Interplay between global and pathway-specific synaptic plasticity in CA1 pyramidal cells

Mechanisms underlying information storage have been depicted for global cell-wide and pathway-specific synaptic plasticity. Yet, little is known how these forms of plasticity interact to enhance synaptic competition and network stability. We examined synaptic interactions between apical and basal dendrites of CA1 pyramidal neurons in mouse hippocampal slices. Bursts (50 Hz) of three action potentials (AP-bursts) paired with preceding presynaptic stimulation in stratum radiatum specifically led to LTP of the paired pathway in adult mice (P75). At adolescence (P28), an increase in burst frequency (>50 Hz) was required to gain timing-dependent LTP. Surprisingly, paired radiatum and unpaired oriens pathway potentiated, unless the pre-post delay was shortened from 10 to 5 ms, which selectively potentiated paired radiatum pathway, since unpaired oriens pathway decreased back to baseline. Conversely, the exact same 5 ms pairing in stratum oriens potentiated both pathways, as did AP-bursts alone, which potentiated synaptic efficacy as well as current-evoked postsynaptic spiking. L-type voltage-gated Ca2+ channels were involved in mediating synaptic potentiation in oriens, whereas NMDA and adenosine receptors counteracted unpaired stratum oriens potentiation following pairing in stratum radiatum. This asymmetric plasticity uncovers important insights into alterations of synaptic efficacy and intrinsic neuronal excitability for pathways that convey hippocampal and extra-hippocampal information.


Results
Spike-timing dependent protocols with 10 ms pre-post delay in adult versus adolescent mice. Former studies examining pathway-specific LTP in CA1 pyramidal neurons of hippocampal slices with spike-timing dependent protocols tested two pathways in stratum radiatum (RAD) (e.g., 23 ). Here, we also tested a pathway in RAD, but similar to former field recordings (e.g., 32 ) the unpaired control pathway was in stratum oriens (OR) (Fig. 1A). First, we tested triplets of APs at a frequency of 50 Hz in adult mice (P75) as previously shown for two RAD pathways 23 . The induction protocol consisted of triplet APs generated by 3 ms somatic current injections preceded (10 ms) by presynaptic stimulation, repeated 60 times at 0.1 Hz for a duration of 10 min 23 . This AP-burst pairing protocol induced pathway-specific LTP in slices of adult mice (Fig. 1B, P75: RAD, 1.67 ± 0.14, p = 0.0007; OR, 1.32 ± 0.24, p = 0.282, n = 13). Increased excitatory postsynaptic potentials (EPSPs) were observed in the unpaired pathway though these were not significant (Fig. 1B, see also Methods of 23 ).
At adolescence (P28), postsynaptic AP-bursts as part of paring protocols with 10 ms pre-post delay do not induce pathway-specific LTP in the paired RAD pathway. Moreover, triplet AP-burst pairing required a frequency above 50 Hz to induce LTP of the paired pathway in young mice (P28), consistent with young rats 33 . Thus, spike-timing dependent protocols established at adulthood cannot readily be applied to adolescence.
Spike-timing dependent protocols with 5 ms pre-post delay in adolescent mice. Recent spike-timing studies often used 5 instead of 10 ms pre-post delay during pairing, e.g. 34,35 , consistent with a former study addressing input specificity of synaptic modification 36 . To examine 5 ms pre-post delay at P28, we chose an AP-burst of 75 Hz which i) is sufficient to induce LTP in RAD (Fig. 1D) and ii) is a compromise between 50 Hz used in some studies 23,37 and 100 Hz in others 12,33 . Thus, single EPSPs were evoked 5 ms before each 75 Hz AP-triplet and repeated 60 times at 0.1 Hz. Pairing in RAD generated pathway-specific LTP ( Fig. 2A: RAD, 1.37 ± 0.12, p = 0.003; OR, 1.07 ± 0.10, p = 0.480, n = 19), since EPSPs in OR increased only transiently (p = 0.003, 0.016 and 0.045 for 5, 10 and 15 min, respectively, n = 19; Fig. 2A). The time course of modulating OR after RAD pairing was not affected in distinct subsets of these 19 experiments with continuation of presynaptic stimulation in OR during RAD pairing (n = 9) or using paired-pulse stimulation before and after induction (n = 6). OR test pulses showed that this unpaired pathway increased gradually during RAD pairing ( Fig. 2A). Notably, when both paired and unpaired pathways were examined in RAD 36 , input specificity was obtained without modulation of the unpaired pathway.
Next, we analyzed the fluctuations of EPSPs to estimate the expression mechanisms of LTP (see our Methods). In brief, the coefficient of variation (CV) was determined as the standard deviation of EPSPs divided by the average EPSP of the 10 min baseline period and 20 to 30 min after induction, respectively. Then the inverse squared is plotted against normalized amplitudes of RAD (left) and OR (right). p slope indicates the probability that the slope of a linear fit through the origin is unequal to 1. The expression mechanism of LTP in RAD fits best with an increase in the number of active synapses (n) (left), whereas the expression mechanism of LTP in OR fits best with an increase in release probability (P r , right, black curve, eq. 1). (D) CV −2 analyses after OR pairing indicate mainly a change in P r for both pathways. Dotted and dashed gray lines illustrate the hypothesis that LTP is due to an increase in the number of active synapses n or in quantal size q, respectively. coefficient of variation (CV −2 ) of the 20 to 30 min interval after induction was normalized to the respective baseline CV −2 and plotted against relative change in EPSP amplitude as in former studies (ref. 38 , their Fig. 4e and their supplements, as well as ref. 39 , their Fig. 11B). Following RAD pairing, the expression of LTP in RAD fitted best with an increase in the number of active synapses n, since the slope of the linear fit was not significantly different from 1 (p slope , Fig. 2C). By contrast, the variability of OR EPSPs after their decay to control level was mainly modulated by release probability consistent with equation (1) (Fig. 2C). Pairing in OR mainly increased the release probability Pr in both pathways (Fig. 2D).
Together, these pairing experiments in OR vs. RAD demonstrate asymmetric plasticity in hippocampal CA1 with pathway-specific LTP selectively in RAD, requiring modulation of the unpaired OR pathway.
Global synaptic and intrinsic LTP induced by postsynaptic action potential bursts without presynaptic pairing. After verifying in adolescent mice (P28) that presynaptic stimulations in the absence of AP triplets did not affect the amplitude of EPSPs up to 50 min (see Methods), we tested whether unpaired postsynaptic AP-bursts generate a global form of LTP, which could be modulated by paired EPSPs in RAD ( Fig. 2A) but not by paired EPSPs in OR (Fig. 2B). Again, we tested AP-bursts at 75 Hz and monitored changes of EPSPs during induction and afterwards. During induction, we alternated presynaptic, electrical stimulations between RAD and OR at a 5 s delay to AP triplets to prevent their influence on EPSPs. Under these conditions, APs alone were indeed capable and sufficient to induce global LTP at apical CA1 dendrites in RAD and at basal dendrites in OR. The gradual EPSP increase during induction reached steady state after terminating induction ( Fig. 3A: RAD, 1.68 ± 0.22, p = 0.0095; OR, 1.70 ± 0.21, p = 0.0051, n = 15). To validate this result, we next recorded excitatory postsynaptic currents (EPSCs) in voltage-clamp before and after generating APs alone in current-clamp. Voltage-clamp improved the stability of our baseline responses and consistently, EPSCs increased in both pathways ( Fig. 3B: RAD, 1.41 ± 0.12, p = 0.009; OR, 1.41 ± 0.13, p = 0.009, n = 10). Notably, presynaptic stimulations were not required during LTP induction to obtain global LTP as tested here (Fig. 3B).
The expression of LTP in RAD fitted best with an increase in the number of active synapses, while LTP expression in OR fitted best with an increase in release probability (Fig. 3C).  Fig. 2C). The expression mechanism of LTP in RAD fits best with an increase in the number of active synapses (n) (left), whereas the expression mechanism of LTP in OR fits best with an increase in P r (right, black curve, eq. 1). As a special case, the plot for OR is hardly correlated (r = 0.38), therefore no p slope was determined. (D) Change of AP firing tested during 500 ms depolarization before and 30 min after induction (left columns and example traces are from a subset of A) or tested during 600 ms depolarization in the presence of glutamatergic and GABAergic receptor blockers without synaptic stimulation (right columns) (**p < 0.01, after vs. before).
SCIeNtIFIC REPORtS | 7: 17040 | DOI:10.1038/s41598-017-17161-z Besides synaptic changes, we also observed persistent changes in excitability leading to increased spiking frequency. More precisely, as part of our EPSP recording in Fig. 3A, we injected constant current of ∼50 to 200 pA for 500 ms in order to evoke 3 to 5 APs following acquisition of baseline EPSPs. The AP firing frequency was then compared 30 min after induction using the same constant current injection ( Fig. 3D left bars: before, 4.03 ± 0.48; after, 6.39 ± 0.91, p = 0.0012, n = 11). To test more directly whether this increase in AP firing is due to an increase in intrinsic excitability, we pharmacologically prevented synaptic activation of excitatory and inhibitory receptors (AMPARs, NMDARs, group I mGluRs, GABA A Rs and GABA B Rs). After induction with APs alone, AP firing frequency increased for at least 30 min ( Fig. 3D right bars: before, 3.43 ± 0.29; after, 6.63 ± 0.94, p = 0.008, n = 10). This increase in AP firing was not observed in control experiments in the absence of AP-bursts (baseline, 3.9 ± 0.3 APs; after, 3.9 ± 0.5 APs; p = 0.90, n = 8).
In summary, AP-bursts globally and persistently potentiated spiking frequency of CA1 pyramidal neurons as well as synaptic efficacy, the latter via increasing the release probability in OR and the number of active synapses in RAD. Of note, pairing in OR or RAD (Fig. 2) did not change these mechanisms in the respective paired pathways, when compared with APs alone. Given that NMDARs are frequently involved in the induction of input-/pathway-specific synaptic plasticity 1 , we tested RAD pairing when NMDARs were antagonized. In the presence of D-APV (50 µM), RAD EPSPs and OR EPSPs potentiated, eliciting global LTP ( Fig. 5A: RAD, 2.19 ± 0.17, p = 0.0013; OR, 1.83 ± 0.30, p = 0.033, n = 9). This NMDAR-independent global LTP is reminiscent of NMDAR-independent LTP in the visual cortex 40 . The higher potentiations compared with Figs 1-4 are consistent with higher pipette series resistances in this set of pharmacological experiments (see Methods). Nevertheless, we performed matching control experiment in the absence of D-APV, finding the unpaired OR pathway modulated as in Fig. 2A (Fig. 5B: RAD, 2.39 ± 0.37, p = 0.007; OR, 1.14 ± 0.10, p = 0.17, n = 8).
To examine whether induction of global LTP involved L-type voltage-gated Ca 2+ channels activated by backpropagating APs, we tested OR pairing in the presence of nifedipine (10 µM). Neither RAD EPSPs nor OR EPSPs were enhanced following OR pairing ( Adenosine receptors. Adenosine enzymatically derived from astrocytic ATP 43 or pyramidal neurons 44,45 is known to regulate the dynamic range for LTP generation, involving the high-affinity A 1 and A 2A adenosine receptors (A 1 Rs and A 2A Rs) 25,27,44,46 , with A 1 Rs having about a twofold higher affinity for adenosine than A 2A Rs 47 . Lower adenosine concentrations decrease glutamate release by predominantly activating A 1 Rs tonically, while higher adenosine concentrations increase glutamate release via facilitatory A 2A Rs 46,48 . Hence, we investigated whether the dualistic nature of these two adenosine receptor subtypes impinged upon the observed plasticity in OR generated by RAD pairing.
Consistent with a previous study 49 , the A 2A R-specific antagonist SCH-58261 (50 nM) did not change basal synaptic transmission (Fig. 6A: RAD, 1.05 ± 0.10, p = 0.79; OR, 0.94 ± 0.09, p = 0.87, n = 5). RAD pairing in the presence of SCH-58261 led to LTP of RAD EPSPs (Fig. 6B: RAD, 1.28 ± 0.10, p = 0.008, n = 9), but not OR EPSPs ( Fig. 6B: OR, 0.92 ± 0.07, p = 0.23, n = 9). Even immediately after the induction period, there was no increase in OR EPSP. Thus, the pronounced pathway-specific LTP suggests that the transient OR EPSP increase apparent in the absence of SCH-58261 (Figs 2A and 5B) was A 2A R-mediated. In the presence of the A 2A R-specific antagonist, A 1 R-mediated tonic inhibition could be emphasized 48 . Consistent with a tonic inhibitory effect, perfusion of the A 1 R antagonist DPCPX (100 nM) increased basal synaptic transmission ( Fig. 6C: RAD, 1.21 ± 0.08, p = 0.028;   GABA B receptors. GABA A Rs were blocked in our experiments and could not contribute to the transient increase in OR EPSPs during RAD pairing. On the other hand, GABA B R activation in astrocytes has been shown to mediate synaptic depression of nontetanized hippocampal synapses within apical dendrites through adenosine 28 . Perfusion of the GABA B R antagonist CGP 55845 (2 µM) under baseline condition suggested that GABA B R activation may be more prominent in RAD than in OR, since RAD EPSPs but not OR EPSPs increased in the presence of CGP 55845 (RAD, 1.31 ± 0.07, p = 0.005; OR, 1.05 ± 0.10, p = 0.696; n = 6; Supplementary Figure S1A). Next, we examined the effects of CGP 55845 on synaptic responses to AP-bursts alone to test whether GABA B Rs were involved in controlling adenosine release that can occur through excitatory autoregulation 44 . In the presence of CGP 55845, OR EPSPs as well as RAD EPSPs remained potentiated throughout 30 min (RAD, 1.34 ± 0.08, p = 0.020; OR, 1.47 ± 0.10, p = 0.0034; n = 7; Supplementary Figure S1B). RAD pairing in the presence of CGP 55845 (Supplementary Figure S1C) still potentiated RAD EPSPs throughout 30 min (1.36 ± 0.09, p = 0.010, n = 11) and OR EPSPs for 4 min (1.23 ± 0.08, p = 0.049, n = 11) but not subsequently (p = 0.055, 0.089, 0.075 and 0.14 (n = 11) for the first 5, 10, 15 and 20-30 min after induction, respectively, n = 11). Thus, LTP of RAD EPSPs and transient plasticity of OR EPSPs was retained in the presence of CGP 55845. This was substantiated by the lack of change in CV −2 analyses in the presence (Supplementary Figure S1D) and absence of CGP 55845 (Fig. 2C). These results suggest that adenosine independent of GABA B R activation mainly modulated the plasticity in OR.

Discussion
Our findings identify a pathway-specific modulation of global plasticity in apical but not basal dendrites of CA1 pyramidal cells. Global LTP was generated exclusively by postsynaptic burst activity. When brief AP-bursts were paired with prior subthreshold stimulation in stratum oriens (OR), global LTP remained largely unaffected, whereas prior subthreshold stimulation in stratum radiatum (RAD) resulted in pathway-specific LTP (with 5 ms but not with 10 ms pre-post delay).
Lack of pathway-specific LTP following OR pairing indicated that postsynaptic burst activity alone remained decisive in inducing global synaptic LTP. Alike, positive as well as negative time delays of burst pairing protocols induced LTP at apical CA1 dendrites 50 . The similarity of global LTP induced via OR pairing and via burst activity alone was further supported by the sensitivity of postsynaptic responses of both pathways to a blocker of L-type voltage-gated Ca 2+ channels, consistent with previous studies. For example, postsynaptic theta-burst spiking alone (5 APs at 100 Hz repeated 10 times at 5 Hz) substantially and simultaneously increased synaptic currents evoked in two independent pathways in apical CA1 dendrites 33 . Similarly, repeated postsynaptic depolarizations or 1 s AP trains at 100 Hz induced global LTP of spontaneous synaptic currents 12 . The latter study suggested pre-and postsynaptic mechanisms in the generation of global LTP evidenced by an effect on CaMKII inhibition; decreased paired-pulse ratios and increased frequency and amplitude of miniature synaptic currents. Our CV −2 analyses following burst activity alone indicated an increase in the number of active synapses in stratum radiatum as well as an increase in release probability in stratum oriens, but no hint for conventional insertion of AMPA receptors into active postsynaptic sites. Thus, LTP is not expressed by an increase in quantal size q, if somatic spikes are generated either by somatic current injection as in our study and others 12,34 or by theta burst stimulation of synaptic inputs 51 . Remarkably, after OR pairing global LTP was preferentially expressed via increased release probability.
In our study, RAD pairing led to a pathway-specific LTP if postnatal development was within the adolescent age (P28) and AP-bursts were immediately (5 ms) preceded by presynaptic stimulation in stratum radiatum. In Xenopus retinotectal connections it is known that LTP pathway specificity emerges with development 52 , which is also evident from studies in rodents and many other species. Buchanan and Mellor 33 failed to induce pathway-specific LTP in juvenile (P14) rat slices though a later developmental stage (P45-P55) resulted in a stronger increase in the test than in the control pathway (their Figs 1C and 2C). Increases in control pathways have been observed previously when postsynaptic AP-burst activity was part of pairing protocols in rat and mouse slices (P42-P70) 8,33,53 . Still, the test pathways paired with theta-burst postsynaptic activity increased to a greater extent than the unpaired control pathways, reflecting pathway specificity. Increases in control pathways are probably underestimated, since control pathways are often not illustrated under all experimental conditions examined 50,54 (see however control pathways in 55 ) or changes in the control pathway lead to exclusion 23 . Thus, postsynaptic burst activity can affect synaptic efficacy in the absence of glutamatergic and GABAergic presynaptic activity, which we confirmed here with brief AP-bursts being part of pairing protocols. By contrast, postsynaptic single spikes are less influential in inducing global plasticity as shown in juvenile slices ( < P14) in which pathway-specific LTP was induced 54 . Interestingly, pairing protocols with 5 ms pre-post delay (and 35 our Figs 2 and 5) allowed the generation of pathway specificity at P28 but exclusively with presynaptic stimulation in RAD (not OR, our Figs 2 and 5).
Pathway-specific LTP following RAD pairing is generally comparable with NMDAR-dependent LTP that is often studied for two CA1 inputs within apical dendrites 1 . Therefore, one wonders why NMDARs in basal dendrites failed to generate pathway-specific LTP in our OR pairing experiments. This was initially very surprising, since pathway-specific LTP can be induced in basal dendrites of CA1 pyramidal neurons as known from extracellular field recordings 24,56,57 and from whole-cell recordings 9 . In the latter study, pathway-specific LTP was assured by local synaptic depolarization and/or dendritic spikes evoked with synaptic stimulation rather than somatic current injection 9,58 . In extracellular field recordings, synaptic stimulation likely generated backpropagating APs with reduced incidence and variable timing precision, since APs generated by repeated high-frequency electrical stimulation in either apical or basal dendrites generated plasticity across compartments 24 . Effects across compartments were also observed in our pairing experiments. Pairing in RAD, but not pairing in OR, generated pathway-specific LTP and thus, our pairing protocols generated asymmetric interactions between the two pathways. Such asymmetric modulation of plasticity has been observed previously in area CA1. High frequency priming stimulations in OR inhibited subsequent LTP in RAD but not vice versa 26 . The latter LTP-weakening effect in RAD involved muscarinic M1 acetylcholine receptor (M1R) activation. Unlike priming, release of acetylcholine following repetitive electrical stimulation in OR, either induced with high frequency stimulation 59 or with a spike-timing dependent protocol 60 , has been shown to enhance LTP in RAD. Consistently, theta burst stimulation in RAD activated M1Rs and potentiated CA1 synaptic transmission that occluded LTP, based on recent experiments with selective M1R agonists and M1R knockout mice 61 . Thus, electrical stimulation of cholinergic fibers unlikely contributed to generate pathway-specific LTP during RAD pairing. This view is supported by the fact that OR EPSPs were either not evoked or temporally separated by a 5 s interval from the AP-bursts during induction in RAD. In contrast, cholinergic modulation required a substantially shorter interval (10 ms) to generate transient depression in CA1 62 .
We pharmacologically characterized pathway-specific LTP following RAD pairing. Antagonism of NMDARs or antagonism of A 1 Rs prevented pathway-specificity and resulted in global LTP in hippocampal CA1 (see 40 for NMDAR-independent global LTP in the visual cortex), since the potentiation in OR synapses following RAD pairing persisted throughout the recording. Similarly, A 1 Rs were reported to destabilize LTP at OR synapses to a greater extent than LTP at RAD synapses 63 . Thus, Schaffer collateral stimulation in RAD may mediate the heterosynaptic plasticity in OR, i.e. across compartments in the basal dendrites via NMDAR-dependent A 1 R activation. Interestingly, NMDARs and A 1 Rs also mediated transient heterosynaptic depression within the RAD pathway 27,28 , whereas A 1 R-mediated heterosynaptic depotentiation in RAD following perforant path stimulation did not depend on NMDARs 64 . Thus, distinct heterosynaptic mechanisms appear to exist within apical dendritic compartments for cortical pathways in stratum lacunosum moleculare versus hippocampal pathways in RAD. The respective heterosynaptic mechanisms including its time dependence (5 ms vs. 10 ms pre-post delay) remain unknown with respect to cortical, hippocampal and septal pathways that converge within the basal dendritic compartment.
The main source of adenosine mediating the heterosynaptic plasticity at OR synapses after RAD pairing is not consistent with previously described NMDAR activation in interneurons and subsequent GABA B R activation in astrocytes 28 , since heterosynaptic plasticity in OR was not prevented by the GABA B R antagonist in contrast to the NMDAR antagonist. Though the transient potentiation in OR was shortened in the presence of a GABA B R antagonist, suggesting a reduced adenosine release and thus reduced A 2A Rs contribution. This points to NMDAR-dependent adenosine release independent of GABA B R activation, e.g. via a direct activation of ionotropic or metabotropic glutamate receptors in astrocytes and/or neurons 65 . As expression of functional NMDARs in hippocampal astrocytes is not confirmed 66 , NMDARs rather mediate adenosine release from neurons. Indeed, neuronal adenosine released by excitatory neurons in this case has been shown to inhibit excitatory inputs through A 1 Rs via an autonomic feedback mechanism within one second 44 . Short-term depression via this auto-A 1 R 44 might lead to LTD, if any long-term plasticity evolves. Under our conditions and following RAD pairing, however, EPSPs increased in OR through A 2A Rs most likely by attenuating the tonic inhibitory effect of A 1 Rs as observed by others 46,48 . The subsequent decay of OR EPSPs to control level within minutes likely reflects restoration of tonic inhibition once A 2A Rs desensitize 67 . By contrast, A 2A R desensitization could be weaker during OR pairing than during RAD pairing, since electrical stimulation in stratum oriens elevates extracellular adenosine less than electrical stimulation in stratum radiatum 68 . Interestingly, adenosine release during OR stimulation involves L-type voltage-gated Ca 2+ channels and/or Ca 2+ -induced Ca 2+ release 68 , and could explain why a blocker of L-type voltage-gated Ca 2+ channels reduced OR pairing induced LTP. Thus, distinct pathways appear capable to elevate extracellular adenosine in CA1 (NMDA in RAD and 'Ca 2+ ' in OR) and could be involved in the timing-dependent, asymmetric plasticity in CA1.
Hebbian synaptic plasticity is associative and usually pathway-specific, and is therefore assumed to support learning and memory storage better than non-associative global plasticity. The latter can represent neuron-wide changes in synaptic efficacy and intrinsic excitability as confirmed here for CA1 pyramidal cells. Global plasticity was not observed for CA3 pyramidal neurons 22 , which express plasticity differently from CA1 pyramidal neurons 69 . However, neuronal network models often consider interactions of pathway-specific and global plasticity 6,13 . These interactions are considered to increase the repertoire of plasticity mechanisms and thereby the possibilities of learning and memory storage mechanisms. Our finding that synaptic activity in distinct CA1 pathways is capable of asymmetrically regulating global plasticity highlights that individual synapses are not regulated in isolation. The interplay between OR and RAD reflects the interaction of contextual and spatial representations important for episodic memory 70 .
CA1 pyramidal cells were identified by firing pattern and had a resting membrane potential (V rmp ) of −66.6 ± 0.3 mV and input resistance (R in ) of 146.0 ± 4.2 MΩ (n = 185) both measured in current-clamp during baseline. Cells were excluded from analysis if V rmp was more positive than −60 mV at the beginning of the recording, if V rmp changed > 5 mV or if R in changed > 20% during the recording. Overall, R in changes which were monitored with hyperpolarizing pulses (−3 pA or −10 pA; 200-500 ms) leveled out.
Baseline EPSPs were recorded at 0.1 Hz by alternating stimulation between the two pathways (interstimulus interval, 5 s; or 0.3 s in Fig. 5). In some experiments, we stimulated paired-pulses with 50 ms interval (Figs 2, 6 and Supplementary Figure 1). Paired-pulse ratios were initially analyzed to determine pre-or postsynaptic effects of GABA B or adenosine receptor antagonists and to address expression mechanisms of LTP. For the latter, we focused on the coefficient of variation (CV −2 ) of the first EPSP amplitudes (see Data Analysis) and omitted results based on paired-pulse ratios. Following baseline recording (10 min; 5 min in Fig. 5), induction protocols were initiated within 20 min after establishing whole-cell configuration. The action potential (AP) alone induction protocol consisted of triplet APs at 75 Hz induced by 3 ms somatic current injections (~1.0-1.5 nA) with 60 repetitions at 0.1 Hz for 10 min. The pairing protocol consisted of an EPSP evoked at one of the two pathways, i.e. either in stratum radiatum (RAD pairing) or in stratum oriens (OR pairing) 5 ms prior to the triplet APs and was also repeated 60 times at 0.1 Hz for 10 min (10 ms pre-post time in Fig. 1). Following induction, recordings of EPSPs at the two pathways were resumed at 0.1 Hz for 30 min. Without induction, EPSPs remained constant if evoked at 0.1 Hz for 50 min (RAD EPSP, 1.03 ± 0.07; p = 0.65, n = 6; OR EPSP, 1.05 ± 0.04; p = 0.32, n = 6; not illustrated). In one set of experiments (Fig. 3B), we recorded under the same conditions excitatory postsynaptic currents (EPSCs) in voltage-clamp (holding potential −70 mV, liquid junction potential was not corrected) during baseline and following induction, and then switched to current-clamp only during induction. Data Analysis. All experiments were analyzed in Fitmaster (HEKA, Lambrecht, Germany), IGOR Pro version 5 and 6 (Wavemetrics, Lake Oswego, OR, USA) and Microsoft Excel. EPSP peak amplitudes were normalized to the average of the 10 min baseline period (norm. EPSP, mean ± SEM). Statistical analysis was performed in GraphPad Prism Version 5.02 (GraphPad Software, Inc., San Diego, CA, USA). Statistical significance for LTP of EPSP/Cs was tested for the last 10 min of recording ('after') relative to baseline ('before'), using two-tailed paired t-test on the absolute values. Differences in AP firing before and after pairing were tested by a two-tailed paired t-test.
To determine the expression mechanisms of LTP, normalized inverse square of the coefficient of variation (CV −2 ) of EPSP amplitudes during baseline ('before') and 20-30 min after plasticity induction ('after') was plotted against normalized EPSP amplitude (cf. Fig. 11B in ref. 39 ). If paired pulses were stimulated, the first EPSP was used. Except for Fig. 3C (right panel), CV −2 analyses (Figs 2, 3 and Supplementary Figure 1) contain linearly correlated data. The averages are either above or on the line through the origin. Thus, without hints for changes in quantal size, we considered changes in the synaptic release probability P r or in the number of active synapses n. To consider LTP through an increase in n (i.e. EPSP norm = n after /n before ), we used t statistics of linear regression statistics (Igor Pro 6.37) to test if the slope of a linear fit through the origin was significantly different from 1, with the p value p slope given in the figures. If p slope < 0.05, we tested for pure changes in P r , (i.e. EPSP norm = P r after /P r before ). In this case, we can fit normalized CV  As usual, CV −2 = n*P r /(1 − P r ) (cf. Fig. 4e and Supplementary Methods of ref. 38 ). In the figures, P r fits are only illustrated if convergent and if the fit parameter P r before is ~30% as in 71 .