Adult Neurogenesis in the Songbird: Region-Specific Contributions of New Neurons to Behavioral Plasticity and Stability

The phenomenon of adult neurogenesis has captivated researchers and the public alike - not only because its acceptance overturned long-held and deeply rooted dogma in our understanding of the brain [Altman and Das, 1965; Goldman and Nottebohm, 1983], but also because of the potential it holds for advancing therapeutic interventions for neurodegenerative diseases and traumatic brain injury [Hallbergson et al., 2003; Galvan and Bredesen, 2007; Brainard and Doupe, 2013; Sun, 2016]. However, there are many perplexing aspects of adult neurogenesis for which we have yet to find satisfactory explanations. For instance, we do not understand adaptive reasons for the wide variations among taxonomic groups in numbers and types of new neurons that continue to be produced throughout life, or variations in the brain regions that receive new neurons. Perhaps one of the most intriguing puzzles is why both numbers and types of constitutive new neurons produced in adulthood appear to be most restricted in mammals [Gould, 2007; Drew et al., 2013], although much needed comparative approaches may reveal more species differences within mammals than has been previously assumed [Bonfanti and Peretto, 2011]. In addition, some brain regions seem to replace existing neurons with new ones [Scharff et al., 2000; Barnea et al., 2006; Quadrato et al., 2014], while evidence suggests that other regions continually pack in additional neurons in the absence of compensatory cell death [Walton et al., 2012]. The functional implications of these different strategies are unclear.

One approach to understanding the evolution of such variations is to examine behavioral correlates of postembryonic neurogenesis across taxa and brain regions. To begin, one must first address the following. (1) For any given neuron type, do new neurons simply replace older neurons and maintain their function? (2) Do brain regions that recruit new neurons need a larger population to serve current needs, so that new neurons join old neurons in contributing similarly to processing? (3) Are new neurons intrinsically unique so that they specialize in particular neural tasks, either replacing older neurons or providing a complementary processing capability?

With respect to the third question proposed, most studies agree, either implicitly or empirically, that the roles of new neurons differ from those of older neurons and that these differences are linked to physiological characteristics of their cellular youth. It has been widely suggested that new neurons enhance plasticity in the circuits into which they are incorporated by various mechanisms, none of which are mutually exclusive. For example, new neurons may be uncommitted to gene expression patterns, which become dedicated as cells age [Nottebohm, 2002a, b]. New neurons may provide more nodes or degrees of freedom in a given circuit, perhaps influencing encoding by providing increased resolution [Aimone and Gage, 2011; Aimone et al., 2011], pattern separation [Clelland et al., 2009; Nakashiba et al., 2012; Johnston et al., 2016], pattern completion [Wilson, 2009], pattern integration [Aimone et al., 2011], or flexibility [Garthe et al., 2009] in memory formation. New neurons may confer plasticity or sensitivity via increased excitability and lowered firing thresholds, due both to smaller membrane surface area [Schmidt-Hieber et al., 2004; Ge et al., 2007] and an excitable response to GABA [Toni and Schinder, 2015]. Regardless, a common supposition is that either via properties inherent in the cells themselves and/or via addition of new nodes to a circuit, new neurons provide increased behavioral plasticity permissive for learning new [Ikrar et al., 2013; Johnston et al., 2016] and/or overwriting old [Yau et al., 2015; Epp et al., 2016] memories. The goal of this review is to explore whether new neurons may contribute to functions other than learning or behavioral plasticity when incorporated into brain regions outside of the hippocampus and olfactory bulb. Because most of our understanding of new neuron function comes from studies of hippocampal and olfactory granule neurons, comparisons with some of this work is incorporated throughout the review. Dubbed ‘ephemeral' neurons by Nottebohm [2004], adult-formed neurons are referred to here as ‘renewable'.

Neuronal proliferation in postembryonic mammals is primarily restricted to the subgranular zone of the dentate gyrus and the subventricular zone of the lateral ventricles and third ventricle [reviewed by Doetsch and Scharff, 2001 and Lipp and Bonfanti, 2016]. Neural precursors then migrate less than ∼100 μm into the granular layer of the dentate gyrus where they become mature granule cells [Amaral et al., 2007], or several millimeters along the rostral migratory stream into the olfactory bulb, respectively, and become primarily granule neurons, and to a lesser extent, periglomerular, and short-axon neurons [Whitman and Greer, 2007; Lepousez et al., 2013]. Neurons born in the wall of the third ventricle migrate to the hypothalamus and become leptin-responsive neurons [Kokoeva et al., 2005; Migaud et al., 2010; Cheng, 2013]. New neurons have also been documented in the adult striatum, amygdala, piriform cortex and neocortex, among other regions, although the conditions under which new neurons migrate to these less reported areas is debated [Bernier et al., 2002; Dayer et al., 2005; Luzzati et al., 2006, 2007; Gould, 2007; Chawana et al., 2013; Ernst et al., 2014; Olaleye and Ihunwo, 2014; Inta et al., 2015; Saul et al., 2015].

Direct tests of the function of new neurons have surged in recent years due to optogenetic control over new neuron activity. In this way, target populations of new neurons can be activated or inhibited while monitoring behavioral performance. To date, this work has taken place primarily in the hippocampus and has been focused on learning and memory tasks [Denny et al., 2014; but see also Bardy et al., 2010]. These findings build on earlier tests of new neuron function based on examining behavioral effects of reducing [Shors et al., 2001; Winocur et al., 2006; Wojtowicz, 2006; Capilla-Gonzalez et al., 2016] or increasing new neuron numbers [Jurado-Arjona et al., 2016; Kapgal et al., 2016]. Converging evidence from such direct manipulations suggests that new neurons may serve equivalent functions in both the hippocampus and olfactory bulb [Ming and Song, 2011], and two general theses have emerged. One is the idea that by enhancing substrate plasticity, new neurons are permissive for learning and the formation of new memories. The distinctions among various permutations of this idea lie largely in the details of memory encoding [Hunsaker and Kesner, 2013; Hersman et al., 2015]. The other idea is that new neurons lead to forgetting of older memories, which may be necessary to permit new learning, and thus is equally adaptive [Frankland et al., 2013; Epp et al., 2016]. Naturally, the two ideas are not incompatible and may even be complementary.

However, apparent similarities among new neuron functions between these regions may be a reflection of similarities specific to new granule cells, similarities in mechanisms of learning and memory in these regions, or something particular about the evolution of neurogenesis in the mammal or rodent brain. Perhaps new neurons in the hippocampus contribute to learning because this is the task of dentate gyrus granule cells - likewise with respect to the addition of new neurons in the olfactory bulb. New neurons of different cell types may contribute to entirely different functions in regions that underlie tasks other than learning. Most of the work discussed below approaches this question indirectly, using correlations between numbers of new neurons and measures of behavior [c.f. Scharff et al., 2000]. Future work should seek to add tools allowing direct manipulations of new neurons to study neurogenesis in cross-specific studies as well. For more details and important critiques of work done in mammals, see Lipp and Bonfanti [2016].

The songbird is particularly well-suited to address questions of functional diversity of adult neurogenesis as several different neuronal cell types are continually incorporated into discrete brain regions that subserve quantifiable, and very different, behaviors related to song learning, perception, and production. Songs are learned by juveniles during a two-phase process of (1) sensory learning, during which adult song(s) are memorized, and (2) sensorimotor integration, during which the juvenile undergoes trial-and-error vocal production, increasingly imitating the adult songs [Doupe and Kuhl, 1999]. Song learning strategies vary widely across species, and differences include the number and degree of adult song models imitated, song repertoire size, duration of song learning, seasonality in song learning and production, as well as genetic and morphological constraints on what is learned. Songbirds also differ in whether they are ‘closed learners', which learn a repertoire only once during a juvenile critical period, or ‘open learners', which continue to learn new songs and song elements throughout life. The two most commonly used songbirds in the laboratory are the closed-learner zebra finch and the canary, which incorporates new syllables seasonally, although it has not been determined that these syllables are learned as adults.

Birds have a postembryonic neurogenic niche located in the subventricular zone, lining the walls of the lateral ventricles [Goldman and Nottebohm, 1983; Alvarez-Buylla et al., 1990b; Scott and Lois, 2007]. As is the case during embryogenesis, different regions of the subventricular zone give rise to different classes of neurons throughout adulthood [Dewulf and Bottjer, 2005; Scott and Lois, 2007]. Ventral and dorsal areas of the lateral wall also contain ‘hot spots' of highly concentrated proliferation [Alvarez-Buylla et al., 1990a; Dewulf and Bottjer, 2005]. After mitosis, neural precursors migrate tangentially, then radially along glia, then leave the glial scaffolding and continue to migrate through more peripheral parenchyma [Doetsch and Scharff, 2001; Vellema et al., 2010].

Neuronal precursors, along with new glial cells, disperse widely throughout the telencephalon [Alvarez-Buylla and Nottebohm, 1988; Vellema et al., 2010]. The same patterns of proliferation, dispersal, and incorporation are seen in both songbirds and non-song birds [Ling et al., 1997]. In addition, groups that learn their vocalizations (songbirds, hummingbirds, and parrots) evolved a network of interconnected nuclei called the ‘song system' underlying cognitive and motor aspects of song behavior [Farries, 2001; Jarvis, 2007; Petkov and Jarvis, 2012; Liu et al., 2013]. In songbirds, new neurons are added to two of the nuclei in the song system, providing a rich arena for studying adult neurogenesis in relation to song-related behaviors.

In temperate songbirds, often only the male sings, either exclusively or predominantly, and the presence of the song system across species in females is proportional to the degree that they learn and produce songs [Brenowitz et al., 1985; Arnold et al., 1986; Brenowitz, 1997; MacDougall-Shackleton and Ball, 1999; Garamszegi and Eens, 2004]. The production of song requires the telencephalic vocal motor pathway nucleus HVC (previously known as the ‘high vocal center', now the acronym is used as a proper name), which contains projection neurons that synapse on cells in the robust nucleus of the arcopallium (RA). Neurons in RA in turn project to the motor neurons in the brain stem hypoglossal nucleus (nXII) that innervate the muscles of the syrinx, the avian vocal organ, via the tracheosyringeal branch of the hypoglossal nerve (NXIIts, fig. 1a). A second population of neurons in HVC are a component of a ‘cortico'-thalamic-ganglia loop, the anterior forebrain pathway [Bottjer and Arnold, 1997; Reiner et al., 2004]. The anterior forebrain pathway is not necessary for song production in adults, but is required for song learning in juveniles [Scharff and Nottebohm, 1991] and is permissive for song plasticity in adulthood [Brainard and Doupe, 2000]. In the anterior forebrain pathway, HVC neurons project to a discrete nucleus in the medial striatum called area X, which sends axons to the medial nucleus of the dorsolateral thalamus, which in turn projects to the lateral magnocellular nucleus of the anterior nidopallium. Neurons in lateral magnocellular nucleus of the anterior nidopallium form a feedback loop, projecting back to area X, and also send axons to nucleus RA of the vocal motor pathway (fig. 1a).

Despite the wide distribution of new neurons throughout the telencephalon (fig. 1b), regions that incorporate new neurons are not random, and it is notable that most song system nuclei do not add new neurons: only HVC and area X receive new neurons throughout adulthood. Moreover, within HVC, it is only the population of HVC neurons that project to RA (HVC-RA) that are renewable, and not interneurons or the population that project to area X (HVC-X) [Scotto-Lomassese et al., 2007; c.f. Scott and Lois, 2007]. Within area X, putative medium spiny neurons (MSN) are added throughout life [Sohrabji et al., 1993; Gale and Perkel, 2010]. As in mammals, MSN are GABAergic; however, unlike in the mammalian striatum, MSN in the songbird do not project out of the striatum. Instead, they project to pallidal-like neurons within the area X nucleus, whereas distinct aspiny projection neurons project out of area X to the thalamus [Gale and Perkel, 2010].

Song perception, discrimination, and storage of song auditory memories occur in the higher-order auditory region of the caudomedial nidopallium (NCM). NCM is specialized for processing conspecific songs and does not respond as strongly to other complex auditory stimuli or heterospecific vocalizations [Chew et al., 1996]. NCM is not part of the song system proper, and instead is found in the auditory forebrain in a position that is analogous to supragranular layers of the auditory cortex of mammals [Mello et al., 1998]. Although there are no known direct connections between NCM and the song system, neurons throughout the song system respond to playback of conspecific songs, with preferential tuning for the bird's own song or that of his tutor. Therefore, it seems that song-specific auditory information is also processed in the song system [Doupe and Solis, 1997; Solis and Doupe, 1999; Prather et al., 2010; Raksin et al., 2012]. Like HVC and area X, NCM continues to receive new neurons throughout the bird's life.

A question relevant to the discussion of new neuron function is whether new neurons fill the niche of a neuron that has died or is dying, thereby replacing an old cell with a new one, or whether new neurons are added to the substrate independent of neuronal death. In the case of the first scenario, dying cells may in fact contribute to the upregulation and/or incorporation of new neurons, although potential chemotaxis signals have not been identified. The plausibility of cell death-regulated neuronal replacement was demonstrated by selectively ablating HVC-RA neurons, and separately HVC-X neurons, in the adult zebra finch using retrogradely transported beads injected into RA or area X, respectively. The beads were cytotoxic when combined with laser stimulation at the soma, resulting in photolysis of the cell [Scharff et al., 2000]. Death of HVC-RA neurons led to a surge in new HVC-RA neurons repopulating the depleted nucleus. Interestingly, ablating HVC-X neurons did not result in the replacement of HVC-X neurons or further upregulate HVC-RA neurons. Thus in this context, the death of naturally renewable neurons led to replacement by the same neuron type, whereas apoptosis of a nonrenewable neuronal type did not induce renewal. Interestingly, this rule is not followed in other cases of neuronal death occurring during brain injury or neurodegenerative diseases, which result in neuronal recruitment to the site of the injury where new neurons are not normally incorporated [e.g. Cao et al., 2002; Zupanc, 2009; Cho and Kim, 2010; Jessberger, 2016].

In a more natural context of neuronal turnover, transferring Gambel's white-crowned sparrows from breeding to nonbreeding conditions resulted in a seasonal-like increase in caspase 3-mediated apoptosis of HVC-RA neurons. This, in turn, led to a rapid increase in neural proliferation in the ventricular zone within 2 days. Blocking caspase-mediated apoptosis in HVC of these same birds prevented the effect [Larson et al., 2014]. This work experimentally supports the observation that naturally occurring seasonal peaks in new HVC neurons seen in canaries may be mediated by the preceding peaks in pyknotic cells [Kirn et al., 1994]. Both seasonal and experimental increases in HVC-RA neuron death coincide with degradation of the song acoustic structure, and increased neuron addition to HVC occurs with recovery of the song [Kirn et al., 1994; Scharff et al., 2000; Larson et al., 2014]. In part, this is not surprising as earlier work established the role of HVC-RA neurons in song production. However, it does indicate that at least some memory for the bird's own song is stored outside of the HVC-RA production pathway. It would be interesting to determine whether auditory feedback during singing is necessary for song recovery in this context, or whether songs would recover in deafened birds following HVC-RA ablation and subsequent HVC-RA replacement.

Other work has questioned whether cell death and new neuron addition are causally related, or even whether they are balanced, in the healthy brain under nonseasonal conditions. For instance, zebra finches are equatorial species and sing year round. Under natural conditions, new neurons added to the zebra finch HVC have been found to accumulate across the bird's lifespan, increasing neuronal packing density without appreciably increasing nucleus volume [Walton et al., 2012]. Perhaps new neuron production and recruitment is partially influenced by cell death [Scharff et al., 2000]. However, neuron addition tends to outweigh neuron death across the lifespan [Walton et al., 2012]. The empirical answer seems to be that both processes occur, and the details of the contexts differentiating which process will occur in a given animal, brain region, or condition remain to be addressed. The question of whether neuron death and new neuron addition are balanced in area X or NCM has not yet been explored.

Findings that under certain conditions new neurons replace dying neurons suggest that one function, at least of some types of neurogenesis, may simply be to replace old neurons that ‘burn out,' misfire, lose precision, or otherwise become senescent and unable to do their job. It has been pointed out [Scotto-Lomassese et al., 2007] that sites of neuron incorporation, including the rodent hippocampus and olfactory bulb, mushroom bodies of certain insects, and HVC of songbirds, all exhibit sparse coding - highly precise and invariant timing of spiking with respect to a behavior or stimulus, generally with a very low overall spike rate [Hahnloser et al., 2002; Perez-Orive et al., 2002; Chawla et al., 2005; Davison and Katz, 2007]. This is also true of the MSNs in the songbird striatal region area X [Scott and Lois, 2007]. It is suggestive that about half of NCM neurons that have been recorded are found to fire phasically with a single action potential when depolarized, and therefore may be sparsely coding as well [Dagostin et al., 2015]. Of these recordings, however, sparse coding has only been documented specifically in identified replaceable neurons in HVC. Perhaps the ultraprecise firing patterns necessary for sparse coding lead to a greater propensity for cell burnout and the need for a constant supply of new neurons. Thus, a processing advantage conferred by firing properties that push the envelope on timing precision may have been a contributing factor in the promotion of postembryonic neurogenesis.

If the strategy of burning out and replacing neurons played a part in selection for adult neurogenesis, it is interesting to speculate on whether the loss of renewable neuron types and the overall decrease in new neuron numbers across phylogenetic history corresponds to a change in the need for neuronal precision, or perhaps increased neuronal durability, leading to decreased burnout and lessening selective pressure for neuronal replacement.

In the burnout-replacement scenario, new neurons may simply perform the function of the neurons they replace, and properties specific to young cells may or may not additionally play a role in the process. Moreover, if cell burnout corresponds to cell usage per se (rather than simply cell aging across time), then the burnout-replacement model is consistent with studies showing an association between the use of a brain region and new neuron addition to that region, independent of whether new neurons contribute to learning or plasticity [Pytte et al., 2010]. Consistent with this idea, it has been shown that blocking activity in the HVC target nucleus RA using the GABA_A agonist muscimol resulted in decreased addition of new HVC-RA projection neurons in HVC [Larson et al., 2013]. It is important to note that in this study activity was altered pharmacologically, without manipulating behavior, and was therefore independent of factors associated with learning. Mechanisms of activity-dependent increases in neural proliferation and new neuron survival include excitatory effects of GABA on young neurons [Ge et al., 2007; Platel et al., 2008] as well as NMDA-receptor activation [Cameron et al., 1995; Nacher et al., 2001; Bruel-Jungerman et al., 2006; Chun et al., 2006].

Several critical periods in a new neuron's lifespan have been identified, during which the cellular environment and the animal's behavioral experience can modify the numbers of neurons that die or survive. The first critical period occurs between proliferation and arrival at the postmigratory destination. For instance, in the dentate gyrus, many more neurons are produced than survive to be integrated into the circuit [Curlik and Shors, 2011; Curlik et al., 2014], and this is assumed to be the case in the song system as well. However, although the origins of new HVC-RA or area X MSN neurons have been roughly mapped, the proportion of new neurons produced that reach their destination has not been calculated [Vellema et al., 2010]. Subsequent critical periods differ by animal and brain region, as well as the time of year that new neurons are born for seasonally singing songbirds [Kirn et al., 1994].

In the mouse dentate gyrus, a critical period for new granule neurons occurs during the neuronal ages of 1-3 weeks and may be mediated by NMDAR activation [Ge et al., 2008; Marin-Burgin et al., 2012]. A second critical period occurs during weeks 4-6 after cell birth during which specifically NMDAR2B receptor activation is required for enhanced synaptic plasticity of glutamatergic inputs onto new neurons [Ge et al., 2007]. During both critical periods, numerous factors increase new neuron numbers, ‘rescuing' new neurons that would otherwise have died [Shors, 2008; Anderson et al., 2011; Curlik and Shors, 2011; Waddell et al., 2011]. At a mechanistic level, factors that rescue new neurons include neurotrophins such as BDNF [Alvarez-Borda et al., 2004], testosterone and its metabolites [Rasika et al., 1999; Yamamura et al., 2011; Alward et al., 2016] and other growth factors [Alshammari et al., 2015; Joppe et al., 2015; Bakos et al., 2016]. At a behavioral level, neuronal lifespan is influenced by exercise [van Praag et al., 1999; van Praag, 2008], stress [Gould et al., 1997], sleep [Guzman-Marin et al., 2005], social conditions [Lipkind et al., 2002; Alward et al., 2014; Holmes, 2016], learning [Shors et al., 2001, 2002], and use of the brain region in the absence of learning [Li et al., 2000; Pytte et al., 2010].

In the canary HVC, about half of the new neurons present at day 15 die by day 25 [Kirn et al., 1999]. After this initial culling, between days 31 and 38 after cell birth dating, activity of HVC measured as the amount of time spent singing is positively correlated with new neuron survival. This effect seems to be mediated at least in part by BDNF, which is upregulated during singing [Li et al., 2000] and also influences seasonally trophic effects of testosterone [Rasika et al., 1999]. In addition, direct infusion of BDNF into HVC during days 14-21 after cell birth increased new HVC-RA neuronal lifespan [Alvarez-Borda et al., 2004]. Other potential mediators between singing behavior and survival of new neurons in HVC include opioids [Riters, 2012] and dopamine [Hoffmann et al., 2016]. In the zebra finch, there is a similarly large culling of roughly two thirds of the new neurons in HVC between neuron ages of 1 month and 4 months after birth dating [Wang et al., 2002]; however, factors that may rescue these cells have not yet been explored and may differ in this nonseasonally singing species.

In the zebra finch NCM, there is a significant decrease in new neuron survival between days 40 and 60 after cell birth dating [Lipkind et al., 2002; Barnea et al., 2006]. Neuron survival in NCM is increased when birds are housed in social groups, and likely factors contributing to this effect include auditory learning of new songs [Lipkind et al., 2002] and also perhaps auditory experience in the absence of new song learning [Pytte et al., 2010]. Differential survival studies of new neuron lifespan have not yet been conducted in area X.

Identifying behavioral variables that rescue new neurons has been used as a proxy for understanding the function of new neurons. The premise is based on the idea that new neurons are maintained through their activity as they are forming synapses, as noted above. Therefore, behaviors resulting in new neuron survival are believed to be behaviors to which the new neurons contribute. The most conspicuous examples are the increases in new neuron survival in the hippocampus and olfactory bulb following engagement in hippocampal- and olfactory-dependent tasks [Ming and Song, 2011]. This line of reasoning has led to extensive and increasingly detailed dissection of types of learning and memory tasks that impact new neuron numbers.

Exceptions, of course, are behaviors that increase new neuron survival indirectly, not by increasing new neuron activity per se, but by upregulating mediating factors such as neurotrophins or hormones, which in turn promote survival. In some cases, such as the positive effects of wheel running on neuron survival in the hippocampus [Voss et al., 2007], it is assumed that new neurons are not contributing to the behavior. In other cases, such as increased survival in enriched social and physical environments [Kempermann et al., 2002; Lipkind et al., 2002; Garthe et al., 2016], and the positive correlation between new neurons and sleep [Kreutzmann et al., 2015], the contributing roles of new neuron activity versus indirect upregulation of survival-promoting mediators are more difficult to parse.

In fact, there is not a simple one-to-one correspondence between the rescue of new neurons and hippocampal-dependent activity [reviewed by Shors, 2008]. Some learning tasks that require the hippocampus do not increase the numbers of new granule neurons, and Shors suggests instead that hippocampal learning must be difficult, and also achievable, in order to promote new neuron survival. It is a murkier issue whether hippocampal-independent learning tasks promote new neuron survival in the hippocampus. Learning new procedural, largely striatal-dependent, behaviors has been shown to increase the numbers of new neurons in the hippocampus independent of overall motor activity [Curlik et al., 2013]. Lesioning the hippocampus did not affect learning the physical skill task; therefore, upregulation of new neurons did not seem to be due to hippocampal involvement in learning. The authors suggest that striatal learning that is difficult and also attainable may upregulate hippocampal neurogenesis perhaps as part of a positive feedback system, which may lead to enhanced learning across brain regions. Neurogenesis in the striatum was not examined in this study, and it would be interesting to see whether procedural learning promotes neurogenesis in the striatum.

However, the degree to which the survival of new neurons reflects the function of new neurons is not clear when new neuron activity is conflated with learning, for instance, as it is in the hippocampus and olfactory bulb. The association between activity in these brain regions and increased neuronal survival initially lead to the conclusion that new neurons function in learning. Although this is an intuitively palatable scenario, it calls for further exploration in contexts and brain regions in which neural activity per se and the act of learning can be dissociated.

New neuron addition to HVC in zebra finches is markedly higher during song learning, between juvenile ages 51 and 90 days after hatching, than throughout the rest of the bird's life [Wilbrecht et al., 2002]. Moreover, prolonging the plastic stage of song learning by rearing birds in isolation without an adult song model prolonged this high developmental level of new neuron addition to HVC [Wilbrecht et al., 2006]. Both results support the idea of a functional link between the availability of new neurons in the HVC-RA song motor pathway and the acquisition of new, learned, motor patterns.

Likewise, seasonal peaks in new neuron addition to HVC in canaries not only coincide with seasonal peaks in cell death and testosterone, but also correspond to the addition of new song elements to the song repertoire [Kirn et al., 1994]. In this study, canaries housed on temperate (New York, N.Y., USA) day length added new song elements in the fall, peaking in September, and showed a second smaller peak in new song elements in February and March. An approximate 7-fold increase in new HVC neurons was seen in October and a 5-fold increase in March compared with neuron addition occurring at the lowest months, May through June [Kirn et al., 1994].

However, seasonal peaks in new neurons are also seen in the song sparrow, which does not learn new song elements after the initial juvenile learning critical period. Instead, high rates of new neuron addition to HVC in the fall correspond to increased variability in the song structure, which then becomes more stereotyped in the spring as new neuron addition decreases. Across all of these contexts, the behavioral correlate that most reliably corresponds to numbers of newly added HVC-RA neurons is the dynamic and declining degree of variability in the song structure as seen in juveniles, seasonally in canaries, and in zebra finches as song becomes more stable with age [Pytte et al., 2007]. Perhaps an important condition, however, is that the variable song eventually becomes stereotyped. In other words, new neurons added to the song system may contribute to achieving a target song irrespective of whether the target song requires new song learning. This idea is also consistent with the finding that experimentally ablating HVC-RA neurons in the zebra finch led to a breakdown of song structure, and the addition of new HVC-RA neurons was associated with recovery of the original stereotyped song [Scharff et al., 2000].

The hypothesis that neuron addition to HVC contributes to the acquisition of a stable song was tested in the zebra finch by rendering songs unstable using Botox® (Allergan Inc.) to partially and reversibly paralyze the muscles of the syrinx [Pytte et al., 2012, following Pytte and Suthers, 2000]. This study found that the rate of song recovery was positively correlated with the numbers of new neurons incorporated into HVC (fig. 2). One explanation for this result is that birds with naturally occurring higher numbers of neuron addition to HVC may have been better able to recover the original song. An alternative explanation for the correlation between song recovery and new HVC-RA neurons is that information about the quality of song structure may feed back into HVC and impact new neuron survival. Better song quality may result in more new HVC-RA neurons being maintained, whereas poor quality songs may increase culling of new incoming neurons. In this way, new neurons may be maintained by the degree to which they contribute to accurate song production [Wilbrecht and Kirn, 2004]. Such a selective process may be mediated by any number of factors; however, dopamine is particularly well suited for this role as it is increased during bird singing [Sasaki et al., 2006] and is associated with neuron survival [Luo and Huang, 2016]. Thus, a correctly produced song may function as a ‘reward,' increasing dopamine, and leading to increased new neuron survival. This conjecture has yet to be tested.

The idea that birds with naturally occurring higher rates of neuron addition to HVC may have been better able to recover the original song after Botox treatment suggests that under natural conditions incoming HVC-RA neurons may contribute to keeping the song motor pattern on track. To further test whether new HVC-RA neurons may contribute to song motor quality after song learning is complete, and to control for a potential role of relearning the song during recovery, Pytte et al. [2012] examined the relationship between the numbers of new HVC-RA neurons and song maintenance after deafening. Like other learned motor behaviors such as speech, song motor structure becomes unstable when sensory feedback is removed [Konishi, 1965; Nordeen and Nordeen, 1992]. Deafening results in song degradation that is commensurate with the bird's age: songs of older adults degrade less than those of younger adults [Lombardino and Nottebohm, 2000; Brainard and Doupe, 2001]. It is known that new neuron addition to HVC declines with age as well [Wang et al., 2002; Walton et al., 2012]. Therefore, it has been proposed that song degradation in deaf birds may be due to the addition of new neurons into the song production circuit, in the absence of auditory feedback necessary to train naive neurons to fire correctly during singing. In keeping with the general idea that new neurons are permissive for plasticity, this model predicts that birds incorporating high numbers of new HVC-RA neurons would be those with the greatest degree of song degradation after deafening. However, this did not turn out to be the case. Instead, birds with the greatest numbers of new HVC-RA neurons showed the least song degradation after deafening (fig. 3) [Pytte et al., 2012]. This supports the idea that new neurons in HVC may contribute to the integrity or maintenance of the motor behavior, rather than contributing to plasticity, drift, and change in the motor structure. In this study, which is consistent with others [Hurley et al., 2008], there was no overall group difference in new HVC-RA neurons between deafened and hearing birds, suggesting that factors related to auditory deprivation or the act of deafening, such as stress, did not alter the number of new neurons in HVC. The result also could not be explained by individual differences in singing rate [Pytte et al., 2012].

One explanation for the findings of Pytte et al. [2012] is that young neurons may maintain song structure after deafening better than older neurons. Although very young hippocampal neurons fire less specifically and with a reduced latency than older neurons [Schmidt-Hieber et al., 2004; Marin-Burgin et al., 2012; Danielson et al., 2016], perhaps there is an intermediate age during which new, but mature, neurons are better able to achieve the precise firing accuracy necessary for ultrasparse coding compared with aging neurons. Alternatively, or in addition, perhaps young neurons in HVC maintain sparse coding, despite high intrinsic excitability, by balancing broad responsivity with low network connectivity as demonstrated recently in the hippocampus [Dieni et al., 2016]. Therefore, birds with greater numbers of new HVC-RA neurons are better able to maintain the precision necessary for song stereotypy.

Another explanation is that the correlation between behavioral stability and numbers of new neurons is, in fact, not a causal relationship. For instance, numbers of new neurons incorporated into HVC may reflect the total numbers of HVC-RA neurons. If so, birds with higher rates of incorporation at a given age may have more overall numbers of neurons in this pathway. Perhaps young and older neurons are equally adept at firing correctly, at least to the degree needed for song production, but the larger the HVC-RA population overall, the greater the stability of song after deafening.

Regardless of the mechanism, this finding is consistent with the idea that new neurons contribute to achieving a target song, whether it be during a natural course of song transition or after experimentally perturbing the song structure. In the context of deafening, new neurons correlate with the maintenance of the bird's song during the process of song degradation. In sum, the addition of new neurons in HVC in zebra finches, which sing a stable song after song learning is complete, does not seem to impart increased plasticity in the song motor program - at least not by our conventional measures of plasticity.

It should be considered that the precise role of the nucleus HVC in song production is not known. Recordings from likely interneurons during singing demonstrate that tonic activity in this population differs with song syllable identity, whereas neuronal activity in RA seems more tightly correlated with the smaller acoustic unit elements within the syllables [Yu and Margoliash, 1996]. HVC-RA neurons, in particular, have been suggested to provide a metronome-like function in maintaining the precise temporal structure underlying the acoustic unfolding of the song [Hahnloser et al., 2002; Long and Fee, 2008]. At a higher-order level, it has been suggested that the vocal motor pathway more generally promotes song stability [Thompson and Johnson, 2007] relative to the anterior forebrain pathway, which in turn confers song variability [Brainard and Doupe, 2000; Nordeen and Nordeen, 2010; Tschida and Mooney, 2012]. Any of these roles is consistent with the idea that more HVC-RA neurons may correspond to greater song stability.

It is interesting to note recent evidence that granule cells of the mouse dentate gyrus play a role in maintaining hippocampal memory traces [Madronal et al., 2016]. In this study, pharmacogenetic inhibition of dentate granule cells resulted in forgetting a learned association. Transgenic mice that expressed an artificial receptor specifically in dentate granule cells were trained on a trace-fear-conditioning paradigm in which they had to learn to associate a tone with a shock presented after the tone. The recall of this learned association has been shown to be hippocampal dependent. Once mice reached an asymptotic level of conditioned responding, transient (2-3 h) inhibition of dentate granule cells with an artificial ligand during presentation of the tone-shock pairing resulted in the rapid, and apparently complete, forgetting of this learned association and reduced downstream, learning-induced plasticity to baseline (prelearning) levels. Moreover, the effect of transient inhibition was persistent, as the forgetting and reduced plasticity were seen both 1 and 5 days after granule cell inhibition. Thus, it seems that dentate granule cell activity was necessary for the maintenance of this hippocampal memory trace [Madronal et al., 2016].

This work appears at odds with evidence demonstrating that new neurons can impair memory stability when new neurons are upregulated following conditioned fear associations [Akers et al., 2014]. However, the two sets of findings are not incompatible when one takes into account that neuron age relative to memory differs between the studies. It may be that the role of new neurons in this region with respect to plasticity and stability is biphasic. The heightened plasticity and low response latency of immature neurons [Schmidt-Hieber et al., 2004; Marin-Burgin et al., 2012] coupled with sparse coding [Dieni et al., 2016] may be characteristics best suited for encoding new memories. Once incorporated into the memory trace, the activity of mature new neurons may serve to stabilize the memory trace. This model of new neuron function in the dentate gyrus suggests at least one way in which new neurons may contribute to both plasticity and stability in the hippocampus. Moreover, this idea is compatible with findings that greater numbers of incoming HVC-RA neurons correspond to greater song stability, perhaps providing a glimpse of more fundamental principles of, or a larger framework for, neurogenesis across taxa and brain regions.

There is strong evidence that new neuron incorporation and survival is correlated with new neuron activity. Because of this, the degree to which new neuron survival reflects the function of new neurons is not clear when new neuron activity is conflated with learning. New neurons in the song system and in the mammalian brain are incorporated into regions that perform various types of learning. It follows that activity in these regions, and in new neurons in particular, leads to increased numbers of new neurons and the conclusion that new neurons function in learning. Although this is a satisfying scenario, it calls for further exploration in contexts in which substrate (or new neuron) activity and learning can be dissociated. Studies of new neurons in the song system motor pathway nucleus HVC suggest that new neurons may not always or necessarily provide a substrate for new learning. Instead, under conditions in which the song motor pattern is experimentally manipulated to become unstable, new neurons correspond to increasing motor stability rather than plasticity. Thus, the role of new neurons may be specific to the brain region that receives new neurons, and may vary as widely as promoting behavioral plasticity or contributing to behavioral stability.

I thank Alice Powers for organizing this Karger workshop and the other participants for stimulating discussions about neurogenesis and plasticity. I thank Jacques Balthazart, Barbara Beltz, Jake Jordan, and Alice Perez for critical comments that greatly improved the manuscript. This work began in the laboratory of John Kirn, to whom I am grateful for many insights and ideas about potential functions of adult neurogenesis. The research was supported by PSC-CUNY Research Awards PSCREG-39-1018, 40-1131, and 42-613, a SOMAS Award, and NIHRO3 NS063182.