NMDA spikes mediate amplification of odor pathway information in the piriform cortex

The piriform cortex (PCx) receives direct input from the olfactory bulb (OB) and is the brain’s main station for odor recognition and memory. The transformation of the odor code from OB to PCx is profound: mitral and tufted cells in olfactory glomeruli respond to individual odorant molecules, whereas pyramidal neurons (PNs) in the PCx responds to multiple, apparently random combinations of activated glomeruli. How these “discontinuous” receptive fields are formed from OB inputs remains unknown. Counter to the prevailing view that olfactory PNs sum their inputs passively, we show for the first time that NMDA spikes within individual dendrites can both amplify OB inputs and impose combination selectivity upon them, while their ability to compartmentalize voltage signals allows different dendrites to represent different odorant combinations. Thus, the 2-layer integrative behavior of olfactory PN dendrites provides a parsimonious account for the nonlinear remapping of the odor code from bulb to cortex.


Introduction
The piriform cortex (PCx) is the main cortical station in olfactory processing. It 36 receives direct odor information from the olfactory bulb, as well as contextual 37 information from higher brain regions, and is thought to be the brain's primary site for 38 odor discrimination and recognition 1 . Given the importance of understanding the cellular mechanisms underlying odor 100 representation in PCx, we revisited the question as to whether pyramidal neuron 101 dendrites in PCx can generate local spikes 21,24,27-31 . We found that robust NMDA 102 spikes can indeed be generated in dendrites of PCx pyramidal neurons, both in layer 103 1a which receives direct LOT input, as well as in deeper layers, and using a model we 104 show that these local spikes can serve to effectively amplify clustered versus 105 distributed LOT inputs forming the basis for a discontinuous receptive field. We also 106 show that supralinear summation of LOT inputs is largely confined to a single 107 dendrite, whereas nonlinear interactions of LOT inputs between dendrites are small. 108 These findings support the idea that a pyramidal neuron in PCx can represent multiple 109 distinct glomerular combinations within its apical dendritic arbor, which fulfills the 110 basic requirements for a discontinuous receptive field 13 . Finally, we show that 111 interactions between LOT and IC inputs are also nonlinear, a fact that will likely be 112 important for understanding the recurrent pattern completion functions of the PCx.

152
The average spike threshold evoked by LOT stimulation and recorded at the soma 153 was 14.1±1.6 mV and the dendritic spike amplitude and area under the voltage curve 154 (hereafter "area") measured at the soma was 27.1 ± 2.4 mV and 3817.8 ± 396.9 155 mV*ms respectively (mean +/-SEM; n=48 cells; stimulation location 276.78 ± 11.45 156 µm from soma).
The average resting membrane potential was relatively hyperpolarized (-80.1 ± 1.43 164 mV), thus in many cases the NMDA spike stayed subthreshold to somatic firing 165 however in other cases we could abserve firing as a result of the NMDA spike. 166 Occasionally we observed spontaneous spike-like events, which resembled 167 synaptically evoked spikes in shape, including a clear inflection at spike initiation 168 (Figure 2g-h). This indicates that the basic circuitry of the piriform cortex can 169 support the initiation of such spikes. The average spontaneous spike amplitude and 170 area were 17.78±2.09 mV and 9941±2640 mV*ms respectively (n=22 spikes).

171
To further study the role of NMDARs in synaptically evoked spikes, we blocked   Using glutamate uncaging, we first tested whether NMDA spikes could be initiated at 202 progressively more proximal sites along the apical dendrites of PCx pyramidal 203 neurons. We found that spikes could not only be generated throughout the LOT-      The second stage question is whether a co-activation of two inputs of comparable 268 magnitude to those used in the experiments above, but split between dendrites, are 269 less effective at driving the cell than the within-branch combination. We found in was weaker still when the bias was delivered to a different branch that was more 276 remote than a sister (6.28 ± 1.88%, n=6 threshold change for the different-branch,

313
To complete the picture, we examined the interaction between a distal LOT input and 314 a proximal IC input either on sister or different branches ( Figure 7). When the bias 315 was provided by an IC input on a sister branch, the nonlinear interaction was evident, 316 though significantly weaker compared to that seen with a same-branch IC bias ( Figure   317 7a-b, e). Threshold reduction was 23.38 ± 2.34% (n=8) for IC locations on sister 318 branches. The interaction was weaker still when the IC bias was on a more distantly 319 related branch, 18.4 ± 2.88% (n=7), approaching the minimal nonlinear interaction   We found a close correspondence between the experimental and modeling results, 341 wherein an LOT bias input activated on the same dendrite produced a much larger 342 threshold-lowering effect than a bias input of the same size (measured at the soma) 343 delivered to the LOT region of a sister or different branch (Supplementary Figure S2). 344 Thus, the model supports our experimental finding that nonlinear synaptic summation 345 of LOT inputs to distal apical dendrites is strongly compartmentalized, with 346 individual apical dendrites acting as well-separated integrative subunits. 347 We also found close correspondence to the experimental data for interactions between 348 LOT and IC inputs on same, sister and different branches (Supplementary Figure S2).

349
The threshold lowering power of an IC bias input on the same branch was somewhat   The 2-layer architecture of a pyramidal neuron in PCx allows it to respond selectively 443 to specific high-order odorant combinationsthose whose LOT activity patterns 444 deliver concentrated (suprathreshold) excitation to at least one apical dendrite -445 without responding to the vast majority of LOT patterns that produce more diffuse, 446 and therefore subthreshold, excitation to multiple branches within the dendritic arbor.

447
This ability to respond to multiple distinct high-order combinations, without 448 responding to re-combinations of the same odor components, may account for this 449 cell type's hallmark physiological property, namely its "discontinuous" receptive field    The softer compartmentalization of IC inputs, and its implications 507 In our exploration of the nonlinear interactions between driver inputs within the LOT 508 and bias inputs delivered either within the LOT or at the IC-receiving regions of the 509 apical tree, we found that the threshold-lowering effects of IC inputs were less well 510 compartmentalized. The effect can be traced to passive cable theory: IC inputs are 511 closer to the branch points where sister and cousin dendrites connect to each other, so 512 that their effects are felt more widely. In quantitative terms, beginning with a 513 "control" input-output curve generated by an LOT input, we found that the threshold-514 lowering power of a second LOT input on the same branch was roughly 10 times that 515 of an LOT input delivered to a different branch. In contrast to this strong 516 compartmentalization, the threshold-lowering effect of an IC bias input on the same 517 branch is only twice that of an IC bias delivered to a different branch (Figure 7e). The 518 observation that IC inputs modulate more globally comes with a caveat, however: it 519 was previously shown that the degree of nonlinear crosstalk between dendritic 520 branches tends to be overestimated in subthreshold summation experiments, 521 compared to a cell operating in the firing regime which enhances subunit 522 independence 50 . This is because the somatic spike-generating mechanism acts as a     Calcium transients were recorded in line-scan mode at 500 Hz.

762
All experiments were performed at 36º C.