Different states of synaptic vesicle priming explain target cell type–dependent differences in neurotransmitter release

To test the dynamic properties of release from CA1 PCs to FSINs, we performed simultaneous whole-cell patch-clamp recordings from CA1 PCs and INs in acute coronal slices obtained from the dorsal hippocampus of adult mice. Postsynaptic INs were visually identified based on the shape and location of their somata using DIC imaging. Their firing properties were tested using DC current injections of variable amplitudes. Once an IN was classified as FSIN (39) and a connected PC was found, recordings were performed within a 10 min window to avoid run-down of EPSCs (Methods). Consistent with a previous study (39), the amplitude of the first unitary EPSC was large (127.4 ± 109.7 pA, n = 106 pairs) and showed prominent cell-to-cell variability [coefficient of variation (CV): 0.87]. Although the majority of the unitary EPSCs had fast rise times (RT), some had 10 to 90% RT values >0.5 ms, which might indicate dendritic filtering of distally located synapses and potential space-clamp problems. To prevent complications arising from these factors, we subselected recordings in which the 10 to 90% RT of the averaged EPSC was <0.5 ms. These subselected well-clamped EPSCs had amplitudes of 160.5 ± 121.2 pA (n = 66 pairs) with a CV of 0.76.

To explore the dynamic properties of the FSIN-innervating synapses, we applied various stimulation protocols, testing both STF and STD, recoveries from facilitation/depression, and the effects of low-frequency conditioning on subsequent high-frequency trains. For the frequency dependence of release, we applied trains of presynaptic stimuli at 5, 20, and 100 Hz and recorded the postsynaptic responses (Fig. 1 A–C). Interestingly, the averaged paired-pulse ratio of the first two EPSCs (PPR_2/1) was frequency independent (PPR_2/1 at 5 Hz: 0.70 ± 0.18, at 20 Hz: 0.74 ± 0.3 and at 100 Hz: 0.74 ± 0.35), but the amplitudes of EPSCs at steady-state toward the end of the stimulus trains showed frequency-dependent depression (Fig. 1D; normalized amplitude from grand total average trace (GTA) at 5 Hz: 0.48, at 20 Hz: 0.37 and at 100 Hz: 0.14). We then tested the recovery of release at 110 ms after a long high-frequency train (15 APs at 100 Hz) and found that the relative amplitude of the first EPSC of the recovery pulses was 0.56 ± 0.23 of the first EPSC of the first train (n = 21), four times larger than that at the steady-state end of the preceding train (Fig. 1E). The recovery was very similar after a short (6 AP) 100 Hz train (relative amplitude of the first EPSC of the test train: 0.58 ± 0.37, n = 13; Fig. 1F). However, when the recovery time after the 6 AP train was increased to 1.5 s, the amplitude of the first EPSC recovered to 0.73 ± 0.21 of its original value (n = 13; Fig. 1F). We also tested these PC–FSIN synaptic connections with two “complex” protocols, in which a preconditioning train (6 APs at 20 Hz) was immediately followed by 15 APs at 100 Hz, which was followed by a short high-frequency test train (6 APs at 100 Hz) either at 110 ms or 1.5 s recovery times (Fig. 1G). The preconditioning 20 Hz train and the following 100 Hz train caused a moderate and robust depression, respectively. The recovery from depression depended on the interval between conditioning and test trains: The first EPSC of the test train recovered to 0.51 ± 0.20 of its original value after 110 ms and fully recovered after 1.5 s (1.16 ± 0.51; Fig. 1 G, Bottom).

To estimate the proportion of well-primed SVs at FS–FSIN synapses, we turned to modeling of the EPSC amplitudes using the recently described sequential two-step priming model including a labile tightly docked SV state (TSL; Fig. 2A adopted from figure 1Ci in ref. 29). We simplified the model slightly by omitting the Ca²⁺ dependence of P_fusion._. We constrained the resting [Ca²⁺] to 50 nM with an increment of effective [Ca²⁺] following each AP of 110 nM (29). The remaining parameters were then fitted (see Methods; model parameters and terms related to the model are given and explained in SI Appendix, Table S1).

First, we performed parameter fitting simultaneously on data of five protocols, as shown in Fig. 2B: three simple trains (100, 20, and 5 Hz), the long train followed by a short one (15 + 6 APs at 100 Hz) and two short trains in sequence (6 + 6 APs at 100 Hz). Fig. 2B demonstrates the quality of the fit to these five protocols and Fig. 2C illustrates the model prediction with these parameters to the two complex protocols. Next, we performed the reverse sequence of analysis. We fitted the model parameters to the two complex protocols plus the 5 Hz steady-state train protocol (Fig. 2D) and tested the model prediction on the other four protocols (Fig. 2E). Visual inspection of the fit revealed an almost identical goodness of fit irrespective of whether the parameters were optimized on the five “simple” protocols (Fig. 2B) or on the complex trains (c.f. Fig. 2 B and E). Quantitatively, when the rmsd was calculated for the seven protocols, very similar values were obtained irrespective of whether the parameters were obtained from fitting the five simple protocols (0.00171) or the two complex protocols (0.00179). We then decided to calculate the mean of each parameter as obtained with the two methods and simulated all seven protocols using these mean values. This resulted in an rmsd (0.00165) that was somewhat smaller than those obtained separately (Fig. 2 F–H). Thus, our modeling demonstrates that similar model parameters are obtained when fitting is constrained to our complex protocols as compared to fitting data from the five simple protocols, offering the advantage of performing much less experiments for obtaining similarly constrained fits. This is of great importance, given the fragile nature of the synapses under study (time-dependent run-down after 10 min). Our model fitting/parameter optimization at PC–FSIN resulted in a P_fusion of 0.6 and a TS fraction (=TS/(TS+LS)) of 0.44, resulting in a Pv of 0.26 that is somewhat lower than that estimated with multiple probability fluctuation analysis (39). All model parameters together with an explanation of terms are provided in SI Appendix, Table S1.

To compare the above results with the properties of CA1 PC–O-LM cell synapses, we performed dual whole-cell recordings in acute slices of transgenic mice in which td-Tomato is expressed in O-LM INs (see Methods and ref. 20). The firing properties of fluorescent INs were examined with DC current injections of different amplitudes. The recorded INs were filled with biocytin and their morphological identity was verified post hoc. The STP of release at PC–O-LM cell connections was tested using the two complex protocols only. Consistent with the results of previous studies (20, 40, 41), the amplitudes of first EPSC of trains were small (14.2 ± 11.9 pA, n = 50 pairs), failure rates were high, and some connections had only failures for the first AP (6 out of 50). The amplitude of EPSCs slightly increased (facilitated) during the 20 Hz preconditioning train and more dramatically during the first few APs of the 100 Hz train episode (Fig. 3 A–C). The normalized amplitude of the first EPSC of recovery test train after 110 ms was 1.50 ± 1.66 (normalized to the first EPSC of the preconditioning train, n = 26), whereas EPSCs fully recovered (0.95 ± 0.54, n = 18 pairs) after 1.5 s (Fig. 3C). Fig. 3D illustrates the superimposed GTA traces of the PC–O-LM and PC–FSIN connections, demonstrating that the >10-fold difference in the first EPSC amplitude disappears during stimulation and that after the ninth AP both synapses can maintain transmission during high-frequency presynaptic activity roughly equally well.

Next, we aimed to determine the key model parameters that are responsible for the differences in the functional properties of PC–FSIN and PC–O-LM cell connection. We started with the parameter values optimized for the PC–FSIN connections (Fig. 2 and SI Appendix, Table S1) and varied parameters one by one to see whether the model predictions fit the PC–O-LM cell data. First, we allowed only a single, then two then three parameters to be changed simultaneously (Methods) and found progressively better and better fits. When k2_0, s2, and P_fusion were simultaneously optimized, the model qualitatively described the initial small facilitation and depression, followed by the large facilitation and depression during the high-frequency EPSC train (Fig. 4A). Recognizing that k2_0 and s2 together determine the Ca²⁺ dependence of k2, we also introduced a scaling factor for k2_0 and s2 of the FSIN fit and optimized this scaling factor together with P_fusion. This resulted in an almost identical goodness of fit to that obtained when the three parameters were fitted separately (Fig. 4B). Finally, we also allowed all model parameters to be fitted, resulting in a slight improvement in the overall fit (Fig. 4B). Fig. 4C illustrates the PC–FSIN experimental data and superimposed best model fits as well as the PC–O-LM cell data with the model fit for which only three parameters were altered.

We then examined the effects of changing these three model parameters and found that a >10-fold reduction of k2_0 and s2 resulted in a dramatic 6.5-fold reduction of the proportion of tightly docked SVs (TS fraction = 0.07 vs. 0.44 for the FSINs), whereas the reduction in P_fusion was only 40% (from 0.60 to 0.36; SI Appendix, Table S1). These results demonstrate that the sequential two-step priming model is capable of describing the release dynamics of both PC–FSIN and PC–O-LM synapses and that altering only three parameters can convert a depressing into a facilitating synaptic “phenotype.” Disregarding small deficits, one could even switch between the two functional “phenotypes” by just changing two parameters: P_fusion and the scaling factor (steepness of the Ca²⁺ dependence of k2).

In the experiments described so far, we reached our conclusions by fitting the sequential two–step priming model to our experimental data obtained from PC–FSIN and PC–O-LM connections and drew our biological conclusions based on model parameters of P_fusion and the TS fraction. In our final set of experiments, we aimed to validate our approach by using selective pharmacological manipulations targeting either P_fusion or the TS pool size. The K⁺ channel blocker 4-AP broadens APs and increases AP-evoked [Ca²⁺] transients (20) and consequently increases P_fusion. The phorbol ester analog PDBU increases release by facilitating priming of SVs through the positive modulation of Munc13s without effecting AP-evoked [Ca²⁺] transients (20), hence, PDBU should not affect P_fusion. Thus, we recorded the effects of these drugs on PC–O-LM cell connections and performed model optimization to investigate which model parameters are altered in the presence of these two drugs.

In the presence of 5 µM 4-AP, the amplitude of the first EPSC of the preconditioning train was 29.1 ± 34.2 pA (n = 23 pairs), which is approximately twofold larger than that in control ACSF (Fig. 5). The pattern of postsynaptic responses remained very similar to that found in control recordings: There was a small facilitation during the preconditioning train and a robust facilitation–depression during the 100 Hz train. However, the recovery at 110 ms was less pronounced; the normalized response of the first recovery pulse was 1.74 ± 1.02 (relative to the first EPSC of the preconditioning train), which was larger than that found in control (1.50 ± 1.66; Fig. 5 A, B, and F).

Postsynaptic responses in the presence of 1 µM PDBU were drastically different. The amplitude of the first EPSC of the preconditioning train was >sixfold larger than that recorded in ACSF (74.9 ± 86.3 pA, n = 15; Fig. 5 C and D) and the EPSCs during the preconditioning train displayed mainly depression rather than facilitation. At the beginning of the 100 Hz train, the facilitation was also smaller than that in control (2.0 ± 1.8-fold vs. 3.1 ± 2.5 in control; Fig. 5 E and F). These pharmacological experiments confirm our previous results (20) showing a much larger effect of PDBU than that of 4-AP on the amplitude of the first EPSC of the train at PC–O-LM cell synapses.

We then fitted our model to the data obtained in 4-AP and PDBU with the following constraints. In our previous study (20), we demonstrated that 5 µM 4-AP resulted in a ~50% increase in the AP-evoked Ca²⁺ influx in axon terminals. Thus, we increased the AP-induced [Ca²⁺] increment from 110 nM to 168 nM in our model. The resting [Ca²⁺] was fixed to 50 nM. We then optimized all other parameters and obtained a fit that qualitatively described well the STP pattern of the data under 4-AP (Fig. 5G). Because our experimental data demonstrated that the application of PDBU did not change the AP-evoked Ca²⁺ influx into hippocampal PC axon terminals (20), we constrained the parameters describing AP-evoked [Ca²⁺] transients to those of controls and fitted the rest of the parameters to the PDBU data. The model with all other parameters as freely variable ones also produced a good fit to the data (Fig. 5H), describing well the pattern of the plasticity during the complex protocol. We then looked at how these drugs affected the two key functional parameters. 4-AP increased the P_fusion by 2.6-fold (to 0.85) without any major change in the proportion of SVs in the TS state (TS fraction: 0.05 in 4-AP vs. 0.08 in control; Fig. 5I). In contrast, the best fit to the PDBU data resulted in an almost identical P_fusion to that obtained in control (0.29 vs. 0.33) with a 4.5-fold increase in the TS fraction (from 0.08 to 0.34; Fig. 5J; see all model parameters in SI Appendix, Table S1). These results verify that the selective pharmacological modification of priming and fusion altered model parameters that influence TS fraction and P_fusion, respectively, in a predictable manner.

Fifty-six adult (P48–94) male and female transgenic mice were used (Chrna2-Cre)OE25Gsat/Mmucd, [RRID:MMRRC_036502-UCD, on C57BL/6 J background (53)] and crossed with reporter line Ai9 or Ai14 [Gt(ROSA)26Sor_CAG/LSL_tdTomato]. The animals were housed in the vivarium of the Institute of Experimental Medicine in a normal 12 h/12 h light/dark cycle and had access to water and food ad libitum. All experiments were carried out in accordance with the Hungarian Act of Animal Care and Experimentation 40/2013 (II.14) and with the ethical guidelines of the Institute of Experimental Medicine Protection of Research Subjects Committee.

Mice were stably anesthetized with a ketamine, xylazine, pypolphene cocktail (0.625, 6.25, 1.25 mg/mL respectively, 10 µL/g body weight) then decapitated, the brain was quickly removed and placed into an ice-cold cutting solution containing the following (in mM): sucrose, 205.2; KCl, 2.5; NaHCO₃, 26; CaCl₂, 0.5; MgCl₂, 5; NaH₂PO₄, 1.25; and glucose, 10, saturated with 95% O₂ and 5% CO₂. Then, 250 µm thick coronal slices were cut from the dorsal part of the hippocampus using a Vibratome (Leica VT1200S) and were incubated in a submerged-type holding chamber in ACSF containing the following (in mM): NaCl, 126; KCl, 2.5; NaHCO₃, 26; CaCl₂, 2; MgCl₂, 2; NaH₂PO₄, 1.25; and glucose, 10, saturated with 95% O₂ and 5% CO₂, pH = 7.2 to 7.4, at 36 °C for 30 min, and were then kept at 22 to 24 °C.

All paired recordings were performed at 32 to 33 °C up to 6 h after slicing in ACSF supplemented with 2 µM AM251 to block presynaptic CB1 receptors and 0.35 mM γDGG to prevent AMPA receptors saturation. Cells were visualized with infrared differential interference contrast (DIC) imaging on a Nikon Eclipse FN1 microscope with a 40x water immersion objective (NA = 0.8). CA1 PCs were identified based on their position and morphology. O-LM INs were identified in the stratum oriens of the CA1 region by tdTomato expression, somatic morphology, and the membrane voltage responses upon de- or hyperpolarizing current injections (600 ms, from −250 to 800 pA with 100 pA steps). FSINs were identified using their position, morphology, and their membrane voltage responses upon de- or hyperpolarizing current injections (600 ms, from −250 to 800 pA with 100 pA steps). Patch pipettes (resistance 4 to 6 MΩ) were pulled from thick-walled borosilicate glass capillaries with an inner filament. For the interneurons, intracellular solution contained the following (in mM): K-gluconate, 130; KCl, 5; MgCl₂, 2; EGTA, 0.05; creatine phosphate, 10; HEPES, 10; ATP, 2; GTP, 1; and biocytin, 7. For the PCs, intracellular solution contained the following (in mM): K-gluconate, 97.4; KCl, 43.5; MgCl₂, 1.7; NaCl, 1.8; EGTA, 0.05; creatine phosphate, 10; HEPES, 10; ATP, 2; GTP,0.4; biocytin, 7 and 10 mM glutamate, pH = 7.25; 290 to 305 mOsm. Paired whole-cell recordings were performed while the PCs were held in current-clamp mode at −65 mV (with a maximum of ±100 pA DC current). Postsynaptic INs were held at −65 mV in voltage-clamp mode (with a maximum of ±200 pA DC current) with access resistance below 20 MΩ using a dual-channel amplifier (MultiClamp 700B; Axon Instruments). APs were evoked in PCs with 1.5 ms long depolarizing current pulses (2.3 nA). Evoked EPSCs were recorded for PC–FSINs pairs using six different protocols: 1) 15 APs at 5 Hz; 2) 15 APs at 100 Hz followed by 6-APs recovery test train after 110 ms at 100 Hz; 3) 6 APs at 100 Hz followed by a 6-APs recovery test train at 100 Hz after 110 ms; 4) 6 APs at 100 Hz followed by a 6-APs recovery test train at 100 Hz after 1.5 s recovery test period; 5) 6-APs preconditioning train at 20 Hz followed by 15-APs train at 100 Hz then 6-APs recovery test train at 100 Hz after 110 ms; and 6) 6-APs preconditioning train at 20 Hz followed by 15-APs train at 100 Hz then 6-APs recovery test train at 100 Hz after 1.5 s recovery test period. For the PC–O-LM pairs, only the last two protocols were recorded. Recording from a given pair was restricted to 10 min to avoid rundown of postsynaptic responses. Sixty-second intertrace intervals were kept except for the third and fourth protocols where the 30 s intertrace interval was used. INs with an increased access resistance (>25%) during the recording period were excluded from analysis. Data were filtered at 3 to 4 kHz (Bessel filter), digitized online at 50 kHz, recorded, and analyzed using Clampfit 10.7 (Molecular Devices). The peak amplitudes, 10 to 90% RTs, and areas under the curves were calculated in Clampfit. Likewise, average traces and their 10 to 90% RTs were calculated from 10 or 20 traces in Clampfit for each PC-FSIN pair. Pairs with RTs <0.5 ms of the first EPSC of the train were subselected and further used in the study.

To perform recordings after a fixed duration of drug application while maintaining short (10 min) whole-cell recording time to minimize EPSC rundown, we modified our experimental protocol. Fluorescent O-LM INs were selected and patched in acute coronal slices of the dorsal hippocampus of transgenic mice and synaptically connected presynaptic PCs were searched. Once a connection was established, the patch pipette from the PC was withdrawn, and the drug was perfused onto the slice. Ten minutes later, the PC was repatched and EPSCs to the complex protocol with 110 ms recovery were recorded for 10 min only. We have previously found that 5 µM 4-AP increases Ca²⁺ influx at PC boutons targeting mGluR1α expressing INs to the same level that was found at PC boutons targeting parvalbumin expressing INs (20).

The sequential two-step priming model (29) was implemented in Berkeley Madonna [version 10.4; (54)]. Michaelis–Menten-like saturation for k1_0 in response to [Ca²⁺] was implemented, but the Ca²⁺ dependence of P_fusion, as described in ref. 29, was omitted. In addition to LS and TS, a “Labile Tight State” (TSL), which contributes to release at high frequencies, was taken into account according to ref. 29. Parameter fitting was done by the algorithm of the software (Euler’s method). Unless otherwise stated, the resting [Ca²⁺] was constrained to 50 nM, and the increment of effective [Ca²⁺] following each AP was constrained to 110 nM according to Lin et al. (29), and these parameters were kept constant during fitting. Parameter values are presented in SI Appendix, Table S1. To quantify the “goodness of fit,” the rmsd was calculated for each fit.

EPSC amplitudes were converted to quantal content, the number of released SVs, by dividing the peak amplitudes by the estimated quantal size. For the PC–FSIN synapse (mean peak amplitude 160 pA), the quantal size is 32 pA (39), resulting in an initial release of 5 quanta. The quantal content of the PC–O-LM connection was estimated to be 1/10th of that of the PC-FSIN connection (20). Therefore, traces were scaled for an initial release of 0.5 quanta.

First, we searched for a single parameter that would change the model from STD to STF. Results showed that only three parameters were capable of converting the model from STD to STF: b2, the backward rate constant of the second priming step, k2_0, the resting value of its forward rate constant, and P_fusion (SI Appendix, Table S1). While the model regimes exhibit STF, none of them describe adequately the data. Hence, we continued our search for model parameters that converted STD to STF, but this time changing two parameters simultaneously. Results showed that, while the two-parameter optimization was better than the one-parameter optimization, there were still great mismatches between the model and data. For example, the solution involving k2_0 and the steepness of its Ca²⁺ dependence, s2, had a reasonably good fit of the first release and the preconditioning EPSC amplitude train, but vastly underestimated the EPSC recovery; the P_fusion+b2 solution produced an acceptable EPSC recovery but started with an initial SV release of zero. Allowing the simultaneous optimization of three parameters revealed parameter constellation that qualitatively described the dynamics of SV release at the PC–O-LM synapse. As shown in Fig. 4A, when k2_0, s2, and P_fusion were simultaneously optimized, the model qualitatively described the initial small facilitation and depression, followed by the large facilitation and depression during the high-frequency EPSC train. Furthermore, the recovery was also reasonably well described, reflected in a robust reduction of the rmsd value. Because k2_0 and s2 values were altered to a similar extent (>90% reduction compared to PC–FSIN values; SI Appendix, Table S1), we introduced a common scaling factor of these two parameters and refitted the data by allowing only changes in the scaling factor and P_fusion. This fit resulted in an rmsd value (0.00057) almost identical to that obtained with separate fitting of k2_0 and s2 (0.00054; Fig. 4B). The rational to use a scaling factor for simultaneous modification of k2_0 and s2 is that their underlying molecular mechanisms might be biologically linked. Finally, we allowed all parameters to be optimized (with the exception of resting [Ca²⁺] and AP-induced [Ca²⁺] increments), which resulted in a further improvement of the goodness of fit (rmsd = 0.00027; Fig. 4B). Notably, the largest improvement involved the first EPSC response of the recovery train. All data are given as mean ± SD.

E.N. and Z.N. designed research; M.A. and Z.N. performed research; M.A., E.N., and Z.N. analyzed data; and M.A., E.N., and Z.N. wrote the paper.

The authors declare no competing interest.