All house crows were weighed prior to starting behavioral training. Experimental crows were housed in an outdoor aviary with natural day and light conditions. Two days prior to training on the visual discrimination experiments, they were transferred to a cage (dimensions: 30″ × 21″ × 34″) in a separate room (visually and auditorily isolated from other crows) also maintained in natural light and dark conditions for habituation. On the second day of this period, they were food-deprived from 4 pm in the evening until the beginning of the pre-training period which started at 9 a.m. the next day. Birds were given ad libitum access to water but the amount of food given post-behavioral training depended on the appetite of each bird, since some of the birds did not participate in the training paradigm if they were fully satiated the previous evening. For the food reward, we used pieces of dried shrimp, chicken sausage, or vitamin-fortified white bread depending on the crows’ preference. A total of two blocks of the shape discrimination task comprising 12 trials each were conducted every day with a gap of 4–5 h, 5 days a week. Over the weekend, experimental birds were transferred back to their aviaries where they were in visual and auditory contact with other crows and provided eggs, bread, dog food (Pedigree), and water ad libitum. There was no reduction in the body weight (∼250 gm) of any of the house crows at the end of training on the behavioral paradigm.

Immunoblotting was performed to check the validity of the DCX antibody [sc-271390, Anti-Doublecortin Antibody (E-6); RRID:AB_10610966, Santa Cruz Biotechnology] using samples of tissue from the anterior striatum of a house crow (n = 1, Female). The tissue was homogenized in SDS lysis buffer and sonicated with 25 pulses of 1 s each (thrice with an interval of 5 s), followed by centrifugation at 12000 rpm at 4°C. The supernatant was collected and protein was estimated using the bicinchoninic acid protein estimation method (BCA, B9643, Sigma-Aldrich). We separated a protein sample (80 μg) on an SDS-PAGE gel (11% Acrylamide-Bisacrylamide gel concentration). The resolved proteins were transferred to a nitrocellulose membrane which was blocked with 5% BSA (Bovine serum albumin) for 2 h. This was followed by incubation in the primary antibody solution (1:1000, anti-DCX) for 13–16 h at 4°C. The primary antibody was rinsed using six washes of TBST (Tris borate saline with 0.1% Tween 20, 10 min each) followed by incubation in a secondary antibody solution (1:3000, peroxidase labeled anti-mouse, PI-2000, Vector laboratories). This was followed by rinsing in TBST, after which the blot was developed with the ECL chemiluminescent reagent (WBKLS0500, Immobilon Western Chemiluminescent HRP Substrate, MERCK, USA). We obtained an intense band at ∼40 kD and a very faint one at ∼32 kD (Supplementary Figure 1B) when a western blot was performed on lysate from a house crow brain using an anti-DCX antibody (sc-271390, Santa Cruz Biotechnology). Bands were obtained at similar molecular weights in western blots performed on mouse brain tissue provided by the manufacturer.

Sections at the level of NCL were stained with TH and demarcated into various subdivisions according to an earlier report from our lab (Sen et al., 2019). In each subdivision, the area of TH-positive profiles was determined using ImageJ (version: 1.52). The contours of different subdivisions in NC (dNC, iNC, mNCL, lNCL, and vNC) determined using TH staining patterns were used to determine the boundaries of NC subdivisions in Arc and DARPP-32-stained sections. We converted the images to 8 bits and thresholded them based on staining in lNCL, followed by quantifying the stained areas (NC subdivisions). We normalized the % area fraction of the five NC subdivisions with a positive control, that is, a band of TH-positive profiles located between LAD (Lamina arcopallialis dorsalis) and AId.

A contour was drawn around the area of interest in serial sections using the Stereoinvestigator software (Microbrightfield, Williston, VT) linked to an Olympus microscope (BX-51). The optical fractionator method (Olesen et al., 2017) was used to count Arc-, DCX-, and DARPP-32-positive cells, which is a stereological counting method and performs a Systematic Uniformly Random Sampling (SURS) by dividing the area of interest into a grid. Cell counts are performed by selecting the sampling fraction which is a proportion of the area (bin) in which counting is performed. Intensely stained Arc / DCX / DARPP-32-positive neurons were counted and considered for analysis. The Arc-positive neurons were counted at 100X using a sampling grid of size 70 μm × 70 μm with an area sampling fraction (asf) of 5 or 10 based on the size of the area and a dissector height of 9 μm with 2 μm guard zones. The DCX and DARPP-32-positive neurons were counted at 40X using a grid size of 175 μm × 175 μm with asf of 50, 30, or 20 and dissector height of 10 μm with 2 μm guard zones. Additionally, we counted the three different neuronal populations stained for DCX, namely, multipolar, fusiform, and spherical. For all fractionator counts, every 6th section was sampled. An estimated count using mean section thickness for Arc, DCX, and DARPP-32 was used to compare the neuronal population across various training groups. To accommodate size differences across brain regions between different birds, we normalized the estimated counts by the number of bins from each area, which is an indicator of the size of the area.

Fluorescence imaging for sections stained for Arc and DARPP-32 were performed using an Apotome microscope (Carl Zeiss1, AxioImager.z1). We imaged z-stacks (magnification: 40X for counting and 63X for imaging) at an interval of 1.5 μm from different subdivisions of NC from all crows used in our experiments on visual discrimination. Manual neuronal counts for Arc and DARPP-32 double-positive neurons were performed from these images using ImageJ software.

We traced 15 DARPP or DCX positive neurons from each subdivision of the caudal nidopallium from each bird using the Neurolucida software (version:11, MBF Bioscience, USA) linked to a microscope (Olympus BX51) at a magnification of 100X. After a careful visual inspection, only those neurons which possessed intact processes were selected for tracing. Since we could not discriminate between the axons and dendrites of neurons based on staining for DARPP-32 or DCX, we have considered all processes as neurites for our study. Traces were exported to the Neurolucida Explorer software and subjected to three morphometric analyses, including (1) Neuron summary analysis for overall changes in the neurite branching and length, (2) Convex hull analysis to quantify changes in the neurite field, and (3) Sholl analysis for quantifying neuronal complexity, wherein the traced neuronal soma was placed at the center of a set of concentric circles at a fixed distance (10 μm). The number of intersections that the neurite tree made with these circles was counted and for each Sholl radius, the total neurite length was analyzed. Statistical analysis was performed on changes observed between the radii from 20 to 70 μm. However, data from changes within the 10 μm Sholl radius were not analyzed since we observed an increase in the size of somata of DARRP-32 and DCX positive neurons in Trained, Undertrained, and No-Association birds compared to Baseline controls and the inclusion of these data might introduce a false estimation of the analyzed parameters. The initial Sholl radius was reduced to 8 μm for DCX neurons from the striatum due to the smaller size of their somata.

Similarly, five Arc and DARPP-32/DCX double-positive neurons each from MSt, mNCL and lNCL regions were traced using the Neurolucida software (version 2020.1.3; MBF Bioscience, USA) linked to a fluorescence microscope (Olympus BX53). The three-dimensional tracings of these neurons were analyzed in the same manner as described above for individually labeled neurons using the Neurolucida Explorer software.

The R (version 4.1.1) software was used to perform statistical tests and visualize the data. Data pertaining to counts (Arc, DARPP-32, DCX and TH) were tested first for normality and homoscedasticity using the Shapiro-Wilk and Levene tests. If the data did not pass these tests, a Kruskal-Wallis test was performed otherwise a one-way analysis of variance (ANOVA) was performed for judging statistical significance. For pairwise multiple comparisons, either Tukey’s (multcomp library; adjusted P-values are generated by the single-step method) or Dunn’s post-hoc test (FSA library using the Holm method for P-value adjustment) were performed.

In order to check the interaction effects between area and experimental conditions for neuronal morphometry data, we performed a two-way ANOVA. For non-normal data, an aligned rank transform was performed, followed by two-way ANOVA (Wobbrock et al., 2011). For group comparisons, data was initially checked for normality and equal variance as described above followed by either one-way ANOVA or the Kruskal-Wallis test and post-hoc tests as described above. For comparisons involving only two groups in some cases, a Welch’s two-tailed t-test or Wilcoxon rank sum test was performed. Details of the statistical analyses for neuronal structural changes (F and/or χ² values, degrees of freedom and exact P-values) are provided in separate Supplementary Tables 1–18 due to the large number of comparisons.

Besides NC (Figure 1A), areas including LMAN, Area X, RA, and AId, generally associated with song control (Brainard and Doupe, 2002) expressed the Arc protein in different groups of house crows (Supplementary Figures 2A, B, 3C). Intensely stained neurons were also present in components of the basal ganglia including LSt, GP (Supplementary Figures 3A, 4A) and MSt, and the midbrain dopaminergic areas, VTA and SN (Supplementary Figure 4C; see Supplementary Figure 3B for a negative control devoid of label at the level of GP and LSt).

Details of the statistical analyses for neuron morphometry data are provided in the Supplementary Tables.

Based on their morphology, DCX-positive neurons in house crows could be categorized as (i) spherical neurons which were devoid of processes, (ii) spindle-shaped unipolar or bipolar fusiform neurons, and (iii) multipolar neurons whose somata were rounded (Brown et al., 2003; Balthazart et al., 2008; Taufique et al., 2018) or triangular (Mehlhorn et al., 2022; Supplementary Figures 12A–C; see Supplementary Figure 12G for the negative control).

Surprisingly, brain regions generally associated with song control (Brainard and Doupe, 2002) were activated after house crows performed the visual discrimination task. The song control area LMAN was significantly activated in No-Association birds vs. the Baseline group of birds. In songbirds such as zebra finches, LMAN is important for generating variability in vocalizations during the sensitive period for song learning (Scharff and Nottebohm, 1991; Ölveczky et al., 2005). Recent studies further demonstrated that LMAN lesions prevented somatosensory-based non-auditory learning which affected vocal output in zebra finches (Mcgregor et al., 2022) and an LMAN-like region in pigeons (NIML) was involved in the execution of sequence learning (Helduser et al., 2013) and serial processing (Rook et al., 2021), but not in generating variability. In our study, trial-by-trial variability would be the highest in No-Association birds since the reward was randomly associated with either of the shapes presented during the task. These findings suggest that besides vocal learning, LMAN is involved in modulating variability associated with learning the visual discrimination task in house crows.

We also observed higher activation in RA and AId of Trained crows vs. other groups. Besides projecting to syringeal musculature for controlling vocalization, RA projects to the ventral respiratory column for controlling respiration (Schmidt and Martin Wild, 2014). Since experimental crows never vocalized during or immediately after training, it is possible that this heightened activity in RA may be important for synchronizing breathing with performing the correct sequence of actions during the task. Another arcopallial region, AId, is involved in motor functions and song learning (Bottjer and Altenau, 2010). It receives topographically organized projections from NCL (Kroner and Gunturkun, 1999; Paterson and Bottjer, 2017) and projects to the optic tectum (Fernández et al., 2020). Increased activation of AId consequent to training on visual discrimination may involve a goal-directed visuomotor pathway beginning in mNCL (unpublished data) which is connected to AId and the optic tectum (Kroner and Gunturkun, 1999; Fernández et al., 2020).

For our study, we were careful not to include neural activity induced due to vocalization. Firstly, all experimental birds were in visual and auditory isolation from other birds starting 2 days prior to the behavioral experiment and also during the experiment. Secondly, we kept all experimental birds in the dark for 90 min before sacrifice, as a result of which there is minimal motor activity and no vocalization. In addition to these factors, the experimental crows never vocalized during the experiment. In earlier studies on zebra finches, period of silence when birds do not sing are associated with a lack of Arc expression in the song control nuclei (Hayase et al., 2018; Hayase and Wada, 2018). Therefore, it is likely that the induction of Arc protein in song control regions is not associated with vocalization but is more likely to be task-driven.

Dopamine plays an important role in motivation, learning, cognition, reward and pleasure, and motor learning and the firing rates of VTA-SNc increase to signal the physical salience of the reward and reward-predicting stimuli (Schultz, 2016). As expected, dopaminergic neurons in VTA-SNc were positive for Arc in all groups of birds other than the unrewarded Baseline controls. Furthermore, NCL contained DARPP-32-labeled dopaminoceptive neurons of which some were positive for Arc, showing that they participated in visual discrimination. Although there were no changes in their number, for the first time, we have demonstrated that the complexity of their neurites and soma size increased significantly across NC, especially in mNCL and lNCL of Trained house crows. These findings suggest that learning to associate specific types of behavior with a reward leads to an increase in the plasticity of dopaminoceptive neurons.

The discovery of experience-induced neurite plasticity began with visual deprivation experiments performed on kittens by Hubel and Weasel (Kandel, 2009). Since then, plasticity in cortical and subcortical circuits was mainly explored in the context of injuries (Chen and Zheng, 2014), experience (Shansky et al., 2009; Bose et al., 2010) and learning (d’Aquin et al., 2022). In adult cats, binocular retinal lesions lead to sprouting of long-range fibers into the reorganized visual cortex (Darian-Smith and Gilbert, 1994). Besides injury, GABAergic non-pyramidal cells demonstrate an expansion and retraction of dendritic tips in layer 2/3 of the visual cortex in mice daily (Lee et al., 2006), suggesting a role for a cell-type specific population in adult neural plasticity. Additionally, in the pyramidal neurons of adult rats, deafness led to a decrease in apical dendrite length but no change in basal dendrites, whereas exposure to an enriched environment led to an increase in the length of basal dendrites but no change in that of apical dendrites in the primary auditory cortex (Bose et al., 2010). Besides sensory deprivation changes in the external environment and learning leads to changes in neuronal plasticity (Comeau et al., 2010). In male rats, complex housing leads to a decrease in the dendritic field of pyramidal neurons in the cingulate cortex (region 3), mPFC (CG3) layer 5 (L5) and an increase in this parameter in layer 3 (L3) of the orbitofrontal cortex (OFC). Furthermore, Comeau et al. demonstrated that training on the T-maze leads to an increase in the dendritic field and spine density of L5 neurons in the CG3 and a decrease in these parameters in L3 neurons in the OFC. They also found that spatial reversal learning in a parallel alley maze causes a decrease in the dendritic field in both areas. This study suggests that the plasticity induced by different kinds of experience varies in different areas and layers of the cortex (Comeau et al., 2010). Additionally, in a forelimb-reaching task, neurons in layer 2/3 of the motor cortex undergo dendritic remodeling. There was a transient increase in dendritic complexity in the distal parts of apical dendrites, whereas there was a decrease in the complexity of basal dendrites (Streffing-Hellhake et al., 2010). These studies suggest that mature neurons in the adult brain are capable of exhibiting plasticity in the active neural circuits.

The findings from our study also demonstrated learning induced plasticity in the mature neurons, these changes were more pronounced in regions such as NCL compared to other NC subdivisions which may suggest that a greater number of neurons in these areas get recruited in the learning process. We found that both active and inactive dopaminoceptive neurons in mNCL and lNCL were more complex in Trained and No-Association birds vs. Undertrained birds, suggesting that training on visual discrimination led to an increase in branching and likely, the number of synapses. Comparisons between these sets of neurons revealed that inactive dopaminoceptive neurons were more complex than active ones in lNCL and mNCL of Trained and No-Association birds. However, active neurons in mNCL were more complex than inactive ones in Undertrained birds (Figure 10A). These findings suggest that the initial phase of learning leads to an increase in complexity and/or synapses of DARPP-32-labeled neurons in mNCL, whereas extensively branched neurites of dopaminoceptive neurons in lNCL and mNCL are pruned to retain only task-specific connections (Hawes et al., 2015) with training. Whereas our results strongly suggest the possible involvement of other NC subdivisions in learning, further studies are needed to address the extent of involvement of other subdivisions and what part of learning and decision-making process are modulated by these regions.

Learning may lead to neurogenesis in house crows (Taufique et al., 2018) and other species (Urakawa et al., 2007; Tronel et al., 2010; Lemaire et al., 2012) even in adulthood. For example, the transient depletion of immature neurons in the mouse hippocampus causes deficits in learning the active place avoidance task (Vukovic et al., 2013) and they are important for reconsolidation of task-induced memories (Lods et al., 2021). Amongst avian species, both fusiform and multipolar DCX-labeled neurons increase in number in the nidopallium of adult house crows (Taufique et al., 2018) and pigeons (Mehlhorn et al., 2022), which was correlated with stress and homing behavior, respectively. In our study, the increase in immature spherical DCX-labeled neurons in the striatum and NC of Trained vs. Undertrained and Baseline categories of house crows (indicative of adult neurogenesis) may be necessary or permissive for learning and decision-making. Furthermore, the significant increase in spherical DCX-labeled neurons in different compartments of the avian basal ganglia (Area X, MSt and LSt) in Trained birds vs. other groups may contribute to habitual learning and habit formation (Malvaez and Wassum, 2018), learning extinction (Goodman et al., 2016; Goodman and Packard, 2018), reinforcement learning (Tanaka et al., 2015) and decision-making (Balleine et al., 2007) besides vocal learning (Scharff and Nottebohm, 1991), as seen in the mammalian dorsal striatum. Interestingly, striatal lesions and exposure to an enriched environment have been linked to an increase in migrating DCX-positive neurons in the dorsal striatum of rodents as well (Urakawa et al., 2007). However, to actually prove that the increases observed in DCX neurons were due to changes in the number of newly generated neurons in adult birds (La Rosa et al., 2020) further studies are required, involving the administration of external S-phase markers such as BrdU (Bromodeoxyuridine) or EdU (5-ethynyl-2′-deoxyuridine) in conjunction with internal markers such as DCX, at specific time points during the training period.

Our results also demonstrated that DCX-labeled multipolar neurons were significantly more complex in different divisions of NC, Area X and MSt of Trained and No-Association birds vs. those of other groups which is suggestive of increased neural plasticity in these circuits. In NCL, inactive DCX-positive neurons were more complex and appeared to be pruned when they became parts of active circuits. Whereas active DCX-labeled neurons in MSt demonstrated fewer changes in complexity across different experimental groups, they were more complex compared to inactive DCX-labeled neurons in MSt of Undertrained birds (Figure 10B). In rodents, the dorsomedial part of the striatum is involved in attentive decision-making, whereas its dorsolateral component underlies automatizing responses during the latter part of the learning process (Whishaw et al., 1987; Yin et al., 2004; Clouse et al., 2009) and neurons in these regions undergo changes in dendritic complexity with learning (Hawes et al., 2015). Changes in the No-Association group were similar to those in the Trained group probably due to the trial-and-error learning method employed by the crows. Overall, these results suggest that inactive neurons become more complex in MSt in Undertrained crows during the initial part of learning the visual discrimination task.

From our results, we cannot provide direct evidence of the differences observed due to learning-induced changes in neuronal complexity between inactive and active neurons. This is because we did not tag neurons which were initially activated during the task and looked for the changes in neurite plasticity over time, as learning progressed. We used the immediate early gene Arc since it is reported to be better correlated to the behavioral task demands as reported in a study involving hippocampal dependent and hippocampal independent behavioral tasks in rats, compared to the expression of other IEGs, zif268 and c-fos (Guzowski et al., 2001). Our results cannot rule out the possibility that immediate early genes other than Arc are expressed in inactive neurons. Moreover, since we did not perform experiments to study the time-course of activation of different sets of neurons and the changes in their complexity in the Trained group, we cannot prove whether inactive neurons had expressed Arc at some earlier time point or if they never expressed Arc. Future studies addressing the time course of these changes and tagging the active neurons are needed to address these questions.

An interesting conundrum is provided by No-Association birds. Despite being exposed only to two blocks of trials (similar to Undertrained birds) of the visual discrimination task, structural changes undergone by DARPP-32- or DCX-labeled neurons in the No-Association group are comparable to those observed in Trained birds. It is possible that employing two extreme strategies for obtaining the food reward, that is, goal-oriented behavior in case of Trained birds and trial-and-error learning in No-Association birds results in similar changes in the complexity of neurons involved in the visual discrimination task. Alternatively, the strengthening of functional connections coding for the correct strategy to obtain rewards may vary across different experimental groups. For example, neural circuits in Undertrained birds, which are at an initial phase of learning may not have been strengthened leading to low and variable neural activity and fewer changes in the complexity of mature and immature neurons in the striatum and NCL. Following extensive training on the task, it is likely that neural circuits responsible for obtaining the food reward may have been strengthened, leading to higher levels of neural activity and more elaborate neurites in neurons within NCL. Finally, in the No-Association group, since rewards are associated randomly with the shapes, the cognitive load may be the highest. Hence, high levels of neural activity as well as morphometric changes in DCX-labeled neurons of No-Association birds are similar to those in Trained birds.

The number of Arc positive neurons was higher in lNCL than mNCL although this difference was not significant. Overall, even under normal conditions with no training, DARPP-32 neurons in mNCL have larger neurite fields compared to those in other subdivisions of NC (Sen et al., 2019). With training, we observed subtle differences between the two NCL subdivisions in different experimental groups. The differences observed in DARPP-32-labeled neurons between inactive and active neurons in the Trained group vs. other groups were more pronounced in case of lNCL than mNCL. However, these differences were not observed in immature DCX-labeled multipolar neurons present in mNCL and lNCL. Hence, it is likely that the mature neurons might be more attuned to learning-related differences than the immature neurons. Alternatively, the dopaminoceptive neuronal population may be more sensitive to task-related differences.