Increased brain lactate levels correlated with decreased pH are associated with poor working memory.

(A) Venn diagrams show the number of strains/conditions of animal models with significant changes (P < 0.05 compared with the corresponding controls) in brain pH and lactate levels in an exploratory cohort. Scatter plot shows the effect size-based correlations between pH and lactate levels of 65 strains/conditions of animals in the cohort. (B) Scatter plot showing the z-score-based correlations between pH and lactate levels of 1,239 animals in the cohort. A z-score was calculated for each animal within the strain/condition and used in this study. (C) Schematic diagram of the prediction analysis pipeline. Statistical learning models with leave-one-out cross-validation (LOOCV) were built using a series of behavioral data to predict brain lactate levels in 24 strains/conditions of mice in an exploratory cohort. (D) The scatter plot shows significant correlations between predicted and actual lactate levels. (E) Feature preference for constructing the model to predict brain lactate levels. Bar graphs indicate the selected frequency of behavioral indices in the LOOCV. Line graph indicates absolute correlation coefficient between brain lactate levels and each behavioral measure of the 24 strains/conditions of mice. r, Pearson’s correlation coefficient. (F–H) Scatter plot showing correlations between actual brain lactate levels and measures of working memory (F), the number of transitions in the light/dark transition test (G), and the percentage of immobility in the forced swim test (H).

Studies in an independent confirmatory cohort validate the negative correlation of brain lactate levels with pH and the association of increased lactate with poor working memory.

(A) Venn diagrams show the number of strains/conditions of animal models with significant changes (P < 0.05 compared with the corresponding controls) in brain pH and lactate levels in a confirmatory cohort. Scatter plot shows the effect size-based correlations between pH and lactate levels of 44 strains/conditions of animals in the cohort. (B) Scatter plot showing the z-score-based correlations between pH and lactate levels of 1,055 animals in the cohort. (C) Statistical learning models with leave-one-out cross-validation (LOOCV) were built using a series of behavioral data to predict brain lactate levels in 27 strains/conditions of mice in the confirmatory cohort. (D) The scatter plot shows significant correlations between predicted and actual lactate levels. (E) Feature preference for constructing the model to predict brain lactate levels. Bar graphs indicate the selected frequency of behavioral indices in the LOOCV. Line graph indicates absolute correlation coefficient between brain lactate levels and each behavioral index of the 27 strains of mice. r, Pearson’s correlation coefficient. (F–H) Scatter plots showing correlations between actual brain lactate levels and working memory measures (F), the acoustic startle response at 120 dB (G), and the time spent in dark room in the light/dark transition test (H).