Genetically-identified neurons in avian auditory pallium mirror core principles of their mammalian counterparts

Summary:

In vertebrates, advanced cognitive abilities are typically associated with the telencephalic pallium. In mammals, the pallium is a layered mixture of excitatory and inhibitory neuronal populations with distinct molecular, physiological, and network phenotypes. This cortical architecture is proposed to support efficient, high-level information processing. Comparative perspectives across vertebrates provide a lens to understand the common features of pallium that are important for advanced cognition. Studies in songbirds have established strikingly parallel features of neuronal types between mammalian and avian pallium. However, lack of genetic access to defined pallial cell types in non-mammalian vertebrates has hindered progress in resolving connections between molecular and physiological phenotypes. A definitive mapping of the physiology of pallial cells onto their molecular identities in birds is critical for understanding how synaptic and computational properties depend on underlying molecular phenotypes. Using viral tools to target excitatory vs. inhibitory neurons in the zebra finch auditory association pallium (CaMKIIα and GAD1 promoters respectively), we systematically tested predictions derived from mammalian pallium. We identified two genetically-distinct neuronal populations that exhibit profound physiological and computational similarities with mammalian excitatory and inhibitory pallial cells, definitively aligning putative cell types in avian caudal nidopallium with these molecular identities. Specifically, genetically-identified CaMKIIα and GAD1 cell types in avian auditory association pallium exhibit distinct intrinsic physiological parameters, distinct auditory coding principles, and inhibitory-dependent pallial synchrony, gamma oscillations, and local suppression. The retention or convergence of these molecular and physiological features in both birds and mammals clarifies the characteristics of pallial circuits for advanced cognitive abilities.

eTOC blurb

Spool et al. use cell-type specific viral optogenetics in bird pallium and observe mirrored physiology, auditory coding, and network roles as in mammalian pallial excitatory and inhibitory cell types. Ancient pallial cell types therefore most likely retained an extensive array of features to support cognitive function in distant vertebrate classes.

Introduction:

The vertebrate pallium is considered essential for advanced cognitive abilities. Across vertebrates, the pallium varies immensely in its cytoarchitecture and elaboration, even within closely related taxa¹. In mammals, for example, dorsal pallium develops into a six-layered structure called neocortex, and neocortical elaboration is associated with advanced cognitive abilities^2,3. Some cartilaginous fishes have a layered pallium, while in other species the pallial organization is not obvious, nor distinct from subpallial structures¹. Large swaths of bird pallium are unlayered^4,5 (Fig. 1A; though see ^6–10 for lamination in auditory pallium and hyperpallium), yet several species of birds, including parrots and corvids, exhibit cognitive abilities on par with or exceeding capabilities of our closest primate relatives¹¹. Deep characterization of the molecular properties and developmental origins of pallial cells^12–22, as well as examination of the physiology of pallial cell types^7,12,23–27 have advanced our understanding of pallial evolution. However, a lack of genetic access to pallial cell types in non-mammalian vertebrates limits the ability to definitively test the links between molecular and physiological identities, and to test hypotheses about the organization of pallial networks. In this study, we begin to address these gaps using promoter-driven viral expression to interrogate the physiological and network phenotypes of two principal cell types in avian pallium.

Fig. 1. Viruses targeting CaMKIIα and GAD1 promoters segregate cell types in avian auditory association pallium.

(A) Sagittal schematic of avian pallium (top left) showing the main thalamorecipient region of auditory pallium (Field L), and auditory association regions of pallium (caudomedial mesopallium, CMM; caudomedial nidopallium, NCM). Left is rostral, top is dorsal. 20x confocal image (top right) shows non-laminar clustered cytoarchitecture of NCM. Bottom: magnified from white box in top right (blue=DAPI; white=NeuN).

(B) Non-overlapping cell types identified by viral expression of fluorophores (left), and 60x images of transduced CaMKIIα (green) and GAD1 (magenta) cells.

(C) Cell soma area at largest cross-sectional diameter.

(D) 60x images of viral and antibody co-expression. Top row: CaMKIIα viral expression (green) co-localizing with CaMKIIα immunolabeling (white). Bottom: GAD1 viral expression (magenta) co-localizing with GABA (gold) and parvalbumin (PV; white) immunolabeling. White arrowheads = co-localization.

(E) 40x images of viral and antibody labeling. Top row: typical CaMKIIα viral expression (green) not co-localizing with GABA (gold) or PV (white); white asterisk shows example of weaker CaMKIIα viral expression (~8% of cells) that co-express GABA and PV. Bottom row: typical GAD1 viral expression (magenta) not co-localizing with CaMKIIα immunolabeling (white); asterisks show example of GAD1 viral expression (~8% of cells) that co-expresses CaMKIIα. White arrowheads in both rows show position of exemplar viral cell across each image.

(F) Typical widefield view of viral injection site in NCM. Shown is mDlx-GFP expression (left), PV immunolabeling (middle), and overlaid images (right) showing specificity (>66% viral cells co-labeled) and efficiency (>88% PV cells co-labeled) of transduction. All scale bars = 50 microns.

See also Figure S1.

Until the latter half of the 20^th century, a majority of avian pallium, the dorsal ventricular ridge, was thought to have evolved from the striatum, and therefore thought to be dissimilar to mammalian pallium⁴. On the basis of circuit connectivity and morphology, Karten and colleagues proposed that avian pallial cell types and circuits are homologous to those in mammalian neocortex^28–30. Subsequently, a number of hypotheses emerged, including the nuclear-to-layer or cell-type homology hypothesis (i.e., homologous circuits are arranged as interconnected nuclei in birds, and as layers in mammals^{5,18,29,31–33}), and the claustrum-amygdala hypothesis (i.e., a majority of avian pallium is derived from the field from which the mammalian cortical amygdala and the claustrum develop, and circuit similarities to mammalian neocortex are the result of convergent evolution^{14,17,22,34–37}). Further complicating matters is that the mammalian cortical amygdala and neocortex each contain similar proportions of principal glutamatergic neurons and interneurons with characteristic physiological properties and local microcircuits^38–43 (likely also true for the mammalian claustrum, reviewed in⁴⁴). This suggests that ancestral pallial cell types shared defining characteristics, which were then deployed in diverse brain regions and networks over evolutionary time. An examination of the physiology of genetically-identified pallial cell types in birds therefore provides an opportunity to test how physiological and molecular identities were retained or dissociated following the last common ancestor of amniotes.

While there remains a lack of consensus over how the avian pallium is related to mammalian pallium, there are unquestionable similarities in pallial function and circuit organization between the two taxa. The avian pallium is thought to support computations important for advanced cognition that are typically ascribed to mammalian neocortex, including fast mapping of complex, ecologically-relevant sensory stimuli, vocal learning, encoding numerical information, cross modal associative learning, and decision-making based on abstract rules^45–55. The connectivity of avian and mammalian neocortex are highly similar, including thalamic input to modality-specific thalamorecipient pallial structures, high interconnectivity of pallial regions, and output to subpallial brain structures^30,56–59. Avian auditory and visual pallia exhibit laminar and columnar organization^6–10. Three-dimensional polarized light imaging of fiber tracts across avian species recently found that, similar to mammalian neocortex, birds have orthogonal radial and tangential fiber organization across pallial lamina⁸. Brain molecular expression analyses have identified striking similarities between birds and mammals at the level of pallial cell types^{13,18,22,31,32} Physiological recordings of cell types in the pallial components of the avian song motor system, a network of novel structures in song-learning birds with many genetic specializations compared to neighboring pallial areas^32,60, have revealed projection neurons and interneurons with similar physiological properties to mammalian pallial cell types^{24–26,61–66}. In avian primary auditory pallium, cell types identified by waveform and coding principles correspond to canonical cell types in mammalian primary auditory neocortex^7,67–69. Even so, the genetic identity of physiological cell types in the avian pallium has not been resolved.

Areas of the avian brain thought to play a key role in advanced cognition, some of which lie in the caudal nidopallium, resist traditional approaches to identify projection neurons and interneurons in vivo (via antidromically stimulating clearly defined projection regions, e.g.^25,63,66). Recent work identifying mammalian cortical-like fiber architecture in the avian forebrain was unable to identify clear architectural logic to the fiber orientations in the caudal nidopallium⁸. The caudomedial nidopallium (NCM) sits in this area, a secondary auditory associative region of pallium (Fig. 1A). NCM is highly interconnected, is involved in auditory memory and individual recognition, and the auditory physiology of neurons in this region is well-described^{50,68,70–73}. Extracellular recordings in NCM consistently identify broad-spiking neurons with sparse firing rates that respond selectively to ecologically-relevant stimuli, and narrow-spiking neurons with higher firing rates that are less selective among stimuli^{7,25,50,63,68,70–75}. The physiology of NCM neurons therefore strongly resembles the principal projection neurons and inhibitory interneurons of mammalian pallium, yet to date this categorization remains putative.

Viral optogenetic tools allow for promoter-specific genetic access to cell types and manipulation of specific populations in vivo. In birds, viral optogenetic tools have been successfully used to target and manipulate avian forebrain neurons (e.g.,^26,76–78). Here, we establish methods for promoter-specific viral optogenetics in the NCM of zebra finches (Taeniopygia guttata) to test the physiological profiles of populations accessed using the molecular phenotypes of mammalian pallial excitatory and inhibitory neurons. Specifically, in mammalian pallium, the calmodulin-dependent kinase alpha (CaMKIIα) promoter is used to access excitatory projection neurons (e.g., ⁷⁹), while the GABA-producing enzyme glutamate decarboxylase 1 (GAD1) is used to access inhibitory interneurons (e.g., ⁸⁰). In this study we characterized CaMKIIα and GAD1 populations using viral constructs with both in vitro and in vivo approaches. We used an mDlx enhancer sequence as an additional, complementary way to access inhibitory populations (previously validated in songbirds^81,82; migration of mDlx cells from the subventricular zone to pallium is similar to tangential migration of interneurons in mammalian pallium²¹).

If avian pallial cell types exhibit similar circuit and computational roles as observed in mammals, then promoters that distinguish mammalian cell types (i.e., CaMKIIα and GAD1) should categorically define viral-driven optogenetic access to broad- and narrow-spiking pallial cells in NCM. Additionally, if avian cell types share similar network roles with their mammalian pallial counterparts, then transient optogenetic manipulation of inhibitory populations in NCM will directly alter local suppression, synchrony of interneurons that drive gamma oscillations in the local field potential, and stimulus selectivity of local target neurons.

Results

Viruses targeting CaMKIIα and GAD1 promoters segregate cell types in avian auditory association pallium

We first injected zebra finch NCM with adeno-associated viruses driving opsin proteins under the control of the CaMKIIα vs. GAD1 promoters, and observed clear segregation of transduced cells (Fig. 1B). The ratio of CaMKIIα vs. GAD1 abundance in co-transduced tissue was 83:16, similar to ratios for projection vs. interneurons in avian nidopallial HVC²⁴ (though interneurons are potentially much sparser in other song system regions ⁶¹), and echoed the distribution of excitatory and inhibitory cell types in mammalian pallium (GABAergic neurons typically comprise ~10-20% of neocortical and cortical amygdalar neurons, e.g.^42,83). In a subset of birds (N=2), we co-labeled tissue with a neuronal-specific marker, NeuN⁸⁴, which co-localized with all virally-transduced cells examined. Qualitatively, we observed that CaMKIIα cells were morphologically variable with respect to dendritic spine density and thickness of dendritic branches, while GAD1 cells were more often aspiny, with thinner processes (fig. S1). CaMKIIα cells and GAD1 cells did not differ in soma size (Welch’s t_29.4 = 0.66, P = 0.51; Fig. 1C). Conventional antibody staining confirmed selective transduction of cell-type targets (Fig. 1D–F). Because imperfect infectivity could create a low estimate of viral co-expression, we also quantified immunolabeled tissue from separate CaMKIIα and GAD1 injections for their anti-targets (Fig. 1E). 91% of CaMKIIα viral expression did not co-localize with GABA immunolabeling (n = 34 viral cells; 3 co-labeled). Similarly, 92% of GAD1 viral expression did not co-localize with CaMKIIα immunolabeling (n = 49 viral cells; 4 co-labeled). This co-labeled population is consistent with the estimated 1-10% of long-range projecting GABAergic neurons in mammalian neocortex^85–89. Thus, promoter-driven molecular cell identity segregates distinct cell types in NCM as it does in mammalian pallium.

CaMKIIα and GAD1 neurons in NCM have distinct physiological properties

We next examined the physiology of genetically-identified cell types in NCM. Mammalian excitatory pallial neurons typically have phasic, accommodating responses to depolarizing current injections and broad spike widths, whereas neocortical interneurons have tonic responses and narrow spike widths^41,90,91. In vitro and in vivo studies of cell types in the avian song system show a similar distinction^61,63,64. In NCM whole-cell recordings (Fig. 2A), depolarizing current steps elicited phasic firing profiles from CaMKIIα cells vs. tonic firing profiles from GAD1 cells (Fig. 2B,C). Similar to excitatory neurons in mammalian barrel and other cortices^92–94, CaMKIIα cells responded with maximum 1-2 spikes (ISIs 48.8 +/− 26.6 ms; range 20-114 ms), in contrast to the tonically-responsive, non-accommodating profile of GAD1 cells (ISIs 48.2 +/− 11.4 ms; range 35-70 ms) that exhibit very low adaptation ratios⁹⁵: 1.105 +/− 0.02. CaMKIIα cells also had broader action potential widths (Mann-Whitney test: W = 162, P = 0.005; Fig. 2D) and afterhyperpolarization half durations (Mann-Whitney test: W = 198, P = 9.5e-06; Fig. 2E) compared to GAD1 cells, but did not differ in other passive membrane properties (all P > 0.1; input resistance (Welch’s t_20.8 = −1.4728, P = 0.1558), rheobase (t_23.8 = 1.3593, P = 0.1868) shown in Fig. 2F,G; resting membrane potential, action potential threshold, peak, and amplitude, and latency to respond to light pulse shown in fig. S2A–E). Next, we conducted in vivo optrode recordings to isolate properties of photoidentified CaMKIIα vs. GAD1 single units (Fig. 2H). Spike widths were broader in CaMKIIα units than GAD1 units (action potential width (peak-to-trough): W = 318, P = 2.7e-05; Fig. 2I; action potential width at quarter height (i.e., spike quarter-width): W = 342, P = 9.9e-07; Fig. 2J; no difference in non-light-evoked units; fig. S2F,G). By contrast, single units transduced with a pan-neuronal viral construct (hSyn1 promoter) spanned the range of action potential widths (fig. S2H,I), further emphasizing that CaMKIIα vs. GAD1 promoter constructs selectively isolate physiologically-distinct populations. Light-evoked spike latencies were longer for CaMKIIα units compared to GAD1 single units (W = 273.5, P = 0.004; Fig. 2K), consistent with their slower spike onset kinetics (fig. S2J) and greater degree of network suppression. The sharp physiological distinctions between CaMKIIα and GAD1 neurons in NCM therefore mirror those of mammalian pallial excitatory neurons and inhibitory interneurons, respectively. Furthermore, our findings in vivo and in vitro indicate that the large volume of prior work that has distinguished crucial features and network computations based on extracellular waveform duration and putative identity in the avian auditory pallium (see Introduction) is now buttressed by molecularly-segregated physiological cell types.

Fig. 2. CaMKIIα and GAD1 single units in NCM have distinct physiological properties.

(A) Transduced cells in in vitro whole-cell current clamp configuration (left) and reliable photopotentials to 25 ms blue light pulses (top row = CaMKIIα cell; bottom row = GAD1 cell).

(B) CaMKIIα cells exhibit phasic responses (top) while GAD1 cells exhibit tonic responses (bottom) to current steps.

(C) Mean ± SEM action potentials for CaMKIIα cells (green; n = 18) and GAD1 cells (magenta; n = 11) in response to current steps.

(D) Spike width of CaMKIIα cells vs. GAD1 cells in whole-cell recordings. *P < 0.05 for Mann-Whitney U test.

(E) Afterhyperpolarization half-duration of CaMKIIα cells vs. GAD1 cells in whole-cell recordings. *P < 0.05 for Mann-Whitney U test.

(F) Input resistance of CaMKIIα cells vs. GAD1 cells in whole-cell recordings.

(G) Rheobase of CaMKIIα cells vs. GAD1 cells in whole-cell recordings.

(H) Raster plots and histograms of exemplar in vivo transduced single units in electrophysiological response to blue light pulses.

(I) Action potential widths of optically-identified CaMKIIα vs. GAD1 single units in vivo. *P < 0.05 for Mann-Whitney U test.

(J) Spike quarter-widths of optically-identified CaMKIIα vs. GAD1 single units in vivo.. *P < 0.05 for Mann-Whitney U test.

(K) Latency to light-evoked response peak of optically-identified CaMKIIα vs. GAD1 single units in vivo. *P < 0.05 for Mann-Whitney U test.

See also Figure S2.

Auditory coding roles distinguish CaMKIIα and GAD1 neurons in NCM

Auditory coding roles are distinct between cell types in mammalian auditory neocortex, in which interneurons have higher auditory-evoked activity and quicker latencies compared to sparse-firing excitatory cells^96–98. Similarly, in songbird pallium, auditory-evoked firing rates are higher in narrow-spiking neurons and song system interneurons as compared to broad-spiking or antidromically-identified projection neurons^{7,63,68,71,75,99}. We therefore tested whether in vivo responses of genetically-identified CaMKIIα vs. GAD1 neurons in NCM segregated accordingly. Conspecific song drove GAD1 units at higher rates compared to CaMKIIα units (Fig. 3A,B; W = 106, P = 0.040) and GAD1 units had a quicker response latency to white noise compared to CaMKIIα units (Fig. 3C; W = 207, P = 0.002). In rat auditory cortex, inhibitory interneurons typically respond promiscuously to sensory stimuli, whereas excitatory neurons in association layers have more selective representations⁹⁸. Likewise in associative auditory avian pallium, putative excitatory neurons are more stimulus-selective compared to putative interneurons⁶⁸. Single-unit spike trains fed to a custom pattern classifier produced higher decoding accuracy values for GAD1 than for CaMKIIα units, indicating that GAD1 units carried information about a variety of auditory stimuli (W = 84, P = 0.006825; Fig. 3D,E; fig. S3). By contrast, when examining stimulus selectivity (see Methods), CaMKIIα single units tended to be more selective for a subset of conspecific song stimuli (W = 241, P = 0.057; Fig. 3D,F). These distinctions in auditory coding roles for CaMKIIα and GAD1 neurons in NCM precisely match the established sensory encoding roles of mammalian pallial excitatory neurons vs. inhibitory interneurons.

Fig. 3. Auditory coding roles distinguish CaMKIIα and GAD1 neurons in NCM.

A) Spectograms and rasterplots of a CaMKIIα single unit (top) and a GAD1 single unit (bottom) in NCM responding to conspecific song.

(B) Song-evoked Z scores of optically-identified single units. *P < 0.05 for Mann-Whitney U test.

(C) Latency in seconds for optically-identified single units to respond to white noise stimuli. *P < 0.05 for Mann-Whitney U test.

(D) Heat map of single unit timing accuracy measures across auditory stimuli. Values closer to 1 represent higher classifier timing accuracy. Histogram insets show density distribution of accuracy metric across stimuli for CaMKIIα (left) and GAD1 single units (right).

(E) Pattern classifier timing accuracy averaged across auditory stimuli for optically-identified single units displayed in D. Accuracy is pattern classifier performance in correctly assigning spike trains to auditory playback stimuli. For details, see methods. *P < 0.05 for Mann-Whitney U test.

(F) Measure of transduced single unit selectivity for subsets of conspecific stimuli (see Methods). #P = 0.057 for Mann-Whitney U test.

See also Figure S3.

GAD1 and/or mDlx neurons in NCM drive suppression, synchrony, gamma oscillations, and stimulus selectivity

One well-described feature of mammalian pallial microcircuits is feedforward suppression, in which incoming excitation immediately drives inhibitory interneurons followed by excitatory neurons in tandem, yielding temporally-precise excitation quenched rapidly by inhibition^{40,96,100–102}. There is clear evidence of feedforward local suppression in the avian song control nucleus HVC^66,103. Consistent with this prediction, genetically-identified GAD1 neurons in NCM had faster auditory onset latencies than genetically-identified CaMKIIα neurons (Fig. 3C). Furthermore, anesthetized optrode recordings showed that GAD1-ChR2 optical stimulation drove short-latency suppression in ~25% of non-transduced single units, compared to 0% for CaMKIIα-ChR2 experiments (fig. S4A,B). To test this further in awake animals, we directed a 32-channel opto-microdrive at the NCM of mDlx-ChR2-transduced birds. At sites with photo-identified mDlx neurons, a separate population of cells were photo-suppressed and quickly rebounded at light-pulse offset (Fig. 4A). We also identified units that rebounded at light-pulse offset using a GAD1-ChR2 construct in an awake animal (fig. S4C). Therefore, both genetically-identified mDlx and GAD1 interneurons suppress local synaptic targets in avian NCM, in a manner similar to mammalian pallium.

Fig. 4. GAD1 and/or mDlx neurons in NCM drive suppression, synchrony, gamma oscillations, and functional auditory response properties.

(A) Instantaneous firing rates of light-evoked single units (magenta; n = 5) and light-suppressed single units (orange; n = 5) from mDlx-ChR2 optical stimulation drawn from n = 36 total units isolated in NCM in vivo.

(B) Violin plots show change in cross-correlation of waveforms following mDlx-ChR2 optical stimulation experiments (top row) and GAD1-archaerhodopsin optical stimulation experiments (bottom). Red line denotes zero change. *P<0.001 for Mann-Whitney U tests.

(C) Heat map (% max LFP power) showing LFP change over time in an mDlX-ChR2 optodrive experiment. Cartoon lasers above represent bins with blue light pulses.

(D) LFP power spectra before and during mDlx-ChR2 optical stimulation experiments (top), and before and during GAD1-archaerhodopsin optical stimulation experiments (bottom). Gray shading represents gamma frequency range for which LFP power is significantly different from baseline according to predictions from mammalian pallium (P < 0.05); $ is range for which LFP power is significantly different from baseline against predicted direction (P < 0.05).

(E) Single units with broad waveforms are plotted with respect to their stimulus-evoked firing rates to various conspecific songs following GAD1-archaerhodopsin optical stimulation (y-axis) compared to no optical stimulation (x-axis). Each point represents a single unit’s response to one conspecific song, and single units are grouped by a unique shape and color.

(F) The same single units as E are plotted with respect to their selectivity across all conspecific stimuli when green laser was on (y-axis) compared to when green laser was off. For E & F, points deviating from the red line denote a non-zero difference between laser off and laser on conditions.

(G) Summary schematic showing features of avian CaMKIIα neurons (green) and GAD1/mDlx neurons (magenta) in NCM in reference to predictions from mammalian pallium. Arrows imply effects of neuron stimulation and do not imply nature of synaptic connectivity.

See also Figure S4.

Inhibitory interneurons in mammalian neocortex and pallial amygdala also coordinate broad-scale spike synchrony and oscillations in the gamma frequency band^104–106, and pairs of interneurons can spike synchronously in the avian song nucleus HVC⁶⁶. With mDlx-ChR2, optical stimulation of NCM increased synchrony of single units with narrow waveforms (narrow single units, W = 4184, P = 2e-05; broad single units, W = 146, P = 0.124; Fig. 4B). Optical stimulation of mDlx cells in NCM also increased the amplitude of the gamma frequency band at 30-34 and 36-43 Hz (All P < 1e-05 after Bonferroni corrections; Fig. 4C,D). Optodrive recordings with a GAD1-ChR2 construct showed similar results (fig. S4D,E). Conversely, with a GAD1-archaerhodopsin construct that hyperpolarizes avian cells (fig. S4F; see fig. S2G,H for green laser controls in a bird without archaerhodopsin transgene expression), optical suppression decreased synchrony of narrow but not broad cells (narrow: W = 5288.5, P = 6e-04; broad: W = 206, P = 0.95; Fig. 4B) and attenuated the gamma band specifically at 39-41 Hz (All P < 0.001 after Bonferroni corrections; Fig. 4C,D). Rapid shifts in NCM inhibitory network activation therefore drive tuning of pallial network synchrony and gamma oscillations.

Finally, a balance of excitation and inhibition is thought to be critical for neocortical function¹⁰⁷. In rodent auditory neocortex, transient disruption of inhibition with optogenetically-targeted suppression of interneurons alters excitatory neuron auditory responses, including firing rates, and stimulus selectivity¹⁰⁸. To test this in NCM, we recorded the responses of broad waveforms to conspecific songs with and without green laser stimulation in GAD1-archaerhodopsin birds. Firing rates during stimulus presentations in single units with broad waveforms were not different as a group between laser off and laser on conditions (Fig. 4E; Wilcoxon signed rank test, V = 55; p = 0.80; Chi-square test for increased/decreased firing, X₁ = 0.8, p = 0.371). However, all single units with broad waveforms had decreased stimulus selectivity during laser on vs. laser off conditions (Fig. 4F; Wilcoxon signed rank test, V = 15; p = 0.063; Chi-square test for increased/decreased selectivity, X₁ = 5, p = 0.025). This suggests that transient disruption of inhibition in NCM has functional consequences for coding of ethologically-relevant auditory stimuli.

Discussion

Our findings demonstrate compelling computational and physiological similarities between corresponding neuronal cell types of mammalian pallial cortex and avian auditory association pallium (i.e., NCM) when brought under optogenetic control (Fig. 4G). Viruses designed to target excitatory vs. inhibitory neurons in mammalian pallium genetically segregate neuronal populations in NCM with predicted intrinsic physiology, auditory coding roles, and network organization that drives local suppression and synchronizes large-scale gamma oscillations, all despite a divergent macro-organization. Our data in NCM support the hypothesis that ancestral pallial cell types of amniotes retained (or less parsimoniously, converged upon) an array of physiological and network features that may be fundamental to advanced cognitive abilities.

The mapping of physiological roles to molecular identity of excitatory and inhibitory neurons in pallium is not a generalized feature of the vertebrate or even the mammalian brain. For example, in contrast to pallium, GABAergic medium spiny neurons in the striatum express high levels of CaMKIIα in both birds and mammals, and are the major source of efferent projections from the region^109–111, and both projection and interneurons in the striatum release GABA^110,112. In the lateral hypothalamus, parvalbumin-positive, glutamatergic neurons are fast-spiking and send long-range projections¹¹³. Relative proportions of glutamatergic vs. GABAergic cell populations also vary widely among brain regions, including many where glutamatergic neurons are the minority^114–116. Neurons in some brain regions can also co-release GABA and glutamate^117,118. Therefore, there are many ways evolution and development have shaped functional circuits using combinations of projection cells, interneurons, excitatory and inhibitory neurotransmission, and genetic identities. Our data demonstrate that genetically- and physiologically-identified CaMKIIα and GAD1 cell populations are distinct in NCM, and that they mirror several physiological and network features of excitatory and inhibitory pallial neurons in mammals. These observations strongly suggest that 1) this set of shared features are core physiological properties that are critical for pallial function, and 2) a shared ancestral origin of cell types and circuit elements constrains the evolution of divergent physiological phenotypes between bird and mammalian pallium.

In vitro, mammalian pallial excitatory neurons can be distinguished from inhibitory interneurons by the electrical phenotype of continuous accommodation to depolarizing current injections^41,119. Similar properties have been attributed to antidromically-identified projection neurons versus interneurons in the avian song system^64,65. In the present study in zebra finch NCM, genetically-identified CaMKIIα neurons exhibit very strong adaptation, whereas genetically-identified GAD1 neurons exhibit a non-accommodating electrical phenotype. In vivo physiological recordings extend previous findings in songbirds demonstrating that auditory coding roles of broad-spiking and narrow-spiking neurons show strong accordance with the roles of excitatory neurons and inhibitory interneurons in mammalian auditory cortex^{7,50,68,70–73}; these putative comparisons can now definitively be attributed to CaMKIIα vs. GAD1 neurons specifically. Additionally, our findings suggest that in avian auditory pallium, GAD1 neurons tune stimulus selectivity of excitatory neurons, a functional outcome predicted directly from recent data in rodent auditory cortex¹⁰⁸. These data strongly imply that in birds as in mammals, the molecular identity of excitatory vs. inhibitory neurons in pallium is tied to a fundamental set of physiological and sensory coding properties, despite remarkable differences in macro-anatomical organization^120,121.

Our data also highlight similarities between CaMKIIα and GAD1 cells in NCM and subclasses of excitatory and inhibitory cell types in mammalian pallium. CaMKIIα neurons were largely non-pyramidal in morphology, contrary to the signature output-neuron morphology in mammalian neocortex and the pallial amygdala⁴², and previous work has not identified projections from NCM that exit pallium (though see ¹²²). NCM instead makes reciprocal connections with other pallial domains, including the mesopallium, a large territory with neurons that recent developmental and genetic data demonstrate to have similar molecular markers to mammalian intra-telencephalic (IT) cells^13,58. A majority of CaMKIIα cells in this study could represent a class of conserved IT pallial cells¹²³, though this requires further study. Our GAD1 population exhibits high variance in auditory coding properties (Fig. 3), suggesting this viral approach captures multiple subclasses of inhibitory interneurons. NCM and other pallial regions contain parvalbumin (PV) and calbindin cells, which likely invade pallium during development in a tangential migration similar to mammals^{16,120,124,125}. In the present study we observed that GAD1-positive cells (and mDlx cells) co-localize with PV (Fig. 1D,F). The physiological roles of GAD1 neurons in NCM that we observed, including a fast-spiking phenotype, faster response latencies, broader selectivity for auditory stimuli, and control over gamma oscillations, are consistent with the role of PV cells in mammalian pallium^96,126,127. A fast-spiking interneuron phenotype in the avian song control region HVC has also been shown to be immunopositive for PV, and single-cell RNA sequencing in HVC has identified GABAergic PV neurons^22,24. Together with the present study, these observations suggest that PV cells existed as a distinct subclass of interneurons in an amniote ancestor and retained several functional characteristics in birds and mammals. While PV is often used a proxy for inhibitory interneurons, there is evidence from both birds and mammals that a subset of excitatory neurons express PV as well^24,128–130. Future studies matching molecular identities to physiological properties will be necessary to determine the extent to which subclasses of pallial cell types have diverged or retained intrinsic physiological properties across amniotes. Additionally, future studies of projection patterns within and outside the telencephalon, combined with dendritic and axonal morphology of CaMKIIα and GAD1 cell types, will further clarify the distinctions and similarities between avian and mammalian pallial cell types.

At the network level, we find several similarities in NCM compared to mammalian pallium in the organization of cell types. Selectively activating GAD1 or mDlx neurons in NCM induces local suppression and synchronizes networks of narrow-spiking cells at the level of single units and broad-scale gamma oscillations in the surrounding pallium. Suppressing GAD1 neurons has the opposite effect on pallial synchrony. Suppressing GAD1 neurons also altered stimulus selectivity of broad-spiking single units, suggesting inhibitory control of downstream excitatory sensory coding. However, contrary to prediction, suppression of GAD1 cells did not alter firing rate of broad-spiking units. This suggests a difference between mammalian auditory cortex and NCM in the role of interneurons in constraining firing rate, but more precise cell-type manipulations of interneuron subclasses could reveal effects on firing rates that are obfuscated in the present study by simultaneous activation of multiple interneuron types with opposing network roles^131–133. Pallial circuits in NCM may therefore exhibit further similarities to mammalian pallial microcircuits, including feedforward suppression^100,101, though our data do not preclude alternative circuit mechanisms for local signal processing.

We obtained qualitatively similar results with the GAD1 promoter and the mDlx enhancer as complementary ways to access inhibitory interneurons in avian pallium. Previous studies in birds have demonstrated that, as in mammals, mDlx appears to mostly define neurons born in the subpallium that invade the pallium during development, and that HVC mDlx cells appear to be GABAergi^16,21,81. Despite the present findings, this does not necessarily mean that the two populations completely overlap in NCM. Futures studies are needed to thoroughly address the proportions of pallial GABAergic neurons that have subpallial origins.

Our current findings highlight both similarities and differences that reveal core intrinsic and network features of CaMKIIα neurons and GAD1 (and mDlx) neurons in the amniote pallium¹³⁴. Our data, however, do not adjudicate between the nuclear-to-layer hypothesis or the claustrum-amygdala hypothesis^17,18. Future studies employing these or similar viral tools can test deeper predictions from these hypotheses. For example, in light of studies showing widespread columnar organization in avian pallium, viral tools could be used to assess whether excitatory and inhibitory cell types exist in different proportions in different layers and/or columns⁸. Additionally, the findings we describe in NCM likely apply to other regions of avian pallium given previous data in the avian song system^{24–26,61–66}, but this is not certain and requires future study. However, the sum total of similarities for genetically-identified cell types in NCM that correspond to mammalian pallium (neocortex, pallial amygdala, and claustrum) appears to most parsimoniously indicate that basal pallial cell types in ancestral amniotes shared a suite of physiological characteristics and network organizations. This idea is further supported by data in lamprey suggesting that some of these characteristics existed in early vertebrates¹². In either case, the retention- or convergent evolution- of certain core physiological and network features of pallial cell types in birds and mammals clarifies essential candidate pallial features for advanced cognition.

STAR METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Luke Remage-Healey (healey@cns.umass.edu).

Materials availability

Custom viral vectors generated in this study are available upon request.

Data and code availability

Original data and code have been deposited to GitHub (https://github.com/HealeyLab/Spooletal2021).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

All Zebra finches (Taeniopygia guttata) used in this study were adults ( > 90 days post hatch date) obtained from a breeding colony in the lab. Both male and female finches were used in experiments, and exact sample size and sex of animals in each experiment are detailed in methods below. Experimental subjects were housed in unisex aviaries prior to experiments under a photoperiod of 14-h light: 10-h dark. Birds were provided with food and water ad libitum, as well as several forms of weekly dietary enrichment (e.g., egg food, fresh millet branches, cuttlebone). All procedures and protocols adhered to the guidelines of the National Institutes of Health Guide for the Care and Use of Laboratory Animals, and were approved by the University of Massachusetts, Amherst Institutional Animal Care and Use Committee.

METHOD DETAIL

Virus injection surgeries

4-6 weeks prior to optogenetic experiments, we injected viruses bilaterally into the caudomedial nidopallium (NCM; the avian auditory association pallium) of male and female zebra finches. Birds were fasted ~30 min prior to surgery, anesthetized with 2% isoflurane in 2 L/min O2, fixed to a custom stereotax (Herb Adams Engineering) equipped with a heating pad (DC Neurocraft) at a 45° head angle, and maintained on 1.5% isoflurane, 1 L/min O2 for the duration of the surgery. Points overlying NCM were marked by scoring the skull lateral and rostral to our coordinates (i.e., a crosshair), and a craniotomy exposed the brain surface (NCM coordinates = 1.1 mm rostral, 0.7 mm lateral of stereotaxic zero, defined as the caudal edge of the bifurcation of the midsaggital sinus). A glass pipette (tip diameter: 20-50 μm) filled with mineral oil was loaded with virus (for constructs and titers see below), attached to a Nanoject (Drummond Scientific Company, Broomall, PA), and lowered into the brain at a depth of 1.5 mm from the surface. We injected 625 nL of virus, 2 nL/sec, 125 nL/cycle, 5 cycles, with a 60 sec wait between cycles. Following injections, we waited 10 min before slowly retracting the injection pipette from the brain. After injections were made in both hemispheres, crainiotomies were filled using Kwik-Cast (World Precision Instruments, Sarasota, FL), and the scalp was fixed around the crainiotomy using cyanoacrylate adhesive.

The viruses and approximate titers used in the present study included: pAAV9-CaMKIIα-hChR2(E123A)-EYFP (titer: 1×10¹³ viral genomes/mL; Addgene: 35505)¹³⁵; pAAV9-hGAD1-FLAG-NLS-Cre (titer: 1.25x10¹² viral genomes/mL; made in-house; rAAV-GAD-SD/SA-YTB plasmid generously gifted by Dr David Lyon; see ¹³⁶ for promoter selectivity in mammals), pAAV9-mDlx-ChR2(H134R)-EYFP (titer: 1.5×10¹² viral genomes/mL; made in-house; mDlx viral construct additionally validated in zebra finch nidopallium in previous work^81,82), AAV9-hSyn-hChR2(H134R)-eYFP-WPRE-hGH (titer: 3.36x10¹²; made in-house), AAV9-pCAG-FLEX-tdTomato-WPRE (titer: 1.5×10¹² viral genomes/mL; Addgene: 51503)¹³⁷, AAV9-FLEX-rev-ChR2(H134R)-mCherry (titer: 5×10¹² viral genomes/mL; Addgene: 18916)¹³⁸, and AAV-FLEX-Arch-GFP (titer: 5×10¹² viral genomes/mL; Addgene: 22222)¹³⁹.

Immunofluorescence

A subset of birds (N = 3; 2 males, 1 female) given injections with viruses targeting both CaMKIIα and GAD1 promoters were transcardially perfused with 4 % paraformaldehyde 4-6 weeks following surgery. Brains were extracted, post-fixed overnight in 4 % paraformaldehyde, then dehydrated in 30 % sucrose in 0.1 M phosphate-buffered saline (PBS). After dehydration, brains were placed in 2 x 2 x 2 inch plastic blocks filled with cryo-embedding compound (Ted Pella Inc., Redding, CA). Brains were sectioned coronally at 40 microns and sections were stored in cryoprotectant solution (30 % sucrose, 1 % polyvinylpyrrolidone, 30 % ethylene glycol, in 0.1 M phosphate buffer) at −20 °C.

We labeled tissue sections for aromatase, a reliable marker for NCM compared to neighboring regions^120,140. At all times during this procedure tissue was processed in a shaded room away from direct light sources. Tissue sections were washed 5 times in 0.1 M PBS, 3 times in 0.1 M phosphate-buffered saline with 0.3% triton (0.3 % PBT), blocked using 10 % normal goat serum, and incubated at 4 °C for two days in rabbit anti-aromatase diluted 1:2,000 in blocking serum (aromatase antibody provided as a generous gift from Dr. Colin Saldanha). In N = 2 birds, tissue was simultaneously incubated in mouse anti-NeuN (Millipore, Danvers, MA; RRID: AB_2298772) diluted 1:2,000 in blocking serum.

Sections were then washed 3 times in 0.1 % PBT and incubated in secondary antibodies diluted 1:500 in 0.3 % PBT. Secondaries used were goat anti-mouse Alexa 405 (Abcam, Cambridge, MA; for N = 2 birds, tissue that was incubated in mouse anti-NeuN above) and goat anti-rabbit Alexa 647 (Invitrogen, Waltham, MA; for N = 3 birds). After 3 more washes in 0.1 % PBT, sections were mounted onto slides, and coverslipped using Prolong antifade mounting medium with DAPI stain in the 405 channel (for bird with tissue incubated in mouse anti-NeuN, Prolong mounting medium contained no DAPI; Invitrogen, Waltham, MA). Slides were dried overnight at room temperature, and stored at 4 °C until imaging.

Sections were imaged on a confocal microscope (A1SP; Nikon, Tokyo, Japan) at the UMass light microscopy core facility. Images were acquired using NIS-Elements software (RRID: SCR_002776). We determined gain and laser intensity separately for each tissue section to minimize background fluorescence. Injection sites were localized within NCM by dense fluorescence in the aromatase-positive field of NCM. To ensure we were only capturing images of transduced cells within NCM, pictures of ventral and dorsal NCM were taken only within the aromatase-positive population of the region dorso-medial to the medial part of the dorsal arcopallial lamina.

Images of tissue sections containing NCM were first taken at 10X to serve as a reference for higher magnification imaging. All cells within an image in NCM were taken at 60X using z-stacks of 1-2 μm through the entire thickness of the tissue section.

For immunofluorescent experiments validating the target-specificity of viral constructs (CaMKIIα virus: N = 1 male, N = 1 female; GAD1 virus: N = 1 male, N = 1 female; mDlx virus: N = 1 male), procedures were identical to above but using the following primary antibodies: rabbit anti-CaMKIIα (GeneTex, Irvine, CA; RRID: AB_10728838) diluted at 1:1,000, rabbit anti-GABA (Sigma Aldrich, St. Louis, MO; RRID: AB_477652) diluted at 1:1,000, mouse anti-PV (Millipore; St. Louis, MO; RRID: AB_2174013) diluted at 1:10,000. Confocal imaging for these experiments were taken at 20X and 40X using z-stacks of 1-2 μm through the entire thickness of the tissue section.

Analysis

For all confocal images, a single observer quantified number of cells using Fiji (RRID: SCR_002285)¹⁴¹. In experiments with tissue transduced with viruses targeting both CaMKIIα and GAD1 promoters, the observer quantified the number of EYFP-positive cells, the number of tdTomato-positive cells, and the number of cells expressing both fluorophores (reflecting transduction of the CaMKIIα and GAD1 promoter, or both, respectively). The total number of transduced cells across all observed slices (n = 122) were tallied. In experiments to confirm specificity of viral transduction, the total number of cells expressing viral-generated fluorescence were tallied, and then co-localization of other protein markers (anti-GABA, anti-PV, anti-CaMKIIα) was assessed by extensive overlap of fluorescence in the cell soma across Z stacks. In all experiments, cells were counted only when positive staining was associated with DAPI or NeuN. In slices that were stained with antibody targeting NeuN, no labeled cells were observed that did not co-localize with NeuN labeling.

In vitro neuronal response properties

Birds were rapidly decapitated and dissected, then the caudal telencephalon was bisected on a petri dish immersed in wet ice and each hemisphere was sectioned at 250 microns on a vibratome. A glycerin-based external cutting solution (substituting for NaCl & sucrose) was used for sectioning to improve slice survival time ¹⁴², containing (in mM): 222 glycerin, 25 NaHCO₃, 2.5 KCl, 1.25 NaH₂PO₄, 0.5 CCaCl₂, 3 MgCl₂•6H₂O, 25 dextrose, 0.4 ascorbic acid, 2 sodium pyruvate, 3 myo-inositol; 310 mOsm/kg H₂O; pH 7.4 when saturated with 95% O₂/5% CO₂. Following sectioning, the slices were warmed to ~ 40 °C in external solution containing (in mM): 111 NaCl, 25 NaHCO₃, 2.5 KCl, 1.25 NaH₂PO₄, 2 CaCl₂, 1 MgCl₂•6H₂O, 25 dextrose, 0.4 ascorbic acid, 2 sodium pyruvate, 3 myoinositol; 310 mOsm/kg H₂O; pH 7.4 when saturated with 95% O₂/5% CO₂. Our potassium-based internal solution contained (in mM): 2.4 potassium gluconate, 0.4 KCl, 0.002 CaCl2, 0.1 HEPES, 0.1 EGTA, 0.06 Mg-ATP, 0.01 Na-GTP, 0.4 C4H8N3O5PNa2•4H2O; 290-305 mOsm/kg H2O; pH 7.4.

Tissue was imaged using a charge-coupled camera (QIClick; QImaging) mounted to a fixed stage microscope (Eclipse FN1; Nikon) that was equipped with a water emersion objective (CFI Fluor; 40X; NA = 0.8; WD = 2.0 mm; Nikon). Glass pipettes were pulled from borosilicate glass capillary tubes (1B150F-4; World Precision Instruments) using a two-stage, vertical puller (PC-10; Narishige International USA). Positive pressure was applied to each pipette while moving through aCSF and tissue. Liquid junction potential was automatically subtracted. Pipettes had a tip resistance of 4-8 MΩ when backfilled with internal solution, which routinely included neurobiotin tracer (Vector Laboratories) for identification post-recording. Once a whole-cell configuration was successfully achieved, series resistance and slow capacitive transients were counterbalanced. Each recording configuration was allowed to stabilize for a minimum of five min before beginning protocols ¹⁴³. Experimental data were acquired at 40 kHz using multichannel acquisition software (PATCHMASTER; HEKA Elektronik), digitized using a patch clamp amplifier (EPC 10 USB; HEKA Elektronik), then exported to Igor Pro (WaveMetrics; RRID: SCR_000325) and processed using a 1 kHz low-pass filter.

Post-recording, tissue sections were drop fixed in 4% Paraformaldehyde dissolved in 0.025 M phosphate buffer (PB), then stored at −20 °C in a cryoprotectant solution composed of 30% sucrose, 30% ethylene glycol, and 1% polyvinylpyrrolidone in 0.1 M phosphate buffer (PB).

Analysis

Recordings were made from N = 5 birds (N = 3 males, N = 2 females), n = 18 cells transduced with CaMKIIα-ChR2 and N = 4 birds N= 2 males, N = 2 females), n = 11 cells transduced with GAD1-ChR2. Steady-state voltage (SSV), action potential kinetics, and action potential width (duration at 50% of action potential depolarization peak from resting membrane potential) were each calculated using custom scripts written in Python.

Anesthetized extracellular recording

Extracellular recording was performed in vivo under urethane anesthesia as in previous studies¹⁴⁴. On the day of recording, birds were injected in the pectoral muscle with 90-120 μL 20% urethane (30 μL every 45 min; specific amount depended on the mass of the bird). Anesthetized birds were fixed to a custom stereotaxic apparatus as above and a stainless steel headpost was fixed to the head using acrylic cement. Birds were then moved to a sound-attenuation booth (Industrial Acoustics) on an air table (TMC, Peabody, MA) for optotagging and extracellular recording experiments. In the booth birds were fixed to a custom stereotax (Herb Adams Engineering) at a 45° head angle using the attached headpost and the craniotomy above NCM was exposed. We recorded preferentially from the left hemisphere but both hemispheres are represented in our dataset.

An optrode consisting of a single tungsten electrode (A-M Systems, Sequim, WA) epoxied to an optic fiber (diameter: ~200 μm; Thor Labs, Newton, NJ; distance between electrode and fiber: 400-600 μm) was lowered into the brain at the NCM coordinates specified above. The optrode was calibrated by inserting the optrode into a light meter (Thor Labs, Newton, NJ) in a dark room, adjusting laser power and obtaining a calibration curve of laser setting to output wattage. To elicit ChR2-induced spike activity locally around our electrode we used a blue laser (447 nm) with an output wattage of ~2 mW. Recordings were made between 1.2 and 1.8 mm ventral to the brain surface to search for characteristic NCM baseline and sound-evoked activity, as well as light-evoked activity. Light stimuli consisted of either 25 msec pulses or 100 ms pulses, and pulses were separated by 4-10 sec. Following light-only trials where light-evoked units were identified (see below), we immediately ran auditory trials in the same location. Auditory trials consisted of seven stimuli: 6 conspecific songs and white noise, all normalized to ~70 decibels. Stimuli were randomly presented during trials and each stimulus was presented 15 times. At a given recording site, we randomly presented 3 of 6 conspecific songs and white noise (i.e., 4 stimuli total). Interstimulus interval was 10±2 sec. Recordings were amplified, bandpass filtered (300 to 5000 Hz; A-M Systems), and digitized at 16.67 kHz (Micro 1401, Spike2 software; Cambridge Electronic Design). Following recordings, birds were transcardially perfused and brains were extracted as above for anatomical confirmation of optrode sites.

Analysis

Data were processed in Spike2 (version 7.04) to identify light-evoked units. Recordings from light-only trials were thresholded above the noise band and peristimulus histograms and raster plots were generated to examine multiunit responses to light stimulation. We only performed auditory playback trials if there was a peak in the histogram during the light stimulus that was consistent across trials (as shown by the raster plot; Fig. 3A). Recordings were made from N = 3 birds (1 male, 2 females), n = 22 cells transduced with CaMKIIα-ChR2, N = 3 birds (3 females), n = 16 cells transduced with GAD1-ChR2, and N = 1 bird (1 male), n = 9 cells transduced with hSyn1-ChR2.

Templates for single units in light-only trials were isolated in Spike2 by their waveform characteristics and filtered so that no units had an interspike interval > 1 ms as in previous studies¹⁴⁴. Individual spikes were assigned to generated templates with an accuracy range of 60%-100%. Principal component analyses were used to confirm well-isolated units (i.e., nonoverlapping clusters in 3-dimensional space). We could reliably obtain 2-4 units from each recording site. We used isolated single unit waveforms from light-only trials to sort multiunit activity in their paired auditory trials. At this point we again checked that none of the units had interspike intervals > 1 ms and that principal component analyses still demonstrated nonoverlapping clusters.

For waveform analyses, we compared physiological properties of light-responsive cells between birds with virus targeting the CaMKIIα promoter and birds with virus targeting the GAD1 promoter. We calculated action potential width as the time between the first (depolarized) and second (hyperpolarized) peaks of the average waveform. Spike quarter-width was calculated as the width of the waveform at a quarter of the height of the action potential.

Z scores were calculated for each single unit’s response to conspecific song and white noise to provide a normalized change in firing during stimulus presentation (S) compared to baseline (B), using the following formula as in previous studies ¹⁴⁵:

$$Z=\frac{Mean\left(S\right)-Mean\left(B\right)}{\sqrt{Var\left(S\right)+Var\left(B\right)-2\left(Cov\left(S,B\right)\right)}}$$

Z scores for conspecific songs for each single unit are reported as average single unit responses across conspecific stimuli presented.

We calculated single unit response latency to white noise by adapting a previously described method⁷³. We calculated the mean and standard deviation of the baseline period (2 s before stimulus onset). Then, peri-stimulus time histograms were created for each single unit’s response to white noise, divided into 5 ms bins and smoothed with a 5-point boxcar filter. We identified the first bin within 400 ms of white noise onset in which firing rate passed a threshold of 3 standard deviations from the baseline mean.

To examine auditory coding properties of isolated single units, we used a custom pattern classifier ^71,146. The classifier uses timing accuracy of stimulus-evoked firing responses of single units to discriminate amongst the different stimulus types, providing a measure of how well stimuli can be distinguished by consistency of evoked firing across trials.

Specifically, for each single unit the classifier pseudorandomly picked one response per stimulus to serve as templates. The classifier compared the selected templates with all other stimulus-evoked responses of that single unit. Based on values of a distance metric (detailed below) between responses and the templates, the algorithm would assign a template identity (e.g. conspecific song 1) to the template that was most similar to a given response. The classifier repeated this procedure 1000 times and then determined the mean accuracy of assigning stimulus-evoked responses to their correct associated stimuli.

For the timing accuracy measure, data were convolved with Gaussian filters prior to template comparison. The optimal standard deviation of the filter for each cell was picked based on which value gave the highest accuracy (values that were used: 1, 2, 4, 8, 16, 32, 64, 128, 256 ms). Comparisons between templates and responses used the R_corr distance metric ^71,147.

$$Rcorr=\frac{S_{trial}\cdot S_{template}}{\Vert S_{trial}\Vert \times \Vert S_{template}\Vert }$$

Where s represents the vectors of the trial and the template responses after filtering, dot multiplied and divided by the product of their lengths. For the timing accuracy measure the classifier matched responses (i.e., trials) to templates based on which template yielded the highest R_corr value.

To statistically test whether the accuracy of the classifier was greater than random chance, we generated confusion matrices (exemplars in fig. S3) and used a trial shuffling approach (modified from ^71,146). Classifier assignments of stimulus-evoked responses in the confusion matrix were shuffled and randomly assigned to stimuli 1000 times. The distribution of the classifier accuracies across the 1000 runs was compared to the randomly shuffled assignments. Accuracies were considered significantly greater than random when Cohen’s d was > 0.2 ^71,148. All single unit accuracies in this study had a Cohen’s d > 0.2. Analyses for conspecific song responses are presented as single unit responses averaged across conspecific stimuli presented to a given unit.

Temporal stimulus selectivity was measured by comparing the mean R_corr values across stimuli repetitions (spike-timing correlation) among the 3 different conspecific stimuli. Specifically, for each single unit, the 3 mean R_corr values (Rx, Ry, Rz) were used to generate a 3-dimensional vector $\left(\overrightarrow{s}\right)$. Then, a second vector was generated such that vector magnitude was identical to $\overrightarrow{s}$ but R_corr values for stimuli were equalized in the 3D-space (Rx = Ry = Rz), i.e. the equal-selectivity vector $\left(\overrightarrow{e}\right)$. The angle (in radians) between each single unit’s 3D vector $\left(\overrightarrow{s}\right)$ and its associated equal-selectivity vector (e) was calculated using the arccosine distance metric: $d=across\left(\frac{\overrightarrow{s}\cdot \overrightarrow{e}}{\Vert \overrightarrow{s}\Vert \times \Vert \overrightarrow{e}\Vert }\right)$, where values close to 0 reflect non-selectivity and larger values reflect greater selectivity.

Optodrive awake recordings

For larger scale recordings, custom optodrives were made by coupling a fiber optic (200 μm diameter) to 8 tetrodes arranged in a single bundle (customized from EIB Open-Ephys drive; Open Ephys Inc., MA). Tetrodes were made from 12.5 μm insulated NiCr wires (Sandvik, Sandviken, Sweden). The optic fiber was placed ~600 μm above the wires. The horizontal distance between the fiber and wire bundle was ~200 μm.

Virally transduced birds were anesthetized with isoflurane, after which bilateral craniotomies were made over NCM (see above for coordinates; sealed with Kwik-Cast) and headposts were secured to the dorsomedial rostral skull with acrylic cement. 2-4 days later, birds were restrained and head-fixed. Optodrives were lowered into NCM 1.5-2.0 mm ventral of the brain surface. After finding a site with characteristic NCM baseline and sound-evoked activity, the optodrive was allowed to stabilize in the brain for ~1 h. Following stabilization, a blue (447 nm) or green (532 nm) laser was used to test network responses to manipulation of transduced cells. Recordings were obtained using the Open-Ephys GUI¹⁴⁹, and amplified and digitized at 30 kHz using Intan Technologies amplifier and evaluation board (RHD2000; Intan Technologies, Los Angeles, CA). Laser delivery was controlled by custom MatLab (MathWorks, Natick, MA) scripts and an Arduino Uno integrated with the evaluation board.

For the GAD1 archaerhodopsin experiment, after testing network responses to manipulation of transduced cells by the green laser, we tested how inhibition of GAD1 cells would affect functional auditory response properties of the network. We randomized playback of 4 conspecific songs. Each song was heard 10 times total, 5 times without the green laser, and 5 times with the green laser turned on specifically for the duration of the stimulus. This resulted in 20 playback trials with laser turned off (and songs randomized within these 20 trials such that each song was played 5 times), followed by 20 playback trials with laser turned on only during the duration of each song playback (and songs randomized within these 20 trials such that each song was played 5 times). We allowed 5 seconds of silence in between playback of different stimuli.

Analysis

Optodrive single unit sorting was done with Kilosort¹⁵⁰. Sorting results were manually curated and only well-isolated units (high signal-to-noise ratio; low contamination; good segregation in waveform principal component analysis space; low frequency in violations of refractory period) were used for these analyses. Data were common median filtered and 300 Hz high-pass filtered.

Sample sizes for optodrive experiments were: mDlx-ChR2: N = 1 female; n = 36 single units. GAD1-ChR2: N = 1 male; n = 30 single units. GAD1-archaerhodopsin: N = 1 male; n = 52 single units. An additional n = 7 single units in the GAD1-ChR2 bird were collected using green light as a control for firing rate changes caused by tissue heating¹⁵¹, or light leakage from the drive that could serve as a visual cue to the awake animal. Recordings were made from both left and right hemispheres.

For local field potential (LFP) analyses, raw traces were 150 Hz low-pass filtered and local field potential power spectra were generated in Python (smoothed and imported using Neo (RRID:SCR_000634)¹⁵² and Elephant (RRID:SCR_003833)¹⁵³ libraries; Welch’s power spectrum frequency resolution at 1 Hz).

We also calculated stimulus firing rate in Hertz and auditory selectivity (as described above in the analysis section for anaesthetized optrode recording) for the GAD1-archaerhodopsin experiment to compare responses to conspecific song when laser was off versus when the laser was on.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analyses included two-tailed t-tests and nonparametric statistics (Mann Whitney U tests) when log₁₀ transformations did not correct violations of parametric assumptions. Bonferroni corrections were used to correct for multiple comparisons. Wilcoxon signed rank tests were used to test paired responses to song in the same cells when laser was off compared to on. Chi-square tests were used to test whether the number of cells that increased or decreased firing rates and auditory selectivity was greater than expected by chance. Sample size of birds is denoted using big “N”, whereas sample size of single units and cells is denoted using small “n”. All statistical tests and interpretations are presented in the Results, and figure legends contain definitions of statistical significance.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
mouse anti-NeuN	Millipore	RRID: AB_2298772
rabbit anti-CaMKIIα	GeneTex	RRID: AB_10728838
rabbit anti-GABA	Sigma Aldrich	RRID: AB_477652
mouse anti-PV	Millipore	RRID: AB_2174013
Rabbit anti-Aromatase	Donated by Dr. Colin Saldanha, American University	N/A
Goat anti-mouse Alexa 405	Invitrogen	RRID: AB_221604
Goat anti-rabbit Alexa 647	Invitrogen	RRID: AB_2535812
Donkey anti-mouse Alexa 488	Invitrogen	RRID: AB_2556542
Donkey anti-mouse Alexa 594	Invitrogen	RRID: AB_2556543
Donkey anti-mouse Alexa 647	Invitrogen	RRID: AB_162542
Donkey anti-rabbit Alexa 488	Invitrogen	RRID: AB_2556546
Donkey anti-rabbit Alexa 594	Invitrogen	RRID: AB_2556547
Donkey anti-rabbit Alexa 647	Invitrogen	RRID: AB_2536183
Normal Donkey Serum	Sigma-Aldrich	Cat# D9663
Normal Goat Serum	Vector Labs	Cat# S-1000
Bacterial and virus strains
pAAV9-CaMKIIα-hChR2(E123A)-EYFP	Addgene	RRID: Addgene_35505
AAV9-pCAG-FLEX-tdTomato-WPRE	Addgene	RRID: Addgene_51503
AAV9-FLEX-rev-ChR2(H134R)-mCherry	Addgene	RRID: Addgene_18916
AAV9-FLEX-Arch-GFP	Addgene	RRID: Addgene_22222
pAAV9-hGAD1-FLAG-NLS-Cre	Donated by Dr. David Lyon, University of California Irvine	N/A
pAAV9-mDlx-ChR2(H134R)-EYFP	This paper	N/A
AAV9-hSyn-hChR2(H134R)-eYFP-WPRE-hGH	This paper	N/A
Chemicals, peptides, and recombinant proteins
Prolong antifade mounting medium with DAPI	Invitrogen	Cat# P36962
Prolong antifade mounting medium	Invitrogen	Cat# P36961
Neurobiotin Tracer	Vector Laboratories	Cat# SP-1120
Software and algorithms
NIS-Elements	Nikon	RRID: SCR_002776
Spike2	Cambridge Electronic Design	RRID: SCR_000903
IGOR Pro	WaveMetrics	RRID: SCR_000325
PATCHMASTER	HEKA Elektronik	RRID:SCR_000034
Fiji	ImageJ	RRID: SCR_002285
Open-Ephys GUI	Open-Ephys	https://open-ephys.org/gui
KiloSort	KiloSort	RRID: SCR_016422
MatLab	MathWorks	RRID: SCR_001622
Illustrator	Adobe Systems	SCR_010279
R	R Core Team	https://www.r-project.org/
Python 3	Python Developers	https://www.python.org/about/
Rstudio	RStudio Team	RRID: SCR_000432
Spyder	Spyder Developers	RRID: SCR_017585
Other
Nanoject	Drummond Scientific Company	Cat# 3-000-207
Stereotax	Herb Adams Engineering	N/A
Tungsten Microelectrodes	A-M Systems	Cat# 573210
Optic Fiber (200 micron diameter)	Thor Labs	Cat# M84L01
Light Meter	Thor Labs	Cat# PM100D
Blue Laser	Opto Engine	Cat# MDL-F-447-2W
Green Laser	ReadyLasers	Cat# MGL-III-532-200mW
Amplifier	A-M Systems	Model 1700
Amplifer	RadioShack	Cat# 3200027
Digitizer	Cambridge Electronic Design	Cat# 1401
Microdrive supplies?	This paper	N/A
Nikon A1SP	Nikon Center for Excellente, University of Massachusetts Amherst	https://www.umass.edu/ials/light-microscopy
QIClick Camera	QImaging	Cat# QIClick-F-M-12
Eclipse FN1	Nikon	ECLIPSE FN1
Borosilicate Glass Capillary tubes	World Precision Instruments	Cat# 1B150F-4
PC-10 Pipette Puller	Narishige International USA	N/A
CleanBench Laboratory Table	TMC	https://www.techmfg.com/products/labtables/cleanbench63series
Amplifier and Evaluation Board	Intan Technologies	Cat# RHD2000
Arduino Uno R3	Arduino	Cat# A000066
Tissue-Tek OCT Compound	Ted Pella, Inc.	Cat# 27050
Kwik-Cast	World Precision Instruments	Cat# KWIK-CAST

Highlights.

• CaMKIIα and GAD1 promoters segregate pallial cell types in birds.

• Physiology and auditory coding roles distinguish these cell types in avian pallium.

• GAD1 and mDlx neurons drive local network synchrony and gamma oscillations.

• These distinctions in birds precisely match their mammalian pallial counterparts.