Time-resolved neurotransmitter detection in mouse brain tissue using an artificial intelligence-nanogap

Abstract

The analysis of neurotransmitters in the brain helps to understand brain functions and diagnose Parkinson's disease. Pharmacological inhibition experiments, electrophysiological measurement of action potentials, and mass analysers have been applied for this purpose; however, these techniques do not allow direct neurotransmitter detection with good temporal resolution by using nanometre-sized electrodes. Hence, we developed a method for direct observation of a single neurotransmitter molecule with a gap width of ≤ 1 nm and on the millisecond time scale. It consists of measuring the tunnelling current that flows through a single-molecule by using nanogap electrodes and machine learning analysis. Using this method, we identified dopamine, serotonin, and norepinephrine neurotransmitters with high accuracy at the single-molecule level. The analysis of the mouse striatum and cerebral cortex revealed the order of concentration of the three neurotransmitters. Our method will be developed to investigate the neurotransmitter distribution in the brain with good temporal resolution.

Figure 1

Detection of single neurotransmitter molecules. (a) Schematic of single-molecule detection. (b) Molecular structures of dopamine, serotonin, and norepinephrine. (c) Experimental setup for mouse brain analysis; brain tissue mounted onto a mechanically controllable break junction (MCBJ) substrate using narrow gold wires (as electrodes) drawn via electron beam lithography and covered with SiO₂. (d) Analysis flow: the neurotransmitter solutions are dropped onto MCBJ substrates for training, the mouse brain tissue is then mounted onto MCBJ substrates and the current measurements are performed while these substrates are bent using piezo to form nanogaps. The pulse signals are classified via supervised machine learning.

Figure 2

Current measurements of single neurotransmitter molecules and classification results via a machine learning-based method. (a) Typical current–time profile of a dopamine aqueous solution via single-molecule measurement; the individual pulse signals were extracted as shown in the Supplementary Information. (b) Typical current histograms of dopamine, serotonin and norepinephrine, obtained under a bias voltage of 100 mV. The bin size is 1 pA. (c) Typical single-molecule signal of dopamine; the red curve and blue dots represent the current–time profile and the signal features, respectively (i.e., the average current values I₁, I₂, I₃, I₄, I₅, I₆, I₇, I₈, I_9, and I₁₀, which were 22.1, 20.8, 27.6, 23.7, 23.5, 21.4, 19.5, 28.8, 26.8, and 31.3 pA, respectively). The black dotted lines divide the pulse signals into 10 regions along the time axis. (d) Typical feature currents (I₄ and I₇) of the single-molecule signals of dopamine (DA), norepinephrine (NE), and serotonin (5-HT). (e) Confusion matrix of the machine learning classification; the ratios and errors represent the average and standard deviation values of 10 classification results. (f) Probability density of the current fluctuation factors of DA, NE, and 5-HT.

Figure 3

Machine learning classification of neurotransmitter mixtures with three different ratios of dopamine (DA), norepinephrine (NE), and serotonin (5-HT), namely 1:2:4 (D1N2S4), 2:4:1 (D2N4S1), and 4:1:2 (D4N1S2). (a) Ternary plot of the classification results; the small dots and diamonds represent the true and predicted mixture ratios. For instance, the blue dot denotes true mixture ratio of DA:NE:5-HT = 1:2:4 (0.14:0.28:0.57), and the blue diamond denotes the ratio of the predicted label counts of D1N2S4; DA:NE:5-HT = 0.12, 0.21, 0.67. (b) Ratio of classification result of mixtures of monoamine neurotransmitters. Each row represents a mixture sample. Each column represents the ratio of the signal counts predicted for each neurotransmitter using ML classification.

Figure 4

Classification of the pulse signals of mouse brain tissues. (a–f) Current signals of dopamine (DA), norepinephrine (NE), and serotonin (5-HT), classified via machine learning, along with noise signals from the striatum (a–c) and cerebral cortex (d–f); the plots on the right show zoomed in images of the current profile portions, as identified by the dotted lines, in (a,d). (g) Mouse brain slice; the slices were separated into 1-mm sections for measurements. (h) Classified signal counts from the striatum and cerebral cortex; the measurement time was 300 s.