Electrophysiological recording

Slices were transferred to the recording chamber and perfused with aerated ACSF (34-35 °C) at a rate of 1.2 ml/min; drugs were applied through bath perfusion. Cortical neurons were visualized under upright infrared differential interference contrast microscope (BX51WI or BX61WI, Olympus). Unless otherwise stated, we chose to perform recordings from large layer-5 PCs, which were identified by their pyramid-shaped soma, a single thick apical dendrite and an axon projecting toward the white matter. The impedances of patch pipettes for somatic and axonal recordings were 3–5 MΩ and 6–10 MΩ, respectively, when filled with the internal solution (in mM): K-gluconate 145, MgCl₂ 2, Na₂ATP 2, HEPES 10, EGTA 0.2 (286 mOsm, pH 7.2). Biocytin (0.2%) was also added for post hoc avidin staining. To obtain dual whole-cell recording from the soma and the axonal bleb of PCs, we formed recording from the soma first with a pipette filled with an internal solution containing Alexa Fluo-488 (100 μM, green), the axonal structure and the terminal bleb formed during slicing procedures could be visualized 3–5 min later, we could then obtain recording from the bleb using another patch pipette containing Alexa Fluo-594 (50 μM, red). Both recordings were in voltage-clamp mode and the holding potential was −70 mV. Individual action currents (or APs) were elicited by voltage pulses (100 mV, 0.7–1.0 ms, 1–2 Hz) at the axonal bleb, changes in current traces were monitored at both recording sites. In some experiments, axons were recorded in current-clamp mode and stimulated by current pulses.

As soon as dual soma-axon recording was achieved, two-photon laser axotomy was carried out at a location just beyond the axon initial segment (60–80 μm away from the soma) to disconnect the axon and the somatodendritic compartments. Thus backpropagating APs could not reach the somatodendritic compartments and therefore aEPSCs arrived at the dendrites could be unmasked. Imaging of recorded cells and axotomy was achieved under a two-photon laser scanning microscope equipped with a water immersion objective (×40, NA 0.8) and a mode-locked Ti:Sapphire laser (Mai Tai DeepSee, Spectra-Physics) set at 840 nm (repetition rate: 80 MHz; pulse width: 80 fs). We increased the laser intensity and applied brief bleaching pulses (50–200 ms in duration) to the selected axon location until axotomy was achieved, as indicated by a sudden change of the holding current and absence of somatic action currents. Because the dye diffusion was blocked after axotomy, the somatodendritic and the axonal compartments were labeled by two distinct dyes respectively. Imaging data were acquired with Fluoview FV 1200 (Olympus) and further analyzed by ImageJ and MATLAB (MathWorks, USA).

In the Sr²⁺ experiments examining autaptic connections, we added 8 mM SrCl₂ to the ACSF but reduced the concentration of CaCl₂ and MgSO₄ to 1 mM. Trains of 4 voltage pulses (1 ms in duration, 100 mV, 20 Hz) every 20 s were applied to the cell through the somatic recording pipette (filled with the K⁺-based internal solution) to evoke action currents. The presence of desynchronized synaptic currents during and after the train stimulation indicates the existence of autapses. Because axons were cut before branching, PCs with short axons (<80 μm) in these experiments were not included for data analysis.

In another sets of Sr²⁺ experiments examining whether autapses express NMDA receptors, we recorded asynchronous aEPSCs in Mg²⁺-free ACSF but in the presence of 10 μM glycine or 100 μM D-serine. We added 8 mM SrCl₂ to the bath but omitted CaCl₂. Tetrodotoxin (TTX, 0.5 μM) was used to minimize synaptic inputs from other cells and thus pharmacologically isolate the recorded PC from the network. Picrotoxin (PTX, 50 μM) was added to block GABA_A receptor-mediated miniature inhibitory postsynaptic currents (mIPSCs). A Cs⁺-based pipette solution (in mM: CsMeSO₃ 138, CsCl 3, MgCl₂ 2, Na₂ATP 2, HEPES 10, EGTA 0.2; 285 mOsm, pH 7.2) was used to block K⁺ conductances, making the cell electrically more compact, and allowing effective propagation of voltage pulses to presynaptic terminals to evoke autaptic responses. At a holding potential of −70 or −90 mV, we applied voltage pulses (100–200 mV, 10–1000 ms) every 25 s to the soma and monitor the occurrence of asynchronous aEPSCs. Properties of the evoked autaptic events were then compared with those in the presence of 50 μM APV, an NMDA receptor antagonist. Individual non-overlapping aEPSC events and mEPSCs (occurring at baseline before voltage pulses) were included for analysis. To obtain desynchronized EPSCs from recurrent synapses, we delivered single electrical shocks (0.1 ms in duration, 10–20 μA) to the neighboring tissue of the recorded layer-5 PCs under similar condition described above but with reduced concentration of Sr²⁺ (3 mM) and no TTX. These experiments were performed at room temperature.

Paired-pulse stimulation (1 ms in duration for each pulse, interval: 20 ms) was employed to examine the responsiveness of PCs with and without autapses. The first pulse was adjusted to evoke APs reliably. To examine the input–output curve of the second pulse, we adjusted the current amplitude to a start level (I_start) that evoked APs at a probability of ~0.9 in control condition and then generated the input–output curve with decrement or increment of 100 pA. The mid-point current that evokes AP with a probability of 0.5 is considered as the threshold current. At the end of each recording, we switched the bath to Sr²⁺-containing ACSF to examine whether the recorded PC possessed autapses. Note that the blockade of autaptic responses by Kyn is reversible.

Simultaneous whole-cell recordings were obtained from two neighboring PCs with a distance less than 70 μm to examine properties of unitary EPSCs in recurrent synapses. Brief voltage pulses (0.7–1.0 ms, 100 mV) were used to evoke single action currents in the presynaptic PC and induce EPSCs in the postsynaptic PC. The short-term synaptic plasticity was also examined with trains of pulse stimulation every 10 s. Properties of PC–PC EPSCs were compared with those of aEPSCs obtained in axotomy experiments. The onset latency was measured as the time between the peak of presynaptic action current and the onset of EPSCs, and the onset was determined by the crossing of the baseline current and the linear fit of EPSC rise phase. The rise time was measured from 20 to 80% of peak, and the decay time constant was obtained by single exponential fit. To remove the preceding artifacts (i.e., residual backpropagating action currents), we fitted individual aEPSCs with an alpha-synapse function and then measured the rise time and onset latency. To obtain the decay time constant of aEPSC, we fitted the raw traces with a single exponential function.

To examine the effect of autapses on burst firing induced by synaptic stimulation, we placed a bipolar electrode at the border of layer 6 and the white matter and delivered single electrical shocks (0.1 ms in duration) to the tissue every 25 s. The stimulation intensity was adjusted to allow burst firing at a frequency of 100–150 Hz. The bath solution contained PTX and APV for blocking GABA_A and NMDA receptors, allowing synaptic transmission via AMPA/kainite receptors. The K⁺-based pipette solution contained 5 mM BAPTA. The firing frequency was monitored since the formation of whole-cell recording (i.e., membrane break-in).

Dynamic clamp was achieved using CED power 1401 and Signal software. We performed whole-cell recording from the PC soma and adjusted the injected 500-ms step currents to evoke APs at a mean frequency of 14 Hz with an initial instantaneous frequency of ~20 Hz. We then monitored V_m changes and AP generation before and after the insertion of autaptic alpha-synapse conductances (g_aut amplitude: 0–10 nS; τ_aut: 3.3 ms; reversal potential: 0 mV). These artificial conductances were applied after each AP with an onset latency of 1.4 ms (from the time when the AP surpasses 0 mV to the onset of g_aut). We measured the initial ISI and its changes after the insertion of autaptic conductance. To avoid the interference from autapses and synapses, we applied CNQX (20 μM), APV (50 μM), and PTX (50 μM) to block the fast synaptic transmission. In the experiments that examined the coincidence detection of synaptic inputs in PCs, the bath contained APV and PTX but no CNQX. To avoid the effect of intact autapses, we recorded PCs with short axons ( < 80 μm) showing no collaterals. The first AP was induced by current injection (1 ms in duration, 2–3 nA) and followed by g_aut injection. We adjusted the kinetics of g_aut to mimic AMPA-only (τ_aut = 3.3 ms) and NMDA-containing autapses (τ_aut = 10 ms). After an interval of ΔT from intracellular stimulation, a single electrical shock (0.1 ms) was delivered by a glass electrode placing in the neighboring tissue (50–80 μm from the soma of the recorded cell) to evoke synaptic inputs. We varied ΔT from 5 to 150 ms and monitored the probability of the evoked second AP.

Voltage and current traces were low-pass filtered at 10 kHz and sampled at 20 kHz using Spike2 or Signal software (Cambridge Electronic Design). The liquid junction potential was not corrected for V_m values showing in figures and the main text.