Multi-functional feedforward inhibitory circuits for cortical odor processing

Feedforward inhibitory circuits are key contributors to the complex interplay between excitation and inhibition in the brain. Little is known about the function of feedforward inhibition in the primary olfactory (piriform) cortex. Using in vivo two-photon targeted patch clamping and calcium imaging in mice, we find that odors evoke strong excitation in two classes of interneurons – neurogliaform (NG) cells and horizontal (HZ) cells – that provide feedforward inhibition in layer 1 of the piriform cortex. NG cells fire much earlier than HZ cells following odor onset, a difference that can be attributed to the faster odor-driven excitatory synaptic drive that NG cells receive from the olfactory bulb. As a consequence, NG cells strongly but transiently inhibit odor-evoked excitation in layer 2 principal cells, whereas HZ cells provide more diffuse and prolonged feedforward inhibition. Our findings reveal unexpected complexity in the operation of inhibition in the piriform cortex.

. In this paper we focus on 29 feedforward inhibition in the primary olfactory (piriform) cortex, and report that a 30 surprising complexity of inhibitory processing is also a feature of this type of inhibition 31 in the paleocortex. 32 Feedforward inhibition is conventionally understood as an input-tracking 33 mechanism that does not depend on local circuit activity (Tremblay et al., 2016). In 34 the primary neocortex, feedforward inhibition is often mediated by parvalbumin- Here we approach this question by using 2-photon targeted whole-cell patch 65 clamping and functional Ca 2+ imaging to record from visually identified NG and HZ 66 cells in layer 1a of the PCx in vivo. We find that the odor-evoked feedforward inhibition 67 provided by these two types of interneurons is strikingly different: NG cells express a 68 powerful and transient inhibition that begins rapidly after odor onset, whereas HZ cells 69 provide a more diffuse and delayed form of feedforward inhibition. Thus, a multi-70 functional feedforward inhibitory circuit exists in the superficial layer of the PCx, 71 where it is well-placed to generate dynamically complex patterns of inhibition in the 72 distal dendrites of principal cells. 73

Interneurons that provide feedforward inhibition in the PCx can be targeted 'in vivo' 76
In this study we took advantage of the simple architecture of the PCx to examine 77 in isolation only those neural circuits that generate feedforward synaptic inhibition. 78 Afferent input from the olfactory bulb via the lateral olfactory tract (LOT) to the PCx 79 is strictly confined to layer 1a (L1a) of the PCx ( Figure 1A; Neville and Haberly, 2004). 80 Hence, only those interneurons with dendrites concentrated in L1a, where they are able 81 to intercept axons from the LOT, are able to generate feedforward inhibition (red cells, 82 Figure 1A). Two such interneuron types have been identified in the PCx, HZ cells and 83 L1a NG cells (Suzuki and Bekkers, 2010a). Conversely, feedback inhibition is largely 84 mediated by interneurons with dendrites concentrated in deeper associational layers, 85 two prominent types being fast-spiking (FS) and regular-spiking (RS) interneurons 86 (blue cells, Figure 1A). In this paper we focus on the feedforward inhibitory neurons, 87 HZ cells and L1a NG cells (here called 'NG cells' for short), located in the anterior 88  (NG:  105 short, thin, highly branched dendrites; HZ: longer, less branched, often spiny; Figure  106 1B, bottom) and distinctive intrinsic electrical properties ( Figure 1C, D), similar to in 107 vitro (Suzuki and Bekkers, 2010a). Thus, we were confident that we could record from 108 identified NG and HZ cells in vivo. 109

NG and HZ cells respond strongly to odors 111
We applied up to 15 odorants from a variety of chemical functional groups to 112 urethane-anesthetized mice and measured the voltage responses of identified NG and 113 HZ cells in the anterior PCx ( Figure 1E, F). Three features were apparent in these 114 responses: first, both cell types responded to many different odors (i.e. they were 115 broadly tuned); second, regular oscillations in the membrane potential (Vm) often 116 became larger in the presence of odor; and third, HZ cells appeared to respond more 117 slowly to odors than did NG cells. Each of these features is examined in the following 118 sections. 119 120

NG and HZ cells are broadly excited by odors 121
Our palette of odorants was drawn from a variety of chemical functional groups 122 intended to span a large part of 'odor space' (although this term is difficult to define; significantly different from each other (NG: 0.78 ± 0.04, n = 28 cells; HZ: 0.66 ± 0.12, 132 n = 9; p = 0.34, Welch's 2-sample unpaired t-test). The odor-averaged index was also 133 large for both cell types but was significantly smaller for HZ cells (NG: 0.75 ± 0.03, 134 n = 15 odors; HZ: 0.60 ± 0.06, n = 15; p = 0.025, Welch's 2-sample unpaired t-test). 135 Similar results were found by using an alternative pair of measures, the lifetime and 136 population sparseness, which are analogous to the cell-and odor-averaged indices, 137 respectively (Willmore and Tolhurst, 2001;Poo and Isaacson, 2009). 138

Respiration-locked oscillations in Vm are prominent in NG and HZ cells 140
Subthreshold oscillations in Vm were frequently seen in both NG and HZ cells 141 (e.g. Figures 2A and 2B, top, show the same cells as in Figures 1E and 1F responding 142 to ethyl-n-butyrate; gray traces are respiration, with upward spikes indicating onset of 143 exhalation). Expanding the traces in windows before (b), during (d) and after (a) odor 144 application revealed that oscillations in Vm were synchronized to respiration and, at least 145 in the NG cell, the amplitude of Vm oscillations appeared to increase during the odor 146 ( Figures 2A and 2B, bottom). These observations were quantified by excising the 147 segments of Vm that lay between successive positive peaks of the respiration trace, 148 linearly warping them to have the same time axis, then averaging together all such 149 segments within each of the three windows (b, d, a) for each odor. The peak-to-peak 150 amplitude of the average Vm and the time of the positive peak of average Vm, expressed 151 as a phase of the respiration cycle, were plotted for each odor application ( Figure 2C; each triplet of connected points is from each odor; only the data from the NG cell in 153 Figure 2A are shown). For this particular NG cell, the peak-to-peak Vm amplitude 154 increased significantly during the odor (b: 7.7 ± 0.8 mV; d: 11.1 ± 1.2 mV; n = 12, p = 155 0.002, 1-way ANOVA with Tukey contrasts), whereas the phase of the peak was 156 unchanged (b: 0.49 ± 0.02; d: 0.45 ± 0.03, both expressed as a fraction of the respiration 157 cycle; p = 0.47). 158 A similar analysis was done for all NG and HZ cells in our dataset and the 159 summary is shown in Figure 2D. The mean peak-to-peak Vm amplitude increased 160 significantly during odors in NG cells but not in HZ cells ( Figure 2D

NG and HZ cells tend to fire early and late, respectively, following odor onset 177
Next, we turned to the apparent difference in the kinetics of the odor response, 178 with NG cells appearing to be excited more quickly following odor application ( Figures  179   1E and 1F). We confirmed this impression by constructing AP raster plots and To test whether this difference in dynamics could be observed using a less 186 invasive approach, we turned to 2-photon Ca 2+ imaging with the red-shifted indicator 187 Cal-590. NG and HZ cells could be distinguished as before from their soma location, 188 soma shape and GFP fluorescence ( Figure 3B We also took advantage of the Ca 2+ imaging approach to examine the effect of 195 anesthetics on this neural circuit. (Because the surgery to expose the PCx is so invasive, 196 our experiments could not use awake animals.) All of the above experiments used 197 urethane (0.7 g/kg). We repeated the imaging experiment in Figure 3B

Different kinetics of odor-evoked EPSPs and EPSCs in NG and HZ cells 207
Having found that NG and HZ cells differ in their AP responses to odors, we next 208 asked whether similar differences could be observed in the underlying EPSPs and 209 EPSCs. Odor-evoked EPSPs were median-filtered to remove APs and notch-filtered at Smirnov [KS] test). Thus, during odors, EPSPs in NG cells tend to decline from an 233 early peak while EPSPs in HZ cells tend to rise to a later peak. These behaviors are 234 consistent with the odor-evoked AP firing observed in NG and HZ cells (Figure 3). 235

Differences in excitatory synaptic input can explain the odor response differences 237 between NG and HZ cells 238
What cellular mechanisms might explain the different odor response dynamics of 239 NG and HZ cells? Perhaps the simplest explanation is that NG cells receive depressing 240 EPSPs from the olfactory bulb via the LOT, whereas HZ cells receive facilitating 241 EPSPs ( Figure 5A). To test this idea, we first recorded from NG and HZ cells in slices

Synaptic inhibition also contributes to the odor responses of NG and HZ cells 269
Although we have so far focused on excitatory synaptic inputs, a likely 270 contribution of inhibitory inputs cannot be excluded. NG  square 2 x 2 contingency test). Synaptic connections were also comparatively rare 287 between NG cells and between HZ cells (11.8% and 5.3% respectively; Figure 6C). 288 Thus, odor-evoked synaptic inhibition of both NG and HZ cells is mostly likely 289 provided by NG cells after they are excited by input from the LOT. 290 291

Feedforward inhibition alters the synaptic responsiveness of SP cells 292
Lastly, we explored the functional consequences of HZ and NG cell inhibition for 293 one of their major targets, layer 2 superficial pyramidal (SP) cells. Because it is difficult 294 to disambiguate these two types of feedforward inhibition in vivo, we conducted the 295 experiments in slices. 296 We began by eliciting in vivo-like IPSPs in SP cells. Extracellular stimuli were 297 applied to layer 1a in patterns obtained from the in vivo odor-evoked firing patterns of 298 NG and HZ cells ( Figure 7A, red traces), and averaged postsynaptic IPSPs were 299 recorded in SP cells while pharmacologically blocking ionotropic glutamate receptors 300 ( Figure 7A, black traces; average of n = 44 or 51 single traces for 13 or 3 different NG 301 or HZ cell stimulus patterns, respectively, while recording from 6 different SP cells). 302 Two further manipulations were done to make the recordings more in vivo-like. 303 First, NG stimulus recordings were made distant (>400 µm) from the LOT to avoid stimulating HZ cell axons, which are clustered around the LOT (Suzuki and Bekkers, 305 2010a). On the other hand, HZ stimulus recordings were made near the LOT, where a 306 mixture of HZ and NG cell axons were likely excited. Second, the stimulus patterns 307 were linearly warped so that the respiration trace recorded for each stimulus pattern 308 matched a reference respiration trace (shown in gray, Figure 7A, bottom). This was 309 done to preserve any respiration-synchronized structure in the stimulus patterns when 310 averaging across stimuli. Such synchronization is apparent in the respiration-locked 311 oscillations in the averaged IPSPs ( Figure 7A, black traces). These results show that 312 NG cells generate a large, rapid IPSP in SP cells, whereas HZ cells generate a smaller 313 and more diffuse IPSP that persists beyond the end of odor application. 314 In a final series of experiments we examined the effect of these two types of 315 synaptic inhibition on spiking patterns in postsynaptic SP cells. The method described 316 in the previous paragraph was employed, except that a dynamic clamp was used to 317 replay into the SP cell an odor-evoked excitatory postsynaptic conductance (EPSG) that 318 had previously been recorded from an SP cell in vivo. Again, the EPSG and all stimulus 319 patterns were linearly warped to match their respiration traces. When injecting the 320 EPSG alone, an in vivo-like train of action potentials was evoked in the SP cell ( Figure

Difference in odor-evoked EPSP kinetics 405
Our main finding is that odor-evoked firing in HZ cells has a delayed onset, 406 contrasting with the rapid onset in NG cells. We showed that this effect can be at least 407 partly explained by the slower time-to-peak of the odor-evoked compound EPSC in HZ 408 cells ( Figure 5C). Surprisingly, however, these kinetic differences cannot be explained by the properties of short-term synaptic plasticity at LOT synapses, assayed in slices 410 ( Figure 5B). What might be alternative explanations? 411 The LOT is a heterogeneous fiber tract that contains the axons of two distinct 412 types of projection neurons in the olfactory bulb, mitral and tufted cells. On the other hand, NG cells also receive odor-evoked synaptic inhibition, presumably 430 from other NG cells ( Figure 6A), yet do not exhibit delayed odor responses like HZ 431 cells. It is possible that inhibition from NG cells is larger in HZ cells but it was not 432 apparent here because the number of experiments was relatively small. 433

Functional significance of feedforward inhibition in the PCx 435
Feedforward inhibition is generally recognized as an input-tracking mechanism 436 which, in the hippocampus and neocortex, can synchronize spike timing (

Two types of feedforward inhibition 485
We have previously shown in slices that layer 1a NG cells and HZ cells generate 486 slow-rising (2-3 ms) and fast-rising (1 ms) feedforward unitary IPSPs, respectively, in 487 layer 2 principal neurons, leading us to suggest that NG cell-mediated feedforward 488 inhibition is slower and more diffuse than that provided by HZ cells (Suzuki and 489 Bekkers, 2012). In the present report, after taking into account the much slower odor-490 In many experiments the dura was left intact, but in some the dura was carefully 552 removed using a needle and fine forceps. For the targeted patching a coverglass 553 fragment was glued in place over the craniotomy, leaving a gap at one edge for electrode 554 access. For the calcium imaging a gap was not required, and for 'blind' patch clamping 555 a coverslip was not used. After completion of the surgery a small chamber made from 556 a plastic weighboat and dental cement was constructed around the site. The chamber 557 was filled with phosphate-buffered saline to keep the area hydrated and allow 558 immersion of the microscope objective. For all the above procedures, as well as during recordings, the animal was placed on an electrically-heated surface at ~37 o C and was 560 kept hydrated by periodic s.c. injections of normal saline with 2% dextrose. 561 Scientific, Sydney, Australia). The reference electrode was a Ag/AgCl wire inserted 580 under the skin. The patch electrode was advanced rapidly to penetrate the dura, then 581 more slowly to approach the selected cell and obtain a gigaseal whole-cell recording in 582 the usual way (Margrie et al., 2002). For current clamp recordings, bridge balance and 583 capacitance neutralization were adjusted and the cell was allowed to remain at its resting potential. For voltage clamp recordings, series resistance compensation was not 585 used. Cells were included in the dataset if they had a mean resting potential more 586 hyperpolarized than -50 mV and were stable enough to allow the recording of responses 587 to at least 5 odors. In addition, cells had to be unambiguously identified as either NG 588 or HZ cells according to the criteria given in the Results. At the end of the recording 589 an image stack of the cell was acquired in both the red and green channels. . Cell identity was also confirmed by fixing the 606 brain at the end of the experiment and recovering the morphology of the recorded 607 neuron as previously described (Suzuki et al., 2014). 608

'In vivo' functional Ca 2+ imaging 610
Imaging used the red-shifted Ca 2+ indicator Cal-590 AM (AAT Bioquest, 611 Sunnyvale, CA), which was prepared and injected as previously described (Tischbirek 612 et al., 2015). Briefly, the dura was removed and dye (1 mM) was pressure-injected into 613 the PCx at a depth of ~200 µm using a glass pipette (tip diameter ~10 µm). A coverslip 614 was glued over the PCx and imaging commenced >1 hour after injection. Imaging 615 frames were acquired at 30 Hz using a custom-modified B-scope 2-photon microscope 616

Patterned stimulation and dynamic clamp experiments in slices 688
In vivo-like patterns of inhibitory synaptic stimulation in slice experiments 689 ( Figure 7) were adjusted to a common respiration timebase as follows. First, a subset 690 of odor-evoked firing patterns from NG and HZ cells, together with their associated 691 respiration traces, was randomly selected from the full dataset for each cell type. For 692 each subset, one firing/respiration combination was chosen as a reference, and the first 693 upward peak in the reference respiration trace after odor onset was defined as t0. Every 694 other odor-evoked firing pattern in that subset was translated in time to align its first 695 respiration peak after odor onset to t0. Working backwards and forwards from t0, the 696 respiration trace (and associated firing pattern) of each other pattern was linearly 697 warped so the peaks in its respiration trace matched those of the reference respiration 698 trace. These warped stimulus patterns (examples in Figure 7A, red traces) were then 699 used as the trigger to the extracellular stimulator in slice experiments. The same method 700 was used to align the EPSG to the inhibitory stimulation patterns ( Figure 7B). In this 701 case the respiration trace for the EPSG was used as the reference timebase. 702 The EPSG used in Figure 7B was obtained from an EPSC recorded blind in vivo 703 from an SP cell in response to a 3 s-long application of ethyl-n-butyrate. For each 704 neuron the conductance magnitude was adjusted to produce a similar firing rate with 705 the extracellular stimulus switched off, then the stimulator was switched on to record 706 the effect of synaptic inhibition. In order to provide a reference firing rate for 707 normalizing the PSTH ( Figure 7B), a fixed conductance stimulus was inserted at the end of the EPSG, well past the odor period (not visible in Figure 7B). The dynamic 709 clamp was implemented using Igor Pro (Wavemetrics, Lake Oswego, OR).  All statistical analysis was done using R (Version 3.6.0) running under RStudio. 776 Sample sizes were not predetermined using a statistical test; we established that our 777 sample sizes were sufficient from the size and statistical significance of the results, and 778 our sizes are similar to those commonly used in the field. Data collection and analysis 779 were not blinded or randomized, but analysis was automated whenever possible. 780 Results are presented as mean ± standard error of the mean (SEM) with associated exact 781 p value (n = number of cells or cell-odor pairs, as indicated  This example shows the response of a NG cell to ethyl-n-butyrate. A, Raw membrane potential (V m ) (top), and V m after median filtering to remove action potentials (middle). Bottom trace is the concurrently-recorded respiration. Peaks in the respiration trace (corresponding to exhalation) were detected (vertical dashed red lines) and the time between consecutive peaks was designated a 'respiration epoch'. B, Mean amplitude of the median-filtered V m , averaged over each respiration epoch. C, z-scored version of the data in panel B. This was calculated by normalizing each point to the mean and standard deviation of the data points in B that occurred during the baseline period, i.e. during the 8 seconds prior to odor application. If the z-score exceeded a detection threshold of 2.5 (horizontal dashed blue line) at any time during the odor application period, then an odor response was counted. This cell responded to 14 out of 15 odors so its odor response index was 14/15 = 0.93. Each odor was applied only once to avoid habituation. A, Same NG cell as in Figure 2A, except that the respiration trace has been replaced by a normalized series of Gaussians representing the peaks in the respiration cycles (bottom). This was done to allow us to compare the cross-covariance between different cells by eliminating differences in the amplitude or shape of the respiration trace. B, Cross-covariance between V m and the Gaussian-normalized respiration for the cell in panel A, calculated for 3 s-long windows starting 6 s before, 0.5 s after and 16 s after odor onset, respectively. Asterisks indicate the peak cross-covariance found within ±0.3 s (approximately ±1 respiration cycle) of the zero lag time. C, Changes in peak cross-covariance amplitude (left) and lag (right) before (b), during (d) and after (a) odor application for the cell in panel A. Each triplet of connected points represents data obtained when applying a different odor to this cell. D, Summary of average peak cross-covariance amplitude (left) and lag (right) for 22 NG cells and 9 HZ cells. Bars show mean ± SEM (black, before odor; red, during; blue, after). ***, p < 0.001; **, p < 0.01; ns, not significantly different; 1-way ANOVA with Tukey. A, Recording from an S1 NG cell while applying odor (here, lavender; red bar). Trace at bottom is a normalized series of Gaussians representing the peaks in the respiration cycles, as in Figure S2. B, Cross-covariance between V m and the Gaussian-normalized respiration for the cell in panel A, calculated for 3 s-long windows starting 6 s before, 0.5 s after and 16 s after odor onset, respectively. Traces are displayed on the same axes as for the piriform NG cell in Figure S2B to facilitate comparison. Note the much weaker cross-covariance for this NG cell in S1. C, Average peak cross-covariance amplitude (left) and lag (right) for the S1 NG cell in panels A and B (n = 11 trials) compared with the PCx NG cells (n = 22) in Figure S2D. ***, p < 0.001; ns, not significantly different; Welch 2-sample unpaired t-test. Top trace shows the odor application kinetics for our olfactometer, as measured using a photoionization detector (PID). The brief transients at times 0 and 3 s are due to the opening and closing, respectively, of the valve in the olfactometer. Note the delay to odor onset (242 ± 6 ms, n = 9 odors measured), and a similar delay at odor offset. (All data presented in this paper were corrected for this delay.) The 20-80% rise time of the odor arrival was 46 ± 2 ms (n = 9). The lower two traces show example recordings from a NG cell and a HZ cell, plotted on the same time scale as the PID trace. The NG cell response rises as rapidly as the odor onset, whereas the HZ response is delayed. These examples are taken from Figure 1 (1-heptanal). Traces show, from top to bottom, the cumulative effect of each step in the analysis. Each row shows the same trace on two different timescales. Raw trace, response of a NG cell to lavender. Median filtered, same trace after median filtering to remove the action potential. Notch filtered, same trace after notch filtering at 2-4 Hz to remove the respiration-synchronized oscillations in V m . Averaged and fitted, average of responses of this cell to 8 different odorants, each response filtered as above, then fitted to a smooth function of the form m 2 h (superimposed red curve). The mean EPSP time to peak and halfwidth were measured from the fitted curve.