Main determinants of presynaptic Ca2+ dynamics at individual mossy fiber-CA3 pyramidal cell synapses.

Synaptic transmission between hippocampal mossy fibers (MFs) and CA3 pyramidal cells exhibits remarkable use-dependent plasticity. The underlying presynaptic mechanisms, however, remain poorly understood. Here, we have used fluorescent Ca2+ indicators Fluo-4, Fluo-5F, and Oregon Green BAPTA-1 to investigate Ca2+ dynamics in individual giant MF boutons (MFBs) in area CA3 traced from the somata of granule cells held in whole-cell mode. In an individual MFB, a single action potential induces a brief peak of free Ca2+ (estimated in the range of 8-9 microm) followed by an elevation to approximately 320 nm, which slowly decays to its resting level of approximately 110 nm. Changes in the somatic membrane potential influence presynaptic Ca2+ entry at proximal MFBs in the hilus. This influence decays with distance along the axon, with a length constant of approximately 200 microm. In giant MFBs in CA3, progressive saturation of endogenous Ca2+ buffers during repetitive spiking amplifies rapid Ca2+ peaks and the residual Ca2+ severalfold, suggesting a causal link to synaptic facilitation. We find that internal Ca2+ stores contribute to maintaining the low resting Ca2+ providing approximately 22% of the buffering/extrusion capacity of giant MFBs. Rapid Ca2+ release from stores represents up to 20% of the presynaptic Ca2+ transient evoked by a brief train of action potentials. The results identify the main components of presynaptic Ca2+ dynamics at this important cortical synapse.


Introduction
Mechanisms of Ca 2ϩ -dependent neurotransmitter release are central to understanding synaptic function. Studies in giant calyceal terminals in the brainstem have provided critical insights into the underlying molecular machinery (Schneggenburger et al., 2002;von Gersdorff and Borst, 2002;Meinrenken et al., 2003). Similar advances with regard to cortical synapses have been made possible in large part by single-synapse imaging in the neocortex (Koester and Sakmann, 2000;Koester and Johnston, 2005) and whole-terminal recordings from giant mossy fiber (MF) boutons (MFBs) in the hippocampus (Geiger and Jonas, 2000;Engel and Jonas, 2005). Synapses formed by giant MFBs on CA3 pyramidal cells have also been intensely studied because they exhibit unusual short-term facilitation (Griffith, 1990) and an NMDA receptor-independent form of long-term potentiation (Zalutsky and Nicoll, 1990).
Several candidate mechanisms of short-term facilitation at these synapses have been proposed. Recordings from giant MFBs show that action potentials (APs) broaden during repetitive spiking, thus boosting presynaptic Ca 2ϩ influx (Geiger and Jonas, 2000;Bischofberger et al., 2002). However, at moderate frequencies (10 -30 Hz), this enhancement, in terms of either Ca 2ϩ cur-rent amplitude or transferred charge, amounts to only Ͻ10% during the first 10 spikes (Geiger and Jonas, 2000). A 10% increase in presynaptic Ca 2ϩ entry is expected to augment postsynaptic responses in CA3 pyramidal cells less than to the second power (Geiger and Jonas, 2000;Blatow et al., 2003;Mori-Kawakami et al., 2003) or by Ͻ21%. AP broadening alone thus cannot fully account for the profound synaptic facilitation routinely observed at similar frequencies.
Another candidate mechanism involves progressive saturation of endogenous Ca 2ϩ buffers (Rozov et al., 2001). In young rodents, increasing the MF buffering capacity with Ca 2ϩ chelators EGTA (high concentration) or BAPTA reduces short-term facilitation or paired-pulse ratios of postsynaptic responses (Regehr et al., 1994;Blatow et al., 2003;Mori-Kawakami et al., 2003). However, the effect is small in mature animals (Mori-Kawakami et al., 2003) and the residual presynaptic Ca 2ϩ facilitates faster than do synaptic responses (Regehr et al., 1994). Furthermore, single-cell measurements of the main MF endogenous Ca 2ϩ buffer calbindin-28K suggest that buffer saturation cannot explain strong facilitation of presynaptic Ca 2ϩ (Muller et al., 2005).
In summary, the machinery of presynaptic Ca 2ϩ signaling at MF synapses remains poorly understood. Because optical recordings from tissue-loaded MFs do not report resting Ca 2ϩ levels supplemental material). In the second test, we measured the baseline fluorescence in the axonal regions at different distances from the soma and found no spatial gradient (supplemental Fig. 4, available at www. jneurosci.org as supplemental material). The results of both tests argued against any significant loss of fluorescence indicators from the axon.
Fluorescence responses were recorded in line-scan mode at 500 Hz (500 or 1000 ms sweeps; intersweep interval, 30 s or 1 min) and stored for off-line analysis. The Ca 2ϩ -dependent fluorescence response ⌬F/F (integrated over the visible MFB width) was routinely calculated as follows: (F post Ϫ F pre )/(F pre Ϫ F 0 ). The values of F pre and F post stand for the line-scan fluorescence averaged over, respectively, 100 ms before the first spike and either 50 ms in the case of single-response amplitude measurements or 250 ms in the case of five-response amplitude measurements (20 Hz train of five APs) after the first spike onset. F 0 denotes the background fluorescence measured outside any cell structures filled with the indicator. Because special care was taken to avoid escape of the indicator from the pipette, and because the site of imaging was hundreds of micrometers away from the pipette tip, F 0 was likely to represent the photomultiplier tube dark current. Image analyses were performed on stacks of stored line-scan images using a set of custom NIH Image macros. False color tables and averaged images were used for illustration purposes but the original (gray level) pixel brightness values in each line-scan image were used for the quantitative analysis. In most experiments, we reconstructed the axon trajectory using a collage of high-resolution Kalmanfiltered z-stacks 15-20 m deep. In total, we obtained full reconstruction of 43 axons, with an average distance between the recorded MFB and the soma of 686 Ϯ 38 m. Throughout the experiments, we observed no failures of spike-driven Ca 2ϩ signals propagating along the main axonal trunk including giant MFBs. This, however, does not rule out the possibility that propagation could fail at higher spiking frequencies and/or in thin axon collaterals.
The two-photon excitation probability profile is proportional to the squared illumination light intensity I 2p 2 (Zipfel and Webb, 2001): where the canonical coordinates u ϭ 4ksin 2 (␣/2)z and v ϭ ksin(␣)r represent the axial distance z; the radial distance r; the numerical aperture of the objective, NA ϭ sin(␣) ϭ 0.9; and wave number k ϭ n(2/) (n ϭ 1.33 is the medium refraction index and ϭ 810 nm is the wavelength); J 0 denotes zero-order Bessel's function of the first kind. This theoretical function, however, represents the lower-limit estimate of the excitation profile: in reality, optical aberrations and imperfect alignment of the experimental optical system are likely to increase the spread of excitation. Similar considerations apply to the emission path. We therefore obtained an estimate of the excitation-emission profile by recording the pointspread function (PSF) of the system using 0.17 m fluorescent beads (PS-Speck Microscope Point Source kit; Invitrogen) as illustrated below.
Estimating the resting concentration of presynaptic Ca 2ϩ . Classically, the resting Ca 2ϩ level ([Ca 2ϩ ] rest ) can be estimated from the following relationship: where K d is the Ca 2ϩ indicator dissociation constant, and F max , F, and F min represent maximum, resting, and Ca 2ϩ -independent fluorescence of the indicator. However, the high dynamic range of Ca 2ϩ indicators such as Fluo-4 implies that the resting fluorescence F could be comparable with the detection threshold of the system. In this case, the useful fluorescence signal could blend with the background noise in a nonlinear manner, which makes it difficult to obtain reliable ratios ⌬F/F or F max /F. This difficulty could be addressed by relating ⌬F to the Ca 2ϩ -independent fluorescence R recorded in the "red" (e.g., Alexa) channel . We have modified this approach by diverting part of the Alexa Fluor 594 emission signal into the "green" Fluo-4 channel using a 560 nm dichroic mirror and a 700 nm short-pass filter. The use of the cross talk Alexa Fluor 594 signal, denoted here as R ct , was advantageous because this (1) elevated emission in the Fluo-4 channel well above the detection threshold of the system (see above) and (2) allowed both the Ca 2ϩdependent signal ⌬F and R ct to be recorded through the same channel. (This method, however, is unlikely to be advantageous when the useful resting green fluorescence is substantially above the background noise.) The average ratio ͗R ct /R͘ was measured in separate experiments, by comparing emission in the two corresponding channels while imaging giant MFBs loaded with Alexa Fluor 594 only (see Results). In these terms, the recorded resting and maximum fluorescence (F and F max , respectively) will include a Ca 2ϩ -indepedent fluorescence fraction R ct ϩ F min . Equation 2 can be therefore recast as follows: In the case of the high dynamic range indicator Fluo-4, F min /F max Ϸ 0 (Maravall et al., 2000;Oertner et al., 2002), implying that estimation of [Ca 2ϩ ] rest simply requires experimental measurements of R ct and R, according to Equation 3. In the case of the lower dynamic range indicator OGB-1 (K d Ϸ 200 nM), this equation also includes the known value of F max /F min Ϸ 6 (Maravall et al., 2000;Oertner et al., 2002;Jackson and Redman, 2003). Nonstationary presynaptic Ca 2ϩ kinetics. To translate the recorded fluorescence into the underlying Ca 2ϩ kinetics, we used a singlecompartment, multicomponent kinetic model (Helmchen et al., 1997;Neher, 1998;Jackson and Redman, 2003) in its nonstationary form, which requires no steady-state approximations (Sabatini and Regehr, 1998;Rusakov et al., 2005). Multicompartmental modeling was deemed unfeasible in the present study because (1) the Ca 2ϩ fluorescence time course was indistinguishable between subregions inside a single bouton (the PSF of the system was too large to distinguish between such subregions) (see Fig. 4) and (2) Ca 2ϩ exchange between multiple space partitions has little effect on the volume-average Ca 2ϩ concentration transient (Sabatini and Regehr, 1998). Parameters of the model were constrained and cross-validated by (1) an independent estimate of [Ca 2ϩ ] rest (see above), (2) an independent estimate of the endogenous buffer (calbindin-28K) concentration (Jackson and Redman, 2003;Muller et al., 2005), and (3) comparing the effects of different configurations of Ca 2ϩ indicators/buffers on the recorded fluorescence signal versus the predicted kinetics.
The finite-difference kinetic scheme explicitly included the rate of Ca 2ϩ entry, j Ca , the binding-unbinding reactions with the endogenous buffer (calbindin-28K) B and Ca 2ϩ indicator I, and the Ca 2ϩ removal rate P, according to the following equation: where brackets denote concentrations, CaI and CaB stand for the Cabound indicator and buffer, respectively, and the kinetic constants (top indices denote buffer B or indicator I; bottom indices indicate on and off constants) are summarized in Table 1. The mass conservation rules are as follows: ͓I͔ tot ϭ ͓CaI͔ ϩ ͓I͔, ͓B͔ tot ϭ ͓CaB͔ ϩ ͓B͔, and ͓Ca͔ tot ϭ ͓Ca 2ϩ ͔ ϩ ͓CaB͔ ϩ ͓CaI͔, (5) where index "tot" denotes total amount. The AP-evoked Ca 2ϩ influx rate j Ca followed the Gaussian (Sabatini and Regehr, 1998;Rusakov et al., 2004) as follows: in line with experimental observations (Meinrenken et al., 2003), with the MFB spike halfwidth Ϸ 0.5 ms (Bischofberger et al., 2002), onset at t 0 , and the time integral of ⌬[Ca 2ϩ ] tot reflecting total Ca 2ϩ entry. Removal of Ca 2ϩ , which is a complex process involving diffusion, pumping out and/or sequestration was approximated by a first-order reaction (Jackson and Redman, 2003) (however, for discussion, see Matveev et al., 2004;Holcman et al., 2005) at the rate P([Ca 2ϩ ] Ϫ [Ca 2ϩ ] rest ) (Eq. 4). The model, therefore, operated with only two adjustable (free) parameters: ⌬[Ca 2ϩ ] tot and P. However, varying either ⌬[Ca 2ϩ ] tot or P had virtually independent effects on the calculated amplitude (⌬F/F ) and decay of fluorescent responses, respectively (see Results) (supplemental Fig. 5, available at www.jneurosci.org as supplemental material). Each of the two parameters could be therefore constrained by a straightforward fitting procedure that would match the calculated and the experimental fluorescence, as discussed previously (Rusakov et al., 2005).
Assessing calbindin-28K washout from the axon. Gradual washout of endogenous buffers in whole-cell mode could alter presynaptic Ca 2ϩ dynamics in the recorded cell (Rozov et al., 2001;Blatow et al., 2003). A previous study has estimated that washout of the endogenous Ca 2ϩ buffer calbindin-28K from the granule cell soma occurs with a time constant of ϳ9.7 min (Muller et al., 2005). To assess its washout rate at different regions of the axon, we used a multicompartmental diffusion model based on a detailed quantitative study of granule cell morphology . The modeled cell included the 19-m-long/10m-wide elliptical soma and a 2000-m-long/0.4-m-thick axon. The average dendritic tree of granule cells (two primary dendrites, each ϳ2.5 m in diameter, giving rise to an arbor spreading over ϳ300 m with the total dendritic length of 3221 m) ) was represented by a single 300-m-long trunk of the matching cross section (9.6 m 2 ) and total volume (2900 m 3 ). The patch pipette (tip diameter, 1.8 m) provided a concentration clamp source linked to a somatic compartment. The initial calbindin concentration was 160 M, and at t ϭ 0, the concentration clamp was imposed through the pipette. Diffusion simulations were performed with the compartment size of 5 m (464 compartments in total) using numerical algorithms described previously (Scimemi et al., 2004). The single unknown parameter, the effective diffusivity of calbindin was adjusted to obtain the best fit between the experimental data on somatic washout (Muller et al., 2005) and the computed calbindin concentration time course in the soma (supplemental Fig. 6, available at www.jneurosci.org as supplemental material). Under these conditions, washout of calbindin from the axonal regions where giant MFBs occur (the average distance from the soma, 686 Ϯ 38 m; n ϭ 43 fully reconstructed axons) is predicted in the range of Ͻ5% within 2-3 h after break-in (supplemental Fig. 6, available at www.jneurosci.org as supplemental material). This is consistent with the stable amplitude of recorded ⌬F/F during that period (supplemental Fig. 2, available at www.jneurosci.org as supplemental material).
Average data are shown as mean Ϯ SEM; a t test was used for statistics.
To reduce the chances of cutting granule cell axons close to the soma, hippocampal slices were sectioned at an angle (close to the parasagittal direction) to the plane of the main MF projections ( Fig. 1 A, B). We then loaded granule cells, in whole-cell mode, with two fluorescence indicators (see Materials and Methods) and monitored APs evoked antidromically in the soma by extracellular stimulation of the stratum lucidum. Although virtually all monitored cells generated APs, only a small proportion (Ͻ10%) showed an AP latency exceeding 3-4 ms in response to the minimum strength stimuli sufficient to generate an AP (Fig. 1C). On subsequent visual inspection in a microscope, only these cells projected their axon into area CA3 (Fig. 1 D) (Alexa Fluor 594 channel); the axons of cells with a short AP latency were truncated at Ͻ200 m from the soma. These observations suggested a simple electrophysiological test to facilitate selection of cells with long axons before imaging. In the microscope, giant MFBs were identified in area CA3, Ͼ300 m from the soma, by their distinctive large size (5-8 m) and thin filopodial protrusions ( Fig. 1 D, E, image panels; supplemental Fig. 1, available at www.jneurosci. org as supplemental material). We used 2 ms somatic depolarizing command voltage pulses to evoke orthodromic APs. These were rapidly followed by fluorescence transients in giant MFBs (Fig. 1 F, G). After a period of equilibration lasting 1 h or more depending on the distance from the soma, both Ca 2ϩdependent and Ca 2ϩ -independent fluorescence in giant MFBs remained stable for several hours (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). In contrast, proximal MF boutons tended to increase their baseline fluorescence, as well as their evoked Ca 2ϩ responses, by the end of 30 -40 min recording sessions in similar conditions (Ruiz et al., 2003); this is consistent with partial washout of endogenous buffers at short distances from the soma (Blatow et al., 2003;Muller et al., 2005), as discussed in Materials and Methods (supplemental Fig. 6, available at www.jneurosci.org as supplemental material).
Neither the amplitude nor the decay time of the AP-evoked fluorescence transients varied systematically with the distance from the soma in n ϭ 43 giant MFBs with fully reconstructed axons ( Fig. 2 A, B). In approximately one-third of all experiments, we compared Ca 2ϩ responses evoked under voltage clamp (V h ϭ Ϫ80 mV) or current clamp (current adjusted to correspond to V m ϭ Ϫ80 mV). Because no consistent differences were found, the data from these experiments were pooled.

Electrotonic control of Ca 2؉ signaling in MFs
We have shown previously (Ruiz et al., 2003) that varying the somatic holding voltage modulates the AP-evoked presynaptic Ca 2ϩ transient up to 40% in MF axonal boutons in the hilus (Fig.  2C). We found, however, that Ca 2ϩ signals at giant MFBs were unaffected by changing the holding voltage ( Fig. 2 D) (n ϭ 8). To determine the extent of the somatic electrotonic influence in our conditions, we repeated these experiments in proximal axonal boutons (Fig. 2C). Changing the somatic voltage between Ϫ110 and Ϫ60 mV resulted in a Ca 2ϩ signal variation, cast in relative terms as var V (⌬F/F ) ϭ (Max ⌬F/F Ϫ Min ⌬F/F )/Max ⌬F/F (Fig.  2 D), of 19 Ϯ 3% (n ϭ 11) (Fig. 2 E). Surprisingly, this influence was only one-half of that detected in the previous study where var V (⌬F/F ) was ϳ40% (Ruiz et al., 2003).
This discrepancy, however, has a simple explanation: the previous study dealt with axons cut, on average, at 100 -150 m from the soma, whereas the present experiments were performed on cells with intact axons. According to cable theory, cutting and sealing the axon increases its electrotonic space constant, thus facilitating remote voltage control. [Although open-ended axons would have a different effect on the length constant, they are unlikely to be compatible with cell survival; in addition, sealed axonal ends, with no detectable escape of fluorescence, were routinely observed in a microscope and the intracellular level of Al- exa Fluor 594 was unchanged throughout the experiment (supplemental Fig. 2, available at www.jneurosci.org as supplemental material).] Indeed, when we recorded from proximal boutons in cells that had their axons cut off 100 -150 m from the soma (n ϭ 7), the values of var V (⌬F/F ) were indistinguishable from the previous results (Ruiz et al., 2003); combining the current and previous data in the cut axons gave an average var V (⌬F/F ) of 39 Ϯ 8% (n ϭ 22) (Fig. 2 E).
We used the observed twofold difference in var V (⌬F/F ) between cut and uncut axons to estimate the electrotonic space constant of the axon. In an equivalent cable approximation, the steadystate voltage V at a distance x from a point clamped at V 0 is given by the following relationship (for discussion, see Jackson, 1992): where X and L represent distance x and the axon length l, respectively, cast in terms of the space constant : X ϭ x/ and L ϭ l/ (Fig. 2 F). This relationship predicts that, to reduce the extent of somatic voltage control approximately twofold, one has to cut the long axon at L ϳ 0.7 (Fig. 2 F, arrows). Because the actual MF axons were cut at L Ϸ 100 -150 m in these experiments, the estimated MF axon space constant was ϭ L/0.7 ϭ 150 -200 m. This value is consistent with electrotonic isolation of Ca 2ϩ transients in the giant MFBs (Fig. 2 D), also indicating that MF synapses in the hilus could be affected by the granule cell resting somatic voltage.

Resting Ca 2؉ concentration at giant MFBs
To estimate [Ca 2ϩ ] rest from Equation 3, we first measured the cross-talk fraction of Alexa Fluor 594 fluorescence in the Fluo-4 (Ca 2ϩ signal) emission channel, R ct /R. Granule cells were loaded with 20 M Alexa Fluor 594, and MFBs were imaged in both channels (Fig. 3 A, B). Comparing direct versus cross-talk signals of Alexa Fluor 594 indicated an average R ct /R ratio of 0.035 Ϯ 0.002 (n ϭ 24) (Fig. 3C). We then loaded cells with 200 M Fluo-4 (in addition to 20 M Alexa) and, after indicator equilibration (supplemental Fig. 2, available at www.jneurosci.org as supplemental material), recorded Ca 2ϩ fluorescence in giant MFBs in response to 50 APs at 20 Hz (Fig. 3D). The resulting fluorescence plateau (Fig. 3D) suggested that the indicator was close to saturation (Jackson and Redman, 2003;Rusakov et al., 2005).
To confirm that this is indeed the case, we compared 20 and 50 Hz trains of APs in seven giant MFBs and found no significant difference in the maximum fluorescence signal F max (difference, 7 Ϯ 5%; n ϭ 7); similar experiments with a lower-affinity indicator Fluo-5F (K d ϳ 1 M; trains of 75-150 APs) gave a qualitatively identical result (difference, 5 Ϯ 3%; n ϭ 7) ( Fig. 3E; supplemental Fig. 7, available at www.jneurosci.org as supplemental material). The saturating trains of APs in Fluo-4 therefore gave average F max /F and R/F ratios for Fluo-4 of 2.63 Ϯ 0.31 and 6.8 Ϯ 1.2, respectively (n ϭ 14). The estimates of [Ca 2ϩ ] rest in each individual giant MFB were obtained by substituting individual F max /F and R/F values in Equation 3; this gave an average [Ca 2ϩ ] rest of 116 Ϯ 20 nM (n ϭ 14) (Fig. 3F ).
An alternative estimate of [Ca 2ϩ ] rest was obtained using another high-affinity Ca 2ϩ indicator, Oregon Green BAPTA-1 (200 M; K d ϭ 205 nM) for which F max /F min ϳ 6 (Maravall et al., 2000;Jackson and Redman, 2003). With a similar protocol, these experiments gave an average F max /F ratio of 2.15 Ϯ 0.14, predicting [Ca 2ϩ ] rest at 103 Ϯ 14 nM (n ϭ 8) (Fig. 3F ). This was in good correspondence with the estimate based on Fluo-4.

Ca 2؉ kinetics at giant MFBs
To quantify the kinetics of presynaptic Ca 2ϩ transients triggered by APs, we routinely evoked five spikes at 20 Hz, a firing pattern compatible with that of individual granule cells in vivo (Henze et al., 2002). The fluorescence response was analyzed using a single-compartment model of Ca 2ϩ dynamics: this approach was relevant because the characteristic point-spread function of the imaging system was comparable with or larger than the characteristic cross section of giant MFBs (Fig. 4 A-C) (see Materials and Methods). To account for the nonstationary kinetics during and shortly after rapid Ca 2ϩ entry, we simulated the bindingunbinding reactions using an explicit finite-difference scheme (Sabatini and Regehr, 1998;Matveev et al., 2004;Rusakov et al., 2005).
The model included several independently estimated quantities: (1) [Ca 2ϩ ] rest Ϸ 110 nM; (2) a direct measure of the main endogenous buffer (calbindin-28K) concentration [B] tot ϭ 160 M (Muller et al., 2005), which is in line with an estimate derived from Ca 2ϩ imaging (Jackson and Redman, 2003); (3) the Ca 2ϩ entry halfduration of ϳ0.5 ms (Bischofberger et al., 2002); and (4) indicator concentration [I] clamped by the whole-cell pipette (see supplemental Fig. 2, available at www.jneurosci.org as supplemental material). The remaining two free parameters were the total Ca 2ϩ influx ⌬[Ca 2ϩ ] tot and the Ca 2ϩ removal rate P. Adjusting these, however, had virtually independent effects on, respectively, the amplitude and the decay constant of the evoked Ca 2ϩ -sensitive fluorescence transient: changing ⌬[Ca 2ϩ ] tot twofold produced a comparable change in the predicted ⌬F/F amplitude but a Ͻ5% change in the decay constant; conversely, changing the Ca 2ϩ removal constant P twofold produced a comparable change in the predicted decay constant but Ͻ5% alteration in the ⌬F/F amplitude (supplemental Fig. 5, available at www. jneurosci.org as supplemental material). In these conditions, fitting the computed fluorescence kinetics to the experimental ⌬F/F is relatively straightforward, as discussed previously (Rusakov et al., 2005). To cross-validate the unknowns of Ca 2ϩ dynamics under different buffer configurations, we conducted similar experiments at two different concentrations of  and also with a lower-affinity indicator Fluo-5F. In all three cases, optimization with the two parameters gave a very good fit, yielding consistent estimates of ⌬[Ca 2ϩ ] tot and P (Fig. 4 D-F ). This consistency further argues that the previous estimates of the endogenous Ca 2ϩ buffer concentration in the soma and proximal boutons (Jackson and Redman, 2003;Muller et al., 2005) are relevant for giant MFBs, as indeed expected from the long-term diffuse equilibration of soluble calbindin-28K. Together, the data thus predict ⌬[Ca 2ϩ ] tot ϭ 51 Ϯ 2 M and an average Ca 2ϩ removal rate, p ϭ 0.37 Ϯ 0.03 ms Ϫ1 .
According to morphological observations, the volume of giant MFBs is in the region of 20 -50 m 3 (Chicurel and Harris, 1992;Acsady et al., 1998), or 2-5 ϫ 10 Ϫ14 L. In electron micrographs, synaptic vesicles and mitochondria appear to occupy ϳ50% of the giant MFB profile (Fig. 4 B, C); this is likely to scale down twofold the volume available to Ca 2ϩ ions. The ϳ50 M step concentration increase (3 ϫ 10 19 ions/L) would then correspond to 3-8 ϫ 10 5 Ca 2ϩ ions per bouton. This is in excellent agreement with electrophysiological observations showing that 3.7 ϫ 10 5 Ca 2ϩ ions flow into a patched giant MFB following an AP (Bischofberger et al., 2002). The estimated Ca 2ϩ dynamics is also in correspondence with the unchanged Ca 2ϩ entry during short 20 Hz AP trains (Geiger and Jonas, 2000). Together, these data provide quantitative insights into the presynaptic Ca 2ϩ kinetics in the absence of exogenous buffering imposed by Ca 2ϩ indicators (see Fig. 6) (see Discussion).

Internal Ca 2؉ stores and presynaptic Ca 2؉ signaling
The role of presynaptic Ca 2ϩ stores at MF synapses remains controversial (Carter et al., 2002;Liang et al., 2002;Lauri et al., 2003;Breustedt and Schmitz, 2004). To address this, we imaged giant MFBs loaded with either Fluo-4 or Fluo-5F (200 M) (Fig. 5 A, B) and, once the Ca 2ϩ responses were stable (supplemental Fig. 2, available at www.jneurosci.org as supplemental material), applied either ryanodine (20 -60 M) or thapsigargin (1 M). These blockers interfere with Ca 2ϩ -induced ryanodine receptordependent Ca 2ϩ release or with the endoplasmic reticulum Ca 2ϩ -ATPase (Ca 2ϩ pump), respectively. Because Ca 2ϩ stores might not respond to a single AP (Lauri et al., 2003), we routinely evoked five or more APs at 20 Hz and measured the integrated ⌬F/F signal over a 500 ms window from the first response onset.
Although ryanodine and thapsigargin interfere with different mechanisms of internal Ca 2ϩ storage, the longer-term consequence is thought to be the blockade/depletion of endoplasmic reticulum Ca 2ϩ stores (Verkhratsky, 2005). Indeed, both ryanodine and thapsigargin produced a small yet significant increase in the resting fluorescence F (11 Ϯ 3%, n ϭ 8, p Ͻ 0.02; and 12 Ϯ 3%, n ϭ 5, p Ͻ 0.02, respectively) and a decrease in ⌬F/F (18 Ϯ 6%, n ϭ 8, p Ͻ 0.02; and 8 Ϯ 7%, n ϭ 5, NS, respectively), recorded in experiments with the highaffinity Fluo-4 ( Fig. 5D-F ). In contrast, ryanodine application in experiments with the lower-affinity Fluo-5F produced only insignificant increases in F, yet it decreased ⌬F/F robustly (17 Ϯ 4%, p Ͻ 0.004; integrated over six APs) (Fig. 5G-I ). In both cases, the significant changes exceeded at least threefold the experimental fluctuations of F and ⌬F/F documented in baseline conditions on a similar timescale (supplemental Fig. 2 B, C, available at www.jneurosci.org as supplemental material).
These results are consistent with an elevation in the resting Ca 2ϩ concentration, [Ca 2ϩ ] rest : the higher affinity Fluo-4 is more sensitive than Fluo-5F to changes in [Ca 2ϩ ] rest , yet it saturates to a greater degree and therefore is less sensitive to changes in ⌬F/F. According to Equation 3, in baseline conditions of Fluo-4 fluorescence, a ϳ12% increase in F corresponds to an increase in [Ca 2ϩ ] rest of 40 -50 nM. In the case of lower-affinity Fluo-5F, however, the same increase in [Ca 2ϩ ] rest should produce a much smaller elevation of F, which is fully consistent with our observations. In line with the buffering role of Ca 2ϩ stores, their blockade also slowed down the decay of the ⌬F/F signal following five or more APs: by 21 Ϯ 8% in Fluo-4 experiments (the apparent decay time constant changed from 730 Ϯ 111 to 856 Ϯ 110 ms; n ϭ 10; p Ͻ 0.03) and by 34 ϩ 12% in Fluo-5F experiments (time constant changed from 1106 Ϯ 121 to 1533 Ϯ 295 ms; n ϭ 65; p Ͻ 0.004); the fluorescent decay in these multi-AP recordings was substantially slower than that following a single AP (Fig. 2), partly because of the indicator saturation.

Discussion
In this study, we have evaluated the main determinants of Ca 2ϩ signals generated in individual giant MFBs by electrical activity in a single granule cell. Somatic voltage, which affects Ca 2ϩ responses in proximal axonal compartments, does not influence rapid Ca 2ϩ transients in giant MFBs. We find that Ca 2ϩ stores take part in maintaining the resting concentration of presynaptic Ca 2ϩ , a mechanism which has not previously been reported in cortical synapses. Rapid Ca 2ϩ release from internal stores also contributes to Ca 2ϩ transients in giant MFBs (this could be masked in multiple MF recordings, at least in part, if other MF compartments were insensitive to Ca 2ϩ store blockade). The data provide a basis for deciphering the kinetics of free Ca 2ϩ at giant MFBs, a key to presynaptic mechanisms of usedependent plasticity at the MF-CA3 pyramidal cell synapse, as discussed below.  Acsady et al. (1998) (C, dark staining indicates rabbit anti-substance P immunoreaction identifying a single presynaptic axon). Comparing these to the PSF (dashed ellipse) suggests that optical recordings normally deal with fluorescence volume-averaged across the MFB lumen. D, F, Fluorescence recordings and computed kinetic reaction components at three Ca 2ϩ indicator settings, as indicated. The top trace inset illustrates five simulated AP-evoked pulses of Ca 2ϩ influx; relative scale. Orange traces, Experimental recordings: the average fluorescence time course (global mean for n ϭ 18, 5, and 5 cells in D, E, and F, respectively). The black lines represent simulated data: the best-fit predictions for the recorded fluorescence [CaI]

Electrotonic control of presynaptic Ca 2؉ entry
The regulatory role of presynaptic ionotropic receptors in neurotransmitter release has attracted much interest (Engelman and MacDermott, 2004). However, to what extent presynaptic electrotonic influences spread along the terminal remains poorly understood. Although the granule cell somatic voltage affects Ca 2ϩ kinetics at MF axonal boutons in the hilus (Ruiz et al., 2003), we detected no such influence at giant MFBs in CA3. This was consistent with the estimated electrotonic length constant of Ca 2ϩ entry control in MFs, ϭ 150 -200 m, also agreeing with the lack of somatic voltage control over MF excitability in stratum lucidum (Schmitz et al., 2000). However, very recent observations of presynaptic EPSCs in giant MFBs have proposed an MF electrotonic length constant of ϳ450 m (Alle and Geiger, 2006). The most parsimonious explanation for the difference is that AP-evoked Ca 2ϩ entry is not sensitive to small changes in the local membrane potential. Indeed, somatic voltage has no effect on Ca 2ϩ currents in giant MFBs (Alle and Geiger, 2006). In any case, the value of suggests that the Ca 2ϩdependent release machinery acting at synapses between granule cells and interneurons or mossy cells in the hilus could be influenced by the resting somatic membrane voltage. Conversely, presynaptic ionotropic action could propagate between neighboring giant MFBs occurring 100 -200 m apart (Acsady et al., 1998).
At the calyx of Held, moderate presynaptic depolarization increases the resting Ca 2ϩ concentration, thus augmenting the release probability (Turecek and Trussell, 2001;Awatramani et al., 2005). The depolarization-induced elevation of resting Ca 2ϩ was also reported in proximal MFBs (Ruiz et al., 2003). If similar phenomena occur locally in giant MFBs, an increase in the postsynaptic responses could follow (Alle and Geiger, 2006). This might reconcile the facilitatory effect of low kainate concentration on MF transmission (Schmitz et al., 2001) with the depolarizing action of kainate receptors. Furthermore, presynaptic depolarization could initiate AP broadening (Geiger and Jonas, 2000), thus boosting Ca 2ϩ entry and the release probability further.

Kinetics of free Ca 2؉ and short-term synaptic facilitation
We have used several Ca 2ϩ indicator configurations to estimate and cross-validate Ca 2ϩ kinetics at giant MFBs. Although our experiments rely on fluorophores that directly interfere with Ca 2ϩ buffering, the resulting kinetic model allows evaluation of the free Ca 2ϩ kinetics in the absence of exogenous buffers (Fig.  6 A). The data predict substantial use-dependent facilitation of both rapid Ca 2ϩ transients and residual Ca 2ϩ rises, with the latter extending beyond 500 ms postpulse (Fig. 6 A, gray line). However, the volume-integrated kinetics could underrepresent concentration microdomains near Ca 2ϩ entry sites, thus underestimating local buffer saturation and hence overestimating usedependent facilitation of local free Ca 2ϩ (Meinrenken et al., 2003). Nonetheless, even modest facilitation would increase the release probability substantially if neurotransmitter release at MFBs depended on Ca 2ϩ in a highly supralinear manner, as reported in the calyx of Held (Schneggenburger and Neher, 2000). This is, however, not the case in giant MFBs: increasing presynaptic Ca 2ϩ transient through either AP broadening, elevation in external Ca 2ϩ , or changed buffering conditions enhances MF-CA3 pyramidal cell transmission with the less than second power The average fluorescence kinetics in response to five APs (global mean; n ϭ 13) in control (blue) and test (red) conditions. E, Relative changes in baseline signal F, ⌬F/F (integrated over five pulses), and fluorescence decay time (single exponent) after application of ryanodine (n ϭ 8; black circles) or thapsigargin (n ϭ 5; green circles) in individual cells. The overall changes are as follows (n ϭ 13; see Results for separate statistics on ryanodine and thapsigargin): F, 12 Ϯ 2% (***p Ͻ 0.001); ⌬F/F, Ϫ13 Ϯ 5% (*p Ͻ 0.03); , 21 Ϯ 8% (*p Ͻ 0.03); the dashed lines connect data points from the same MFB. relationship (Regehr et al., 1994;Geiger and Jonas, 2000;Blatow et al., 2003;Mori-Kawakami et al., 2003). The question therefore arises whether the kinetics of free presynaptic Ca 2ϩ evaluated here could explain short-term synaptic facilitation at these synapses.
To address this, we calculated use-dependent increases of residual [Ca 2ϩ ] and rapid [Ca 2ϩ ] transients (average [Ca 2ϩ ] elevation over 4 ms during an AP) in giant MFBs and compared them with the average facilitation of EPSCs evoked in CA3 pyramidal cells by short trains of APs at 10 and 20 Hz (elicited by extracellular stimuli in dentate gyrus) (Fig. 6 B). The comparison suggests that, in conditions of our experiments, synaptic facilitation follows more closely rapid [Ca 2ϩ ] transients than residual [Ca 2ϩ ] rises (with an exception of the second pulse). Although these data suggest a causal link between facilitation of presynaptic Ca 2ϩ signals and postsynaptic responses, it would be important to establish how the presynaptic Ca 2ϩ kinetics reflects synaptic adaptation to different firing patterns (Zucker and Regehr, 2002).

A role for Ca 2؉ stores
We found that blockade of internal Ca 2ϩ stores in giant MFBs elevates resting Ca 2ϩ by ϳ40%. Although this phenomenon has not previously been reported in cortical synapses, it is not sur-prising: the Ca 2ϩ concentration inside endoplasmic reticulum stores is critical for cytosolic Ca 2ϩ homeostasis (Verkhratsky, 2005). Indeed, Ca 2ϩ leakage from the stores is negatively coupled with Ca 2ϩ uptake, producing a kinetic equilibrium (Solovyova et al., 2002), which is likely to reflect the resting Ca 2ϩ level. By taking Ca 2ϩ stores out of the equilibrium, ryanodine blockade should reduce the Ca 2ϩ removal capacity of giant MFBs, thus elevating resting Ca 2ϩ concentration. The present data allow us to evaluate this quantitatively.
Ca 2ϩ removal depends primarily on the two families of Ca 2ϩ pumps, the plasma membrane Ca 2ϩ -ATPase and the sarco/endoplasmic reticulum Ca 2ϩ -ATPase; mitochondria and a low-affinity Na ϩ /Ca 2ϩ exchanger also contribute to Ca 2ϩ removal, albeit on a slower timescale (for review, see Parekh, 2003;Mata and Sepulveda, 2005;Verkhratsky, 2005). Because pumps operate through binding and translocation, Ca 2ϩ uptake kinetics could be generally represented by the following two-stage reaction: where P and Ca t 2ϩ denote pumps and intralumen Ca 2ϩ , respectively, CaP stands for the Ca-bound pump complex, and k on , k off , and k u stand for the average rate constants of Ca 2ϩ binding, unbinding, and translocation (coupled with the reappearance of available pumps P), respectively. These reactions (Eq. 9) evoke the following simple kinetic equations: with the mass conservation law ͓P͔ ϩ ͓CaP͔ ϭ ͓P͔ tot , where brackets denote concentration, L Ca is the rate of overall Ca 2ϩ leakage into the cytosol, and [P] tot is the total amount of available pumps. In the steady-state case (time derivatives zeroed), Equations 10a, 10b, and 10c reflect equilibration between uptake and leakage as follows: , and the average experimental EPSC amplitude (filled circles) in CA3 pyramidal cells in response to five APs (n ϭ 5 cells). All values are normalized with respect to the first response. C, Same as in A but the blockade of Ca 2ϩ -induced Ca 2ϩ release from stores is mimicked as a reduction in evoked Ca 2ϩ entry by 20% and an increase in resting Ca 2ϩ by 40% (see Results). D, The difference between A and C illustrating net contribution of Ca 2ϩ store releases to free Ca 2ϩ transients.