Averaging sleep spindle occurrence in dogs predicts learning performance better than single measures

Sleep spindles, brief bursts of oscillatory brain activity during non-REM sleep, are believed to play a crucial role in memory consolidation. A substantial body of research in humans has demonstrated a positive correlation between sleep spindle activity and post-sleep recall, indicating a link between sleep spindles and learning success. However, the robustness of this relationship has been challenged. Studies employing larger sample sizes have failed to replicate the observed effect, raising concerns about the reliability and generalizability of previous findings. One significant methodological issue is the variability in algorithms used for automated sleep spindle detection. Different algorithms yield different estimates of spindle occurrence, potentially confounding results across studies. Furthermore, publication bias may inflate the apparent strength of the association. To address these concerns, the present study leverages a well-established, consistent method for detecting sleep spindles in dogs, (*Canis familiaris*), a model species increasingly employed in sleep research. The advantage of using dogs lies in the standardization of sleep spindle detection methods within the research group, minimizing the variability observed in human studies. This study aims to evaluate the relationship between sleep spindle occurrence and learning success in dogs by analyzing multiple datasets using a unified analysis pipeline, thereby assessing the prevalence of positive findings and the conditions under which these relationships emerge. By comparing the associations between single and averaged measurements of sleep spindle density, the researchers hope to determine whether the effect reflects a true underlying relationship or is merely an artifact of measurement error.

The literature concerning the relationship between sleep spindles and learning is extensive, but inconsistent. Studies in humans have reported a positive correlation between sleep spindle density (number of spindles per minute) and post-sleep recall, particularly for verbal memory tasks. However, a notable study involving a large sample size failed to find a significant association, highlighting the need for further research. Methodological inconsistencies, such as differences in sleep spindle detection algorithms, are frequently cited as a source of variability across studies. Invasive studies on animals have revealed potential mechanisms linking sleep spindles to memory consolidation, suggesting causality rather than merely correlation. The existing literature points to the importance of using standardized methods, large sample sizes, and a nuanced understanding of the various factors that might influence the strength of the spindle-learning relationship.

This study uses data from three independent datasets of dogs undergoing a novel word learning paradigm. All datasets employed the same sleep spindle detection method, reducing inter-study variability. The learning paradigm involved teaching dogs novel English words associated with previously learned Hungarian commands. Following a period of sleep, performance on the novel task was assessed, and learning gain was calculated as the percentage improvement in performance after sleep. Data were collected from 46 dogs (23 females, ages 1–9 years, representing 16 breeds). EEG recordings were obtained using standard polysomnographic techniques. Sleep spindles were automatically detected using a previously validated algorithm, which involved a two-step process: an initial detection based on amplitude and frequency criteria followed by an individual adjustment using maximum likelihood estimates for each dog. The algorithm focuses on the 9–16 Hz frequency range, reflecting the overlap in spindle frequency definitions across various species. Analyses used Pearson correlations to assess the relationships between sleep spindle density (spindles/minute) and learning measures (final performance and learning gain). Two types of analyses were conducted: one using single measurements from individual learning conditions and another using averaged measurements across conditions to approximate underlying traits and reduce measurement error. Missing values, mainly due to insufficient non-REM sleep, were excluded from analyses. The study adhered to Hungarian animal experimentation regulations, obtaining necessary ethical approvals and obtaining owner consent.

The study's key findings can be summarized as follows: 1. **Single Measurement Analysis:** When analyzing single measurements of sleep spindle density and learning outcomes, a significant positive correlation was observed between spindle density and learning gain in one dataset and one condition (dataset 0 and condition 1a). However, this association wasn’t found in other conditions or datasets. 2. **Averaged Measurement Analysis:** When analyzing averaged sleep spindle density across all recordings (approximating a trait-like measure of spindle density), and averaged learning outcomes, the study found a positive association between sleep spindle density and learning gain in one dataset and between average spindle density and final performance in another. These findings were significant and were also observed when considering slow spindles recorded over the frontal midline electrode (Fz). 3. **Consistency of Spindle Detection:** The consistency of using the same spindle detection algorithm across datasets strengthened the validity of the findings, minimizing potential methodological biases associated with different detection methods. 4. **Electrode Specificity:** The significant correlations were found primarily using data from the Fz electrode, consistent with observations in human studies on verbal memory. 5. **Slow Spindle Dominance:** The study found that the spindle-learning associations were more pronounced for slow spindles, in line with similar observations in humans using verbal learning tasks.

The study's findings provide further support for the existence of a positive relationship between sleep spindle activity and learning, but with an important nuance: averaging across multiple measurements appears to increase the statistical power and visibility of this effect. This suggests that individual variability in spindle density may obscure the relationship when using single-session measurements. Averaging across sessions may provide a more reliable estimate of an individual's underlying “sleep spindle trait”, making it a more robust predictor of learning ability. The observed association between slow, frontal spindles and verbal learning in dogs mirrors findings in human studies, adding to the cross-species generalizability of this phenomenon. The study acknowledges limitations due to relatively small sample sizes and the possible influence of factors like emotional arousal or specific sleep stages. Nevertheless, the consistency of method and focus on an averaged, trait-like measure of sleep spindle density enhances the reliability of the findings.

This study reinforces the link between sleep spindle activity and learning success, but highlights the importance of considering measurement error when evaluating this relationship. Averaging sleep spindle density across multiple measurements provides a more robust and reliable indicator of an individual's learning potential than relying on single measurements. Further research with larger samples and controlled conditions is needed to fully delineate the role of sleep spindles in learning and memory consolidation.

The study's relatively small sample sizes in each dataset limit the generalizability of the findings. While the use of the same spindle detection method across datasets strengthens internal validity, external validity could be improved by replicating the findings using different learning tasks or spindle detection algorithms. The analysis focuses primarily on slow spindles and frontal electrodes, potentially overlooking other relevant aspects of spindle activity. Moreover, the absence of a control group without learning demands limits the ability to differentiate between memory consolidation and general cognitive ability related to spindle density.