The Mauthner-cell circuit of fish as a model system for startle plasticity

Highlights

• Vital for survival, startle has to be robust but able to adapt to change.

• Temperature, social status or sensory filtering can modulate startle in fish.

• Behavioral changes are paralleled by changes in the M-cell system.

• Mechanisms of plasticity operate on the excitation/inhibition balance of the M-cell.

• The M-cell is a valid model to study multimodal integration of single neurons.

Abstract

The Mauthner-cell (M-cell) system of teleost fish has a long history as an experimental model for addressing a wide range of neurobiological questions. Principles derived from studies on this system have contributed significantly to our understanding at multiple levels, from mechanisms of synaptic transmission and synaptic plasticity to the concepts of a decision neuron that initiates key aspects of the startle behavior. Here we will review recent work that focuses on the neurophysiological and neuropharmacological basis for modifications in the M-cell circuit. After summarizing the main excitatory and inhibitory inputs to the M-cell, we review experiments showing startle response modulation by temperature, social status, and sensory filtering. Although very different in nature, actions of these three sources of modulation converge in the M-cell network. Mechanisms of modulation include altering the excitability of the M-cell itself as well as changes in excitatory and inhibitor drive, highlighting the role of balanced excitation and inhibition for escape decisions. One of the most extensively studied forms of startle plasticity in vertebrates is prepulse inhibition (PPI), a sensorimotor gating phenomenon, which is impaired in several information processing disorders. Finally, we review recent work in the M-cell system which focuses on the cellular mechanisms of PPI and its modulation by serotonin and dopamine.

Introduction

All animals display some form of defensive reflexes to avoid potential injury. One prominent example is startle behavior. In addition to its vital importance, study of startle or other protective reflexes has contributed to fundamental advances in neuroscience. For example, study of the gill withdrawal reflex in Aplysia (Croll, 2003, Glanzman, 2009, Kandel, 1976), the tail flip in crayfish (Edwards et al., 1999), the eye blink response in humans (Graham, 1975), and the C-start in fishes (Eaton et al., 1991, Korn and Faber, 2005, Zottoli et al., 1999) provided critical insights to issues ranging from the behavioral and neural basis of habituation, sensitization, fear conditioning and sensorimotor gating (Koch, 1999) to the advancement of the command-neuron concept (Eaton et al., 2001, Edwards et al., 1999), the cellular and molecular basis of learning and memory (Glanzman, 2009), and the research on neural networks implementing decision-making (Edwards et al., 1999, Korn and Faber, 2005).

The startle response typically involves fast and massive activation of head and body muscles in response to threatening and intense sensory stimuli. As such, startle is a protective reflex that also constitutes often the initial phase of a more elaborate escape behavior that involves other motor systems, although the latter function is less clear in mammals (Yeomans and Frankland, 1995, Yeomans et al., 2002). Despite its vital role, frequent or unnecessary startles need to be avoided since they disrupt other important behaviors. These constrains are reflected in the structure of startle networks, which are typically centered around large, (i.e. high-threshold) ‘decision’ neuron/s that integrate vast excitatory and inhibitory inputs from multiple sense organ, and control the activation of large muscle areas (Eaton, 1984). Startle can be an all-or-none behavior mediated by a pair of bilateral decision neurons [e.g. crayfish (Edwards et al., 1999, Wine and Krasne, 1972), squid (Otis and Gilly, 1990), teleost fish (Eaton et al., 1977)], or a graded response mediated by the sequential recruitment of numerous (50–60) decision neurons in distinct brain nuclei [mammals (Lingenhöhl and Friauf, 1994, Yeomans and Frankland, 1995), see also below]. In that context, it is interesting to note that even all-or none startle systems are typically complemented by parallel multifiber pathways that modulate either the later parts of a startle response and/or produce graded yet flexible startle-like behaviors by themselves (Bhatt et al., 2007, Fetcho and Faber, 1988, Fetcho and O’Malley, 1995, Herberholz et al., 2004, Otis and Gilly, 1990, Preuss and Gilly, 2000, Wine and Krasne, 1972).

Startle behavior is distinct, relatively easy to quantify, and the large size and small number of startle circuit neurons allows in many cases their identification in the CNS for anatomical, electrophysiological and molecular studies (Cachope and Pereda, 2012, Curti and Pereda, 2010, Eaton, 1984, Korn and Faber, 2005, Pereda et al., 2004). Particularly important for this review however, is the fact that startle circuits provide an excellent preparation and readout for studying the sensory integration processes that underlie the initiation of startle behavior including its modification by environmental context and physiological state of an animal.

Indeed, startle plasticity is widespread and subject to intense research. Startle response can be increased by conditioned or unconditioned aversive manipulations as an electrical foot shock (Boulis and Davis, 1989, Davis, 1974), habituated by repeated presentation of the startling stimulus (Aljure et al., 1980, Davis et al., 1982, Typlt et al., 2013, Valsamis and Schmid, 2011) and it can be enhanced by fear, anxiety and related states [reviewed in Fendt and Koch (2013)]. Failure to adjust startle threshold levels has been connected to several fear and anxiety disorders (Dreissen et al., 2012, Ganser et al., 2013, Grillon, 2002, Grillon, 2008) and startle testing is a well established assay to investigate anxiety-like behaviors in several species (Pittman and Lott, 2014).

One of the most intensively studied aspects in startle plasticity is prepulse inhibition (PPI) of the auditory startle response. In the PPI paradigm, the startle response to a strong stimulus is reduced when it is preceded by a weak prepulse of the same or a different modality by 30–500 ms (Campeau and Davis, 1995, Hoffman and Ison, 1980, Weber and Swerdlow, 2008). The difference on the intensity (or probability) of the startle response with or without a sensory prepulse provides an operational measure of the inhibition induced by the prepulse. This reduction is thought to reflect the subject́s sensorimotor gating levels (Braff et al., 2001a). It has been proposed that the functional role of PPI is protection from a disruptive event such as startle at an early stage of stimulus information processing (Graham, 1975). Underlining its importance as a basic filtering mechanism, PPI of startle response has been extensively studied in rodents (Braff et al., 2001a, Swerdlow et al., 2008) but also in sea slugs (Frost et al., 2003, Lee et al., 2012, Mongeluzi et al., 1998), teleost fishes (Burgess and Granato, 2007, Kohashi and Oda, 2008, Neumeister et al., 2008) and birds (Schall et al., 1999). These studies suggest cross-species similarities for some of the mechanisms that regulate startle plasticity and PPI (Siegel et al., 2013). PPI has also attracted considerable attention from biomedical research as schizophrenia patients show deficits in PPI although these deficits are not unique of schizophrenia but are also present in bipolar mania, Huntington’s disease, panic disorder and other sensory processing disorders (Braff et al., 2001b, Van den Buuse, 2010, Siegel et al., 2013).

Given the biological and medical relevance of understanding startle and startle plasticity mechanisms, the importance of developing animal models to study startle behavior and PPI has been repeatedly acknowledged (Koch, 2013, Siegel et al., 2013). Great progress has been made in elucidating the circuits, neuropharmacology, and genetics of PPI in rodents and linking these findings to a range of information processing disorders (Braff et al., 2008, Swerdlow et al., 2008). However, some methodological limitations continue to constrain the field. For example, reliably accessing the startle circuitry relevant to PPI with in vivo electrophysiology remains difficult in rodents (Lingenhöhl and Friauf, 1994). In vivo experiments are critical, however, since they allow physiological stimulation of the inhibitory pathway/s active during PPI, a requirement to identify the effector mechanisms underlying PPI.

The thesis of the current review is that the startle system of teleost fishes, the Mauthner-cell (M-cell) is ideally suited to advance such mechanistic studies of startle plasticity.

Several recent reviews have focused on aspects of plasticity in the M-cell circuit (Cachope and Pereda, 2012, Curti and Pereda, 2010, Kano, 1995, Korn and Faber, 1996, Korn and Faber, 2005, Pereda et al., 2004, Zottoli and Faber, 2000, Zottoli et al., 1995) but here we will specifically focus on recent findings on cellular mechanisms regulating startle plasticity and particularly PPI in the primary auditory startle circuit of teleost fishes. We start describing the startle circuit in fishes and mammals to stress their common organizing principles, followed by an account of main sensory inputs to the Mauthner cell. Next we review environmental factors capable of modulating the startle response and the role of dopamine and serotonin in M-cell plasticity. A description of PPI and its modulation by dopamine follows, and we conclude with an overall discussion of the results presented and open questions for the future.

Escape behaviors are critical for survival as they allow predator avoidance, and most vertebrates, including mammals, have highly developed neural networks for detecting an approaching predator, deciding when to initiate the escape and the trajectory to follow.

The startle response pattern is suggestive of a protective function against injury from potential threats and of a preparatory phase for the flight/fight response (Koch, 1999, Yeomans et al., 2002), however, its functional role in mammals is not always clear as startle itself does not necessarily imply an escape (Yeomans and Frankland, 1995, Yeomans et al., 2002). In many species of teleost fish, however, startle is part of a true escape behavior with a clear-cut function: interrupt all ongoing behaviors to follow a relatively stereotyped motor sequence resulting in an escape reaction to the threat (Batty, 1989, Dill, 1974, Eaton et al., 1991, Faber et al., 1989, Kohashi and Oda, 2008, Neumeister et al., 2010, Preuss and Faber, 2003, Whitaker et al., 2011). This escape sequence starts with a fast and massive unilateral contraction of trunk muscles resulting in the fish assuming a C-shape (stage 1) followed by a return stroke in the opposite direction (‘return flip’) where the tail straightens propelling the animal away from potential danger (stage 2) (Domenici and Blake, 1997, Eaton et al., 1977, Eaton et al., 2001, Zottoli, 1977).

Stage one of the C-start is initiated by a pair of large brainstem neurons, the Mauthner cells (M-cells), which are prototypical integrate-and-fire neurons that receive massive sensory inputs from the acoustic-lateralis, vestibular, visual, and somatosensory systems (Canfield, 2003, Furukawa and Ishii, 1967, Korn and Faber, 2005, Mirjany and Faber, 2011, Preuss and Faber, 2003, Preuss et al., 2006, Szabo et al., 2006, Szabo et al., 2007) (Fig. 1). A single action potential (AP) in one of the two M-cells is sufficient to activate motor networks on the contralateral trunk muscles and simultaneously inhibit those on the ipsilateral side (Eaton and Farley, 1975, Eaton et al., 2001, Faber et al., 1989, Fetcho and Faber, 1988, Nissanov et al., 1990, Weiss et al., 2006, Zottoli, 1977). Ablation of M-cells in goldfish or zebrafish (DiDomenico et al., 1988, Eaton et al., 1982, Issa et al., 2011, Liu and Fetcho, 1999, Zottoli et al., 1999) or evolutionary loss of M-cells (Greenwood et al., 2010) eliminates short latency C-start escapes.

Stage two of the C-start determines the final escape trajectory, which is influenced by for example, stimulus direction and obstructions in the environment (Eaton and Emberley, 1991, Eaton et al., 1988, Eaton et al., 2001, Foreman and Eaton, 1993, Mirjany et al., 2011, Nissanov et al., 1990, Preuss and Faber, 2003). This flexibility depends at least partly on the activation of other reticulospinal neurons, such as the M-cell homologous that in conjunction with the M-cell are collectively known as the brainstem escape network (BEN) (Canfield, 2006, Eaton et al., 2001, Gahtan et al., 2002, Weiss et al., 2006).

Indeed, the Mauthner cell homologous are electrically connected with the M-cell (Neki et al., 2014), and receive auditory inputs although their firing thresholds and projection patterns are different from the M-cell (Nakayama and Oda, 2004). In addition to their putative role for stage 2, thay can also produce C-start type behaviors, albeit with longer latencies when the M-cell is eliminated from the circuit (Kohashi and Oda, 2008, Kohashi et al., 2012, Liu and Fetcho, 1999).

In summary, activity in either of the two M-cells decide the likelihood, latency, and initial turn direction of the response (Hatta and Korn, 1999, Korn and Faber, 2005, Preuss and Faber, 2003, Preuss et al., 2006, Zottoli et al., 1999), with other neurons in the BEN adding necessary plasticity to the expression of the behavior.

The M-cell dendritic morphology is relatively simple (Fig. 1): two primary dendrites, one lateral and one ventral, each extending up to 500 μm from the soma (Korn and Faber, 2005, Faber and Korn, 1978). Because of its size, morphology, and electrophysiological signature (an unusually large extracellular field potential), it is possible to record reliably from the M-cell soma and its dendrites in vivo (Faber and Korn, 1978, Furshpan and Furukawa, 1962, Korn and Faber, 2005, Preuss et al., 2006). The excitatory sensory inputs to the M-cell include afferent inputs from the auditory (Furshpan and Furukawa, 1962, Preuss and Faber, 2003, Szabo et al., 2006), vestibular (Zottoli and Faber, 1979), visual (Preuss et al., 2006, Zottoli et al., 1987), somatosensory systems (Chang et al., 1987) and lateral line (Faber and Korn, 1975, Mirjany and Faber, 2011). However, the two sensory systems that most often have been shown to trigger a startle response with physiological stimuli are an intense sound or a gradually increasing visual or auditory loom stimuli (Eaton et al., 1988, Preuss et al., 2006, Weiss et al., 2006, Zottoli, 1977). The fact that subthreshold LED flashes can bias the direction of an escape evoked by an abrupt sound highlights multimodal aspects of sensory processing in the M-cell (Canfield, 2003, Canfield, 2006). Similar modulatory functions have been shown for lateral line inputs to the M-cell, which by themselves are insufficient to evoke an action potential in the M-cell (Faber and Korn, 1975, Mirjany and Faber, 2011). Moreover, the relative efficacy of a given sensory modality in the M-cell can change during development. Zebrafish shows a transition from somatosensory to statoacoustic nerve as the preferred input for the M-cell during development (Kohashi et al., 2012).

The sensory input to the M-cell studied in most detail is mediated by the monosynaptic connection from a group of large afferents of the posterior branch of the eight (auditory) nerve (Furshpan, 1964, Nakajima, 1974, Tuttle et al., 1986). These saccular fibers terminate on the lateral dendrite as single club endings, 10–15 μm in diameter, and they have mixed electrotonic and chemical synapses with the dendrite (Pereda et al., 2003, Pereda et al., 2004). This constitutes a rapid pathway with a latency of 1–2 ms when using acoustic stimuli in air (Canfield, 2003, Casagrand et al., 1999, Preuss and Faber, 2003, Szabo et al., 2006), contrasting with the slower polysynaptic visual pathway coming from the retina through the optic tectum and contacting the M-cell in its ventral dendrite, which has a latency of 12–22 ms when using short (10–20 ms) visual pulses (Canfield, 2003, Canfield, 2006, Weiss et al., 2006).

Inhibition has been shown to serve two major roles in the M-cell. Feedforward inhibition sets the threshold for escape, regulating M-cell excitability in response to sensory stimuli, to ensure that only sufficiently strong and abrupt stimuli trigger the M-cell (Faber and Korn, 1978). Feedforward inhibition is also involved in information processing by restricting the spatial spread of excitation and its duration, favoring the detection of temporal changes in a signal (Faber et al., 1991, Preuss and Faber, 2003, Preuss et al., 2006). For the auditory pathway this is attained by commissural passive hyperpolarizing potential (PHP) exhibiting neurons, which receive mixed electric and synaptic inputs from primary afferents and provide chemical inhibition to the lateral dendrite and M-cell soma within a synaptic delay (Fig. 1). Interestingly, the same pathway also mediates an instantaneous electrical field (ephaptic) inhibition coincident with presynaptic action potentials at the M-cell axon hillock (Furukawa and Furshpan, 1963, Furukawa et al., 1963, Takahashi et al., 2002, Weiss et al., 2008) (Fig. 1). Another source of inhibition is a recurrent pathway triggered by the M-cell spike that avoids repetitive firing of the activated M-cell as well as firing of the contralateral M-cell (reciprocal inhibition) (Faber et al., 1989, Takahashi et al., 2002). This feedback pathway involves cranial relay neurons (CR, Fig. 1), which in turn bilaterally activate inhibitory PHP neurons, but also cranial motor neurons that evoke startle-related opercular, ocular, jaw and pectoral fin movements (Auerbach and Bennett, 1969, Diamond, 1971, Hackett and Buchheim, 1984, Hackett and Faber, 1983).

In summary, in the M-cell startle network, threshold to startling stimuli is determined ultimately by the balance between excitatory and inhibitory mechanisms acting in the startle “decision-making” circuit (Faber et al., 1989). The downstream network of the M-cell includes interneurons and motorneurons forming cranial and spinal networks that ultimately mediate the execution of the escape response (Bhatt et al., 2007, Faber et al., 1989, Fetcho and Faber, 1988, Fetcho and McLean, 2010). Plasticity in the M-cell output synapses or in the downstream elements could be translated to behavioral plasticity of startle behavior (Aljure et al., 1980, Gelman et al., 2011) but here we will concentrate on plasticity affecting the ‘decision making’ process (e.g. plasticity affecting the firing probability of the M-cell).

The circuit for the mammalian startle reflex shares many organizing principles with those for fast escape responses in lower vertebrates and invertebrates. In mammals, the primary acoustic startle circuit encompasses two central relay stations between the sensory periphery (cochlea) and the motor and premotor neurons that execute the motor response. Auditory input enters the cochlear nuclei where the cochlear root neurons are, which in turn excite a small cluster of about 60 giant neurons (PnC neurons) from the caudal pontine reticular formation that innervate cranial and spinal motoneurons (Lingenhöhl and Friauf, 1992, Lingenhöhl and Friauf, 1994). The graded nature of the escape mammalian startle responses is partially a consequence of startle being the result of the activation of a population of neurons (and not a single neuron).

Similarly to M-cells, PnC neurons are activated by high-intensity acoustic stimulation at latencies of 3–8 ms (Lingenhöhl and Friauf, 1992) and have a relatively low resistance and large membrane time constant (Faber et al., 1989, Wagner and Mack, 1998). Lessions of the PnC neurons block the startle response elicited by acoustic or air-puff stimuli (Koch et al., 1992, Lee et al., 1996). Trigeminal or auditory stimuli reach the PnC neurons through large-diameter, myelinated axons projecting directly from the cochlear, vestibular and spinal trigeminal nuclei of rats (Lee et al., 1996, Lingenhöhl and Friauf, 1994). In turn, PnC giant neurons project directly and indirectly to motoneurons in the brain stem and spinal cord via a large and fast-conducting axon (Wu et al., 1988). One of the major difference between the M-cell circuit and the PnC circuit is that whereas only one M-cell is activated producing unilateral contraction of the trunk muscles, the mammalian response is bilateral, with reticulospinal neurons (PnC neurons) on both sides of the midline being excited (Lee et al., 1996). Besides this fact, the mammalian startle networks parallels the fish auditory startle network at every stage, the properties of this circuit performing a very fast response minimizing the number of synapses between the sensory input and the motor output and using large caliber myelinated fibers, electrical synapses and establishing direct contacts with motoneurons.

Up to 100 eight nerve (auditory) afferent inputs impinge onto the mid distal M-cell lateral dendrite and provide massive excitatory input via mixed electrical and chemical synapses (Lin and Faber, 1988, Pereda et al., 2004). The size of these primary afferent terminals (up to 15 μm) and the M-cell dendrite allows for simultaneous pre- and postsynaptic recordings and the characterization of their role in sensory processing of acoustic stimuli. Interestingly, paired recordings suggest that >80% of the chemical connections are silent (Lin and Faber, 1988), i.e. a presynaptic AP produces a postsynaptic electrotonic potential but no chemically mediated excitatory postsynaptic potential (EPSP). The chemical component seems to be recruited only when a large fraction of the afferent fibers is simultaneously activated as with direct electrical stimulation to the eight nerve fibers (Pereda et al., 1992, Pereda et al., 1994). Accordingly, M-cell EPSPs evoked by short sound pips (single sine waves, Fig. 2A) and longer lasting frequency (FM) or amplitude (AM) modulated tone bursts (Fig. 2B) show two components namely, one that is phasic and phase locked to stimulus frequency and its first harmonic (Fig. 2C and D), and a sustained underlying depolarization with a slower onset that tracks the amplitude of the stimulus (Szabo et al., 2006). Blocking of the chemical synaptic transmission revealed that the fast components are essentially electrotonic coupling potentials (Fig. 2B) of eight nerve APs, whereas the underlying slow EPSP is largely glutamatergic (Szabo et al., 2006). Together with the demonstration of silent synapses (see above) these results suggest that the chemical component of the mixed synapse is weak or functionally silent when single sound pips or short tone bursts are used (Szabo et al., 2006). Chemical transmission becomes more prominent with longer lasting stimuli and with repetitive stimulation (Cachope et al., 2007, Curti and Pereda, 2004).

The M-cell also receives input from the anterior and posterior lateral line nerves (Faber and Korn, 1975). These fibers contact the M-cell proximal lateral dendrite and soma with mixed (chemical and electrical) synapses and provide weak but essential modulatory information for the left–right directionality of goldfish startle response (Mirjany and Faber, 2011, Mirjany et al., 2011).

Each M-cell receives binocular visual input from both ipsi and contralateral optic tectum projections that end in the distal ventral dendrite (Zottoli et al., 1987). In goldfish, brief visual stimuli produced by a flash or a black disk projected on top of the animal typically do not elicit startles but do produce subthreshold EPSPs in the M-cell (Fig. 3A and B). As noted, similar stimuli are capable of modulating the auditory startle response in the African cichlid fish Astatotilapia burtoni (Canfield, 2003, Canfield, 2006).

Startle responses however, can be reliably evoked in goldfish by visual loom stimuli, i.e., by motion stimuli that stimulate an approaching object on collision course (Batty, 1989, Dill, 1974, Preuss et al., 2006, Webb, 1986). Intracellular in vivo M-cell recordings show graded depolarizing EPSPs (Fig. 3C), and chronic recordings in freely moving animals confirmed that these escapes are initiated by a M-cell AP (Preuss et al., 2006, Weiss et al., 2006).

As noted, acoustic and visual information converge in the M-cell and both modalities and most predator strikes will typically involve visual and mechanosensory components, e.g. a diving bird breaking the water surface. Indeed, the M-cell has been shown to integrate visual, statoacoustical and somatosensory inputs for an appropriate startle behavior (Canfield, 2003, Mirjany et al., 2011 and Medan and Preuss unpublished results).