American crows that excel at tool use activate neural circuits distinct from less talented individuals

Tool use occurs across diverse taxa but is neither universal nor common within many lineages that possess the morphology to use tools. In humans and nonhuman primates, two neural circuits support tool behavior: semantic knowledge of tools and learned motor skills. Whether other tool-using animals possess comparable functional circuitry is unclear. Corvids exhibit advanced cognition comparable to great apes, yet habitual tool use is largely confined to New Caledonian crows; other corvids, including American crows, are only occasional tool users despite being able to learn tool use in captivity. This study investigates which brain regions American crows recruit when learning and using a novel tool and how neural activity shifts with proficiency. The authors test naïve and trained crows on an Aesop’s fable water-raising task and use FDG-PET to compare pre- and post-training brain activity, asking whether proficiency relates to a shift from associative/executive to motor-learning circuits and which individual factors (age, sex, size, brain volume, body condition, nervousness) predict success.

The paper situates tool use across vertebrates and some invertebrates, noting that learned tool use often co-occurs with cognitive flexibility. Prior work in humans and nonhuman primates distinguishes neural circuits for tool semantics and motor skills. Corvids have forebrain neuronal counts comparable to great apes and display innovation and flexibility, yet habitual tool use is rare beyond New Caledonian crows. Avian pallial regions such as the nidopallium caudolaterale (NCL) are analogized to mammalian prefrontal cortex (executive function, working memory, planning). Previous PET imaging in crows has mapped neural responses to faces and vocalizations. Lateralization is prominent in avian brains and relates to novelty, spatial cognition, and object processing. Comparative neuroanatomy suggests tool-using birds may have enlarged telencephalon, striatal complex, septum, and tegmentum. Studies of human athletes show a shift from prefrontal/executive to motor/cerebellar regions with expertise, a pattern the authors hypothesize will parallel increases in crow tool-use proficiency.

Subjects: 16 wild American crows (10 males, 6 females; adults and subadults equally represented) were captured and housed in an outdoor aviary. All birds underwent >7 habituation sessions (mean 10.2 ± 4.5 days) to reduce novelty-related neural activity during imaging. Task apparatus: A clear acrylic tube (internal diameter 5.1 cm) mounted on a polycarbonate base was baited with floating, highly visible food rewards that remained buoyant >3 h. Plastic aquarium stones (green/blue) served as tools; each stone displaced ~5 mm of water when dropped. Training protocol (Aesop’s fable paradigm): After baseline imaging, crows were incrementally trained to retrieve the food by raising the water level with stones. Investigators first established each bird’s maximum reachable depth by lowering water level; training stones were initially introduced at the rim and then progressively made more challenging (e.g., hanging stones, increasing string length, lowering stones along tube exterior) to encourage deliberate pickup and dropping into the tube. Trials were recorded and birds were food-motivated by withholding food before sessions. Imaging design: FDG-PET was used to assess whole-brain activity. Pre-training scans occurred at first exposure to the task materials (naïve condition). Post-training scans followed weeks of training. During the FDG uptake (stimulus) phase, crows viewed the baited apparatus in a controlled stage with sliding doors that alternately revealed the apparatus for 60 s and concealed it for 30 s over 10 min (7 exposures, 6 breaks). Birds fasted ≥14 h pre-scan. Approximately 26 min post-injection, anesthetized birds underwent a 20 min microPET scan followed by a co-registered CT. Two birds were excluded from imaging analyses due to mechanical/software issues. Image processing: PET data were reconstructed with Siemens Inveon 3D OSEM/MAP (2 OSEM + 18 MAP iterations), target 2.5 mm FWHM, with attenuation and scatter correction. DICOMs were aligned to an adapted jungle crow brain atlas and a carrion crow atlas using NEUROSTAT; uptake values were globally normalized. Behavioral and individual measures: Sex, age, morphometrics (e.g., culmen length), body condition, nervousness (movements/min), brain volume, and relative brain volume were recorded. Weight change in captivity was tracked. Behavioral attention (gaze/eye use) during the stimulus phase was monitored. Statistical analysis: Voxel-wise pre- vs post-training subtractions were converted to Z-score maps. Regions of interest (VOIs) centered on peak coordinates provided normalized uptake values. Proficiency groups (low, medium, high) were compared post-training. Linear models, multimodel inference (AICc weights), and correlations were conducted in R 3.6.3.

Sample and proficiency: Of 16 crows, 4 fully mastered the task (high proficiency), 8 reached medium proficiency, and 4 remained low proficiency. Imaging analyses included n = 14 due to two scan exclusions. Pre- vs post-training (all crows): Naïve (pre-training) scans showed higher FDG uptake in associative/sensory regions compared to post-training, including bilateral mesolimbic areas (left +8.2%, Z = 4.53; right +8.6%, Z = 4.22), right dorsal mesolimbic (+7.4%, Z = 4.82), right dorsal nidopallium (+7.2%, Z = 5.24), and a more ventral nidopallium focus (+6.9%, Z = 4.31). Activity patterns suggested broader right-hemisphere involvement. Some nidopallium activity did not exceed the Z-threshold in certain slices. High-proficiency birds uniquely showed increased activity medially in left segmental cortex after training (+8.2%, Z = 4.16), just caudal to the arcuate nucleus. Proficiency-based differences (post-training): High-proficiency crows, compared to low-proficiency birds, showed significantly higher activity in motor-learning and task-control regions: right caudal cerebellum (+9.1%, Z = 4.44), dorsal caudal cerebellum bilaterally (left +9.8%, Z = 4.89; right +15%, Z = 4.23), and left nucleus basalis (+3.3%, Z = 3.77). During task performance, higher activity was also observed in left mesopallium ventrolateralis (MVL; +10.3%, Z = 3.92) and left hyperpallium apicale (did not exceed Z-threshold across analyses). Low- and medium-proficiency birds maintained elevated mesopallial activity relative to high-proficiency birds. Lateralization: Naïve birds exhibited right-hemisphere biases in mesopallial/nidopallial activity, consistent with novelty processing. Low-proficiency birds showed left-hemisphere bias in hyperpallium apicale and MVL, aligning with left-dominant visual object processing; high-proficiency birds’ patterns were consistent with spatial memory recall and motor control engagement. Individual measures: Crows gained significant weight during captivity (387.5 ± 31.4 g at capture vs 415.0 ± 37.6 g at release; t15 = 6.12, p < 0.001). Nervousness tended to decrease (p = 0.07). Higher body condition correlated with lower nervousness (r = -0.58, p = 0.017). Predictors of proficiency: High-proficiency birds were exclusively adults and primarily females. Among adults, successful birds were mostly female, had smaller culmen length (solved: 46.5 ± 3.0 mm vs not solved: 51.3 ± 0.96 mm) and smaller absolute brain volume (solved: 7.05 ± 0.45 cm³ vs not solved: 8.00 ± 0.04 cm³). Some models (e.g., sex) failed to converge due to all adult females solving the task. Subadults generally achieved medium proficiency but none reached high proficiency.

The study demonstrates that, as American crows learn to use a novel tool, their neural activity shifts from associative/sensory pallial regions (mesopallium, nidopallium) during initial exposure to circuits supporting motor learning, memory, and task control (cerebellum, nucleus basalis, hippocampus/tegmentum) with proficiency. This pattern parallels human expertise development, where novices rely more on prefrontal/executive networks and experts shift toward motor and cerebellar regions. Lateralization patterns fit known avian specializations: right-hemisphere engagement with novelty and cognitively demanding tasks, and left-hemisphere dominance for object processing. Behaviorally, smaller adult females likely face greater competitive constraints, potentially selecting for cognitive flexibility and strategy use, which may explain their higher proficiency. The results support the idea that while many vertebrates can learn tool use, widespread habitual tool use depends not only on morphology but also on ecological pressures and the neural shift from executive to motor circuits through practice.

This work provides whole-brain, longitudinal evidence in a nonhabitual tool-using corvid that tool-use learning transitions neural engagement from associative/executive pallial regions to motor learning and task-control circuits with increased proficiency. Adult female crows were most likely to master the Aesop’s fable task, consistent with hypotheses that subordinate, smaller individuals adopt cognitively flexible strategies to access food. The findings align with comparative evidence that tool-using birds have enlarged telencephalon, striatal complex, septum, and tegmentum, and show how these regions are differentially utilized across learning stages. Future research should extend longitudinal neuroimaging across taxa and tasks, incorporate finer neuroanatomical metrics (e.g., neuron counts/densities, regional connectivity), and manipulate ecological context to test how social dominance and resource competition shape the adoption of tool use.

- Small sample size (n = 16; imaging analyses n = 14) with uneven proficiency groups limits generalizability. - Some statistical models failed to converge (e.g., sex, age) due to group compositions (all adult females succeeded; no subadult reached high proficiency), constraining inference. - Body condition likely improved between capture and first imaging, potentially confounding its association with proficiency. - The habituation and training design may have minimized detectable effects of nervousness and body condition on proficiency. - Some reported regional differences did not exceed Z-thresholds in all comparisons; atlas disagreements and alignment limits (1–2 pixels) could affect regional labeling. - PET provides indirect metabolic measures with limited temporal resolution; scans were conducted under anesthesia after uptake, which may not capture moment-to-moment task execution.