All animal handling procedures were approved by the Institutional Animal Care and Use Committee of the University of Maryland, Incheon National University, and City University of Hong Kong. Animals were treated in accordance with the National Institutes of Health Guide for the care and use of laboratory animals, and the Animal Welfare Act (7 U.S.C. et seq.). Sprague-Dawley rats and C57BL/6 mice (postnatal age: 3–6 weeks) for brain slices were deeply anesthetized with halothane. The brains were quickly removed and placed into chilled (4°C), oxygenated (5% CO₂ and 95% O₂) slicing medium containing (in mM): 212 sucrose, 5 KCl, 1.23 NaH₂PO₄, 26 NaHCO₃, 11 glucose, 1.5 MgCl₂, 2.5 CaCl₂. Transverse slices (300 μm) were cut normal along the septo-temporal plane. Brain slices was then transferred to a holding chamber containing oxygenated physiological saline made up of (in mM): 124 NaCl, 4 KCl, 1.23 NaH₂PO₄, 26 NaHCO₃, 10 glucose, 1.5 MgCl₂, 2 CaCl₂. Also, brain slices of primary auditory cortex were prepared according to our established method (Yang et al., 2011b). After ~1 h recovery, individual slices were transferred to a recording chamber. Oxygenated physiological saline was continuously superfused at a rate of 1.5 ml/min at 32–33°C temperature.

Whole-cell patch recordings were obtained using an Axon instruments Axoclamp 700B Amplifier (Molecular Devices), and recording pipettes had tip resistances of 3–7 MΩ when filled with a solution containing (in mM): 135 K-gluconate, 5 KCl, 1 MgCl₂, 0.02 CaCl₂, 0.2 EGTA, 10 HEPES, 4 Na₂-ATP, 0.3 Na-GTP. Alexa 594 (50 μM) was included in the internal solution for visualization. The pH and osmolarity of intracellular solution were adjusted to 7.3 and 290 mOsm, respectively. pClamp Version 10.2 software (Molecular Devices) or Igor Pro (WaveMetrics) was used for data acquisition. Hyperpolarizing current pulses (10–20 pA; 0.5 s duration; 5–10 s interval) were continuously applied to monitor cells’ input resistance (Rin). During recordings, the access resistance was often monitored; it was typically 20–30 MΩ. Recordings were excluded if Rin changed by >15%.

The procedures for digital holographic photolysis have been described (Yang et al., 2011a). The holographic beam was brought into the optical axis of an upright fluorescence microscope (Olympus BX51) below the epi-fluorescence unit, with a long-pass dichroic mirror. The output beam of a 150 mW, 405 nm diode laser (CNI Laser) was expanded by a beam expander (3×) to fill the short axis of a reflective spatial light modulator (SLM; LCOS Hamamatsu, model X10468–05). The SLM plane was projected onto the back aperture of the microscope objective through a telescope (L1, f1 = 750 mm; L2, f2 = 500 mm). The magnification of the telescope was chosen in order to match the SLM short axis with the diameter of the objective’s back aperture (Olympus, 60×, W 0.9 NA). The undiffracted component (zero order spot) was removed by placing a small (<0.5 mm) anodized metal plate on antireflective coated glass plate at the focal plane of L1. The algorithm for the phase hologram calculation and calibration of the temporal spatial resolution were previously described (Yang et al., 2011a). MNI-caged-L-glutamate was prepared fresh each day at final concentration in physiological solution. Glutamate uncaging pulse duration was set to 0.5 ms. Tetrodotoxin (TTX; 1 μM) was added to examine synaptic responses over a wider range of input intensity except when investigating a role for intrinsic excitability as a gating signal.

The input-output curve of gating and priming responses was tested before STPP experiments as previously established (Santos et al., 2012). The proper intensity of a gating signal was determined based on the input-output curve of gating responses. The laser power (or duration) just before laser intensity giving rise to a dendritic spike was chosen for gating intensity (Figure 1A, 1.3~1.4 ms in this case). Once a dendritic spike occurred, its amplitude usually became saturated. The laser power of priming intensity was adjusted into a level to potentially induce a dendritic spike upon the given gating input. A wide range of priming intensity (0.6~1.2 ms in this case) produced the read-out response efficiently (Figure 1B). This scheme was also applied to experiments with two-photon imaging.

A two-photon imaging system (Prairie Technologies) was used as described previously (Yang et al., 2014, 2016). Briefly, A Chameleon Ultra II laser (Coherent) was tuned to 810 nm for Ca²⁺ imaging. Epi- and trans-fluorescence signals were captured through a 60×, 1.4 NA oil immersion condenser (Olympus). Fluorescence was split into red and green channels using dichroic mirrors and band-pass filters (T560LPXR, ET525/50, ET620/60; Chroma). Green fluorescence (G, Fluo-5F) was captured on an H10770PA-40 photomultiplier tube (PMTs, Hamamatsu). Red fluorescence (R, Alexa 594) was captured with an R9110 PMT. Data were presented as averages of 10 events per site, and expressed as Δ(G/R)/(G/R)_sat*100 (simply, ΔG/G_sat), where (G/R)_sat was the maximal fluorescence in saturating (2 mM) Ca²⁺. The other laser split from a Ca²⁺ imaging laser path was tuned to 720 nm for uncaging of MNI-caged-L-glutamate. Glutamate uncaging pulse duration was set to 0.5 ms. All experiments were done at ~32.5°C temperature.

Fluo-5F and Alexa Fluor 594 were from Molecular Probes. MNI-caged-L-glutamate, Ifenprodil, Ro 25-6981 and TTX were purchased from Tocris (Ellisville, MO). NVP-AAM077 was from Sigma. Imaging dyes and MNI-caged-L-glutamate/blockers were introduced to the pipette and the artificial cerebrospinal fluid, respectively.

All data are shown as mean ± standard error (SEM). An analysis of variance (ANOVA) of Fisher’s PLSD post hoc test was performed for between-group comparison, while a paired t-test was done for within-groups (significance, *P < 0.05; **P < 0.01).

In order to demonstrate STPP on a single basal dendrite, whole cell patch recordings were made on CA1 neurons in hippocampal slices. Alexa 594 (20 μM) in the patch electrode was dialyzed into the dendritic arbor. When a fluorescence signal of a basal dendrite became visible, photolysis sites on spine areas of the dendrite were identified and targeted according to our established method (Yang et al., 2014, 2015). Two sets of photo-stimulation using 3D digital holography were employed to a single dendrite separated temporally and spatially. TTX and MNI-caged-L-glutamate were employed to identify a postsynaptic component of synaptic responses. A priming signal was created by glutamate uncaging to multiple synaptic locations over ~100 μm length (blue circles and cyan trace on the blue arrow, Figure 2A). Subsequently, a gating stimulus was directed at neighboring spots on the same dendrite 200 ms later (green circles and trace on the green arrow, Figure 2A). The gating response was significantly facilitated when preceded by a priming stimulus in all six cells tested (Figure 2A, gating: 3.61 ± 0.30 mV vs. gating + priming: 6.83 ± 0.25 mV, paired t-test, p < 0.001, n = 6 cells). Such facilitation was completely blocked by the application of an NMDAR antagonist, AP5 (1 μM), an NR2B (an NMDAR subunit) antagonist, Ifenprodil (1 μM) and Ro 256981 (1 μM), but not an NR2A antagonist, NVP-AAM077 (0.5 μM) which was known to preferentially block the NR2A subunit of NMDARs at this concentration (Frizelle et al., 2006; Bartlett et al., 2007). With the same gating intensity, facilitation under various antagonists was normalized with that of each control case (Figure 2B, AP5: −0.02 ± 0.10; Ro 25-6981: 0.16 ± 0.13; Ifenprodil: 0.02 ± 0.03; NVP-AAM077: 0.66 ± 0.10, Fisher’s PLSD Post hoc test, *p < 0.05; **p < 0.01). This result confirmed our previous demonstration that postsynaptic facilitation can largely occur through an NR2B-dependent mechanism.

Next, we asked whether, in addition to synaptic input, membrane excitability can serve as a gating signal. A single sinusoidal wave for mimicking membrane fluctuation was elicited by injecting a biased sine wave current, while a priming signal was photostimulated on a basal dendrite (Figure 3A). Action potentials (APs) were triggered only when the membrane fluctuation was coupled with the preceded priming input. Then, a gating signal was generated by injecting a sinusoidal current into the soma to mimic an oscillatory sinusoidal wave (~8 Hz, a frequent rhythmic form in CA1 hippocampus) while a priming signal was photostimulated on dendritic spines. APs were triggered only when the membrane fluctuation was coupled with the priming signal in the absence of TTX (cyan trace, Figure 3B). They were triggered with 71% success rate when the photolysis-induced priming signal co-occurred a postsynaptic depolarization (Figure 3B, 5 neurons of 3 animals, AP, 17; no AP, 7). These data suggest that the gating signal can be read out by either synaptic or intrinsic excitability irrespective of the location of input sources.

In order to test the expression of STPP at the spine level, we stimulated individual spines of a CA1 pyramidal neuron using the two-photon calcium imaging method. We identified spines of a basal dendrite that were co-labeled with Alexa 594 (20 μM) and a dynamic Ca²⁺ indicator Fluo 5F (100 μM, Figure 4A). For a priming stimulus, 10 spines of a basal dendrite (blue circles) were targeted for photolysis, while a gating stimulus targeted five of these spines (green outlines). The gating stimulus alone produced an electrical response in the soma (the green trace on the green arrow head, Figure 4Bi, n = 9 cells) and Ca²⁺ transients on the spines (the green trace on the green arrow head, 5 cells among 9 cells, Figure 4Ci). When a priming stimulus preceded the gating stimulus, both the electrical and calcium signals were remarkably facilitated (Figure 4Bii, gating alone: 2.3 ± 0.35 mV vs. gating/priming: 7.17 ± 0.73 mV, paired-t test, p < 0.001, n = 9; Figure 4Cii, normalized ΔG/G_sat of gating alone: 1.78 ± 0.21 vs. gating/priming: 2.94 ± 0.20, paired-t test, p < 0.001, n = 5). This result suggests that STPP is a response integrated by dynamic activity of individual spines.

Next we asked whether STPP is present in other brain areas. To this end, layer 4 pyramidal neurons of primary auditory cortex were tested with the same experimental strategy as CA1 pyramidal neurons. Ten spines of a cortical neuron were identified and targeted to photolysis of MNI-glutamate (Figure 5A). Similar to the results from hippocampal neurons, the electrical responses in the soma were remarkably facilitated when a priming stimulus preceded the gating stimulus (Figures 5Bi,ii, gating alone: 3.3 ± 0.35 mV vs. gating/priming: 4.86 ± 0.62 mV, paired t test, p < 0.001, n = 12). The facilitation was completely blocked by the application of an NR2B selective blocker, Ifenprodil (1 μM; Figures 5Ci,ii, gating alone: 3.10 ± 0.63 mV vs. gating/priming: 3.20 ± 0.59 mV, paired t test, p > 0.05, n = 5). These results implicate that NMDAR-dependent activity within individual spines mediate STPP in cortical neurons.

Response facilitation in a temporal sequence has been previously suggested as a core model of feature detection and discrimination (Calford and Semple, 1995; Brosch and Schreiner, 1997, 2000; Fortune and Rose, 2000; Wehr and Zador, 2005; Schreiner and Polley, 2014). PPF tested with in vitro brain slices has been suggested as a potential synaptic mechanism of the observed in vivo response facilitation (Brosch and Schreiner, 2000). However, the causal relationship between in vitro PPF and in vivo response facilitation still remains speculative. It is noteworthy that the NMDAR-dependent STPP is similar to the in vivo response enhancement phenomenon in light of the temporal scale (~600 ms) and all-or-none expression pattern (Brosch and Schreiner, 2000; Yang et al., 2014). Hence, we propose that STPP acts along with PPF as a cellular process underlying sensory perception.

STPP is determined by a combination of several factors such as dendritic morphology, the spatiotemporal pattern of synaptic inputs and active intrinsic conductances. STPP was observed when a priming and gating stimulus were co-localized on the same distal dendrite. By contrast, it was not observed when one of two stimuli was separately directed on an apical trunk or adjacent dendrite and two inputs were separated by more than 50 μm even on the same dendrite (Santos et al., 2012). At the spine level, however, we showed that priming and gating synaptic inputs can be neighboring ones for STPP induction as in Figure 2A. Also, a gating stimulus of rhythmic membrane depolarization (which does not necessarily share the same synapses with the priming stimulus) consistently induced STPP, showing independence of input sources (Figure 3). It is notable that different input patterns of a gating stimulus, whether spatially distributed or clustered within a single dendrite, were capable of producing facilitation, which is consistent with a previous study (Losonczy and Magee, 2006). Taken together, these data provide evidence that facilitation likely occurs only when a priming and gating input are situated on the same dendrite unless they are not far apart from each other within the dendrite. Consistent with our findings, there is physiological evidence that single dendrites are tuned to different orientations of object movement in pyramidal neurons of visual cortex (Jia et al., 2010).

STPP can be induced by its characteristic regenerative Ca²⁺ influx through NMDARs (i.e., NMDA spikes). Although the basal level of a synaptic event is largely initiated by non-NMDAR-mediated cationic currents, its amplification begins with NMDA spikes in thin dendrites where most excitatory synaptic events occur. An earlier synaptic input (“priming stimulus”) may not produce a depolarization large enough to remove the Mg²⁺ block of the NMDARs. However, when a subsequent stimulus (“gating stimulus”) which by itself would produce only a modest depolarization is delivered to the same dendrite, its response can be markedly facilitated by promoting Mg²⁺ liberation and initiating an NMDA spike. The question then arises whether STPP can be a false-positive facilitation due to prolonged dendrite depolarization following a priming input (Milojkovic et al., 2004). However, the 200 ms interval between two stimuli, in our study, was sufficient to prevent voltage and calcium deflections from overlapping. Priming responses were fully recovered into baseline as shown in both somatic recording and spine calcium imaging (Figure 4). Also, STPP was induced even at longer intervals up to 600 ms which could guarantee temporal separation (Yang et al., 2014). These results support the STPP mechanism of the “glutamate-bound but Mg²⁺ blocked” state rather than electrical interference by prolonged dendritic depolarization. Those physiological characteristics of STPP are distinct from those of the PPF with respect to location of action (PPF, presynaptic glutamate release vs. STPP, postsynaptic NMDARs), activation pattern (PPF, graded vs. STPP, NMDA spike-driven all-or-none) and spatial integration spectrum (PPF: synapse-specific vs. STPP, dendrite-specific).

Pyramidal neurons are connected to one another in an associational fashion along the longitudinal axis (e.g., CA1-CA1 connection) in hippocampus (Yang et al., 2014) and also along the representation of stimulus features (e.g., isofrequency corticocortical line) in cortex (Imig and Reale, 1980; Lee and Winer, 2008; Oswald and Reyes, 2008; Brown and Hestrin, 2009). The connectivity of the pyramidal-pyramidal neurons and their biophysical properties (holding and amplifying electrical traces) allow signals to pass through a linearly connected brain circuit. Considering that neurons receive streams of information in a temporal sequence, the brain must have a capacity to process a temporal sequence of information for a short period of time; the individual information cannot be properly interpreted without a form of short-term mnemonic buffer. Signals bearing temporal information (e.g., priming, including subthreshold responses) can be, for example, read out by pronounced neuronal activity (i.e., gating, including a certain frequency of brain oscillation), and subsequently converted into a spatial pattern of information along the linear circuit (Hasselmo and Stern, 2006, 2014; Santos et al., 2012). This signal conversion of time-to-space may be necessary for object recognition in the brain.