Synaptic Output of Individual Layer 4 Neurons in Guinea Pig Visual Cortex

Abstract

More than 90% of geniculocortical axons from the dorsal lateral geniculate nucleus of the thalamus innervate layer 4 (L4) of V1 (primary visual cortex). Excitatory neurons, which comprise >80% of the neuronal population in L4, synapse mainly onto adjacent L4 neurons and layer 2/3 (L2/3) neurons. It has been suggested that intralaminar L4–L4 connections contribute to amplifying and refining thalamocortical signals before routing to L2/3. To unambiguously probe the properties of the synaptic outputs from these L4 excitatory neurons, we used multiple simultaneous whole-cell patch-clamp recording and stimulation from two to four neighboring L4 neurons. We recorded uEPSCs (evoked unitary synaptic currents) in response to pairs of action potentials elicited in single presynaptic L4 neurons for 102 L4 cell pairs and found that their properties are more diverse than previously described. Averaged unitary synaptic response peak amplitudes spanned almost three orders of magnitude, from 0.42 to 192.92 pA. Although connections were, on average, reliable (average failure rate, 25%), we recorded a previously unknown subset of unreliable (failure rates from 30 to 100%) and weak (averaged response amplitudes, <5 pA) connections. Reliable connections with high probability of neurotransmitter release responded to paired presynaptic pulses with depression, whereas unreliable connections underwent paired-pulse facilitation. Recordings from interconnected sets of L4 triplets revealed that synaptic response amplitudes and reliability were equally variable between independent cell pairs and those that shared a common presynaptic or postsynaptic cell, suggesting local perisynaptic influences on the development and/or state of synaptic function.

Introduction

Visual information is transmitted from the retina through the mammalian LGN_d (dorsal lateral geniculate nucleus) (Sherman and Guillery, 1996; Sherman, 2001) to primary visual cortex (V1). Thalamocortical projections primarily innervate dendrites within layer 4 (L4) (Friedlander and Martin, 1989; Friedlander et al., 1991; Stratford et al., 1996; Castro-Alamancos and Connors, 1997; Gil et al., 1999; Erisir and Dreusicke, 2005). Approximately 90% of the geniculocortical synapses terminate onto L4 neuron dendrites (Gilbert and Wiesel, 1979; Peters and Payne, 1993; Peters et al., 1994; Stratford et al., 1996; Binzegger et al., 2004). Intracortical information processing has serial and parallel components (Hubel and Wiesel, 1962; Gilbert, 1983, 1993; Callaway, 1998; Martin, 2002; Thomson and Bannister, 2003), through vertical interlaminar and horizontal intralaminar projections (Lund et al., 1979; Ahmed et al., 1994; Lübke et al., 2000). Information is then sent from V1 to other cortical areas primarily by L2/3 pyramidal cells (Horton, 1984; Felleman and Van Essen, 1991; Gilbert, 1993; Schmidt et al., 1997; Sincich and Blasdel, 2001).

There is considerable synaptic integration within L4 (Gilbert, 1983; Martin, 2002), including both horizontal excitatory and inhibitory projections (Hirsch and Martinez, 2006). Although the properties of intralaminar synaptic connections in visual cortex within supragranular and infragranular layers have been studied (Markram et al., 1997; Mercer et al., 2005; Song et al., 2005; Feldmeyer et al., 2006; Hardingham et al., 2007), less is known about the functional synaptic properties within L4 in visual cortex (Stratford et al., 1996; Tarczy-Hornoch et al., 1999), although these connections represent a substantial fraction of the synapses. It has been estimated that ∼45% of the dendritic spines of L4 excitatory neurons receive input from adjacent L4 excitatory cells (Ahmed et al., 1994; Bruno and Sakmann, 2006), compared with ∼10% that receive input from thalamocortical fibers (Ahmed et al., 1994). Moreover, L4 has been hypothesized to act as an amplifier of thalamocortical inputs (Martin, 2002). To directly evaluate the synaptic outputs of individual L4 cells (unitary synaptic connections), we used multiple simultaneous patch-clamp recording from two to four L4 neurons. Over 100 pairs of such unitary synaptic connections were assessed this way. Quantal analysis was applied in a subset of experiments. We also recorded from sets of triplets of synaptically connected L4 cells and from a smaller group of L4 to supragranular connected cell pairs.

There was considerable variability among unitary synaptic connections between L4 excitatory neurons. Individual synaptic strengths ranged from 0.42 to 192.92 pA. Connections were, on average, reliable, with an average failure rate of 25% but a subset of weaker connections (<5 pA average peak amplitude) had failure rates >50%. Most connections (79%) were best fit best by a quantal or binomial model of synaptic transmission. Connections between L4 and supragranular layers were similar to those within L4 in their reliability and strength. Multiple release sites on the axon of a common presynaptic L4 neuron behaved independently on individual simultaneously recorded trials, whereas properties of synaptic connections from or onto a common cell were no more functionally similar than connections between separate cell pairs.

Materials and Methods

Slice preparation.

All experiments were performed according to guidelines of the Institutional Animal Care and Use Committees of Baylor College of Medicine and followed the National Institutes of Health's Guidelines for the Care and Use of Laboratory Animals. Tricolor guinea pigs of ages postnatal day 6 (P6) to P14 were deeply anesthetized with a mixture of 0.85 mg/kg ketamine and 0.15 mg/kg xylazine and decapitated. The brain was rapidly removed and cooled for at least 90 s in artificial CSF (aCSF) containing the following (in mm): 124 NaCl, 2 KCl, 2 MgSO₄, 2 CaCl₂, 1.25 KH₂PO₄, 26 NaHCO₃, and 11 dextrose, and saturated with 95% O₂/5% CO₂ to a final pH of 7.4. Coronal slices of the visual cortex were cut at 300 μm with a Vibratome 1000 Plus (Technical Products International). Slices were incubated at 33–35°C for 45–60 min in a holding chamber in a heated water bath (Thermo Fisher Scientific) and then transferred to a room temperature bath until being transferred to a submerged recording chamber (Warner Instruments) and perfused continuously at 2–3 ml/min with oxygenated aCSF at 32–34°C. Neurons were visualized with a Zeiss upright microscope (Axioskope FS1; Zeiss) equipped with an Achroplan 40×, 0.8 numerical aperture water-immersion lens set up for differential interference contrast (DIC) microscopy.

Glass micropipettes (Corning 7056 glass; outer diameter, 1.5; inner diameter, 1.12; A-M Systems) were pulled on a vertical puller (PP-830; Narishige) to an open tip resistance of 2.5–4.0 MΩ and filled with a pipette solution containing the following (in mm): 115 K-gluconate, 20 KCl, 10 HEPES, 4 NaCl, 4 Mg-ATP, 0.3 Na-GTP, and 4 phosphocreatine-Na, with the pH adjusted to 7.4 by KOH. Osmolarity was adjusted to 280–290 mOsm with mannitol. Biocytin (0.3–0.5%; Sigma-Aldrich) was added to this solution to identify recorded cells and confirm their position. The relative positioning of recorded cells was noted to allow for matching of anatomical and electrophysiological data. After the experiment was completed, the slices were transferred to 4% paraformaldehyde at 4°C. After overnight fixation, sections were incubated in an avidin–biotin–HRP complex overnight at 4°C with 0.1% Triton X-100 (Vector Elite; Vector Laboratories) and developed using DAB (diaminobenzidine) (Vector Laboratories). The tissue was counterstained with cresyl violet and coverslipped with distrene plasticizer xylene (EM Science). Laminar location of the cells was then determined. Whenever a pair of biocytin-stained presynaptic and postsynaptic neurons were unequivocally identified in the slice and matched with the corresponding electrophysiological recording, their axonal and dendritic arbors were reconstructed using Neurolucida software (MBF Bioscience). Putative synaptic contacts were identified with light microscopy with a high-magnification (100×) oil-immersion objective as the close juxtaposition of a presynaptic bouton on an axonal branch belonging to an identified presynaptic neuron and a dendritic branch or spine belonging to a corresponding postsynaptic neuron. The number of such putative synaptic contacts was counted, and the dendritic distance from the putative contact to the soma of the postsynaptic cell was also calculated. These estimates at the LM level are a maximum number of potential contacts as EM analysis would be necessary to verify actual synapses.

Electrophysiology.

All recordings were made with two MultiClamp 700B amplifiers (Molecular Devices), and signals were digitized at 20 kHz with a Digidata digitizer 1440A and recorded using Clampex 9 or 10 software (Molecular Devices). Recordings were filtered on-line at 4 kHz with a four-pole Bessel low-pass filter. Layer 4 was identified under light and DIC microscopy on the basis of its differential opacity to transmitted light and the smaller size and greater density of L4 somata compared with L5 somata. Cells with membrane potentials more positive than −60 mV and recordings with high access resistance (>40 MΩ or >20% the value of the input resistance for that cell, whichever was lower) were discarded. After a pair of cells was patch clamped, one of them was held under current clamp and pairs of action potentials were elicited with a 30 ms interspike interval with two 5 ms square-pulse depolarizing current injections (typically 300–800 pA) at 0.2 Hz. The putative postsynaptic cell was held under voltage clamp at −70 mV. We chose to record the unitary synaptic currents under voltage clamp (vs recording synaptic potentials under current clamp) for better signal-to-noise and detection of small events for quantal analysis. If an evoked postsynaptic unitary current (uEPSC) response was not apparent immediately, at least 50 (and often >100) trials were collected, spike-trigger averaged phase locked to the peak of the action potential in the putative presynaptic cell, and examined again. If a response was detected, data collection proceeded for at least 10 min. If no response was observed in response to the first or second presynaptic action potential, the configuration was reversed with the original putative postsynaptic cell tested for input onto the original putative presynaptic cell. If a response was not observed in either direction, the cells were considered not functionally connected and a third cell was patched and examined as a potential presynaptic and postsynaptic partner to the other two. If necessary, a fourth cell was patched and tried as many times as necessary until at least a synaptic connection was found. Occasionally, multiple synaptic connections were found and simultaneously (if possible) or sequentially recorded. Connections were analyzed off-line by placing a 0.4 ms window (noise measurement) on the postsynaptic current trace 2–3 ms before the action potential onset in the presynaptic channel and a second window, time-locked to the first one, around the peak of the average uEPSC response (EPSC measurement). Latency was measured from the presynaptic action potential peak to the onset of the uEPSC deflection as defined by 10% of the rise time to uEPSC peak. Half-width of the uEPSC trace, 10–90% rise time and decay time of the synaptic connection were analyzed on a trial-by-trial basis and then averaged after discarding trials containing synaptic failures. Several methods were used to estimate failures (supplemental Fig. S1, available at www.jneurosci.org as supplemental material). Quantal analysis was the most rigorous method for failure rate estimation, but since it was limited to connections with ≥100 trials, we used two other measures of failure rate as well: (1) the proportion of trials with uEPSC amplitudes <2 SDs of the noise and (2) double the proportion of trials with positive uEPSC measurements (assuming that noise measurements are symmetrically distributed around 0). We compared these estimates with those from quantal analysis for those pairs with >100 recorded trials and found that the first method resulted in a significant overestimation of failure rates (p < 0.01, Wilcoxon's test) (supplemental Fig. S1, available at www.jneurosci.org as supplemental material), whereas the second one did not (p = 0.18, Wilcoxon's test). Therefore, unless otherwise noted, we used the latter method. Analyses were performed in Clampfit (Molecular Devices) and R 2.71 (R Development Core Team; Foundation for Statistical Computing, Vienna, Austria).

Differences in synaptic parameters in triplet recordings.

In a subset of experiments, we recorded from pairs of connected cells with a common presynaptic (divergent triplets) or postsynaptic (convergent triplets) cell. We computed the differences in synaptic parameters between pairs of connections in divergent or convergent triplets. We compared uEPSC strength (averaged peak amplitude of all trials including failures), failure rate (see above), and uEPSC potency (averaged peak amplitude of only successful synaptic transmission trials excluding failures). These differences were statistically compared with a similar number of differences obtained by randomly selecting pairs of strength and failure rate values from our whole L4–L4 database. This statistical comparison between the differences in triplet pairs and differences between random pairs of connections was repeated 10⁵ times, and probability of obtaining a significant difference (p < 0.05) between the two sets was recorded.

Analysis of simultaneous triplet recordings.

In some of the divergent triplets, the three connected cells were recorded simultaneously such that a presynaptic action potential could evoke a postsynaptic response in both of its postsynaptic partners simultaneously. In these experiments, we performed trial-by-trial analysis of synaptic failures by determining the proportion of presynaptic activations in response to which a response could be observed in both of the postsynaptic cells. Subsequently, the observed proportions of coincident successful events [p(r_aUr_b)] were compared with those calculated from the failure rate for each connection assuming independent probabilities in either connection [p(r)_ap(r)_b].

Quantal analysis.

Unbiased quantal analysis methods were used to analyze the PSC evoked by the first action potential in which ≥100 trials were collected. A full account of the analysis methods used in this study has been described in studies by others (Stricker and Redman, 2003; Cowan and Stricker, 2004). Briefly, both the noise and uEPSC measurements were converted into probability density functions (PDFs) using a Gaussian kernel estimator by convolving each measurement with a normal distribution. The noise PDF, which is frequently skewed presumably because of contamination of the measurement by spontaneous currents, was fit by two Gaussian distributions by using an expectation–maximization (EM) algorithm. The resulting fit was considered as a “model peak,” which accounts for intracellular and experimental noise. After current models of synaptic transmission in the cortex, the observed uEPSC response distribution was assumed to be attributable to a real fluctuation of the response between a number of possible discrete amplitudes (because of probabilistic release at an unknown number of release sites) superimposed with recording and intracellular noise. We therefore sought to determine the optimal number of underlying discrete amplitudes that account for the observed distribution. This was achieved first by sequentially fitting an increasing number K of different model peaks using an EM algorithm. Increasing the number of peaks increases the likelihood of the fit at the cost of introducing additional degrees of freedom in the model. Because we sought to obtain the minimal number of peaks that correctly fit the observed distribution without introducing unnecessary complexity, we compared the different models in pairs to determine whether the fits were significantly different using a Wilks statistic computed from the log-likelihoods of the H₀ (the model with the lower number of peaks) and H₁ models. Since this is a non-nested comparison between mixture models with different number of peaks, the Wilks statistic is not asymptotically distributed as a χ² random variable, so we performed Monte Carlo simulations to generate its distribution. By comparing pairs of models, eventually an optimal model of peaks is obtained. This initial fit was the unconstrained model, since no restrictions were imposed on the locations or weights of the peaks. Once this optimal K was obtained, a quantal model (in which the peaks are forced to be equally distanced by a separation q, corresponding to the quantal content of a synaptic vesicle) was fit to the uEPSC PDF and compared with the unconstrained model. Typically, the compared quantal and unconstrained models had a similar number of peaks, and the comparison could be handled with a χ² test instead of a Monte Carlo simulation. If H₀ could not be rejected, a quantal model was assumed. Finally, a third model was tested in which the release sites mediating the connection were assumed to be independent and have the same release probability p (binomial model). Since no valid estimates could be made about the real number of release sites mediating connections with a failure rate of 0, these were not fitted to a binomial model. Again, a Monte Carlo simulation was used to obtain the distribution of the Wilks statistic. In this way, every analyzed synaptic connection was assigned to one of three models (unconstrained, quantal, or binomial) and a different number of synaptic parameters were extracted depending on the model: number of release sites N, failure rate, quantal size Q (for quantal and binomial connections), and release probability p (for binomial connections).

Results

Database

We recorded from 117 L4 to L4 and 9 L4 to supragranular (L2/3, n = 8 or L1, n = 1) cell pairs. Based on several criteria (see below), 28 of the L4–L4 pairs and 2 of the L4–supragranular pairs were excluded from the analysis. Reasons for exclusion were as follows: (1) anatomical reconstruction identified one of the cells to be in L5 (n = 15); (2) one of the cells was a presumptive inhibitory neuron based on fast spikes, lack of spike frequency adaptation in response to a sustained depolarization, and high input resistance (McCormick et al., 1985; Thomson and Deuchars, 1997; Contreras and Palmer, 2003) (n = 3); (3) <50 stable trials were recorded or one of the cells had high access resistance (>20% of the input resistance of the cell) or was otherwise unhealthy (i.e., maintaining voltage clamp at −70 mV required injection of >200 pA or access resistance changed by >20%; n = 12). The remaining 96 connected cell pairs (89 connected L4–L4 cell pairs and 7 connected L4-supragranular cell pairs) form the database used in this study. Reciprocal connections were rare (n = 3 L4–L4 connections) in our study, compared with a previous report (Lübke et al., 2000). There are several possibilities for this difference including the species used (guinea pig vs rat), possible differences between connectivity of spiny stellates (Lübke et al., 2000) and pyramidal neurons (our study), or the search protocol. In our study, once a connection was validated, the responses were recorded for an extended period to perform the quantal analysis (and, in many cases, also were subjected to a plasticity induction protocol for purposes of another study) before the reciprocal connection was tested. Thus, in many cases, we were simply not able to adequately evaluate the reciprocal connection, making our estimates a lower bound on the number of reciprocal connected pairs.

Example results

An example of a pair of synaptically connected L4 neurons that were also successfully filled with biocytin and reconstructed is illustrated in Figure 1. Four cells that were patched, filled with biocytin, and tested for connectivity are visible in the field of view (Fig. 1A, photomicrograph; Fig. 1B, drawing). Of all 12 possible interactions tested (6 sets of putative bidirectional interactions), 3 were functionally connected. Figure 1C shows example individual trial responses (top 25 traces) and the averaged response from 120 trials (second from bottom trace) recorded from the postsynaptic cell (Fig. 1B, green cell) in response to pairs of action potentials (bottom trace) evoked in the presynaptic cell (Fig. 1B, orange cell). The quadruple patch-clamp recording setup is schematized in Figure 1D. The peak amplitudes of the evoked responses to the first spike on individual trials over 10 min (120 trials) are represented as a time plot in Figure 1E. Note that, in this particular example, the unitary EPSC peak amplitudes (uEPSCs) (Fig. 1E, black dots) on individual trials vary between −5 and −20 pA with very few failures of synaptic transmission (Fig. 1E, gray dots). Throughout the study, we refer to the average peak amplitude of the uEPSC evoked by the first action potential for all trials as the “strength” and to the average peak amplitude of the uEPSC evoked by the first action potential only on trials with successful transmission events (excluding failures) [for comparison of methods of failure analysis, see Materials and Methods and supplemental Fig. S1 (available at www.jneurosci.org as supplemental material)] as the “potency.” All four cells in this experiment (including the connected pairs) had pyramidal morphology, consistent with most L4 cells that were successfully recorded and filled with biocytin. Only a very small proportion of the anatomically reconstructed cells had a stellate morphology (2 of 118) (see Discussion). To compare our data from guinea pig visual cortex to results from other studies of L4 cortical neurons in different species and sensory cortical areas and to estimate the functional implications of intralaminar synaptic convergence (see Discussion), we counted the number of dendritic spines on a subset (n = 6) of our cells (mean, 2524 ± 413 SD; n = 6) and found them to be similar to reports from L4 cells in rat barrel cortex (2800) (Lübke et al., 2000; Bruno and Sakmann, 2006) and in cat visual cortex (∼3000 for pyramidal cells) (Ahmed et al., 1994; Binzegger et al., 2004).

Figure 1.

Example experiment. A, B, Photomicrograph (A) and anatomical reconstruction (B) of a set of four biocytin-filled cells that were patched simultaneously and recorded in an acute slice from visual cortex. C, Example traces of individual sequential trials of uEPSCs recorded from the green-colored cell in response to paired action potential stimulation of the orange-colored cell; 25 consecutive individual trials (top) and the averaged response for all 120 recorded traces (middle) are aligned to the presynaptic action potential peak (bottom). D, Diagram showing the recording configuration. All four cells were patched simultaneously, and all possible connections (12) were tested, indicated by the black and gray arrows. In this particular experiment, the green cell was responsive to activation of both the orange and blue cells (example of a convergent triplet), and it was also presynaptic to the orange cell (example of a reciprocal connection) for a total of three unitary connections in this particular multiple patch set, as indicated by the black arrows. The gray arrows indicate tested connections that did not show a postsynaptic response. E, Time plot of uEPSC peak amplitude recordings for the unitary connection shown in C.

Analysis of evoked uEPSCs

Example records illustrating the measurement of uEPSC waveform parameters are shown in Figure 2. The measurement criteria for rise time (10–90% RT), decay time constant (τ_DT), half-width (HW), and latency are indicated for the response evoked by the first presynaptic spike for this L4–L4 cell pair (Fig. 2A). The distributions of these parameters including the peak amplitudes of the evoked uEPSCs, the transmission failure rates, and responses to paired presynaptic stimuli of the 89 connected L4–L4 cell pairs are summarized in Figures 2B–H and 3, E and G. The uEPSCs had fast rise times consistent with excitatory glutamatergic transmission (10–90% rise time average, 1.42 ± 0.90 ms; median, 1.15 ms) (Fig. 2B); the decay phase of the uEPSCs was adequately fit by a single exponential [average decay time constant, τ (τ_DT) = 4.88 ± 3.22 ms; median, 4.30 ms] (Fig. 2C); the average half-width of the uEPSCs was 2.53 ± 1.74 ms with a median of 2.06 ms (Fig. 2D). The average strength (peak amplitudes of all trials including failures) for these L4–L4 unitary connections was 21.18 ± 31.65 pA, with individual values ranging over almost three orders of magnitude, from 0.42 to 192.92 pA. The distribution of strengths, however, is skewed (Fig. 2E), and the median strength was only 7.55 pA. When failures are excluded from the analysis, the average peak amplitude (potency) was 23.65 ± 30.87 pA with a range from 3.03 to 192.92 pA (Fig. 2F); and the median potency was 10.48 pA. For the largest average peak amplitude recorded, strength equals potency of 192.92 pA, indicative of the perfect fidelity of transmission in the strongest synaptic connections (as described in more detail in Fig. 3E below). We analyzed the anatomical distance of the putative synaptic inputs of our pairs (supplemental Figs. S2, S4, available at www.jneurosci.org as supplemental material) relative to the rise time of the uEPSCs (see Discussion). We found no significant relationship, either when the average distance from the soma to all of the putative anatomical synaptic sites for each pair was measured (R² = −0.15; p = 0.89; n = 9 pairs) or when the distance of the closest putative anatomical synaptic site (which might dominate rise time) was measured (R² = −0.14; p = 0.95; n = 9 pairs). Interestingly, all of the putative anatomical synaptic contacts were located along basilar dendrites within 150 μm of the soma.

Figure 2.

Waveform analysis. A, Schematic of waveform analysis, showing how the different waveforms parameters were calculated for an example single trial. Latency is calculated from the action potential peak to the 10% of peak uEPSC rising phase; the decay time constant τ (τ_dt) is obtained from a simple exponential fit to the decay phase of the uEPSC; the rise time (RT) is taken as the interval between 10 and 90% of uEPSC peak amplitude; and the half-width (HW) is 50% of the time interval between the rising and decaying phase of the waveform at 50% peak amplitude. The values summed in the histograms in B–F are averages for all trials for each experiment in which a functionally connected pair was recorded (n = 89): RT (B), uEPSC decay exponential constant (τ_dt) (C), uEPSC HW (D), and uEPSC peak amplitude including (E) and excluding (F) synaptic failures. G, Scatter plot showing the relationship between the average evoked uEPSCs in response to the first and the second action potential elicited in the presynaptic cell (interstimulus interval, 30 ms) for all connections (n = 89) (R² = 0.97; p < 0.01). H is an expanded view of the area of the scatter plot in G ranging from 0 to 10 pA, corresponding to connections with smaller EPSC peak amplitudes.

Figure 3.

Relationship of uEPSC strength and failure rate. A, B, Example traces (A) and time plot (B), respectively, of a highly reliable and strong L4–L4 connection. A shows 21 example consecutively recorded trials (top; blue), the average of all recorded trials (middle; blue), and an example action potential evoked in the presynaptic cell (bottom; black). B shows a time plot of all recorded uEPSC peak amplitudes for the pair shown in A. C and D show a similar analysis for a much weaker, unreliable connection. Failures are shown as gray traces (C) or points (D), respectively, whereas successful events are shown as black traces (C) or points (D). C shows 21 example consecutively recorded traces (top), two average traces from the postsynaptic cell (middle; in dark gray, average trace of all trials including failures; in black, average trace only of the trials with successful transmission), and an example action potential evoked in the presynaptic cell (bottom). E, Distribution of failure rate versus uEPSC peak amplitude including (strength; black points) or excluding failures (potency; red circles) for all L4–L4 connections analyzed (n = 89). uEPSC strength and potency values for each connection are linked by a solid horizontal line for clarity. The two example connections shown in A and C are indicated by the blue and black arrows in E, respectively. F, Superimposed average evoked responses of all trials for the connections shown in A (blue trace) and C (black trace). G, Inset of E showing the pairs of average uEPSCs including (strength; black points) and excluding failures (potency; red circles) for connections with strengths <25 pA.

The paired-pulse behavior of the L4–L4 connections was dominated by depression [paired-pulse depression (PPD)] (data points below the unity line in Fig. 2G,H) (n = 60 of 89; or 67% of the L4 pairs tested), whereas a minority of connections (n = 29 of 89; 33%) showed paired-pulse facilitation (PPF). The correlation between the peak amplitudes of the first and second uEPSC (R² = 0.97; p < 0.01) is illustrated in Figure 2G. The region of smaller amplitudes is expanded for clarity in Figure 2H in which it is evident that PPF occurred primarily for the smaller uEPSCs, whereas connections >10 pA almost exclusively exhibited PPD.

Reliability of connections

Since quantal analysis was limited to connections with >100 recorded trials (n = 75 of 89 L4–L4 connected cell pairs), we used the method of double the number of positive (outward) currents in the evoked uEPSC measurement window as an estimate of failure rate. The two methods for estimating failures (quantal analysis and the method of doubling the number of events with positive outward currents) were not significantly different (26 ± 26% for double proportion of positive events and 19 ± 22% for quantal analysis; p = 0.18, Wilcoxon's test) (supplemental Fig. S1, available at www.jneurosci.org as supplemental material). Synaptic transmission failure rates ranged from 100% (connections in which no response was elicited during the recording period but for which we were able to detect an evoked uEPSC in response to the second presynaptic spike in the paired-pulse paradigm) to 0%. Those connections with perfect fidelity of transmission showed a postsynaptic response every time the presynaptic cell produced an action potential, although this does not imply that glutamate release occurred in response to every action potential at all the release sites mediating the connection. Indeed, quantal analysis indicates that individual release sites are not very reliable (see Fig. 6C), and so the reliability of these connections might be explained by the existence of numerous release sites (see Fig. 5D). Reliable connections were predominant: only 17% of the connections had failure rates ≥50% with an average failure rate of 25 ± 26% (median, 17%). In a subset (21 of 89; or 24%) of the connections, no failures occurred (example in Fig. 3A,B) (strength and potency, 88.90 pA). This was particularly apparent in connections with large average amplitude responses (n = 15 of 17 connections with responses >30 pA); the other two connections with responses >30 pA were still very reliable with failure rates of 1 and 4%, respectively. The average strength (and potency) for connections with a failure rate of 0% was 61.11 pA.

Figure 6.

Quantal analysis results and uEPSC correlation. A–C, Histograms of distribution of quantal analysis parameters across L4–L4 unitary connections: N (A) (number of release sites; n = 75), Q (B) (quantal size; n = 59), and binomial p (C) (n = 29). D, Correlation between strength and number of release sites as estimated by quantal analysis (orange points) and putative number of synaptic contacts as estimated from anatomical reconstructions of biocytin-filled cells (gray points). E, Correlation between unitary synaptic strength and quantal size. Only pairs with failure rates >0 were included in this analysis (see text). F, Correlation between strength and binomial release probability. G, H, Correlation of paired-pulse ratio with connection release probability and binomial release probability, respectively. Two outlier values corresponding to very unreliable cells (failure rates, >95%) have been excluded from the linear fit. I, Correlation of binomial and connection release probabilities.

Figure 5.

Quantal analysis models. A1–A3, PDFs (in black) and fitted models (colored lines) for a single L4–L4 connection, showing the likelihood of the different model fits. B1–B3, Time plots for three different L4–L4 connections that were best described by each of the three different models of synaptic transmission. B1 shows the PDF (black line) and fitted peaks (in orange) of a connection optimally fitted by an unconstrained model. B2, Same for a quantal connection, in which the response peaks are evenly spaced by a quantal amplitude Q. The analysis shown in B2 is for the same connection shown in A1–A3. B3, Same for a binomial connection. C, Proportion of L4–L4 connections that were optimally fitted by unconstrained (n = 16), quantal (n = 30), and binomial (n = 29) models; each model allowed extraction of different sets of synaptic parameters (N, p, Q) as shown on the right. D, Scatter plot showing the relationship between connection failure rates (FR) and number of release sites calculated by quantal (orange points) and morphological (black points) analysis across L4–L4 connections.

For the smaller amplitude connections (0.42 to 25.01 pA; n = 67), the failure rates were higher ranging from 5 to 100% (example illustrated in Fig. 3C,D) (strength, 4.57 pA; failure rate, 41%). More than one-third (n = 31 of 89; 35%) of the L4–L4 connections had strengths <5 pA. The relationship between failure rate and strength and potency for the entire population of connections studied is shown in Figure 3E with the area around the smaller uEPSCs (up to 25 pA) expanded in Figure 3G to show the individual strength and potency as a function of failure rate for each uEPSC. Any particular set of synapses between a pair of connected neurons with moderate to high failure rates may have high potencies (e.g., the average uEPSC amplitude only when successful transmission occurs). The interaction of failure rate and potency gives rise to the overall average strength of the unitary synaptic connections as evaluated over multiple trials. Thus, it is of interest to evaluate these parameters independently to separate their respective contributions. An example of this type of analysis is illustrated in the two bottom traces in Figure 3C in which the averaged response of all trials (synaptic strength with failures included) is shown above in dark gray and the averaged response for only the trials with successful transmission (synaptic potency) is shown below in black. The stronger connection illustrated in Figure 3A responded to paired-pulse inputs with depression (Fig. 3F, blue trace), whereas the weaker connection illustrated in Figure 3C responded to paired-pulse inputs with facilitation (Fig. 3F, black trace).

Latency of unitary responses

To better understand the temporal dynamics of information transfer within the excitatory networks of layer 4, we measured the latency (Fig. 4) of unitary synaptic transmission. Individual trials from pairs with weak connections were too noisy to extract meaningful data on an individual trial basis, so we limited this analysis to pairs with average strengths >10 pA. Latency was measured on a trial-by-trial basis as the time from the presynaptic action potential peak to the 10% level of the rising phase of the evoked response. The grand mean ± SD of the mean latencies of each connected cell pair was 1.30 ± 0.91 ms; the median was 1.12 ms (Fig. 4A), consistent with monosynaptic connections (supplemental Fig. S2, available at www.jneurosci.org as supplemental material). The average uEPSC peak amplitude and average latency were correlated for individual connected cell pairs, such that weaker connections tended to show longer latencies (Fig. 4B) as has been previously described for L5 cortical neurons (Boudkkazi et al., 2007). When the latency and peak uEPSC amplitude values from all trials for each connected L4 cell pair are normalized to the average values for that pair, pooled, and then averaged after binning in normalized uEPSC increments (Fig. 4C), the data for uEPSC peak amplitude versus latency are best fit by a logarithmic decay function [y = −0.32 * log(x) + 0.982; p < 0.01]. The correlation with uEPSC peak amplitude is not limited to the mean latency value as the SD or jitter of the trial-by-trial latency measurements are also dependent on the average size of the uEPSC peak (Fig. 4D). For example, the mean and SD of the latency was highest for small (<20 pA) connections (mean, 2.07 ± 1.02 ms). In contrast, the mean value for pairs with responses >30 pA was 0.85 ± 0.46 ms (p < 0.01, Student's t test). Therefore, large reliable connections (>30 pA) between L4 cells produce responses that precede smaller inputs (<20 pA) by 1.22 ms on average and are less temporally variable. Interestingly, neither connection strength, potency, failure rate, nor latency (Fig. 4) was related to the distance between the presynaptic and postsynaptic somata (supplemental Fig. S3, available at www.jneurosci.org as supplemental material).

Figure 4.

Latency analysis. A, Histogram showing the average latency for all pairs with connection strength >10 pA. Latency was calculated on a trial-by-trial basis and then averaged. B, Scatter plot showing the relationship between connection strength and average latency. C, Relationship between normalized strength and normalized latency for all trials from all pairs with strength >10 pA. For every recorded trial, the peak EPSC and latency values were normalized to the average values for that connection. All trials from all connections were then pooled together and binned in 0.5 normalized EPSC increments, the values for each bin averaged, and the resulting data points fitted by a logarithmic function [y = a*log(x) + b]. Error bars indicate SEM. D, Scatter plot showing the relationship between connection strength and latency SD.

Failure rate and quantal parameters

In our quantal analysis protocol, every analyzed pair was fitted to three different models: unconstrained, quantal, and binomial. The three different models were statistically compared (see Materials and Methods) to find out which most adequately explains the data with the minimal assumptions. Figure 5A1–A3 shows the three different fits to the PDF for an example L4–L4 connected cell pair for which the quantal model was eventually chosen. The unconstrained model, which allows extraction of the failure rate and estimated number of release sites, was the chosen model for 16 of the 75 analyzed cases (21%). Results from an example connection that was optimally fit by an unconstrained model are shown in Figure 5B1. The quantal model allows extraction of quantal size (in picoamperes) that corresponds to a single vesicle of neurotransmitter; this model was chosen for 30 of 75 cases (40%). An example is shown in Figure 5B2. Note the evenly spaced distribution of the modeled peaks, corresponding to the connection quantal size. The binomial model was optimal for 29 (39%) of the studied connections, for which the individual release site probabilities could be extracted. An example is illustrated in Figure 5B3. Figure 5C illustrates the distribution of optimal models for all connections (n = 89), as well as what information (quantal parameters) was extracted for each model.

All quantal models allowed for determination of connection failure rate (supplemental Fig. S1, available at www.jneurosci.org as supplemental material). The failure rate of a connection is partially determined by the number of release sites in the connection, such that connections with a higher number of release sites are more reliable. In an attempt to try to relate the quantal analysis results on the number of release sites modeled to the putative anatomical synaptic connections between cells, we carefully reconstructed the axonal arborizations and dendritic arbors of the functionally connected cell pairs. An example of the results of this analysis is illustrated in Figure 5D showing the relationship of failure rate to the number of putative release sites (black points) as determined by anatomical reconstruction of putative synaptic contacts (supplemental Fig. S4A, available at www.jneurosci.org as supplemental material) and as determined by quantal analysis (orange points). We are very sensitive to the limitations of this approach in terms of potentially grossly overestimating the number of anatomical synaptic connections as the light microscope is only a limited tool for this type of approach (see Discussion). However, since high-quality electron microscopic reconstruction of synapses was not possible because of the whole-cell, submerged brain slice preparation, we were limited to this approach.

Relationship of quantal parameters to uEPSC peak amplitude

The distributions of the three quantal analysis parameters [number of release sites (N), quantal size (Q), and the binomial probability of neurotransmitter release (p)] are shown as histograms in Figure 6A–C. The average number of release sites estimated by quantal analysis was 4.29 ± 2.66 (n = 75), with the majority (59 of 75; or 79%) of connections in the range of one to five release sites, as has been described previously (Feldmeyer et al., 1999, 2002; Silver, 2003; Cowan and Stricker, 2004). The average quantal size as estimated from the 59 pairs best described by the binomial or quantal model (Fig. 6B) was 5.02 ± 1.52 pA. The average binomial probability of release (n = 29) (Fig. 6C) was only 0.41 ± 0.13, suggesting that, despite the reliability of L4–L4 connections (see 3E), individual release sites only respond to approximately two in five action potentials. These quantal parameters were related to the strength of synaptic connections. Greater connection strength was positively correlated with the number of modeled release sites from quantal analysis (Fig. 6D, orange points) (R² = 0.65; p < 0.01), and probability of release (R² = 0.46; p < 0.01) (Fig. 6F), but not to quantal size (Fig. 6E). We considered pairs with failure rates of 0% to be subject to additional errors in quantal size (since our quantal analysis method relies on the existence of a failure peak), and therefore they were discarded from the correlations shown in Figure 6,E and F. Similarly, we observed a positive correlation between synaptic strength and the number of putative synaptic contacts as determined by anatomical reconstruction of biocytin-filled pairs (Fig. 6D, gray points) (R² = 0.87; p < 0.01). Only in one case were we able to perform an anatomical reconstruction of a strong connection with perfect reliability of transmission (failure rate, 0.0%). Unlike in the case of weaker pairs, this one connection was mediated by numerous putative synaptic contacts (n = 16) (supplemental Fig. S4, available at www.jneurosci.org as supplemental material).

Paired-pulse responses

Short-term plasticity characteristics have been suggested to be under presynaptic control (Thomson, 2000; Oertner et al., 2002; Schneggenburger et al., 2002). L4–L4 connections in visual cortex behave similarly, with PPF predominating at connections with release probability <75% and PPD at more reliable connections (R² = 0.50; p < 0.01) (Fig. 6G) (PPR ratios for two cells with failure rate >95% have been excluded from the fit). Short-term plasticity is also correlated with the binomial probability of individual release sites as estimated by quantal analysis (R² = 0.41; p < 0.01) (Fig. 6H), suggesting that the binomial release probability and the overall release probability should correlate, as is the case (R² = 0.59; p < 0.01) (Fig. 6I).

Assay of potential contribution of common cellular elements to synaptic function

Synaptic connectivity within local neocortical networks is not random (Song et al., 2005; Yoshimura and Callaway, 2005; Yoshimura et al., 2005). Thus, it is of interest to know to what degree common anatomical elements within such microcircuits (e.g., a common presynaptic neuron that diverges to innervate multiple postsynaptic targets or a common postsynaptic neuron that is innervated by multiple presynaptic elements) share functional synaptic properties. Moreover, such an analysis can be used to evaluate the potential contribution of a cortical postsynaptic target to the expression of presynaptic properties and vice versa. To address these questions, we performed triple recordings within L4 from a common presynaptic cell to two postsynaptic cells (simultaneous, n = 7, and sequential, n = 2, recordings) and from a common postsynaptic cell with two presynaptic cells (sequential recordings only; n = 7). Indices of presynaptic and postsynaptic function, including strength, potency, and failure rate, were measured for each set of convergent and divergent connections and compared with similar results from randomly selected sets of connected L4–L4 cell pairs.

Divergent triplets recorded simultaneously

In a subset (n = 7 of 9) of divergent triplets, both connections were recorded simultaneously, permitting individual trial-by-trial analysis of the responses evoked in two postsynaptic cells by activation of a single presynaptic cell. One example of a simultaneously recorded divergent triplet is shown in Figure 7A. Figure 7B shows example traces of one such experiment in which the simultaneous responses of the two postsynaptic cells to each action potential elicited in the presynaptic cell are illustrated. The time plots of the uEPSC peak amplitudes in response to the first of the two spikes elicited in the presynaptic cell for each connection are shown in Figure 7, C and D, with the superimposed averaged responses to both presynaptic spikes shown as an inset. Note that cell A (Fig. 7B, top green traces; Fig. 7C, time plot) showed paired-pulse facilitation, whereas cell B (Fig. 7B, bottom blue traces; Fig. 7D, time plot) showed paired-pulse depression. To determine whether the responses at the two sets of connections were functionally linked, we plotted the peak uEPSC amplitude of one connection versus the other on a trial-by-trial basis. Figure 7E shows this analysis for the set of connections illustrated in Figure 7A–C; no correlation was found between the trial-by-trial peak amplitude for either cell (Fig. 7G) (R² = 0; p = 0.33). Similarly, no such correlation was found for any of the seven simultaneously recorded divergent triplets (average R² = 0.009; all values of p > 0.05; n = 7). We also analyzed the proportion of simultaneous synaptic events (Fig. 7F) in each connection in the simultaneously recorded divergent triplets p(r_aUr_b) and compared it with what would be predicted if the two connections behaved independently. If this was the case, the rate of simultaneous failures would be the product of the independently estimated success rate p(r) for each connection p(r)_ap(r)_b. If presynaptic terminals that belong to the same presynaptic cell tend to release neurotransmitter simultaneously, a deviation from this estimation should occur consistently across experiments. We were able to calculate these expected values for five simultaneously recorded divergent triplets. The other two included one pair with a failure rate of 0, prohibiting extracting a meaningful value with this analysis. The expected and observed proportions of simultaneous releases and failures were not significantly different (p = 0.86, Student's t test) (Fig. 7G), suggesting that individual presynaptic terminals in different parts of the axonal tree behave independently. Figure 7H shows the strength, potency, and failure rate for each triplet in which a divergent connection from a common presynaptic cell drives two separate postsynaptic cells, arranged from the pair of connections that are the most similar in strength to the left (triplet number 1; strength_a, 4.08 pA; strength_b, 4.80 pA; normalized difference in strength, 0.08) to the one with the greatest absolute difference in strength to the right (triplet number 7; strength_a, 9.32 pA; strength_b, 53.17 pA; normalized difference in strength, 0.7). There were considerable differences in strength, potency, and failure rate with the maximal differences being 43.85 pA, 43.11 pA, and 43.8%, respectively (average differences were 13.67 ± 14.61 pA, 12.89 ± 13.98 pA, and 20.5 ± 14.89%, respectively).

Figure 7.

Simultaneous recordings from divergent triplets. A, Anatomical reconstruction of three L4 pyramidal cells; one of these cells (black) was presynaptic to the other two (green and blue). Putative synaptic contacts are indicated by green and blue circles for cells A and B, respectively. The time plots for these two connections are shown in C and D. B, Example consecutive traces recorded simultaneously from the common presynaptic cell and its two postsynaptic cells (A, green; B, blue). Each pair of colored traces shows the response in either cell to the same presynaptic stimulation (top). The middle traces show the average uEPSC responses including failures for cells A (green) and B (blue). C and D show the time plots of both simultaneously recorded connections. Synaptic failures are indicated as gray data points. The inset shows the superimposed average postsynaptic trace for all recorded trials (n = 120) for both pairs. p(r_a) and p(r_b) refer to the proportion of successful synaptic transmission event in pairs A and B, respectively. E, Scatter plot showing the lack of a relationship (R² = 0) between the uEPSC peak amplitude simultaneously evoked in both cells in the example triplet in A–C on a trial-by-trial basis. F, Expanded time plot showing 24 simultaneously recorded trials in both cells. Successful synaptic transmission events are shown as green (for cell A) or blue (for cell B) points, and synaptic failures are shown as gray points (for either cell). Trials in which a successful synaptic event was evoked in both cells are indicated by a solid black line, and their proportion over the total number of trials [p(r_aUr_b)] was calculated and compared with the expected proportion of simultaneous events calculated by multiplying the proportion of successful events of both connections [p(r_a)p(r_b)] for all simultaneously recorded divergent triplets in which both pairs showed failures (n = 5) (G). There was no significant difference between the two (p > 0.05, Student's t test). H, Bar plots showing the strength, potency, and failure rates for all pairs of connections (shown as green and blue bars for each connected pair, respectively) in all recorded divergent triplets (n = 9). The boxed values correspond to the example triplet shown in A–D. I, The normalized [(X_A − X_B)/(X_A + X_B)] difference in strength (top), potency (middle), and failure rate (bottom) of the two recorded connections for each divergent triplet (gray points; the circled point indicates the values for the triplet shown in A–C) were compared with differences in strength, potency, and failure rate values for pairs of connections randomly extracted from all recorded connections (n = 89). One hundred such sets are shown; black points indicate sets not significantly different to the actual triplet data, and red points indicate the rare sets that are significantly different (p < 0.05) from the actual triplet data (for details, see Materials and Methods).

To determine whether the connections in a divergent triplet set were more similar between them than expected from random sampling of all L4–L4 functionally connected cell pairs, we performed multiple comparisons between the differences in strength, potency, and failure rate for the divergent triplet database (n = 9) with 100,000 similarly sized (n = 9) sets of pairs of uEPSC and failure rate values obtained from randomly sampling our entire database of L4–L4 connection parameters (see Materials and Methods). One hundred of these randomly generated normalized difference values are shown for comparison in Figure 7I as black (p > 0.05) or red (p < 0.05) points. In all three cases, only a small minority of comparisons (<5%) showed a significant difference between observed and simulated values. Thus, neither overall connection strength, potency, nor failure rate are more similar between pairs that share a common presynaptic cell.

Potential contribution of a common cellular postsynaptic target to synaptic properties

In some experiments (n = 7), we obtained paired recordings from two connections impinging onto a common postsynaptic cell. The anatomical reconstructions of three such neurons and the individual time plots of the peak amplitudes of each pair's evoked uEPSCs are shown in Figure 8A–C. Unlike in the cases of recorded divergent triplets (Fig. 7), the data for the two pairs in the convergent triplet were obtained sequentially, ruling out trial-by-trial analysis of the two uEPSCs. Figure 8D shows the relationship between the strength, potency, and failure rate for pairs of connections, similar to Figure 7H, but in this case for different inputs to a common postsynaptic cell. As for the divergent triplets, the histogram is arranged from the pair of connections with the most similar strength values on the left (triplet number 1; strength_a, 2.74 pA; strength_b, 2.93 pA; normalized strength difference, 0.03) to the one with the greatest absolute difference in strength on the right (triplet number 7; strength_a, 10.26 pA; strength_b, 83.79 pA; normalized strength difference, 0.78). The maximum observed differences in strength, potency, and failure rate were 72.43 pA, 73.43 pA, and 20%, respectively. The average differences in strength, potency, and failure rate across all convergent triplets (n = 7) were 23.71 ± 24.61 pA, 23.59 ± 24.22 pA, and 13.01 ± 0.06%, respectively. The average values were not significantly different from those obtained for divergent triplets (p > 0.05, Student's t test). We compared the difference in strength, potency, and failure rate for the connections in convergent triplets with similarly sized (Fig. 8E) (n = 7) sets of differences obtained from randomly sampled L4–L4 connection parameters (see Materials and Methods). We conclude that, as was the case for divergent triplets (Fig. 7I), convergent triplets do not behave more similarly than expected from random sampling of pairs of L4–L4 connections.

Figure 8.

Convergent triplets. A, Anatomical reconstruction of three L4 pyramidal cells; two of these cells (green and blue drawings) are presynaptic to the third cell (shown in black). Putative synaptic contacts are indicated by green and blue solid circles for cells A and B, respectively. The time plots of the evoked uEPSCs for these two pairs are shown in B and C (evoked unitary responses are indicated by green and blue circles and failures by gray circles). Insets, Averaged postsynaptic traces for all trials (120) including failures. D, Bar plots showing the strength, potency, and failure rates for all pairs of convergent triplet connection studied (shown as green and blue bars for each connected pair of the convergent triplet sets, respectively; n = 7). The boxed values correspond to the convergent triplet shown in A–C. E, The normalized difference in strength (top panel), potency (middle panel), and failure rate (bottom panel) of the two recorded connections for each divergent triplet (gray points) were compared with differences in strength, potency, and failure rate values for pairs of connections randomly extracted from all recorded connections. One hundred such sets are shown; black points indicate sets not significantly different to the data, and red points indicate the rare sets that were significantly different (p < 0.05) (for details, see Materials and Methods).

L4 outputs to L2/3 and L1 cells

Although the primary focus of this study is the intralaminar excitatory unitary synaptic connections between pairs of L4 neurons in the visual cortex, we were also interested in the output properties of L4 neurons onto other postsynaptic interlaminar targets. Thus, in several cases, we recorded from pairs of presynaptic L4 neurons and postsynaptic neurons located in other layers, including layer 2/3 and layer 1. We tested 289 connections between putative L4 presynaptic neurons and postsynaptic cells located in supragranular layers. Of these, we successfully recorded from six L4–L2/3 connected cell pairs and one L4–L1 cell pair. A photomicrograph and anatomical reconstruction of the L4–L1 connection are shown in Figure 9, A and B. The L4–L1 connection was very weak (strength, 0.85 pA; potency, 4.44 pA; failure rate, 74%) and underwent prominent paired-pulse facilitation (PPR, 4.13). Example traces, the averaged evoked response to both presynaptic spikes, and the time plot of the peak PSC amplitudes are shown in Figure 9, C and D, respectively. A photomicrograph and reconstruction of a L4–L2/3 connected cell pair, along with similar recordings are shown in Figure 9E–G. This connection had a moderate strength of 3.88 pA, a potency of 7.99 pA, and a failure rate of 29%. Strengths of L4–L2/3 connections ranged from 2.53 to 60.03 pA (mean, 15.63 ± 22.36 pA), and their failure rates ranged from 0 to 77% (mean, 36 ± 30%). The individual traces, averaged responses, and presynaptic spikes are shown in Figure 9G and the time plot in Figure 9H. These values are not significantly different from those of L4–L4 connections (p = 0.65 and p = 0.51, respectively; Wilcoxon's test). Despite the limited sample size for these interlaminar connections, the relationship between their strength and failure rate is similar to that for L4–L2/3 connections. Figure 9I shows the strength/failure rate values for the L4–L2/3 pairs (black points) as well as the L4–L1 pair (red point) in relation to the similar plots for the larger sample of L4–L4 pairs (light gray points).

Figure 9.

L4 to L2/3 and to L1 cells. A–D, Photomicrograph (A) and anatomical reconstruction (B) of a pair of synaptically connected (L4 to L1) biocytin-filled cells. C, Thirty consecutive example traces (top; red), average of 100 recorded trials (middle; red), and example presynaptic action potential (bottom; black) corresponding to the L4–L1 cell pair shown in A and B. D, Time plot of uEPSC peak amplitude for the same cell pair. E–H, Photomicrograph (E) and anatomical reconstruction (F) of a pair of synaptically connected (L4 to L2/3) biocytin-filled cells. G, Thirty consecutive example traces (top; black), average of 100 recorded trials (middle; black), and example presynaptic action potential (bottom; black) corresponding to the L4–L2/3 cell pair shown in E and F. H, Time plot of uEPSC peak amplitude for the same cell pair. I, Relationship between strength and failure rate for all L4–L2/3 connections (black; n = 6) and L4–L1 connection (red; n = 1). Values for L4–L4 connections (gray points) are included for comparison purposes (n = 89).

Discussion

Summary

L4–L4 excitatory synaptic connections are variable in strength, failure rate, and short-term plasticity behavior. These properties are interrelated: reliable connections (failure rates, <30%; n = 57 of 89; 64%) are strong (average potency, 34.07 ± 36.61 pA), and undergo paired-pulse depression (average PPR, 0.73 ± 0.19), whereas unreliable connections (failure rates, >30%; n = 32 of 89; 36%) are weak (average potency, 8.17 ± 6.19 pA) and undergo paired-pulse potentiation (average PPR, 3.01 ± 6.61). Synaptic properties of multiple cells with a common presynaptic or postsynaptic element are independent, indicative of local influences that are perhaps developmentally regulated (Lohmann et al., 2002; Pelkey and McBain, 2008) or dependent on experience and local synaptic history (Abraham and Bear, 1996; Le Bé and Markram, 2006).

Range of functional properties

Average uEPSC strength and potency ranged from 0.42 to 192.93 pA, and from 3.03 to 192.93 pA, respectively, with 97% of both measures <100 pA. Failure rates ranged from 0 to 100% (the 100% failure rate connection only responded to a second paired-pulse spike). Interestingly, among perfectly reliable connections (0% failure rates; n = 21 of 89), the average strength (and potency) was large, ranging from 14.24 to 192.93 pA with a median of 56.19 pA. Also, connections with no failures generally had large potencies (14.24–192.93 pA; median, 56.19 pA) (Fig. 3E), whereas connections that exhibited failures had small potencies (range, 3.03–60.11 pA; median, 7.60 pA; p < 0.01, Wilcoxon's test). This did not necessarily have to be the case, as neurotransmitter release probability and potency may not be interdependent (Stevens and Wang, 1994). Paired-pulse responses were mostly determined by presynaptic reliability (Fig. 6G). Similar to other synapses (Mennerick and Zorumski, 1995; Hashimoto and Kano, 1998; Oleskevich et al., 2000), unreliable connections exhibited paired-pulse facilitation, whereas reliable connections exhibited paired-pulse depression.

Quantal analysis

We used unbiased quantal analysis (Stricker and Redman, 2003; Cowan and Stricker, 2004), making no a priori assumptions about the behavior of individual release sites. The binomial model assumes that release probabilities across sites are uniform and independent (Kullmann, 1989; Clements, 2003). Although most connections (59 of 75; 79%) were quantal, only a minority (29 of 75, or 39%) were best described by a binomial model. Thus, probabilities across release sites may not always be uniform, perhaps because of differences in presynaptic densities or readily releasable pools (Harris and Sultan, 1995), vesicle fusion machinery (Augustin et al., 1999; Rosenmund et al., 2002; Yang et al., 2002), or activity histories.

Our anatomical (n = 9) and quantal analysis (n = 75) estimates for the average number of release sites agree with each other (4.11 ± 4.56 and 4.29 ± 2.02, respectively) and are close to previous estimates for L4 connections to other L4 [6.6 ± 2.1 (Cowan and Stricker, 2004); 3.4 ± 1.0 (Feldmeyer et al., 1999)] and to L2/3 [4.5 ± 0.5 (Feldmeyer et al., 2002); 5.1 ± 0.9 (Silver et al., 2003)] cells, although our results suggest that the range of N might be greater than previously thought (Fig. 5D). The average individual site release probability for binomial pairs (p) was 41.3 ± 13.4%, suggesting that individual release sites respond, on average, to approximately two of five spikes and that reliability is primarily attributable to factors such as the number of release sites (Fig. 5D). This is in contrast with previous estimates of release probabilities at individual L4–L4 sites [rat S1, average P_r = 79 ± 2% (Silver et al., 2003); cat V1 range of P_r = 68–98% (Tarczy-Hornoch et al., 1999)]. However, in these studies, a binomial release model was assumed and weak unreliable connections were not found. Alternatively, the difference may be attributable to age or species.

Contribution of common presynaptic or postsynaptic cells

By recording from triplets of connected cells, we evaluated the potential contribution of common presynaptic and postsynaptic cellular elements (Figs. 7, 8). We found no evidence for common functional properties among such cellular triads (Figs. 7I, 8E), suggesting that synaptic reliability and potency at connections with a common cellular element are likely to be under local perisynaptic regulation. Although there may be some functional relationship among synapses with common cellular elements during development, these similarities may become obscured because of local plasticity differences (Bramham, 2008; Turrigiano, 2008) such as the recent history of an individual synapse or dendritic segment (Engert and Bonhoeffer, 1997).

Morphology of excitatory L4 cells

Most (116 of 118) V1 L4 excitatory neurons in guinea pig cortex were pyramidal in contrast with the abundance of spiny stellate cells reported in cat (Anderson et al., 1994; Stratford et al., 1996; Tarczy-Hornoch et al., 1999) and macaque V1 (Lund et al., 1979; Saint Marie and Peters, 1985; Yabuta et al., 2001) and in rat S1 (Feldmeyer et al., 1999; Schubert et al., 2003; Feldmeyer et al., 2005). Possible explanations include the following: denser interconnections between pyramidal cells, whole-cell recording selection bias, delayed development of spiny stellates (Vercelli et al., 1992) or species/cortical area differences (McCormick et al., 1985). The difference in connectivity would have to be several orders of magnitude greater than in rat S1 (Feldmeyer et al., 1999; Cowan and Stricker, 2004) and cat V1 (Tarczy-Hornoch et al., 1999; Binzegger et al., 2004) to account for our low proportion (<2%) of spiny stellates. When neurons without an apparent apical dendrite were patched, they had properties consistent with inhibitory interneurons, including high R_in and little spike frequency adaptation (Connors and Gutnick, 1990). It is unlikely that late development of spiny stellates was a factor because we used guinea pigs [which are more precocious than other rodents (Rice et al., 1985)] in their second postnatal week, when spiny stellates are abundant within L4 in other preparations such as rat S1 (Feldmeyer et al., 1999, 2005; Schubert et al., 2003). Since the morphology was mostly similar (pyramidal) for most cells in our study, we were unable to relate functional properties to morphological cell type.

Analyses of uEPSP rise time can indicate relative distance of synaptic inputs from the recording site (Redman, 1990; Magee and Cook, 2000; Williams and Stuart, 2003; Letzkus et al., 2006). However, under voltage clamp, limited effective space clamp restricts the accuracy of similar analysis of uEPSCs (Williams and Mitchell, 2008). All the putative anatomical synaptic contacts that we observed were located on proximal basilar dendrites (distance range, 33–151 μm from the soma; n = 9 connected pairs). Such putative contacts are often anatomically verifiable synapses [93% (Silver et al., 2003); 86% (Lübke et al., 2003)], although other studies suggest the proportion may be much smaller (32%) (DaCosta and Martin, 2007). Because of the narrow range of synaptic distances and the limitations of effective space clamp, we were unable to reliably infer the location of synapses from the uEPSC characteristics alone.

Functional implications

Our results indicate that L4 to L4 neuron synapses have more heterogeneous failure rates than previously reported (Feldmeyer et al., 1999) or to L2/3 neurons (Feldmeyer et al., 2002; Silver et al., 2003). Assuming a postsynaptic R_in of 100 MΩ, E_m of −70 mV, and threshold of −40 mV, a Monte Carlo sampling of L4–L4 connection strengths indicates that, on average, simultaneous activation of 15.7 ± 5.7 presynaptic L4 cells could elicit postsynaptic spiking, assuming linear input summation. In cat V1, if the distribution of inputs onto pyramidal cells is similar to that onto spiny stellates, then they would receive ∼3000 excitatory inputs onto spines (11, 44, 24, and 21% from thalamocortical, L4, L6, and unknown sources, respectively) (Ahmed et al., 1994). In rat barrel cortex (Lübke et al., 2000; Bruno and Sakmann, 2006), L4 neurons have ∼2800 dendritic spines and a similar distribution of excitatory innervation (21, 45, and 34% maximum for thalamocortical, L4, and L6 inputs, respectively). Excitatory L4 neurons in guinea pig V1 have ∼2500 spines (see Results). Based on these anatomical data and our estimate of the number of release sites mediating individual L4–L4 pyramidal cell connections derived from quantal analysis (4.29 ± 2.65) (Fig. 6A), if the relative fraction of inputs to spines are similar to cat V1 or rat S1 (∼45%) and if few of these synapses are on dendritic shafts (∼25%) (Ahmed et al., 1994), the average L4 to L4 neuron convergence ratio would be ∼350:1. Approximately 125 (36%) of these connections would be expected to show relatively high failure rates (>30%). Although weak, if these connections were to fire nearly synchronously, they could drive L4 postsynaptic cells to spike in vivo. For example, highly synchronized thalamocortical inputs, which are considerably less numerous than L4–L4 connections (see above), are efficient drivers of L4 in vivo despite being in a continuously depressed state because of the continuous barrage of synaptic input (Bruno and Sakmann, 2006). Interestingly, unreliable L4/L4 connections undergo PPF in vitro (average PPR, >3) (Fig. 6G), suggesting that they operate in a continuously facilitated state, potentially enhancing their contribution to postsynaptic depolarization. Alternatively, activation of this set of weak connections could contribute to ongoing subthreshold synaptic activity, which has been shown to be important for modulating gain of cortical neurons (Chance et al., 2002), cortical processing of attention (Salinas and Abbott, 1997; Reynolds and Chelazzi, 2004), integration of eye movement commands into visual processing (Moore and Armstrong, 2003), and light adaptation (Ohzawa et al., 1982).