, tBrain Research Institute and the §Division of Neurological Surgery, UCLA School of Medicine, University of California, Los Angeles, CA 90024-1769, U.S.A.
Al~traet--This study was designed to identify whether synaptic reorganizations occur in epileptic human .hippocampus which might contribute to feedback excitation. In epileptic hippocampi, (n = 21) reactive synaptogenesis of mossy fibers into the inner molecular layer of the granule cell dendrites was demonstrated at the light microscopic and electron microscopic levels. There was no inner molecular layer staining for mossy fibers in autopsy controls (n = 4) or in controls with neocortex epilepsy having no hippocampal sclerosis (n = 2). Comparing epileptics to controls, there were statistically significant correlations between Timm stain density and hilar cell loss. Since hilar neurons are the origin of ipsilateral projections to the inner molecular layer, this suggests that hilar deafferentation of this dendritic zone precedes mossy fiber reafferentation. Quantitative Timm-stained electron microscopy revealed large, zinc-labelled vesiclesin terminals with asymmetric synapses on dendrites in the inner molecular and granule cell layers. Terminals in the middle and outer molecular layers did not contain zinc, were smaller and had smaller vesicles. These histocbemical and ultrastructural data suggest that in damaged human epileptic hippocampus, mossy fiber reactive synaptogenesis may result in monosynaptic recurrent excitation of granule cells that could contribute to focal seizure onsets.
It has been well established that following deafferentation of mammalian dendrites, "new" synapses are formed within a few weeks. 38-4° This process of axonal growth into vacated dendrites has been well studied in the hippocampus (for review see Ref. 19), and referred to as reactive synaptogenesis because the new synapses form in response to loss of normal afferents. Granule cell dendrites in the fascia dentata (FD) receive afferents to the outer molecular layer (OML) from lateral entorhinal cortex, to the middle molecular layer (MML) from medial entorhinal cortex and to the inner molecular layer (IML) from a variety of regions, most prominently the hilar ipsilateral associational and commissural system. 35 Most of the reports on the hilar cell denervation and IML reactive synaptogenesis used transection or removal44'47'48of only the commissural inputs to the IML layer in adult or neonatal r a t : 7 Electron microscopic studies showed that terminal degeneration was followed by recovery of synapses; however the origin of these fibers was not fully determined. The authors suggested that the ipsilateral associational axons in the IML sprouted and innervated the vacated "com:~To whom correspondence should be addressed. Abbreviations: CA3/CA4, Cornu Ammonis regions; FD,
fascia dentata; GLD, gray level difference; H and E, hematoxylin and eosin; HS, hippocampal sclerosis; IML, inner molecular layer; KA, kainic acid; MF, mossy fiber; MFT, mossy fiber terminal; MML, middle molecular layer; OML, outer molecular layer; Pb, lead salts; PM, polymorph layer; R, correlation; SG, stratum granulosum; Ur, uranyl acetate.
missural" zones."45 The functional significance of this new synaptic rearrangement remained a puzzle, even though this new circuit possibly could increase the hilar ipsilateral projections back to the granule cells and thereby form a stronger disynaptic excitatory feedback system: i.e. granule cell mossy fibers would excite hilar neurons which would then excite IML dendrites of granule cells. Laurberg and Zimmer36used the Timm's technique to demonstrate that following IML denervation, mossy fibers from granule cells sprout back into the IML, apparently forming a monosynaptic feedback circuit. This circuit was confirmed using Golgielectron microscopy)° Using the kainic acid (KA) lesion of CA3/CA4, Tauck and Nadler 6° demonstrated that this mossy fiber reorganization produces calcium-dependent monosynaptic multiple population spikes in granule cells after single activation by antidromic stimulation. They found that the incidence of multiple firing was positively correlated with the greatest loss of hilar neurons and with the highest rating of Timm's staining density in the IML. Other studies using KA hippocampal lesions have shown faster hippocampal kindling27 and occasional spontaneous seizures17 which have been associated with IML sprouting.2° Hence, the available evidence suggests that following denervation of the IML there is a functional reinnervation that results in hyperexcitability. Our interest is in the anatomy of synaptic reorganization in the IML because those dendrites would be selectively denervated in human hippocampal
351
352
T.L. BABBet al.
lepsy by loss o f hilar neurons. Because e n t o r h i n a l cortex n e u r o n s are spared in h u m a n h i p p o c a m p a l epilepsy, neither the M M L n o r O M L is denervated (cf. Ref. 6). The hilar n e u r o n s are considered to be the cells o f origin for ipsilateral associational axons to the IML, 13'35a n d they would be reduced at the time t h a t severe h i p p o c a m p a l sclerosis occurred. We hypothesized t h a t loss o f hilar n e u r o n s occurred at the time of severe h i p p o c a m p a l sclerosis, deafferentation o f the I M L followed, a n d reactive synaptogenesis by mossy fibers would form a b n o r m a l m o n o s y n a p t i c recurrent excitatory synapses o n the I M L o f granule cells, a circuit t h a t could contribute to epileptogenicity. Previous reports have described a n o m a l o u s inn e r v a t i o n of the h u m a n epileptic I M L by mossy fibers 9'10'16'32'5s a n d by peptide-containing axons. 23,32
buffer for 48 h then stored at -80°C. Sections (30 # m thick) were cut at -70°C, returned to buffered ethylene glycol and glycerol for six months at -70°C, then placed in routine Timm solution. Although paraformaldehyde fixation was used prior to the sulphide solution, the low concentration of paraformaldehyde and high concentration of sodium sulphide resulted in a Timm reaction comparable to the routine procedure. Hence, the procedures with this hippocampus differed in many ways; however, the routine highsulphide Timm stain pattern was comparable to other non-epileptic FD. (3) A12189 was a 56-year-old male who died of an adenocarcinoma of the sigmoid colon and was autopsied 20 h post mortem. A 1.0 cm block of unfixed normal hippocampus was placed in neo-Timm, lowsulphide-glutaraldehyde fixative within 30 rain and processed with the neo-Timm procedure. (4) A12430 was a 59-year-old male who died of acute myocardial infarction without p r i o r ischemic changes and no evidence of gross brain damage. The brain was removed 24 h post mortem, and a 1.0 cm block of the hippocampus was processed with the neo-Timm procedure.
EXPERIMENTAL PROCEDURES
Criteria for identified epileptic hippocampus All 21 epileptic hippocampi had sclerosis typical of hippocampal epilepsy (cf. Ref. 5). Following hippocampal resection all patients had either no seizures or rare seizures, indicating that the hippocampus was epileptogenic. Additionally, intrahippocampal recordings from macroelectrodes26 or microelectrodes 8 verified that seizures originated in the hippocampus within 1-2cm of the tissues subsequently processed for the current studies. Electrodes in orbitofrontal gyrus and over neocortical regions approximating the International 10-20 System 3a provided a physiological detection of extrahippocampal seizure activity (for further details see Ref. I 1). In nine of the 21 epileptic patients, temporal lobeetomy was performed without intrahippocampal seizure recordings (for diagnostic criteria see Ref. 26). These nine hippocampi were also sclerotic, and the surgical resection which included the anterior 3.5 cm of the hippocampus resulted in seizure relief. Control comparison hippocampi The density of Timm's sulphide silver stain processed by either routine Timm (n = 9) or neo-Timm (n = 12) in identified epileptic hippocampus was compared to six nonepileptic hippocampi. Two patients (4295, 4406) had seizures arising outside the hippocampus and mild neuron loss not typical of hippocampal epilepsy. Patient 4406 had a grade II astrocytoma in the superior temporal gyrus, and all seizures were recorded from lateral temporal cortex. Patient 4295 also had seizures originating from lateral temporal cortex; however, no specific pathology was found in the resected temporal lobe. These hippocampi were stained by the routine high-sulphide Timm method exactly as the epileptic hippocampi. Four hippocampi removed at autopsy were also studied. The fixation protocols prior to Timm stain differed. (1) A11702 was a 64-year-old female who died from intestinal metastatic adenocarcinoma and was autopsied 4 h post mortem. A 1.0 cm block of the unfixed normal hippocampus was placed into the routine Timm solution within 30 min and processed identically to the epileptic hippocampus. This was a normal brain, and hence the Timm procedure differed from the epileptic's procedure only in the extra 4 h delay. (2) C587-L was a hippocampus kindly provided to us by Drs David Amaral and Ricardo Insausti, The Salk Institute, San Diego, U.S.A. The brain of a 63-year-old male who died of epidermoid carcinoma was removed 90 min after death, flushed through one carotid and the basilar artery with buffered heparinized normal saline followed by the following fixative: 4% paraformaldehyde, 0.002% picric acid and 2.7% NaCI in 0.1 M phosphate buffer. One-centimeter blocks were placed in 20% glycerol in 0.1 M phosphate
Hippocampal fixation and development Two methods were used to visualize zinc and other metals in the hippocampus. The first method (routine Timm) was used for light microscopic analysis, while the second method (neo-Timm) allowed both light and electron microscopic analysis from a single block of tissue. Routine Timm stain. The solution consisted of 1.2% sodium sulphide in 0.I M sodium phosphate buffer at pH 7.4. Tissue blocks were immersed at 4°C for 24-96 h. Blocks were than transferred to a cold 3% glutaraldehyde, 15% sucrose solution pH 7.4 for a maximum of 24 h. Tissue was taken directly from the glutaraldehyde-sucrose solution, quick-frozen and sectioned with a cryostat (-15°C) at 30#m. Sections were mounted on chromiumalum-gelatin-coated slides and allowed to air dry. Mounted sections were immersed in a "physical developer" in darkness at 21°C. The physical developer was prepared as follows: to 180ml gum arabic (500 g/1 water) add 30ml aqueous solution of 7.65 g citric acid and 7.05 g sodium citrate; just prior to use add (in the dark) 5.0 g hydroquinone in 90 ml distilled water and 1.5 ml of 15% silver nitrate solution. Two batches of different staining density were made for each patient based on visual inspection of staining, staining times varied from 30-120 min. After removal from developer, sections were washed first in distilled water for 2 min, in running tap water for 10 min, and finally dehydrated and coverslipped. Routinely, alternate sections were processed for hematoxylin and eosin (H and E) staining to localize anatomical fields and provide cell counts using the Abercrombie I correction, as described previously. 6 Hilar cell counts were limited to a region between stratum granulosum (SG) and the pyramids of CA4 or CA3c3 This region was at all times at least 50/~m from SG and 100 #m from CA4 (see hilus in Fig. 1A). The entire areas of the hilus counted ranged from 0.3 to 1.5 mm 2. The hilar cell counts were used for correlations with routine Timm staining density by quantitative densitometry. The hilar cell losses adjacent to neo-Timm-stained tissues were rated from 0 (no loss) to 3 (greatest loss) for subsequent correlations with neo-Timm staining density which was rated 0-3. Neo-Timm stain. To increase the specific staining of zinc and improve visualization of the mossy fiber system at both light and electron microscopic levels, a second Timm's staining method was used. 2~'5°The fixative consisted of 4% glutaraldehyde, 0.1% sodium sulphide, and 0.002% calcium chloride in 0.12M Millonig's buffer at pH7.3. Tissue blocks, approximately 1.0 cm thick, were immersed in fixative for 24-72 h at room temperature. Blocks were then transferred to cold 0.12 M Millonig's buffer with 0.002% calcium chloride at pH 7.3 for periods ranging from 2 h (room temperature) to two days (6°C). Next, blocks were
Fig. 1. Photomicrographs of human hippocampus to demonstrate patterns of Timm stains in control (B) and epileptic (C-E) FD (A, box). (A) Cresyl Violet stained autopsy normal hippocampus (case C587-1) shows normal cytoarchitecture of coronal section. Dashed line segregates the CA4 pyramidal cells from the hilus of the FD. (B) Timm stain of normal section adjacent to A showing dense stain only in polymorph or hilus (h) but not in SG, IML, MML, or OML. There is a faint irregular trace of Timm stain in parts of SG. (C) Timm stain in epileptic, sclerotic hippocampus showing dense stain in IML. (D) Timm stain puncta in IML from boxed area in C. (E) Timm stain puncta in SG from boxed area in C. Note that there is less Timm stain in PM layer of epileptic (C) than control PM (B) because there is a 55% granule cell loss in the epileptic (C). g, Granule cells; hf, hippocampal fissure; PM, polymorph layer. Scale bars in A = 500 gm; in B, C = 100 gm; in D, E = 20/~m. 353
354
T.L. BABBet aL
embedded in agar and sectioned with a Vibratome at 40 p m for light microscopy or 500#m for electron microscopy, Light microscopic sections were mounted, dried and developed as described previously for the routine Timm procedure. Alternate 40-pm sections were stained with H and E for routine cytology and measures of cell density. Neo- Timm's electron microscopy Electron microscopic sections (500 p m thick) were transferred to fresh buffer for 2 h, then into 15 ml glass vials containing a developing solution. The developer was prepared as follows: to 180 ml gum arabic (500 g/1 water) a 30 ml aqueous solution of 7.65 g citric acid and 7.05 g sodium citrate was added; just prior to use 2.5 g hydroquinone dissolved in 45 ml water and 45 ml 0.73% silver lactate were added in darkness. Development took place in darkness with constant agitation in a 26°C water bath for 60-80min; then sections were washed for 15min, wetmounted and coverslipped. Using low power light microscopy, areas of interest with optimal staining were selected and recorded by drawings. These areas (approximately 4 mm 2) were microdissected from the section and transferred to fresh buffer. Sections were post-fixed for 1 h in a solution of 2.0% osmium tetroxide, 0.12 M Millonig's buffer, 0.002% calcium chloride and 0.54% glucose, at pH 7.3. After two 10-min washes in 2.4% sodium chloride, the tissue was dehydrated through alcohols to propylene oxide then embedded in epoxy. Initial polymerization was in BEEM capsules for three days at 60°C. Tissue was observed by light microscopy in the block, trimmed to the F D and again recorded by drawing. Sections ( 5 ~ p m thick) were cut and dropmounted on glass slides, and those with optimal anatomy and Timm's stain were selected for re-embedding. Optimal Timm's staining was always within 30 # m of the tissue surface due to limits of the penetration of reagents. To obtain a fiat tissue surface for light microscopic observation and thin sectioning, the tissue was re-embedded in a BEEM capsule directly on the glass slide. After one day of polymerization at 60°C the block was removed from the glass slide by sliding a razor blade between the block and the glass. After two or three more days of polymerization, blocks were trimmed to contain the FD from the sub-granular polymorph layer through the granule cell and molecular layers to the hippocampal fissure. Once again, reference drawings were made. Thin sections (60 nm, silver interference color) were cut with a diamond knife and mounted on 200- or 300-mesh copper grids. Tissue contrast was increased by drop staining for 1 h in uranyl acetate (Ur) and 3-5 min in lead salts (Pb). Photographs were taken on a Siemens model 1A at an accelerating voltage of 80 kV. Anatomical distinctions between layers of the FD were made by correlating reference drawings with the thin section in the electron microscope. Data analysis To quantify the density of routine Timm's stain in the FD, low-power light microscopic densitometric measurements were made of the various layers using a Ziess IBAS interactive image analysis system. A video image of a light microscope field including SG, IML, M M L and OML was digitized into the system and each layer outlined using a cursor. The mean densitometric gray level was measured in
each layer, and the measurement repeated at the same location on six tissue slides for each patient; three slides were stained lightly (short development time) and three slides were stained heavily (long development time). Light and dark staining provided a control for the effect of staining density on relative density differences within each tissue section. A length of approximately 1 mm of FD was sampled in each slide. The sample location was selected as being representative of the overall staining pattern in F D prior to any knowledge of the patient's hippocampal epilepsy or pathology. A total of 13 hippocampi with routine Timm staining (nine epileptic, four control) were examined. IML staining positively correlated with hippocampal sclerosis (HS), whereas M M L staining was relatively light, increasing slowly with development time but not correlated with HS. A normalized measure of IML staining density was computed by subtracting the gray level difference (GLD) of the M M L from the IML. To quantify the density of neo-Timm staining, the 0-3 rating procedure of Tauck and Nadler 6° was used on coded slides by three independent scorers (T.L.B., W.R.K. and J.K.P.) and averaged. Similarly, hilar cell loss was scored 0-3 on H and E-stained slides coded without knowledge of the relation to the Timm slides and averaged. Pearson product moment correlations (R) were made of the Timm densitometry or the neo-Timm density scores against variables such as hilar neuron loss, age and background staining density. Student's t-tests were used for statistical significance (Ref. 25, p. 78). Ultrastructural aspects of neo-Timm-stained axon terminals [presumed mossy fiber terminals (MFTs)], and synapses were analysed in each layer of FD for: (i) the number of stained terminals, (ii) mean terminal size, (iii) average vesicle size and (iv) average vesicle density. Stratified electron microscopic sampling from the layers of FD was made by excluding sampling in transition areas between those layers, a procedure similar to that reported by McWilliams and Lynch. 44 Ten photographs ( × 17,000) were taken of axon terminals in each layer of the FD. If Timm's staining was rare or absent in a layer, photographs were taken randomly of well-preserved unstained axon terminals. The criteria used to identify terminals in any F D layer included: (i) a continuous spherically shaped membrane containing vesicles and mitochondria, (ii) a pre-synaptic density adjacent to a synaptic cleft, and (iii) a postsynaptic density. In addition MFTs were categorized by the presence of zincspecific silver in or adjacent to vesicles. The number of terminals used for each layer was SG (n = 17), IML (n = 23), and M M L (n = 13). Terminals were measured by planimetry on a Zeiss IBAS image analyser. Vesicle size and density were measured on photographic prints at a final magnification of × 55,000. Vesicle diameter was measured for the five largest vesicles per terminal because larger diameters more closely approximate a section through the center of a vesicle, i.e. the true diameter of a vesicle. Vesicle density was measured by counting all vesicles contained in a 30 x 30-mm square overlay, or a 30 × 15-mm rectangular overlay in smaller terminals. To avoid sampling errors in vesicle density measures, terminals smaller than either the square or rectangular overlay were not sampled. For statistical comparisons between layers, a full design Fisher least significant difference ANOVA was computed and differences were accepted at the 0.05 level with a one- or two-tail test, depending on the hypothesis being tested.
Fig. 2. Photomicrographs of control (A, B: A12430) and epileptic (C, D: 4855) hippocampus stained with H and E (A, C) and neo-Timm (B, D). In control (B) there are no IML (arrowheads) mossy fiber terminals above the SG (small circles), but in epileptic (D) there is a dense Timm stain around the entire IML (arrowheads). Note the hilus (h) is more densely stained in control (B) than epileptic (D) because there is a 69% loss of granule cells (see C, small circles). Note in B the characteristic infra- and suprapyramidal mossy fiber termination zones (straight arrows) in CA3 (curved arrow). Scale bar = 500 #m.
Fig. 2.
T.L. BABBet al.
356 RESULTS
Light microscopy of normal and epileptic Timmstained hippocampus Control F D processed with the routine (high sulphide) Timm procedure was always darkly stained in the hilus, CA3 dendritic zones, OML near the hippocampal fissure, and in granule cells. The IML and M M L had little or no staining. Control tissue processed with the neo-Timm (low sulphide) procedure was darkly stained in the hilus (Figs 1B, 2B) and CA3 (Fig. 2B) and showed no staining in the IML, M M L (Figs 1B, 2B) or granule cells. Figure 1A shows a coronal section of a control hippocampus (autopsy C587-L) Cresyl Violet-stained to illustrate the anatomy of the F D [hilus, SG, hippocampal fissure (h0] for comparison with CA3 pyramids proper (CA3) and the extension of pyramidal cells into the F D (termed CA3c or CA4 pyramids). The dashed line is included to show where the hilar neurons would be distinct from the CA4 pyramids. The box spans the polymorph zone to the hippocampal fissure, the area magnified 55 times in Fig. 1B (control, A12189) and C (epileptic, 4727). Figure 1B shows the normal pattern of Timm staining of mossy fibers, where only the polymorph-hilar (PM) region is well stained. By contrast, in epileptic hippocampus (Fig. 1C) the inner molecular layer is Timmstained, and these MFTs are shown at higher magnification as dark puncta in Fig. 1D (arrowheads). The 40-/~m-thick section of the SG in Fig. IC shows very little Timm stain compared to the high magnification of Fig. 1E where Timm-stained puncta appear among the granule cells (g). These puncta in SG may be located on cell bodies, but our ultrastructural studies indicate that they are probably terminating on dendrites passing through SG, arising from or actually on the apical shafts of granule cells (see Fig. 4). In all the physiologically and pathologically identified epileptic hippocampi (n = 21) the only reliable difference from controls (n = 6) in the pattern of Timm stain was a dense band in the IML (Figs 1C, D; and 2D). The width of the band was variable, but usually encompassed the proximal third of the molecular layer. For example, Figure 2D shows the dense wide band (arrowheads) in a sclerotic epileptic hippocampus, and the IML staining varies in thickness and density around the dentate gyrus. The H and E stain (Fig. 2C) is shown to indicate that there is severe hippocampal gliosis and cell loss in the hilus (Fig. 2C). Figure 2A is an H and E stain from a control autopsy (A12430) to show the normal cell density in the hilus and CA3. The neo-Timm-stained section (Fig. 2B) is 200#m away and shows the normal pattern of mossy fibers projecting from the granule cells (Fig. 2B, small circles), through the hilus, to terminate above and below (Fig. 2B, straight arrows) the CA3 pyramids (Fig. 2B, curved arrow). These supra- and infrapyramidal terminations are typical of mossy fibers.
Ultrastructure of mossy fiber terminals & epileptic fasc& dentata Electron microscopy of neo-Timm-stained F D in five epileptic hippocampi was studied quantitatively. Figure 3C shows a typical M F T labeled with zincspecific silver granules (arrowheads). This terminal, located in the upper SG, is very large, has large zinc-containing vesicles, and forms an asymmetrictype (possibly excitatory) synapse (arrow) on a dendrite (d). These characteristics are similar to those of MFTs in the hilus and CA3. All mossy fiber synapses in epileptic F D were the assymmetric type. We have not yet found a M F T synapsing on a granule cell body. Such axosomatic synapses may simply be very rare compared to the mossy fiber synapses on dendrites coursing through the SG. Figure 3A shows a large, zinc-labelled (arrowheads) N F T surrounding a dendrite (d) in the IML and forming three asymmetric synapses (arrows) in this ultrathin section. The presence of a mossy fiber synapsing on a dendrite in either SG or IML would not necessarily prove that a "recurrent" circuit exists. These mossy fibers could synapse on dendrites not originating from granule cells. However, it is most likely that granule cell dendrites are the mossy fiber targets, based on the light microscopic evidence of the dense terminals in only SG and IML where there are comparatively few dendrites of non-granule neurons. Figure 4 is an example of electron microscopic evidence of a granule cell whose apical shaft (Fig. 4B) and dendritic processes (Fig. 4A, B) have zinc-labeled MFTs. There are also numerous MFTs in close contact with the apical shaft (Fig. 4B) that may make synapses out of the plane of this section. When visually sampling SG and IML in the electron microscope it appeared that the Timm-positive synaptic density was greater in the IML than in the SG; however, no significant differences could be found when computing the actual sampled regions of Table 1. The average synapse density was 17.09 + 1.88/200/lm 2 in the IML and 13.97 + 1.44/200 # m 2 in the SG. It is important to note that in the SG all the measures of zinc-labeled terminals had to be made in the neuropil between granule cell bodies which would occupy a greater portion of the area sampled than would be the case for dendrites in the IML. In contrast to Fig. 3A and C, Fig. 3B (from the MML) shows a smaller, zinc-free terminal (T) and smaller dendrite (d) with an asymmetric synapse (arrow). There were only a few zinc-labeled terminals found in the M M L of our study, and none were found in the OML. Zinc-label was never observed in dendrites (e.g. Fig. 3) or cell bodies (e.g. Fig. 3C). Table 1 summarizes the quantitative analyses of Timm-positive and -negative terminals in three layers of Timm-stained epileptic FD: (i) SG, (ii) IML, and (iii) MML. These comparisons provide further
Synaptic reorganization in human fascia dentata
Fig. 3. Neo-Timm electron micrographs from epileptic sclerotic FD to show zinc-label (arrowheads) in MFTs with asymmetric synapses (thick arrows) on dendrites (d). Label is present in upper SG (C) and IML (A) but not in terminal (T) in MML (B). cm, cell membrane of granule cell. Ur and Pb stain. Scale bar = 0.5 #m.
357
O0
Synaptic reorganization in human fascia dentata
359
Table 1. Mean terminal densities and terminal areas in neo-Timm-stained electron microscopy of the fascia dentata in five epileptics to document the location and characteristics of Timm-positive and -negative terminals FD layer MML IML SG
Timm-positive Terminals/200/~m 2 Mean terminal area [3.74 + 1.76] 17.09 + 1.88 13.97 __+1.44
1.38 +__0.51 #m 2 [1.29 + 0.30 #m 2] 2.32 -1-0.31 #m 2
evidence for the identification of M F T s innervating the SG and I M L and less so the MML. There were no M F T s detected in the OML. The most important findings between layers or columns are bracketed in Table 1: (i) significantly fewer Timm-positive terminals in M M L compared to both SG (P < 0.005, two-tail test) and to I M L (P < 0.002, two-tail test), (ii) significantly more Timm-negative terminals in M M L compared to both SG (P < 0.002, two-tail test) and to I M L (P < 0.05, two-tail test), and (iii) slightly smaller Timm-positive terminal areas in the IML compared to the SG (P < 0.05, with a one-tail test). It is important to note that the mean area of Timmpositive terminals in the I M L was smaller than the other layers, which was an unexpected finding. There were no significant differences in the Timm-negative terminal areas between any layer. The Timm-negative mean terminal area in SG was larger but much more variable than in the IML and MML. Valid statistical comparisons cannot be made between Timm-positive and -negative ultrastruetures because of a possible sampling bias. Electron micrographs were taken to include Timm-positive terminals in each layer; hence the incidence of positive to negative terminals in the total data set could be biased toward detecting more Timm-positive terminals. Nevertheless, the random sampling of terminals in M M L yielded fewer than one positive to every five negative terminals (Table 1). By contrast, in both the SG and I M L fewer than one negative terminal was detected for every four positive terminals. Note also that the standard errors of the means for each of these conditions were relatively low, indicating that the ratios of positive to negative terminals across each of the three layers were similar between the five patients (Table 1). There was no a priori sampling bias in measuring mean areas between Timm-positive and negative terminals. As predicted, Timm-positive terminals were larger than negative terminals in SG (P < 0.05, using one-tail test), in I M L (P < 0.05, two-tail test) and in M M L (P < 0.05, two-tail test). Further evidence that the Timm-positive terminals in SG and IML are characteristic of MFTs, based on
Timm-negative Terminals/200 #m 2 Mean terminal area [21.7 _ 3.62] 4.42 __+1.10 3.01 -I-0.88
0.41 _ 0.07 #m 2 0.41 + 0.06 #m 2 1.11 + 0.54 gm 2
vesicle sizes being significantly larger than vesicles in terminals of the M M L or O M L is provided by Table 2. Additionally, the larger MFTs (see Table 1) have significantly fewer vesicles per/~m 2 in SG and I M L than in M M L or OML. Inner molecular layer Timm staining and hilar cell loss Correlation studies between routine Timm I M L densitometry and percentage hilar cell loss or between neo-Timm I M L rating and hilar cell loss rating were both statistically significant and supported the hypothesis that hilar cell loss is positively correlated with the density of I M L Timm staining only when both controls and epileptics are included (see Table 3). However, for epileptics alone the positive correlations were not statistically significant (see Table 3); that is, our study did not demonstrate a linear relation between extent of local hilar cell loss and the nearby extent of MFTs in the IML within the epileptic tissue. There was no significant correlation between I M L stain density and age of epileptics. DISCUSSION The present study was designed to determine if there are synaptic reorganizations in damaged epileptic human hippocampus. It was likely that hippocampal sclerosis in the F D and CA45'7 was associated with depopulation of hilar neurons and deafferentation of the IML; however, the cells of origin for possible reactive synaptogenesis into the I M L were not known. In fact, Golgi studies of dendrites of granule cells in hippocampal epilepsy showed degenerated branches and spines s3 and loss of complex spines in both apical and basilar proximal dendrites (see Ref. 4, Figs 3 and 4), which might indicate failure of I M L reinnervation. However, new data strongly indicate that dendrites of the "epileptic" IML are reinnervated with mossy fibers from granule cells, forming putative monosynaptic recurrent excitatory synapses. 6,TAe,32,s8The importance of this type of synaptic reorganization is that monosynaptic reexcitation is probably much more effective than
Fig. 4. Electron micrographs to illustrate zinc-labeled MFTs synapsing on granule cell dendritic processes (A, B) and apical shaft (B). (A) Low magnification of granule cell body (N, nucleus; n, nucleolus), apical shaft (AS) and mossy fiber synapse (thick arrows) on dendrite (d). Inset is magnified 2.6 times to show that the MFT forms an asymmetric synapse. (B) Electron micrograph of zinc-labeled mossy fiber synapses on granule cell apical shaft (AS, large arrow, on dendritic processes (arrowheads), and other MFTs in close contact with the apical shaft (small arrows). Some glial fibrils are marked with white asterisks. Ur and Pb stain. Scale bar in A = 2 # m ; in B = 1.0/~m.
360
T.L. BABBet al.
Table 2. Mean vesicle sizes (nm) and vesicle densities (per pm 2) in four layers of fascia dentata in two epileptic hippocampi processed with neo-Timm electron microscopy simultaneously to control for possible shrinkage differences FD layer
Mean vesicle size (nm)
OML MML IML SG
56 _ 2 51 + 3 68 __+2 76_+4
Vesicle density (#m ~) 190 + 15 196 + 18 141 _ 12 161 _+ 11
OML
MML IML SG Statistically significant differences between layers (P < 0.05 two-tail Fisher ANOVA test) shown by closed arrows; open arrows not significant. disynaptic feedback or feedforward excitation. For example, in vitro studies of CA3 neurons have demonstrated that current-evoked action potentials in one neuron may elicit excitatory postsynaptic potentials and/or discharges in a second (postsynaptic) neuron. 41 Such recurrent or local excitation detected intracellularly in vitro was not c o m m o n (6% of 88 paired neurons); however, Lebovitz et al. 37 had used extracellular techniques in the deafferented fornix in vivo preparation to demonstrate shortlatency recurrent excitation in CA3 pyramids. After a fornix-induced antidromic spike, an orthodromic spike could be detected 1 ms later, just before the massive recurrent inhibition blocked further CA3 firing (Ref. 37, p. 104). Miles and Wong 46 demonstrated that blockade of recurrent inhibition in vitro would lead to even greater numbers of neurons being synchronously discharged by activation of a single neuron. These results suggest that in CA3 monosynaptic recurrent excitatory circuits exist but are normally controlled by the more powerful recurrent inhibition. There are very few papers on the physiology of normal dentate granule cells; however, Fricke and Prince 28 have concluded that there was an " . . . absence of significant recurrent excitatory circuitry within the dentate gyrus itself (p. 195)." Using
the in vitro hippocampal preparation, they were unable to evoke more than one granule cell action potential with supramaximal orthodromic stimulation and did not find burst discharges after blockade of inhibitory postsynaptic potentials. By contrast, granule cells exhibited typical recurrent inhibition in 27 of 34 cells, which is a comparable incidence to that of hippocampal pyramidal cells. These physiological results suggest that recurrent excitatory circuits are not normally present or not very effective in rat dentate gyrus. Golgi-electron microscopic studies have demonstrated in primates that M F T s may normally synapse on dendrites in the FD, suggesting a normal monosynaptic feedback circuit25 Timm studies of primate indicate, however, that there are very few mossy fibers in SG or I M L ) 8 Golgi studies in rats have suggested a disynaptic recurrent excitatory path where mossy fibers excite hilar mossy cells that secondarily re-excite the IML of the FD. zS~ Hence, it appears that normal F D has weak monosynaptic or disynaptic re-excitation and would not be a site for epileptogenesis, as previously demonstrated in vivo 24 and in vitro. ~ However, with a more potent monosynaptic recurrent excitatory synapse, granule cell discharges and re-excitation would more likely contribute to epileptogenicity. Species differences and ultrastructure in fascia dentata Timrn staining
Electron micrographs revealed Timm's silver precipitate only in vesicles and free intervesicular space of apparently typical M F T s in the IML, SG and a few in the MML. With longer development times, silver capsules grow on the zinc, rupture vesicles and eventually fill the entire terminal. Hence, given the longer development allowed for light microscopy, each Timm puncta seen with light microscopy usually represents a terminal seen with electron microscopy. The distribution of silver in h u m a n M F T s (see Fig 3A, C) appears to be similar to that previously described in the rat hippocampus. 33,5° For example, their silver- or zinc-labeled giant boutons had round clear vesicles with only about 10% of them containing the silver label, and all labeled terminals made only asymmetric synapses on dendrites. The two
Table 3. Results of two independent studies correlating extent of hilar cell loss with extent of inner molecular layer Timm staining
Expt 1 Routine Timm IML density Expt 1 Routine Timm IML density
Controls and epileptics (n = 13) Percentage hilar cell loss r = 0.61, P < 0.05 (two-tail) Epileptics only (n = 9) Percentage hilar cell loss r =0.16, P
