To test the importance of different temporal structures in PR fire models, we assumed that ~30% of PRs did not respond to a particular odor, and divided the remaining 70% into four equally large groups with the excitation (E) and inhibition (I) sequences induced by the following odors: EEI, EIE, IEI and IIE, the last epoch occurring in the balancing of stimuli (Laurent and Davidowitz, 1994a). Within each group, the times of excitement had alternating sets of active PRs, with new PRs recruited during each LFP cycle (Mazor and Laurent, 2005). Each group began its exciting era with the activation of 70% of its members and 10% were recruited in each of the next three LFP cycles. In this scheme, a maximum of about 30% of PRs were active at some point during odor presentations. This means that, just like when converting glucose into ATP, cells need to find a way to efficiently use their raw materials to make exactly what they need at any given time. And just like ATP, they use feedback regulation to make sure they only produce the amino acids they need at any given time. In men, a positive feedback mechanism is noticed during childbirth, which is caused by the baby pressing against the ovarian wall. The brain receives the sensation of pressure through several nerves, and the pituitary gland is stimulated to produce oxytocin in response. The feedback loop of oxytocin is responsible for contractions of the uterine muscles, which cause the fetus to move closer to the cervix, thereby increasing stimulation. Until the baby is born, the positive feedback loop continues.

Because GGN plays an oversized and central role in the olfactory system, studying its responses to odors can shed light on broader questions of olfactory function. To better understand the determinants of the network of GGN responses to odors, we took it in vivo while providing the animal with a variety of odors, and then developed a large-scale model that included CCIs, GIs, and realistic OLIC inputs to study the types of network activity needed to create the membrane potential models. that we observed in GGN. We have identified two new characteristics in the olfactory network. First, to produce the types of membrane potential models we observed in GGN, the forces of synaptic compounds on KCs must be heterogeneous. Second, our in vivo images of GGN revealed new complex reaction patterns that had not been previously documented, including periods of hyperpolarization that vary with odorant. Although GGN receives mutual feedback from the IG (Papadopoulou et al., 2011), the periods of hyperpolarization could not be explained by a disinhibition of the GI from GGN. Instead, our model predicts that this behavior could occur if IG receives direct excitation from another currently unknown odor-activated pathway in addition to receiving input from GGN.

In addition, our model reproduced emergent features of the olfactory network that were not explicitly programmed into it, such as the appearance of a small portion of CTO spitting at relatively high rates. The olfactory locust system is well studied because it is accessible, robust and relatively simple. In the absence of environmental odors, the olfactory receptor neurons (ORNs) of the antenna are spontaneously active. Odorous substances trigger an increase or decrease in spontaneous ornate trigger rates of ORNs and simple peak patterns with excitation and inhibition sequences that are heterogeneously distributed across the ORN population. These reactions vary depending on the smell and its concentration. Spontaneous activity in ORNs leads to background peaks in NPs (Joseph et al., 2012). NPs respond to the variety of odor-induced reactions in ORNs with sometimes sophisticated fire patterns that are formed by inhibition by local interneurons (LNs) of the antenna lobe (Raman et al., 2010). Peaks in NPs are also brought to rhythmic waves by rapid reciprocal interactions between exciter NPs and inhibitory NLs (MacLeod and Laurent, 1996; Bazhenov et al., 2001).

Odour-induced fire patterns in the NP population are instructive about the identity, concentration and timing of the odour (Laurent et al., 1996; Stopfer et al., 2003; Brown et al., 2005). This information is transmitted by the PRs to the fungal body and the lateral horn. a) Schematic representation of the fungal body network model. Each of the 50,000 KC receives the input of 50% of the 830 PRs modelled as peak trains. All CCIs excite GGN in its rag branch α and receive inhibition from a random sepal branch of GGN. (b–d) The model with simplified and homogeneous cooking patterns in NPs and uniform synaptic forces produces an unrealistic membrane potential in GGNs. (b) Grid diagram of the PN peak trains of the model (67 opposite); Points in each line mark peak hours in a PR. c) Raster diagram of peak trains caused in KCs (397 shown) when all receive identical GGN inhibitor compounds. d) The unrealistic membrane potential in GGNs shows some peaks corresponding to highly synchronized bursts of activity in KCs. Light gray bar: 1 s of odor stimulation.

(e–g) The model in which NP subpopulations exhibit different temporal patterns of peaks during and after odor stimulation produces an unrealistic membrane potential in the GGN. (e) The grids show different patterns of inflammation in different PRs. (f) Model simulation with the PN activity model in panel e) as well as uniform synaptic forces on KCs produce a synchronized peak in KCs and (g) an unrealistic membrane potential in GGNs. Simulation of a model with structured models of PN activity in panel e) and heterogeneous synaptic forces leads to (h) diffuse peaks in time in the DE KC population and (i) sustained depolarization of GGN, similar to that observed in vivo (e.g. Figure 4 Tier 1). Dark gray bar: 1 s of odor stimulation; Light grey bar: “out of response” period of 0.5 s. We used our model to test other basic properties of the GGN and the olfactory network that surrounds it. In vivo, GGN feedback inhibition dilutes KC odor-induced reactions by increasing KC`s peak threshold and limiting KC`s peak to short time windows defined by the oscillator cycle specified in the AL (Papadopoulou et al., 2011; Gupta and Stopfer, 2014). Large-scale advance inhibition has already been shown to extend the dynamic range of cortical neurons (Puglia et al., 2009), and in fruit flies, inhibition of APL feedback, the analogue of GGN, extends the dynamic range of CCIs (Inada et al., 2017).

It is not known whether inhibition of GGN feedback has a similar effect on CCIs. To test this, for simplicity, we added a single KC to our model that receives feedback from GGN (Figure 3a). To simulate KC in this test, we used a single-compartment model with hodgkin-Huxley ion channels (Wüstenberg et al., 2004). Since a single KC would have negligible effects on the GGN, we applied its peak output of more than 50,000 synapses to the lobe branch α the GGN. To avoid unrealistic and strong synchronous inputs to GGN, we staggered incoming peak times by random synaptic delays between 0 and 60 ms. Thus, after each peak generated by the KC model, GGN received 50,000 EPSPs, spread over a 60 ms time window. We ran the KC model with a series of tonic current injections and compared its reactions to those of an isolated KC model that received the same input without feedback inhibition. As expected, initial inhibition by spontaneous activity in the GGN increased KC`s threshold for skyrocketing.

Notably, however, the GGN-coupled KC continued to increase over a much wider current injection range than the isolated KC, which was rapidly saturated to a level where it could no longer increase (Figure 3b, c). A similar result was obtained when we tested KC by applying simulated inhibition of GGN from an olfactory network model described below (Figure 3 – Figure Supplement 1). These results suggest that inhibiting GGN feedback allows a single KC to operate efficiently over a wider dynamic range of inputs. To quantify our results, we used a standard analysis of the control systems literature in which the dynamic range is characterized by the slope of the input response curve, which quantifies the efficiency of the input to trigger the output. By extending the dynamic range, the slope of the input-response curve becomes flatter, as we observed in our model (Figures 3c and d; see Materials and methods for calculating the slope). Therefore, our model predicts that inhibition of GGN feedback extends the dynamic range of CTO. Feedback inhibition is usually achieved through a so-called “allosteric site” – a site on an enzyme that changes the shape of an enzyme and subsequently the behavior of the active center. The process of a positive feedback loop consists of a control system consisting of various components that work in a circular path to stimulate or inhibit each other.

The overall process can be described using the system components. The GGN extends over a large part of each hemisphere of the brain and branches very widely (Figure 1a). It is reported that it receives an excitatory input of 50,000 KC to synapses in the lobe α of the fungal body and in turn provides inhibitory feedback to all CCIs at a distance of 400-500 μm in the calyx. In addition, GGN receives an inhibitory input from a well-named rising neuron “GGN Inhibitor” (GI), which is itself inhibited by GGN (Figure 1a, right) (Papadopoulou et al., 2011). GGN is an interneuron that does not spit. Odor presentations, CCI peaks, and intracellular current injections have been shown to depolarize all GGNs, but none of these stimuli cause GGN spikes; even large injections of intracellular current into the GGN trigger only passive reactions (Leitch and Laurent, 1996; Papadopoulou et al., 2011); and our own observations).