Do Mossy Fibers Release GABA?


Address correspondence and reprint requests to Dr. M. Walker at Department of Clinical and Experimental Epilepsy, Institute of Neurology, Queen Square, London WC1N 3BG, U.K. E-mail:


 Purpose: Mossy fibers are the sole excitatory projection from dentate gyrus granule cells to the hippocampus, forming part of the trisynaptic hippocampal circuit. They undergo signficiant plasticity during epileptogenesis and have been implicated in seizure generation. Mossy fibers are a highly unusual projection in the mammalian brain; in addition to glutamate, they release adenosine, dynorphin, zinc, and possibly other peptides. Mossy fiber terminals also show intense immunoreactivity for the inhibitory neurotrasnmitter γ-aminobutyric acid (GABA), and immunoreactivity for GAD67. The purpose of this review is to present physiologic evidence of GABA release by mossy fibers and its modulation by epileptic activity.

Methods: We used hippocampal slices from 3- to 5-week-old guinea pigs and made whole-cell voltage clamp recordings from CA3 pyramidal cells. We placed stimulating electrodes in stratum granulosum and adjusted their position in order to recruit mossy fiber to CA3 projections.

Results: We have shown that electrical stimuli that recruit dentate granule cells elicit monosynaptic GABAA receptor–mediated synaptic signals in CA3 pyramidal neurons. These inhibitory signals satisfy the criteria that distinguish mossy fiber–CA3 synapses: high sensitivity to metabotropic glutamate-receptor agonists, facilitation during repetitive stimulation, and N-methyl-D-aspartate (NMDA) receptor–independent long-term potentiation.

Conclusions: We have thus provided compelling evidence that there is a mossy fiber GABAergic signal. The physiologic role of this mossy fiber GABAergic signal is uncertain, but may be of developmental importance. Other evidence suggests that this GABAergic signal is transiently upregulated after seizures. This could have an inhibitory or disinhibitory effect, and further work is needed to elucidate its actual role.

Mossy fibers, the axons of dentate granule cells, are considered one of the main excitatory inputs to the hippocampus proper, and part of the excitatory trisynaptic loop (1). Mossy fibers also form synapses with inhibitory neurons in the hilar and CA3 regions, and activation of the mossy fiber pathway results in a large recruitment of inhibitory interneurons (2).

Mossy fibers are unusual in that the terminals form large boutons with multiple release sites that envelop spines on the proximal apical dendrites of CA3 neurons. Proximity to the CA3 pyramidal cell soma and multiple release sites result in a high probability that granule cell activation produces a large somatic excitatory potential. Transmission reliability is further reinforced by the marked frequency facilitation of mossy fiber to CA3 synapses that occurs at even low frequencies (3).

The mossy fiber pathway has other properties that distinguish it as a highly unusual projection in the mammalian brain. In addition to releasing glutamate, mossy fibers have been shown to release adenosine, dynorphin, and zinc (1); other neuropeptides also are present within the mossy fibers (1). These substances can modulate mossy fiber signaling through pre- and postsynaptic receptors and transporters.

The complexity of this pathway was further confirmed with growing evidence in the 1980s and 1990s that mossy fibers contained, in addition to glutamate, γ-aminobutyric acid (GABA).

Anatomic and biochemical evidence for GABA in mossy fibers

In 1983, with the advent of immunocytochemical techniques for visualizing GABA and glutamate in brain tissue (4), came the observation with light microscopy that GABA-like immunoreactivity was present in mossy fibers and the mossy fiber boutons (5). Electron microscopy with immunocytochemistry established intense GABA-like immunoreactivity within mossy fiber terminals at the ultrastructural level (6). With serial sections, the mossy fibers were shown to be immunoreactive for both glutamate and GABA. A consistent finding in inhibitory neurons is the presence of glutamate decarboxylase (GAD), the major GABA-synthesizing enzyme. Is GAD present in mossy fibers? Early studies of GAD messenger RNA (mRNA), and GAD immunoreactivity at the ultrastructural level using electron microscopy suggested not (7,8). Close scrutiny of the figures in these articles, however, reveals faint staining, and the methods used may not have had the necessary sensitivity to detect low levels of this enzyme or its mRNA. Indeed, low levels of GAD67 mRNA were subsequently found in the granule cell layer (9). Sloviter et al. (1996) later provided overwhelming evidence of GABA and GAD within mossy fiber terminals (10). They found GABA- and GAD67-like immunoreactivity within dentate granule cells and mossy fibers. These findings using light microscopy were confirmed with electron microscopy and in situ hybridization, which demonstrated induction of GAD65 and GAD67 mRNA.

Further biochemical evidence of releasable GABA within mossy fiber boutons came from synaptosomes (11). Potassium-evoked release of GABA and glutamate occurred in a synaptosomal fraction that consisted of mossy fiber boutons. This same fraction in rats that had been irradiated at a neonatal stage to destroy granule cells and mossy fibers showed a significant decrease in both glutamate and GABA release (11).

Thus there has been substantial anatomic and biochemical evidence for the coexistence of GABA and glutamate in mossy fiber terminals. A further element, necessary for fast ionotropic GABAergic transmission (although not metabotropic signaling), is the presence of postsynaptic GABAA receptors. At least one GABAA-receptor subunit (γ3) is relatively selectively expressed in the mossy fiber termination zone (stratum lucidum) of the CA3 hippocampal subfield (12). This issue has not, however, been addressed at an ultrastructural level with antibodies against different subunits.

Electrophysiologic evidence

Ultimately the most direct evidence for GABAergic transmission would be an electrophysiologic demonstration that inhibitory postsynaptic currents (IPSCs) can be elicited at mossy fiber synapses. An unambiguous demonstration that GABAergic IPSCs occur at mossy fiber synapses would require paired recordings from identified pairs of granule cells and postsynaptic neurons. It has been estimated, however, that a single granule cell has less than a 0.05% probability of making a synaptic contact with an individual CA3 pyramidal neuron (13). We thus turned to an indirect approach to provide compelling evidence that there is a mossy fiber GABAergic signal. Our approach was to show that we could elicit a monosynaptic GABAA-receptor mediated signal from dentate to CA3 that fulfilled the physiologic and pharmacologic criteria for mossy fiber-to-CA3 pyramidal cell neurotransmission (Fig. 1)(14).

Figure 1.

Diagram showing physiologic and pharmacologic criteria for mossy fiber-to-CA3 glutamatergic signaling in guinea pig, including inhibition by group III metabotropic glutamate receptor agonist, L-AP4;frequency-dependent facilitation; and N -methyl- d -aspartate (NMDA) receptor–independent long-term potentiation (LTP). If γ-aminobutyric acid (GABA) also is released from mossy fibers, then it would be expected that the GABAergic signal would fulfil the same criteria (see text). Released GABA may act not only on postsynaptic GABA A receptors, but also on presynaptic GABA B receptors to inhibit neurotransmitter release.


Hippocampal slices (450 μm thick) were obtained from guinea pigs (3–5 weeks) and were maintained at room temperature in artificial cerebrospinal fluid (aCSF) containing (in mM) NaCl (119), KCl (2.5), MgCl2(4), CaCl2(4), NaHCO3 (26.2), NaH2PO4(1), and glucose (11), gassed with 95% O2/5% CO2. We placed bipolar stainless-steel stimulating electrodes in stratum granulosum of the dentate gyrus to recruit mossy fibers and in stratum radiatum to recruit local interneurons as shown in Fig. 2A (20 μs, 20–80 V). Whole-cell voltage-clamp recordings at −60 mV were made from CA3 pyramidal cells by using pipettes (3–5 MΩ) containing (in mM) CsCl (135), HEPES (10), NaCl (8), EGTA (2), MgCl2 (0.2), Mg adenosine triphosphate (ATP) (2), guanosine triphosphate (GTP) (0.3), QX314 Br (5), at pH 7.2, osmolarity corrected to 295 mOsm. Series resistances were ≤20 MΩ and were not routinely compensated. Minimal stimulation was delivered via a monopolar glass electrode (1 MΩ resistance, 50-μs pulse width) filled with a CSF, advanced into stratum granulosum as described by Jonas et al. (15). Drugs were obtained from Tocris Cookson or Sigma. We initially identified mossy fiber projections in guinea-pig hippocampal slices by recording field EPSPs with an extracellular electrode positioned in stratum lucidum. To minimize the risk of recruitment of interneurons, we stimulated granule cells with an electrode positioned in stratum granulosum of the dentate gyrus, rather than in stratum lucidum (Fig. 2A). Mossy fiber–mediated field EPSPs were identified by marked frequency facilitation on switching the stimulation frequency from 0.05 Hz to 1 Hz (3). Once this criterion was met, we obtained a whole-cell voltage-clamp recording from a CA3 pyramidal neuron close to the original recording site with a pipette containing a CsCl-based solution.

Figure 2.

Dentate-to-CA3 evoked γ-aminobutyric acid (GABA)ergic signals show the physiologic and pharmacologic characteristics of mossy fibers. A: Position of stimulating electrodes: stratum granulosum of dentate gyrus and stratum radiatum (s. rad.) of CA3. B: A residual dentate-evoked postsynaptic current (PSC; later shown to be GABA A -receptor mediated) remains after blockade of ionotropic glutamate receptors (results from one neuron). C: 10 μ M L-AP4 depresses stratum granulosum-evoked IPSC. D: Inhibitory postsynaptic currents (IPSCs) evoked by stratum granulosum (but not stratum radiatum) stimulation exhibit frequency-dependent facilitation as frequency was increased from 0.05 to I Hz at arrow. E: Mean PSC amplitudes evoked by dentate (n = 9: solid circles) or stratum radiatum (n = 4: open circles) stimulation. Both pathways were tetanized at time 0 (2 × 100 Hz, 1-s, 10-s interval) in 10 μ M NBQX and 100 μ M APV, resulting in long-term potentiation (LTP) of dentate-evoked PSCs, but not stratum radiatum-evoked PSCs. PSCs evoked by both stimuli were subsequently abolished by picrotoxin (100 μ M ). Calibration bars in B, 1 nA, 50 ms; in C, 100 pA, 50 ms; in D, 500 pA, 50 ms; in E, 400 pA, 50 ms. Traces are average of seven to 10 successive PSCs (adapted from ref. 14, with permission.)


We adjusted the intensity of the stratum granulosum stimulus to elicit a large inward postsynaptic current (PSC) when the CA3 neuron was held at −60 mV. This PSC contained both monosynaptic and polysynaptic components (Fig. 1B). We then blocked ionotropic glutamate receptors with the N-methyl-D-aspartate (NMDA) receptor antagonist dl-2-amino-5-phosphonovalerate (APV, 100 μM) and the AMPA/kainate-receptor antagonist 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo[f]quinoxaline-7-sulphonamide disodium (NBOX, 10 μM). This reduced the PSC by 95%, implying that the original PSC was mediated mainly by ionotropic glutamate receptors (Fig. 2B). The residual PSC could be blocked by picrotoxin or bicuculline, and reversed at the reversal potential for GABAA-receptor–mediated responses, and was thus a GABAA-receptor–mediated IPSC (14). The presence of ionotropic glutamate-receptor blockade with 100 μM APV and 10 μM NBQX prevents recruitment of local interneurons (15), and thus this IPSC is monosynaptic. That it comes from mossy fibers is supported by the finding that its latency was no different from the latency of the synaptic current elicited in the absence of glutamate-receptor antagonists (14). This finding, however, does not guarantee that the responses are mediated by mossy fibers, because electrical stimuli in the dentate gyrus might also recruit axon collaterals of GABAergic interneurons, some of which have been shown to project into the granule cell layer (16,17).

A more compelling test that this GABAergic IPSC arises from mossy fibers would be to show that it has several highly unusual characteristics of mossy fiber synapses, established from recordings of glutamatergic excitatory postsynaptic currents (EPSCs) or potentials (Fig. 1). The conventional “signature” of CA3-mossy fiber synapses is threefold: profound sensitivity to group II (18), and in guinea pig, group III (19,20) metabotropic glutamate-receptor agonists; marked facilitation in response to modest increases in stimulation frequency (3); and NMDA receptor–independent long-term potentiation (LTP) (21,22). One or more of these phenomena have frequently been used as criteria to identify mossy fiber EPSPs and EPSCs.

The use of guinea-pig hippocampal slices permitted us to exploit the high sensitivity of mossy fiber synapses to low concentrations of the group III metabotropic glutamate agonist l(+)-2-amino-4-phosphonobutyric acid (L-AP4)(19), which is not seen in rat (20). In the continued presence of NBQX and APV, L-AP4 (10 μM) significantly decreased the IPSC amplitude (Fig. 2C). The high sensitivity of the residual PSCs to L-AP4 is in marked contrast to the lack of effect of higher concentrations on interneuron-mediated synaptic signals in pyramidal neurons (15,23). We re-examined the effect of L-AP4 on interneuron–CA3 pyramidal neuron synapses by positioning a second stimulating electrode in stratum radiatum (Fig. 2A). L-AP4 (10 μM) had no effect on IPSCs elicited via this electrode (data not shown) (14). L-AP4 thus selectively depresses dentate stratum granulosum–evoked PSCs, compatible with a mossy fiber origin.

In the same experiments, we examined the frequency dependence of dentate stratum granulosum–evoked PSCs, to establish whether these responses show the pronounced facilitation with modest increases in frequency exhibited by mossy fiber signals (3). Increasing the stimulation frequency from 0.05 to 1 Hz caused marked frequency facilitation (Fig. 2D)(14). The average size of the frequency-dependent facilitation met the criterion initially used to identify mossy fibers. Again, because little is known about the frequency dependence of nonglutamatergic transmission over similar stimulation rates, we delivered the same increment in frequency via the second electrode in CA3 stratum radiatum. Stratum radiatum–elicited IPSC amplitudes were halved (Fig. 2D)(14). An unusual feature of mossy fiber synapses to CA3 pyramidal cells is that they exhibit NMDA receptor–independent LTP, which is expressed presynaptically (21,22). We therefore examined the effect of tetanization on the GABAergic IPSCs evoked by electric stimulation in stratum granulosum of the dentate gyrus in the presence of NBQX (10 μM) and APV (100 μM). Two high-frequency tetani (100 Hz, 1 s, 10-s interval) were followed by a very pronounced initial potentiation, which gradually decayed to reach a plateau of 188 ± 35%(Fig. 2E). This plateau was accompanied by a decrease in the coefficient of variation consistent with a presynaptic site of expression, as has been reported for mossy fiber LTP of glutamatergic signals. We were able to mimic and occlude this LTP by application of the adenylate cyclase activator forskolin, consistent with mossy fiber LTP (24).

Thus dentate stimulus–evoked GABAergic IPSCs show high sensitivity to L-AP4, pronounced frequency-dependent facilitation, and NMDA receptor-independent LTP, which is characteristic of LTP of the mossy fiber EPSC (Fig. 1). These results are consistent with the release of GABA from mossy fibers and were in direct contrast to results obtained through stimulation in stratum radiatum.

Are GABA and glutamate co-released?

In our experiments, the initial PSC elicited in the absence of glutamate blockers showed relatively little trial-to-trial fluctuation (CV, <20%), whereas the GABAergic IPSCs recorded in APV and NBQX was accompanied by intermittent failures of transmission. This implies that the quantal content of dental-evoked GABAergic IPSCs was smaller than that of the mainly glutamatergic signal, and could be explained by either a subset of mossy fibers releasing GABA, GABA release having a lower probability than glutamate release from individual terminals, or lack of postsynaptic GABAA receptors at some terminals. To investigate these possibilities further, we used minimal stimulation. Unitary granule cell–evoked responses can be elicited by minimal stimulation in stratum granulosum (25). With this technique, we obtained all-or-none GABAergic responses in DL-APV (100 μM) and NBQX (10 μM) that appeared abruptly with a failure rate of 10–30%, at stimulation intensities similar to those required for glutamatergic responses recorded in the same conditions (Fig. 3A). This failure rate is similar to that reported for the glutamatergic response (25) and further supports the hypothesis that the GABAergic response is monosynaptic. L-AP4 (10 μM) depressed the minimal stimulation-evoked IPSCs (Fig. 3A), further confirming a mossy fiber origin. The similar failure rates for IPSCs and EPSCs evoked by minimal stimulation imply that the release probability from single-fiber stimulation does not differ for the GABAergic or glutamatergic response. The increased failure rate of multifiber evoked IPSCs compared with EPSCs is thus consistent with the hypothesis that GABA is released by only a subset of mossy fibers.

Figure 3.

Minimal stimulation evokes unitary γ-aminobutyric acid (GABA)ergic postsynaptic currents (PSCs) in CA3 pyramidal neurons that respond to L-AP4 and have the same threshold as glutamatergic PSCs. A: Stimulation (50-μs pulse width) via a monopolar glass electrode (1 MΩ) positioned in stratum granulosum (Jonas et al., 1993) in the presence of NBQX (10 μ M ) and APV (100 μ M ) elicits a response in a CA3 pyramidal cell with an abrupt stimulus threshold (left). The plot shows mean ± SEM current amplitude (including failures) averaged at each stimulation voltage. The failure rate decreased from 100% at 20–30 V to 27% at 40–60 V. The response was reversibly depressed by 10 min of L-AP4 (10 μ M ) application (right). Traces are averages of 10 successive PSCs. B: Minimal stimulation via a monopolar glass electrode (3 MΩ; 25-μs pulse width) in stratum lucidum elicits unitary GABAergic responses in 5 m M kynurenic acid that have the same threshold as unitary glutamatergic responses in 10 μ M bicuculline and 100 μ M DL-APV. C: In the same cell, in the presence of DL-APV (100 μ M ) at a holding potential of −40 mV, stimulation at 40 V elicits failures and inward (glutamatergic), outward (GABAergic), and biphasic (dual-component) PSCs. Calibration bars in A, 10 pA, 50 ms; in C, 5 pA, 50 ms (adapted from ref. 14, with permission.)

This now begs the question whether both glutamate and GABA are released from the same terminals, and if so, whether they are packaged in the same vesicles, as has been argued for glycine and GABA at some spinal interneuron synapses (26). We applied minimal stimulation in stratum lucidum, with 5 mM kynurenic acid used to block ionotropic glutamate receptors. We then adjusted the position of the stimulating electrode to elicit a GABAergic response with an abrupt stimulus threshold and plateau consistent with single-fiber stimulation (Fig. 3B). Once such a response was obtained, we washed out kynurenic acid, but left NMDA receptors blocked with DL-APV (100 μM). We then held the cell at a membrane potential between the reversal potentials for GABAA and AMPA receptors. In each of four cells, the responses fluctuated between failures, and inward (glutamatergic), outward (GABAergic), and biphasic (dual-component glutamatergic and GABAergic) synaptic currents, with relative frequencies suggestive of independent release of the two transmitters (Fig. 3C). Washing in bicuculline (10 μM) abolished the outward PSCs, confirming that they were GABAergic, and left inward (glutamatergic) PSCs. These pure glutamatergic PSCs had the same latency and the same threshold as the GABAergic PSCs, consistent with an identical presynaptic origin (14). These results imply that a single mossy fiber can release both GABA and glutamate, albeit from distinct populations of vesicles.


What is the physiologic and pathologic relevance of mossy fiber GABA release?

The physiologic role of GABA release from mossy fibers is largely a matter for speculation. GABA release could have developmental importance in establishing glutamatergic synapses (27). Another possibility is that the postsynaptic GABAA receptor–mediated signals are of little quantitative importance, and that the major target of GABA is actually presynaptic GABAB receptors (28,29). This could contribute to synapse autoregulation, with GABAB-receptor activation inhibiting release of both glutamate and GABA (Fig. 1).

Importantly, the balance of glutamatergic and GABAergic signaling at mossy fiber synapses is under dynamic control. Independent regulation of the two signals thus provides an additional degree of freedom in the transmission of information from the dentate gyrus to the hippocampus. After seizures or prolonged stimulation, GAD mRNA expression, GAD protein expression, and the GABA content of dentate granule cells and mossy fibers increase (9,10,30). A further mechanism by which the mossy fiber GABAergic signal could be increased is by an increase in the affinity or number of postsynaptic GABAA receptors. This has been observed after kainic acid–induced limbic seizure, after which there is a striking upregulation of the γ3 subunit of the GABAA receptor restricted to stratum lucidum (31).

Although there are many anatomic data to support the induction of mossy fiber GABA with seizures, it is only recently that physiologic corroboration has been forthcoming (32,33). After kindling, there is a monosynaptic GABAA receptor–mediated potential in CA3 pyramidal cells from stimulation in the dentate granule cell layer that was not detectable in nonkindled animals (Fig. 4)(33). That this is due to mossy fiber GABA release rather than a function of synaptic rearrangement is supported by the findings that (a) this monosynaptic IPSP wanes with time after the last stimulation in the kindled animals (Fig. 4), only to reappear if the kindled animals are restimulated (32,33), and (b) a similar effect can be seen in animals immediately after a pentylenetetrazol seizure at a time when synaptic rearrangement is unlikely to have taken place (32). In neither study is the possibility that these GABAergic potentials are due to stimulation of interneurons rather than mossy fibers completely excluded, but the results would be in keeping with seizure-induced augmentation of a mossy fiber–mediated GABAergic signal. So what is the role of an augmentation of the mossy fiber GABAergic signal after seizures? It is possible that it has minimal or no effect, and that its presence is an epiphenomenon from seizure-induced expression of early genes. Alternatively, it could play a compensatory role. After prolonged seizures, inhibition of principal cells is decreased because of inadequate recruitment and loss of interneurons (34). Upregulation of a proximal GABAergic input could inhibit action-potential initiation, and would be necessary to inhibit the excitatory mossy fiber input (35,36). Alternatively, it is possible that the upregulation of the GABAergic signal could have a proepileptogenic effect and could underlie the development of hyperexcitability after seizures. Because the major output of mossy fibers is onto inhibitory interneurons (2), then decreasing the effectiveness of mossy fiber transmission could decrease the recruitment of interneurons and so could have a disinhibitory effect. The upregulation of the GABAergic signal is a transient phenomenon, so it is possible that it is responsible for postictal amnesia and memory problems (33).

Figure 4.

Synaptic responses of CA3 pyramidal cells to mossy fiber stimulation in control conditions, and 24 h (freshly kindled) and 1 month after the last generalized convulsive seizure before (upper row) and during perfusion with NBQX + APV (lower row) at resting membrane potential. Note that application of NBQX + APV blocks both excitatory postsynaptic potentials (EPSPs) and inhibitory PSPs (IPSPs) in control preparations, but a fast IPSP remains in freshly kindled preparations. A month after the last kindled seizure, perfusion with NBQX + APV blocked the EPSPs and IPSPs. Traces are averages of six responses. Calibration bars apply to all traces (from ref. 33, with permission.)

Further work is thus needed to elucidate both the physiologic and pathologic role of this mossy fiber–mediated GABAergic signal.

Acknowledgment: This study was supported by the MRC, the Wellcome Trust, and the Brain Research Trust.