Neuromodulation adjusts the biophysical and synaptic properties of neurons, allowing fine control over network dynamics (Kupfermann, 1979; Kaczmarek and Levitan, 1987; Katz, 1999). Foundational work from invertebrate motor systems (Harris-Warrick and Marder, 1991; Katz, 1995; Katz and Frost, 1995, 1996) identified two major categories of neuromodulation that modify network processing under different circumstances; intrinsic neuromodulation vs. extrinsic neuromodulation. Intrinsic neuromodulation is exerted by neurons that are within a neural network and participate in information processing undertaken by that network. The amount of intrinsic neuromodulation depends on past or ongoing network activity and provides a “memory” of the network state. Extrinsic neuromodulation is exerted by neurons that originate in independent networks and therefore provide information based on the activity of other neural networks. These can be centrifugal neurons innervating a given network or endocrine cells that release humoral factors. The amount of extrinsic neuromodulation therefore depends on the activity of other networks, rather than the network that is being modulated. Thus, extrinsic neuromodulation provides a context about the broader state of the animal. The buccal ganglion of Aplysia californica, which coordinates motor output to control biting movements, illustrates the influence of both intrinsic and extrinsic modulation within a single network (Morgan et al., 2000). The cerebral interneuron “CBI-2” initiates and directly participates in biting motor programs, making it an intrinsic element of the feeding central pattern generator (CPG). With each motor program, CBI-2 improves bite articulation via several neuromodulators (Morgan et al., 2000; Koh and Weiss, 2005, 2007; Friedman and Weiss, 2010; Dacks et al., 2012b). Biting motor programs can also be modulated based on prior exposure to food (Kupfermann, 1974) via the serotonergic metacerebral cells (MCCs) which are external to the feeding CPG. The MCCs do not initiate biting motor programs, yet their activity decreases latency of motor program initiation (Kupfermann, 1979; Kupfermann and Weiss, 1982; Morgan et al., 2000) and lesioning the MCCs reduces the frequency of biting (Rosen et al., 1989). Finally, the MCCs (the extrinsic modulators) shorten feeding motor program latency by increasing CBI-2 (the intrinsic modulator) quantal content (Proekt and Weiss, 2003). This “metamodulation” demonstrates how extrinsic modulators can target intrinsic modulatory components to exert contextual control over network activity (Katz, 1999).

While this conceptual dichotomy (intrinsic vs. extrinsic neuromodulation) has been extensively explored in motor systems, it has received less attention in sensory systems. Here, we take advantage of the depth of work on neuromodulation of olfaction to discuss mechanisms of intrinsic and extrinsic modulation within the first processing stage of the olfactory systems of mammals and insects. The intent of this review is to discuss the concepts of intrinsic and extrinsic modulation to the olfactory system and is by no means exhaustive. As an exemplar of intrinsic modulation, we discuss GABAb-mediated presynaptic inhibition as a means of gain-control modulating the dynamic range of the olfactory system based on previous and ongoing network activity. For extrinsic modulation, we discuss serotonergic inputs from outside the olfactory system that target specific neuron classes via differential receptor expression. We then discuss how serotonergic modulation of GABAergic interneurons provides an elegant mechanism by which metamodulation can use pre-existing gain control mechanisms to efficiently adjust network dynamics. Finally, we discuss the heterogeneous nature of neuromodulation throughout this review, as populations of neurons, and even the arbors of a single neuron, display a surprising degree of molecular and synaptic heterogeneity.

There are many parallels between the insect and vertebrate olfactory systems (Hildebrand and Shepherd, 1997; Ache and Young, 2005). Notably, for this review, both are subject to a broad suite of neuromodulators. In the insect antennal lobe (AL), odorants bind to chemosensory proteins expressed by input neurons called olfactory sensory neurons (OSNs; Figure 1A). Individual OSNs generally express a single chemosensory receptor protein (Vosshall et al., 1999; Vosshall, 2000; Goldman et al., 2005; Joseph and Carlson, 2015) and all OSNs that express the same chemosensory receptor protein project into the same sub-structure in the AL called a glomerulus. Within a glomerulus, OSNs synapse upon output neurons called projection neurons (PNs). PNs then send olfactory information to higher order brain centers like the mushroom bodies (involved in learning/memory; reviewed in Zars, 2000; Owald and Waddell, 2015) and the lateral horn (involved in odor valence; Gupta and Stopfer, 2011; Sachse and Beshel, 2016; Schultzhaus et al., 2017). Finally, a diverse population of local interneurons (LNs; Seki and Kanzaki, 2008; Chou et al., 2010; Seki et al., 2010; Reisenman et al., 2011) refines the input/output relationship of OSNs and PNs. All three principal neuron types, OSNs, LNs and PNs, are subject to both intrinsic and extrinsic sources of neuromodulation. LNs release GABA, dopamine (DA) and a suite of neuropeptides (Homberg et al., 1990; Kirchhof et al., 1999; Berg et al., 2007; Utz et al., 2008; Carlsson et al., 2010; Chou et al., 2010; Siju et al., 2014; Fusca et al., 2015; Hamanaka et al., 2016; Tedjakumala et al., 2017), while the AL is innervated by centrifugal neurons releasing serotonin (5-HT), DA and octopamine (OCT) which act as extrinsic modulators (Kent et al., 1987; Rehder et al., 1987; Salecker and Distler, 1990; Ignell, 2001; Dacks et al., 2005, 2006, 2012a; Sinakevitch et al., 2005; Sinakevitch and Strausfeld, 2006).

Similarly, in the vertebrate olfactory system odorants activate OSNs in the olfactory epithelium which project into glomeruli in the olfactory bulb (OB; Figure 1B). OSNs synapse onto two types of output neurons called mitral and tufted (M/T) cells, which send olfactory information in part to the piriform cortex, olfactory tubercle and other secondary targets. Much like insects, the OB relies on heterogeneous populations of LNs to refine M/T cell output. Subtypes of LNs in the glomerular layer (“juxtaglomerular neurons”) exhibit complex connectivity with the major OB cell types (Wachowiak and Shipley, 2006). Juxtaglomerular neurons can be subdivided into three classes. GABAergic periglomerular cells (PG) synapse onto M/T cells and have an unconventional inhibitory relationship with OSNs in which PG cells may influence OSNs via GABAergic spillover and not via a traditional inhibitory synapse (Pinching and Powell, 1971; Aroniadou-Anderjaska et al., 2000; Wachowiak et al., 2005). Glutamatergic external tufted cells (ET) synapse onto PG cells, M/T cells and superficial short axon cells (sSA). There is evidence that subsets of PG and ET cells may also release DA (Kosaka and Kosaka, 2016). Finally, GABAergic/DAergic sSA cells synapse onto OSNs, and ETs, and widely interconnect both neighboring and distant glomeruli in the glomerular layer (Aungst et al., 2003; Kiyokage et al., 2010; Liu et al., 2013). In the granule cell (GC) layer of the OB, GABAergic GCs provide feedback inhibition onto M/T cells (Shepherd et al., 2007; Burton, 2017). Furthermore, GABAergic deep short axon cells (dSA; Eyre et al., 2008; Burton et al., 2017) synapse onto themselves and reciprocally synapse upon GCs (Burton, 2017). PG and ET cells also appear to express a wide variety of neuropeptides, including NPY, VIP and CCK (Seroogy et al., 1985; Gall et al., 1986). The OB is also subject to extrinsic sources of modulation including 5-HT, norepinephrine (NE) and acetylcholine (ACh), that are released from centrifugal neurons outside of the OB (McLean and Shipley, 1987; Linster and Cleland, 2002, 2016; Kiselycznyk et al., 2006; Matsutani and Yamamoto, 2008; Fletcher and Chen, 2010; Steinfeld et al., 2015).

The olfactory system must efficiently encode odorant information over a wide concentration range to produce reliable representations of odor identity. Heterogeneous populations of LNs and juxtaglomerular neurons alter the input/output relationship between principal cell types to accomplish much of this computation. LNs can release a variety of transmitters including many neuromodulators that act over a range of timescales and in Drosophila, lateral input from inhibitory LNs scales with overall network activation (Olsen and Wilson, 2008). Thus, LNs intrinsically modulate odor coding within the context of current and previous network activation. Intrinsic modulation of olfactory processing can: (1) alter network output based on the strength of odor input; (2) mediate long-lasting temporal effects via metabotropic receptors; and (3) regulate the dynamic range of output neurons, allowing for reliable coding of odor identity across a range of stimulus intensities.

The information transferred from individual OSNs to PNs is highly reliable (Murphy et al., 2004) yet non-linear (Wilson et al., 2004; Bhandawat et al., 2007; Olsen and Wilson, 2008). At high stimulus intensities, output neuron activity can saturate such that further increases in OSN output do not result in a concomitant increase in output neuron activity. To avoid saturation, the olfactory system relies on presynaptic inhibition as a means of gain control. Gain control adjusts the signal strength between input and output neurons to filter out noise from spontaneous firing of OSNs (Wilson, 2013), avoid saturation (Martin et al., 2011) and adjust the dynamic range of output neurons (Olsen and Wilson, 2008; Root et al., 2008). It does this, broadly, by activating GABAb receptors on presynaptic terminals to reduce presynaptic calcium levels, resulting in reduced transmitter release from the presynaptic neuron (Figure 2A; Wang, 2012). While early studies on GABAergic inhibition in the insect AL focused on the role of ionotropic GABAa mediated lateral inhibition (Waldrop et al., 1987; Christensen et al., 1993, 1998a,b; Lei et al., 2002), GABAa blockade in insects does not fully block a slower form of inhibition (MacLeod and Laurent, 1996; Christensen et al., 1998b; Bazhenov et al., 2001; Wilson et al., 2004; Wilson and Laurent, 2005). A slower GABA component had also been suggested as GABAb receptor agonists reduce the responses of M/T cells responses to olfactory nerve stimulation (Nickell et al., 1994). Together this suggested that: (1) GABA may also act presynaptically; and (2) a slower, metabotropic mechanism is at play.

In Drosophila melanogaster, GABAb blockade on OSN terminals increases presynaptic calcium influx, broadens odor tuning of PNs and decreases the range of OSN input over which PN firing rate can change, ultimately resulting in PN saturation (Figure 2B; Olsen and Wilson, 2008; Root et al., 2008). In normal conditions, PN responses are normalized via increased lateral inhibition which scales with ORN activity (Figure 2C; Olsen et al., 2010). Overall, this suggests that interglomerular presynaptic inhibition adjusts the dynamic range of PNs to avoid saturation and refines the breadth of odor tuning across a wide range of stimulus intensities (Wang, 2012). It is important to note that GABA is not the sole modulator of gain control in this system, as the neuropeptides tachykinin (Ignell et al., 2009), and short neuropeptide F (Root et al., 2011; Ko et al., 2015) also mediate presynaptic inhibition.

In the OB, electron microscopy and anatomical studies in rats revealed metabotropic GABAb and D₂ receptor expression on vertebrate OSNs (Bonino et al., 1999; Koster et al., 1999) and direct physiological evidence demonstrated that DA modulates the olfactory nerve synapse (Hsia et al., 1999; Berkowicz and Trombley, 2000) and that presynaptic inhibition is mediated by GABAb receptors (Aroniadou-Anderjaska et al., 2000). Broadly, presynaptic inhibition suppresses calcium influx at OSN axon terminals (Wachowiak and Cohen, 1998, 1999) via both GABAb (Aroniadou-Anderjaska et al., 2000; Wachowiak et al., 2005) and D₂ receptor activation (Ennis et al., 2001; Vaaga et al., 2017), potentially decreasing M/T firing rates. Additionally, GABAergic presynaptic inhibition appears to have both a tonic component that is consistent across stimulus strength, as well as a feedback component that alters glomerular input in an activity dependent manner (Pirez and Wachowiak, 2008). There are a few hypothesized roles for presynaptic inhibition in the OB: (1) that it functions as an adaptive gain control mechanism (Nickell et al., 1994; McGann et al., 2005; Vučinić et al., 2006; Wachowiak and Shipley, 2006; Banerjee et al., 2015; Vaaga et al., 2017); (2) it suppresses OSN input during sniffing (Aroniadou-Anderjaska et al., 2000; reviewed in Wachowiak and Shipley, 2006); and (3) it sharpens the representations of odors across the glomerular map (Vučinić et al., 2006).

However, it is still unclear whether presynaptic inhibition mediates gain control in the OB. A hallmark of gain control is that inhibition is stronger at higher stimulus intensities and weaker at lower stimulus intensities (Robinson and McAlpine, 2009; Saalmann and Kastner, 2009; Martin et al., 2011; Wang, 2012). To demonstrate that presynaptic inhibition adjusts the dynamic range of OB output, presynaptic inhibition must scale with odor concentration. One study found that both weak and strong odor-evoked inputs are subject to the same amount of GABAergic inhibition, and tonic inhibition does not scale with the strength of OSN activation (Pirez and Wachowiak, 2008). This suggests that presynaptic inhibition may alter OSN sensitivity, rather than adjust the dynamic range of M/T cells. Another study demonstrated that juxtaglomerular GABA/DAergic cells, likely sSA cells, exert concentration dependent gain control onto M/T cells (Banerjee et al., 2015). However, it is unclear whether this gain control has a presynaptic component, as this study focused on the synaptic interactions of ET, sSA and M/T cells. Finally, activation of GABA/DAergic sSA cells inhibit presynaptic OSNs, resulting in decreased spiking in M/T cells, and M/T attenuation is blocked by GABAb and D₂ receptor antagonists (Vaaga et al., 2017). However, this does not rule out a GABAb-dependent postsynaptic mechanism of gain control, as sSA activation could act via multiple synapses to control M/T output. Finally, it is still unclear which exact subpopulation(s) of neurons mediate presynaptic inhibition. Potential sources include GABA spillover from GABAergic PG neurons (Aroniadou-Anderjaska et al., 2000; Wachowiak et al., 2005), direct inhibition of OSNs via GABA from sSA (Vaaga et al., 2017), or altered glutamatergic ET cell activity to indirectly influence OB output (Pirez and Wachowiak, 2008; Banerjee et al., 2015).

Animals must constantly adjust their sensory processing to meet the ongoing demands of a dynamic internal and external environment. Both insects and vertebrates heavily rely on their sense of smell to find mates, acquire food and avoid harmful threats in their environment. However, the relative importance of different odors varies with current physiological demands. Extrinsic modulatory neurons from other networks can therefore adjust activity within the OB and AL to provide the context of current internal demands of the individual animal. The olfactory system is subject to a number of extrinsic sources of neuromodulation including 5-HT, DA (in some insects), ACh and NE that have been associated with broad physiological states like waking state, aversion, attention and learning/memory (McLean and Shipley, 1987; Mandairon et al., 2006; Matsutani and Yamamoto, 2008; Fletcher and Chen, 2010; Wasserman et al., 2013; but see Linster and Cleland, 2016). Here, we will focus on the effects of 5-HT as both the OB and AL receive 5-HT innervation from extrinsic sources, and there are many similarities between the cellular and molecular features of serotonergic modulation in both networks.

Metamodulation, or the modulation of modulation, allows extrinsic neurons to exert global control over already existing intrinsic modulatory circuits (Katz, 1999). Centrifugal neurons that innervate the olfactory system often target LNs and juxtaglomerular neurons (Matsutani and Yamamoto, 2008; Mouret et al., 2009) as an efficient mechanism for altering network processing. Presynaptic inhibition provides powerful, odor-specific control over the dynamic output of the olfactory system. Some studies have suggested that this mechanism may be further modulated via extrinsic inputs in the context of behavioral state (Pirez and Wachowiak, 2008; McGann, 2013). Specifically, widely projecting serotonergic neurons (as detailed above) target sub-populations of LNs to indirectly influence presynaptic activity of OSNs and alter stimulus dependent inhibition. In insects, 5-HT increases lateral GABAergic input to OSNs to adjust GABAb mediated presynaptic gain control (Dacks et al., 2009). In the OB, 5-HT depolarizes ETs via the 5-HT2a receptor, sSAs via 5-HT2c, and indirectly excites both sSAs and PGs via glutamatergic ETs. This increases GABAergic/DAergic modulation onto presynaptic terminals of OSNs, reduces OSN output, and ultimately may reduce M/T firing rate to alter OB output (Petzold et al., 2009; Liu et al., 2012; Brill et al., 2016). Overall, these studies suggest that extrinsic serotonergic modulation exerts global control over olfactory network dynamics, in part, by targeting an intrinsic modulatory network.

KML and AMD both conceived of the ideas for this review and co-wrote the manuscript. Author order was determined by degree of caffeination.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.