Microcircuits Mediating Feedforward and Feedback Synaptic Inhibition in the Piriform Cortex

Two types of feedforward inhibition can be observed in the upper layers of the PC

We began by focusing on feedforward inhibition. It seemed likely that the interneurons involved (Fig. 1A, IN1) would possess dendritic trees that ramified significantly within layer Ia, where they could intercept excitatory aff fibers. We have previously reported that there are two types of interneurons that fit this description: HZ cells and layer Ia NG cells (Fig. 3C; Suzuki and Bekkers, 2010b). Both of these cell types receive strong excitatory input from the LOT, and their axonal arbors are dense within layer I, where they likely form inhibitory synapses on the distal dendrites of layer II/III principal neurons (Suzuki and Bekkers, 2010b). To test whether these axons could provide feedforward inhibition of principal neurons, whole-cell recordings were made from SL or SP cells while applying extracellular stimulation to the LOT/layer Ia border. These experiments were done in the presence of blockers of excitatory synaptic transmission (10 μm DNQX + 25 μm d-APV) to prevent contamination by feedback inhibition (Fig. 1A, i.e., IN2).

Open in a separate window

Figure 3.

Pair recordings confirm that feedforward inhibition is provided by two classes of GABA-releasing interneurons in layer Ia, HZ cells and NG cells. A–C, Data from the same HZ/SL cell pair (left column) and the same NG/SL cell pair (right column), all located near the LOT. A, Upper traces show the presynaptic AP in the HZ cell (left) or NG cell (right); lower traces show IPSCs (three single episodes superimposed) in the postsynaptic SL cells. Holding potential for the SL cells was +3 mV. B, Upper traces show a train of five APs at 20 Hz applied to the HZ cell (left) or NG cell (right); lower traces show the resultant IPSCs recorded in the SL cells (average of 20 episodes). C, Morphologies of the cells from which the recordings in A and B were obtained. Blue, Presynaptic (HZ or NG) cell dendrites; red, presynaptic cell axon; gray, postsynaptic SL cell dendrites. D, Summary of mean (±SEM) 20–80% rise time of IPSCs, measured for IPSCs obtained using either extracellular stimulation (Far LOT and Near LOT, as in Fig. 2; filled bars) or pair recordings (NG → SL, HZ → SL, as in A and B; open bars). n.s., Not significantly different (p > 0.05).

When recording and stimulating far from the LOT (>400 μm) in coronal slices, a slow-rising synaptic current was recorded in SL and SP cells for a wide range of stimulus strengths (Fig. 2A, left; stimulus range 5–60 μA for different cells; rise-time data for this cell are shown in Fig. 2B, left; mean 20–80% rise time of IPSCs in n = 8 cells, 2.39 ± 0.14 ms). In contrast, when recording and stimulating close to the LOT (<100 μm), stronger stimulation (>15 μA) increasingly recruited a fast-rising synaptic current in addition to the slow-rising response (Fig. 2A, right; arrowheads indicate top of fast-rising current and arrows point to superimposed slow-rising current; rise-time data for this cell are shown in Fig. 2B, right; mean 20–80% rise time of fast current in n = 10 cells, 1.23 ± 0.16 ms; significantly different from slow current, p < 0.001). Both slow- and fast-rising synaptic currents were completely blocked by bicuculline (10 μm; data not shown), confirming that both were mediated by GABA_A receptors. Moreover, both occurred at a brief latency after the stimulus, indicating that they likely arose from a monosynaptic connection (slow: 2.16 ± 0.08 ms, n = 7 cells; fast: 1.61 ± 0.08 ms, n = 8 cells). Thus, the fast- and slow-rising IPSCs were both probably due to the direct stimulation of GABA-releasing axons within layer Ia.

Pair recordings confirm that horizontal and neurogliaform cells produce feedforward inhibition

What could be the explanation for the two kinetic classes of IPSCs (fast-rising near the LOT, slow-rising both near and far)? We have previously reported that HZ cells are strictly localized in layer Ia close to the LOT, whereas NG cells are uniformly distributed throughout layer Ia (Suzuki and Bekkers, 2010b). Hence, we hypothesized that the fast- and slow-rising IPSCs were generated by HZ and NG cells, respectively. This was tested by making connected pair recordings from a presynaptic HZ or NG cell in layer Ia and a postsynaptic SL or SP cell in layer II (Fig. 3A,B; morphologies of the same cells are shown in Fig. 3C). In each case the postsynaptic response was completely blocked by bicuculline (10 μm; data not shown), confirming that it was an IPSC mediated by GABA_A receptors.

As predicted, the IPSCs recorded from these unitary connections were fast-rising when the presynaptic neuron was an HZ cell (Fig. 3A, left; mean 20–80% rise time 1.15 ± 0.18 ms, n = 3) and slow-rising when the presynaptic neuron was an NG cell (Fig. 3A, right; 2.95 ± 0.25 ms, n = 7; p < 0.01). These rise times were not significantly different from the rise times of the fast and slow IPSCs evoked with extracellular stimulation (Fig. 3D). The results suggest that the distinctive synaptic properties of HZ and NG cells can explain the two kinetic variants of feedforward IPSCs.

Note that the extracellular stimulation experiments in the previous section (Two types of feedforward inhibition can be observed in the upper layers of the PC) left open the possibility that we were exciting GABA-releasing axons that sometimes ascend into layer Ia from interneurons located in deeper layers (Suzuki and Bekkers, 2010b). However, the pair recording experiments in this section indicate that axons from deeper neurons are not necessary to account for our findings.

NG → SL/SP cell unitary connections were readily found in our slices (13 connections out of 35 pairs tested; 37% connectivity), while HZ → SL/SP cell connections were less common (3/50; 6% connectivity). These connectivities were significantly different (p < 0.001; χ² 2 × 2 contingency test). The mean postsynaptic IPSC conductance was similar for both HZ and NG cells (HZ: mean conductance, 920 ± 250 pS, n = 3 cells; NG: 680 ± 110 pS, n = 7; p = 0.43). These results, together with the observation that HZ cells are restricted to regions close to the LOT (Suzuki and Bekkers, 2010b), suggest that NG cells provide a more ubiquitous form of feedforward inhibition. The application of trains of presynaptic action potentials (5–100 Hz) showed that both HZ and NG cells produced depressing IPSCs in postsynaptic principal neurons, but the depression was more profound for NG cells. For example, in a 20 Hz train (Fig. 3B) the amplitude of the fifth IPSC, normalized to the amplitude of the first IPSC, was reduced to 18.7 ± 4.0% (n = 7) for NG cells but to only 57.5 ± 3.3% (n = 3; p < 0.001) for HZ cells.

We have previously described three other classes of GABAergic interneurons in the PC (Suzuki and Bekkers, 2010b), one of which, the bitufted (BT) cell, can also extend its dendrites into layer Ia. Thus, it is possible that BT cells could also receive aff excitation and generate feedforward inhibition. However, using extracellular stimulation we previously found that the layer Ia excitation received by BT cells (17 ± 4 pA, n = 7) is considerably weaker than that received by HZ cells (1130 ± 256 pA, n = 6, p < 0.01) and layer Ia NG cells (212 ± 57 pA, n = 4, p < 0.01; all measured at a holding potential of −77 mV; Suzuki and Bekkers, 2010b). Moreover, as shown below, BT cells fire action potentials much more reluctantly following LOT stimulation than do other cell types, and the connectivity from BT cells to layer II principal neurons is weak. For all of these reasons, it is unlikely that BT cells make a significant contribution to feedforward inhibition.

In summary, feedforward inhibition engages two kinds of interneuron, namely, HZ cells, which provide inhibition close to the LOT, and layer Ia NG cells, which provide a more diffuse and slowly rising form of feedforward inhibition throughout the aff layer.

A tissue cut enables isolation of feedback inhibition in deeper layers of the PC

We next turned to feedback synaptic inhibition, which is hypothesized to arise from interneurons found in deeper assn layers (IN2, Fig. 1A). To remove contamination by feedforward inhibition, a cut was made across the LOT and layer Ia (Fig. 4A; the LOT is visible as a dark band at the upper margin of the parasagittal slices used in these experiments). A recording was then made from an SL or SP cell and a stimulating electrode was placed in the LOT either on the same side of the cut (Fig. 4A, left) to stimulate all inputs, or on the far side of the cut (Fig. 4A, right) to eliminate feedforward inhibition within superficial layers. For all recordings, the postsynaptic cell was voltage clamped at +3 mV to minimize the EPSC amplitude, and complete blockade of the IPSC by bicuculline (10 μm) was confirmed at the end of the experiment. In other experiments, addition of glutamate receptor antagonists (10 μm DNQX + 25 μm d-APV) also blocked the IPSC, confirming that it involved feedforward synaptic excitation (not illustrated). Weak (“minimal”) extracellular stimulation was used to excite one or a few axons (Materials and Methods).

First, with the stimulating electrode on the proximal side of the cut (Fig. 4, left column), a complex polysynaptic IPSC was recorded in the SL/SP cell (Fig. 4B, left), similar to that observed without a tissue cut (Fig. 1B). The delay between the stimulus and the foot of all clearly discernable components of this IPSC (arrows, Fig. 4B, left) was >3 ms and appeared to contain at least three distinctive latency components in this example (3 peaks in latency histogram in Fig. 4C, left; overall mean latency 5.41 ± 0.09 ms, n = 107 trials). We hypothesized that the earlier components arose from disynaptic feedforward inhibition following LOT stimulation (IN1, Fig. 1A; not to be confused with the monosynaptic feedforward inhibition in Fig. 2 in which interneuron axons were directly stimulated; such direct-axon stimulation was not present in Fig. 4). We further hypothesized that the longer-latency IPSCs (Fig. 4B, left) resulted from polysynaptic feedback inhibition involving interneurons in deeper layers.

These hypotheses were tested by moving the stimulating electrode to the distal side of the tissue cut to eliminate direct connections within the aff layer (Fig. 4, right column). Now the synaptic response was much simpler, i.e., only the longest-latency IPSCs were present (Fig. 4B, right), occurring with a delay-to-onset >6 ms (Fig. 4C, right; mean latency 7.53 ± 0.06 ms, n = 115 trials; significantly greater than overall mean latency in Fig. 4C, left, p < 0.001). This result suggests that feedforward inhibition can, indeed, be selectively ablated using a tissue cut, leaving a delayed synaptic response that is mediated by feedback inhibition. Similar results were obtained in 13 experiments of this kind, yielding a mean feedback latency of 7.00 ± 0.20 ms. These feedback IPSCs had a mean amplitude of 108 ± 25 pA (with minimal stimulation; mean conductance 1540 ± 360 pS), and their mean 20–80% rise time was 0.60 ± 0.04 ms (n = 13), significantly faster than both forms of feedforward inhibition (cf. Fig. 3D; HZ: 1.23–1.15 ms; NG: 2.39–2.95 ms; p < 0.001).

Stimulus–response plots implicate fast-spiking cells in feedback inhibition

Having identified IPSCs that likely result from feedback inhibition, we next sought to confirm and extend this finding by identifying the interneurons responsible. It is likely that these interneurons would be located in deeper layers of the PC, where they would preferentially receive excitatory input from assn fibers. We have previously reported that there are four main classes of GABAergic interneurons that match this description, namely, fast-spiking multipolar (fMP) cells, regular-spiking multipolar (rMP) cells, BT cells and deep-layer NG cells (Suzuki and Bekkers, 2010b). Our aim was to find which of these were the predominant sources of feedback inhibition.

One of these cell types, deep-layer NG cells, can be ruled out immediately. Recordings of unitary connections between deep-layer NG cells and SL/SP cells (below; see also Fig. 3A, right) showed that the 20–80% rise time of IPSCs generated by NG cells is slow (3.48 ± 0.24 ms, n = 12), much slower than the observed rise time of feedback IPSCs (0.60 ± 0.04 ms, n = 13, p < 0.001; see A tissue cut enables isolation of feedback inhibition in deeper layers of the PC). Hence, we focused on the remaining three types of deep-layer interneurons, fMP, rMP and BT cells.

As illustrated schematically in Figure 1A, we are seeking a type of interneuron (IN2) that is reliably driven by a layer II principal neuron (SL or SP cell) following LOT stimulation. We have recently shown that SL cells receive the strongest aff input from the LOT (Suzuki and Bekkers, 2011; see also Wiegand et al., 2011); hence, it is likely that SL cells drive IN2 to generate feedback inhibition (Fig. 1A). (Further evidence for this statement will be given in Figure 6.) Therefore, in an initial series of experiments we measured the excitability of SL cells in response to LOT stimulation while simultaneously measuring the excitability of different classes of interneuron. We reasoned that interneurons that were as excitable as SL cells under these stimulus conditions were the most likely to be driven by SL cells, and hence the most likely candidates for providing feedback inhibition.

Focusing first on fast-spiking fMP cells, dual whole-cell current-clamp recordings were made from an SL cell and an fMP cell while stimulating the LOT (Fig. 5A, left). Increasing the stimulus strength (numbers above traces) eventually caused each cell to fire one or more APs. The response of each kind of cell to LOT stimulation was quantified by plotting the probability of firing at least one AP versus the stimulus strength (Fig. 5A, right; open symbols, SL cells; filled symbols, fMP cells; averages of n = 5 dual recordings). This analysis showed that fMP cells started to fire at stimulus strengths that were only slightly greater than those required to fire SL cells (mean stimulus current required to generate at least one AP with a probability of 0.5 was 1.47 ± 0.18-fold higher in fMP cells compared with SL cells, n = 5 dual recordings). This result is consistent with the existence of a strong excitatory connection from SL cells to fMP cells.

In contrast, when the same experiment was done with an SL/rMP cell pair (Fig. 5B), it was found that rMP cells were much more reluctant to fire (10.6 ± 3.5-fold higher stimulus current required in rMP cells than in SL cells; n = 7 pairs; significantly different from fMP/SL cell pairs, p < 0.05; Fig. 5B, right). This result suggests that regular-spiking rMP cells are not effectively recruited following LOT stimulation and, therefore, are less likely to contribute feedback inhibition. Similar results were obtained for BT cells (24.3 ± 6.6-fold higher stimulus current required in BT cells than in SL cells; n = 4 pairs; significantly different from fMP/SL cell pairs, p < 0.05; not illustrated). Hence, these results support the idea that fMP cells are responsible for generating most of the feedback inhibition under these stimulus conditions.

Synaptic and firing latencies confirm the involvement of fast-spiking multipolar cells in feedback inhibition

If LOT → SL → fMP connections generate feedback inhibition, the time delay between stimulation of the LOT and the appearance of a feedback IPSC in a layer II principal cell (∼7 ms, Fig. 4C, right) should be consistent with the chain of synaptic delays along this pathway. This idea was tested by measuring synaptic and firing latencies in the proposed circuit.

As noted earlier, SL cells are most likely the first neurons in the feedback inhibitory circuit because they receive the strongest LOT input (Suzuki and Bekkers, 2011; Wiegand et al., 2011). Additional evidence comes from measurements of firing latency. A whole-cell recording was made from either an SL or an SP cell while stimulating the LOT at an intensity just above threshold for eliciting a synaptically evoked action potential (Fig. 6A, left). As we have previously reported (Suzuki and Bekkers, 2006), the latency from the stimulus to the peak of the AP (arrows, Fig. 6A, left) was much briefer in SL cells (black trace) than in SP cells (gray trace). Histograms of this latency (Fig. 6A, right, redrawn from Suzuki and Bekkers, 2011) confirm that the AP latency in SL cells (mean 3.87 ± 0.07 ms, n = 4 cells) is significantly smaller than that in SP cells (8.69 ± 0.13 ms, n = 4; p < 0.001). Critically, the mean AP latency in SP cells (8.69 ms) is greater than the mean latency of feedback IPSCs (∼7 ms; Fig. 4C, right, and red dashed histogram, Fig. 6C, right), i.e., SP cells only fire after the appearance of feedback inhibition. These results confirm that SL cells, not SP cells, must drive feedback inhibition under these stimulus conditions.

The next neuron in the feedback inhibitory cascade (Fig. 1A, IN2) is proposed to be an fMP cell (see Stimulus–response plots implicate fast-spiking cells in feedback inhibition). A whole-cell voltage-clamp recording was made from an fMP cell while applying minimal stimulation to the LOT (Fig. 6B, left). Large, all-or-none EPSCs were observed (mean amplitude, −123.2 ± 29.1 pA, holding potential −77 mV, n = 10 cells). The latency between the stimulus and EPSC onset (arrow, Fig. 6B, left) had a mean value of 4.60 ± 0.02 ms (Fig. 6B, right; combined data from n = 10 cells). Hence, this fMP EPSC appears, on average, 4.60 − 3.87 = 0.73 ms after the mean time of occurrence of an AP in SL cells (3.87 ms; Fig. 6A), consistent with monosynaptic transmission from the SL to the fMP cell. (See also Fig. 7)

Next, a similar experiment was performed, but this time recording from the fMP cell in current-clamp mode. With minimal LOT stimulation, a large EPSP was elicited, and this could reliably evoke an AP in the fMP cell (Fig. 6C, left). The latency between the stimulus and the AP peak (arrow, Fig. 6C, left) had a broad distribution when averaged across different fMP cells (Fig. 6C, right; mean latency, 7.09 ± 0.13 ms, combined data from n = 8 cells). This may be compared with the latency distribution for feedback IPSCs obtained from tissue-cut experiments like those in Figure 4, right (red dashed histogram, Fig. 6C, right; mean latency, 7.00 ± 0.2 ms, combined data from n = 13 cells). Although these two histograms are dispersed, reflecting latency variability between the different cells included in the average, their similarity is consistent with the idea that monosynaptic transmission from fMP cells is able to generate the feedback inhibition in SL/SP cells that we observed in the tissue-cut experiments (Fig. 4, right).

In summary, measurements of synaptic and firing latencies, together with measurements of synaptic strength (Fig. 5), suggest that fMP cells, driven by SL cells, are the most likely candidates for providing feedback inhibition of layer II principal cells following minimal single-pulse stimulation of the LOT.

Pair recordings confirm that layer II principal cells can provide feedforward excitation of layer III fast-spiking cells

If SL cells excite fMP cells to provide feedback inhibition, it should be possible to find unitary SL → fMP cell synaptic connections. Pair recordings showed that such connections were, indeed, present (5 of 31 pairs tested = 16% connectivity; Fig. 7A,B, left; morphologies of the same cells are shown in Fig. 7C, left). The postsynaptic response in the fMP cell was completely blocked by DNQX (10 μm; data not shown), confirming that it was a glutamatergic EPSC. The unitary EPSC in the fMP cell was large (−109.7 ± 47.2 pA at −77 mV holding potential; mean conductance 1425 ± 613 pS, n = 5 pairs) and not significantly different in amplitude from the EPSC measured in fMP cells above (see Synaptic and firing latencies confirm the involvement of fast-spiking multipolar cells in feedback inhibition) using extracellular LOT stimulation (−123.2 ± 29.1 pA; p = 0.81). Moreover, the mean latency from the peak of the AP in the SL cell to the foot of the EPSC in the fMP cell was 0.83 ± 0.10 ms (n = 5 cells), which is similar to the value estimated in the previous section (Synaptic and firing latencies confirm the involvement of fast-spiking multipolar cells in feedback inhibition) using extracellular stimulation (0.73 ms). SL-evoked unitary EPSCs in fMP cells also exhibited short-term depression (e.g., 20 Hz stimulus train, Fig. 7B, left; amplitude of the fifth EPSC normalized to the amplitude of the first EPSC, 26.2 ± 4.0%, n = 5).

We also looked for unitary SP → fMP cell connections to test whether SP cells in layer IIb would also be capable of providing feedforward excitation of fMP cells. Such unitary connections were found (6 of 33 pairs tested = 18% connectivity; not significantly different from SL → fMP connectivity, p > 0.5, χ² 2 × 2 contingency test; Fig. 7A–C, right). Interestingly, these SP cell-generated EPSCs were very similar to those produced by SL cells: the mean amplitude was −111.3 ± 53.7 pA at −77 mV (mean conductance 1445 ± 697 pS, n = 6 pairs; not significantly different from SL cell EPSCs, p = 0.98), and the normalized depression after five stimuli at 20 Hz was 31.8 ± 8% (not significantly different from SL cells, n = 5, p = 0.53). These results suggest that SP cells should be as capable as SL cells of exciting fMP cells to produce feedback inhibition, if all other factors were equal. However, we find here that SL cells preferentially drive feedback inhibition when the LOT is minimally stimulated, because SL cells are more strongly excited by LOT inputs under these conditions (Suzuki and Bekkers, 2011; Wiegand et al., 2011).

Unitary synaptic connections confirm the role of fast-spiking cells in feedback inhibition

Finally, we examined the last step in the proposed feedback inhibitory circuit (interneuron → SL or SP cell) by making recordings from synaptically connected pairs of these neurons. As expected from the extensive axonal arborization of fMP cells in layer II (Fig. 8C, left; Suzuki and Bekkers, 2010b), fMP → SL or SP connections were comparatively common (17 of 53 cell pairs tested = 32% connectivity) and powerful (mean synaptic amplitude at +3 mV, 137.6 ± 31.4 pA, giving a mean conductance of 1965 ± 449 pS, n = 16 pairs; Fig. 8A, left). The 20–80% rise time of IPSCs generated by presynaptic fMP cells was rapid (0.69 ± 0.05 ms, n = 16), as expected for perisomatic basket-type connections, and similar to the rise time measured for feedback IPSCs in the tissue-cut experiments (0.60 ± 0.04 ms; not significantly different, p = 0.18). Trains of fMP IPSCs evoked at 20 Hz depressed moderately during the train (Fig. 8B, left; amplitude of the fifth IPSC normalized to the amplitude of the first IPSC, 52.4 ± 4.0%, n = 9).

In contrast, both rMP → SL/SP and BT → SL/SP connections were less common than fMP connections (rMP: 3/31 = 10%, p = 0.02; BT: 4/44 = 9%, p < 0.01; χ² 2 × 2 contingency test) and, when found, were weaker (rMP: 29.2 ± 8.4 pA at +3 mV, conductance 420 ± 120 pS, n = 3 pairs, p = 0.004, Fig. 8, right; BT: 20.2 ± 0.3 pA at +3 mV, conductance 290 ± 4 pS, n = 4 pairs, p = 0.002, not illustrated). Their IPSC 20–80% rise times were also slower than those found for fMP cells (rMP: 1.56 ± 0.52 ms, n = 3, p = 0.002; BT: 1.54 ± 0.22 ms, n = 4, p < 0.001). Trains of IPSCs evoked at 20 Hz in rMP and BT cells showed similar depression to that seen for fMP cells (normalized amplitude of fifth IPSC: rMP, 61.5 ± 18.6%, n = 3, p = 0.45, Fig. 8B, right; BT: 53.9 ± 10.7%, n = 3, p = 0.38, not illustrated).

Connections from deeper (layers Ib, II, III) NG cells to SL/SP cells were, as in layer Ia, relatively easy to find (19/31 = 61%). However, the rise times of their IPSCs were, like those of layer I NG cells (Fig. 3A, right), very slow (3.48 ± 0.24 ms, n = 12), ruling them out as possible contributors to fast feedback inhibition of the kind observed in the tissue-cut experiments (Fig. 4).

Comparison with other work

Our findings broadly confirm classic work that provided evidence for feedforward and feedback inhibition in the piriform cortex (Satou et al., 1983; Haberly and Bower, 1984; Haberly et al., 1987; Tseng and Haberly, 1988; for review, see Neville and Haberly, 2004). Our results are also consistent with a recent report describing inhibitory microcircuits in the anterior PC (Stokes and Isaacson, 2010). Like us, Stokes and Isaacson (2010) identified feedforward inhibition mediated by interneurons in layer Ia, as well as feedback inhibition provided by fast-spiking interneurons in deeper layers. However, by using our knowledge of the different types of PC interneurons and layer II principal neurons (Suzuki and Bekkers, 2010b, 2011), we have extended their findings in three ways. First, we were able to identify two types of feedforward inhibition. Interestingly, the corresponding inhibition reported by Stokes and Isaacson (2010) appears to be a mixture of our two types. Second, we confirm the involvement of fast-spiking cells in feedback inhibition, but also explicitly address the possible involvement of other neuronal types (regular-spiking, bitufted and neurogliaform cells; Figs. 5, ​,8).8). Finally, we implicate SL cells, rather than generic pyramidal cells, in generating the excitation that drives feedback inhibition following LOT stimulation (Fig. 6). This is consistent with recent reports that SL cells receive stronger afferent input (Suzuki and Bekkers, 2011; Wiegand et al., 2011). Thus, our data provide more explicit detail about the architecture of these circuits.

Another clear difference between our two studies is that Stokes and Isaacson (2010) focused on the short-term dynamics of inhibition by using trains of LOT stimulation at a range of different strengths. In contrast, we mainly used single, weak shocks to the LOT to elicit minimal synaptic responses and avoid complications due to polysynaptic excitation. However, we did use trains when recording from synaptically connected pairs (Figs. 3, ​,7,7, ​,8).8). Our pair recordings confirmed the conclusion of Stokes and Isaacson (2010) that feedforward inhibition depresses during trains (Fig. 3B). More puzzling, however, is their finding that feedback inhibition facilitates during trains of LOT stimulation. This was attributed by them to increased excitatory drive of the feedback fast-spiking cells by pyramidal cells, because the latter receive facilitating input from the LOT. However, we report here that fast-spiking cells are (under our weak-stimulation conditions) mainly driven by SL cells, not SP cells, and the former do not receive facilitating LOT input (Suzuki and Bekkers, 2006, 2011). A possible explanation for the discrepancy is that, by applying stronger stimulation to recruit feedback inhibition, Stokes and Isaacson (2010) also recruited more SP cells than was the case under our minimal stimulation conditions. It will be important to establish which stimulus strength more closely replicates physiological conditions.