Automated algorithm for measurement of spontaneous adenosine transients in large electrochemical data sets

Spontaneous adenosine release events have been discovered in the brain that last only a few seconds. The identification of these adenosine events from fast-scan cyclic voltammetry data has been performed manually and is difficult due to the random nature of adenosine release. In this study, we develop an algorithm that automatically identifies and characterizes adenosine transient features, including event time, concentration, and duration. Automating the data analysis reduces analysis time from 10-18 hours to about 40 minutes per experiment. The algorithm identifies adenosine based on its two oxidation peaks, the time delay between them, and their peak ratios. In order to validate the program, 4 data sets from 3 independent researchers were analyzed by the algorithm and then verified by an analyst. The algorithm resulted in 10 ± 4% false negatives and 9 ± 3% false positives. The specificity of the algorithm was verified by comparing calibration data for adenosine triphosphate, histamine, hydrogen peroxide, and pH changes and these analytes were not identified as adenosine. Stimulated histamine release in vivo was also not identified as adenosine. The code is modular in design and could be easily adjusted to detect features of spontaneous dopamine or other neurochemical transients in FSCV data.


Acknowledgement
I would like to acknowledge my advisor, Dr. Jill Venton, for helping me grow as a scientist by thinking more critically about my work and by teaching me how to communicate my ideas better.I also would like to acknowledge everyone one in my lab, whom I bothered with endless amounts of questions and always took their time helping me understand the answers.I would especially like to thank Dr. Michael Nguyen for entertaining this hair-brained idea of automating adenosine detection.It started as a joke (and now it's a thesis) because we were not experts in computer programming.Also, I would like to thank Dr. Ashley Ross for editing many of my papers and helping me develop a clever experiment for this thesis.Lastly, I would like to think Ying Wang and Scott Lee for taking their time verifying the validity of the program with their data.
Ultimately, I would like to thank my parents, John and Julie Borman, who have been very supportive of my decisions and I could not ask for better parents.When I was down, they picked me up and always kept things in perspective.

Adenosine biological role
Adenosine is an important biological nucleoside that is involved in many metabolic processes including cell signaling 1 , neuromodulation 1,2 , and neuroprotection 1,2 .
Adenosine is a constituent of all cells since it is a byproduct of ATP catabolism 1 and is found in many regions of the brain including the caudate-putamen 3 , hippocampus 4 , nucleus accumbens 3 , and cortex 5 .Endogenous adenosine acts as a neuromodulator and plays an active role in the regulation of cerebral blood flow 6,7 .Neuromodulators have been typically thought to act on slow time scales, minute to hours, by volume transmission.Volume transmission happens within the brain extracellular fluid and can include both short and long distances 8 .Adenosine modulates by binding to one of four G protein-coupled receptors to cause either an inhibitory or stimulatory regulation of neurotransmitters 6 .Release of adenosine has been shown to protect heart cells during ischemia, where there is a deficiency in oxygen delivery 9 .The ability to measure adenosine is important in understanding the role it plays in neuromodulation and homeostatic regulation.Furthermore, since it is pervasive throughout the central nervous system and found in every cell, understanding the function adenosine plays in physiological disorders like hypoxia and ischemia could help with new treatments for these disorders.Adenosine traditionally has been studied as a slow acting molecule but recent evidence suggests a more rapid spontaneous mode of signaling 7 .The discovery of rapid spontaneously released adenosine in the brain on the minute timescale has generated a need for automatic feature detection due to hundreds of events being released in a single animal experiment.In this study a straightforward algorithm was designed to identify and characterize random adenosine transients from FSCV color plots.

Adenosine production
The production and regulation of adenosine is a highly complex system involving both intracellular and extracellular formation mechanisms (Figure 1.1).Adenosine can be formed intracellularly in the central nervous system by the catabolism of AMP, due to metabolic stress, or by cytosolic 5'-nucleotidase.
The intracellular formation of adenosine can also be formed from the hydrolysis of s-adenosylhomocysteine (SAH in .1) 2 .Adenosine can be released to the extracellular space by two different

Borman | 3
mechanisms: bi-directional nucleoside transporters or exocytosis of ATP that metabolizes to adenosine.

Adenosine regulation and function
There are four adenosine receptors expressed in the brain, which are A 1 , A 2A , A 2B , and A 3 .A 1 and A 2A are high affinity for adenosine with binding affinities in the 1-30 nM range 10 .A 1 and A 2A receptors are thought to be active at normal physiological conditions since the basal level of adenosine is at low nanomolar concentrations 1 .The A 1 receptor is an inhibitory G-protein coupled receptor (G i ).Alternatively, A 2A is an excitatory Gprotein coupled receptor (G e ).The A 2B and A 3 receptors have lower binding affinities; in the 1-20 μM range 11 .The activation of A 2B and A 3 receptors is thought to occur under stressful physiological conditions like hypoxia or ischemia due to higher concentrations of adenosine 6 .Overall, adenosine is a highly complex signaling molecule in the brain and many questions are still unanswered on how adenosine is regulated.

Adenosines physiological role
Adenosine modulates numerous important physiological functions including sleep 12 , breathing 13 , and heart rate 14 .Furthermore, adenosine is directly involved in pathologies like inflammation and cerebral ischemia.A 2A adenosine receptors increase immunosuppressive cAMP in the immune cells of mice and play a role in the attenuation of inflammation and tissue damage in vivo 15 .Moreover, adenosine has been studied for it role as a neuroprotectant against damage in cerebral ischemia and as a possible therapeutic for stroke 16 .In response to energy depletion induced by ischemia, extracellular concentrations of adenosine can increase 1000 fold.During pathological events like cerebral ischemia and inflammation the increase of adenosine typically lasts minutes to hours 5 .Recently, adenosine was shown to rapidly modulate stimulated dopamine release in the caudate-putamen by A 1 receptors 8 .

Adenosine release
Metabolic processes and the breakdown of ATP during energy consumption can cause a buildup of adenosine in the extracellular space.Typically, these extracellular adenosine concentrations have been studied use techniques with only minute to hour temporal resolution.One group used electrophysiological techniques to explore a rapid modulatory role for adenosine in the brain 4 .Electrically stimulated adenosine regulates glutamate receptor-mediated excitatory postsynaptic potentials (EPSPs) in the hippocampus.The measured duration of the EPSPs were 2 seconds, suggesting that adenosine causes changes at sub-minute timescales.This suggests that traditional measurement of adenosine, formerly delegated as a slow acting molecule, does not sufficiently describe adenosines role in rapid signaling.
Subsecond adenosine changes have also been directly measured from electrical stimulations in striatal rat brain slices by fast-scan cyclic voltammetry 17 .The results suggest that adenosine release is activity-dependent.Stimulated adenosine has been studied in vivo using carbon-fiber microelectrodes as adenosine sensors 6 .The purpose of this experiment was to determine stimulated adenosine release in rat caudateputamen, after electrical stimulation.Results show that adenosine increased in the extracellular space and was cleared in about 15 seconds.
Recently, a new form of adenosine release has been found for the first time 5 .An adenosine transient is a non-stimulated occurrence of adenosine with event duration on the order of seconds.Spontaneous release of adenosine was measured in rat caudateputamen and the prefrontal cortex.Average concentrations of adenosine release were 0.18 μM and had a range of 0.04-3.2μM, in both brain regions.The study illustrates that adenosine is rapidly released and cleared in the brain, which suggests that adenosine is involved in rapid neuromodulation in addition to the longer term neuromodulation described above.The frequency of spontaneous adenosine events for a single transient was every 2-3 minutes.This study demonstrates that adenosine events are spontaneous and do not follow any regular pattern and therefore are random events 5 .
Due to these adenosine transients being spontaneous and random, an analyst must find each adenosine transient by "hand", which is very time consuming.Depending on the type of experiment (i.e.brain slice/in vivo models and pharmacological/stroke experiments), 2-4 hours of data are obtained and the number of adenosine transients varies widely.Furthermore, the more transients in a data set the longer time it takes for data processing.A user who is experienced in data analysis can find one adenosine transient approximately every 1.5 minutes, therefore a data set containing 700 transients will take a user 18 hours to analyze.Automation of the data analysis process will allow an analyst to more than double their experimental production when high transient counts are measured.

Measuring adenosine by microdialysis
Historically, microdialysis coupled with HPLC has been one of the most employed techniques for measuring adenosine 12,18,19 .Microdialysis is a sampling technique that is used frequently in neurobiology because it is minimally invasive, has the ability to sample continuously, and measures basal levels of analytes.During ischemia, acutely implanted microdialysis probes (300 μm in outer diameter), measured an increase in the dialysate levels of adenosine compared to chronically implanted probes 19 .Acute and chronic measurements of adenosine were taken at 2 and 24 hours, respectively.Microdialysis probes are relatively large and therefore can disrupt the accuracy of the measurement of adenosine.

Measuring adenosine with biosensors
Enzyme based sensors are used for measuring adenosine at sub-minute temporal resolution.The Dale group developed a three enzyme biosensor for the detection of adenosine 20 .It works by breaking down adenosine to inosine via adenosine deaminase, subsequently to hypoxanthine via purine nucleoside phosphorylase, finally to xanthine, urate, and hydrogen peroxide via xanthine oxidase.These enzymes work to metabolize adenosine to a final product of hydrogen peroxide, which is amperometrically detected.Adenosine biosensors are 25 -100 μm in diameter, have a detection limit of 12 nM, and have a temporal resolution of 2 seconds.A null sensor that does not contain adenosine deaminase is positioned next to the biosensor to measure any interfering downstream metabolites.The null detector signal can be subtracted from the biosensor to obtain the biosensors response to adenosine 7 .Adenosine biosensors are smaller than typical microdialysis probes, have low detection limits, and sub-minute temporal resolution.An alternative to biosensors for the detection of adenosine is fast-scan cyclic voltammetry, an electrochemical technique that directly measures adenosine at carbonfiber microelectrodes.

Measuring adenosine by fast-scan cyclic voltammetry
Fast-scan cyclic voltammetry (FSCV) is an electrochemical technique that was developed to measure real-time changes in dopamine levels in vivo 21 .FSCV is similar which allows it to be placed in specific brain regions while minimizing potential tissue damage compared to larger microdialysis electrodes or enzyme biosensors.An advantage of FSCV is its ability to measure rapid changes in electroactive neurotransmitters.To measure adenosine by FSCV a triangle waveform is applied to a CFME that scans from a holding potential of -0.4 V to a switching potential of 1.45 V versus a Ag/AgCl reference electrode at a scan rate of 400 V/s.The limit of detection for fast-scan cyclic voltammetry at a carbon-fiber microelectrode is 15 nM, which is similar to detection limits at enzyme biosensors of 12 nM 7 .
Figure 1.2:The applied potential waveform for in vivo adenosine measurement.The waveform is scanned from -0.4 V to 1.45 V at a scan rate of 400 V/s.Each scan is 10 ms and scans are repeated every 100 ms Adenosine is an electroactive molecule that can go through three successive, two-electron oxidations (Scheme 1.1) 23 .When a triangle waveform is applied to a CFME, an adenosine molecule undergoes a two-electron primary oxidation to form product II in Scheme 1.1.This primary oxidation product is observed at 1.4 V. Subsequently, a secondary oxidation occurs to form product III at 1.0 V.These first two oxidation steps are irreversible and no reduction peak is observed in its corresponding cyclic voltammograms (CV) as can be seen in Figure 1.3B.Typically, the tertiary oxidation product III is not detected using FSCV at our CFMEs.
The oxidation of adenosine at our carbon-fiber microelectrodes is a two-step process.The primary product is produced in an irreversible oxidation and further oxidizes to a secondary product at our CFMEs 5 .This can be visualized in a color plot of the start of the primary peak, shows the primary peak forming and the secondary peak absent.In the second CV, taken at the primary peak maximum, the secondary peak has

Secondary Oxidation Tertiary Oxidation
Scheme 1.1: Adenosine (I) undergoes a two-electron primary oxidation at 1.4 V to form product II.Subsequently, product II is involved in a secondary oxidation at 1.0 V to form product III.Product IV is normally not observed in FSCV.In this scheme R is ribose.

Borman | 9
started to form.In the third CV, the primary peak is waning and the secondary peak is at a maximum.Adenosines signal changes as a function of time with the secondary peak evolving only after the primary peak is produced.The lag time between the primary and secondary peak max can be utilized to identify spontaneous adenosine transients.seconds.In large data sets it is very time consuming to identify adenosine transients and many hours are devoted to identifying and characterizing these transients in our lab.
The oxidation voltages for the primary and secondary peaks are specific to adenosine and the difference in time between these peaks is a way to identify adenosine.Therefore, the primary oxidation product must be produced before the secondary product is able to be formed.One existing technique for peak identification is principal components regression (PCR) but it is not fully automated in identifying transient peaks and their features.

Principal components analysis
Principal components analysis (PCA) is a multivariate statistical technique that reduces dimensionality by retaining relevant and discarding non-relevant information provided in large data sets.The combination of principal components analysis with inverse least-squares regression is known as principal components regression (PCR) 25 .
PCA is used to determine relevant information from non-relevant noise.PCR then can use residual analysis and remove noise from unknown data.Succinctly, residuals are the difference between an observed value and an estimated value, which is the value of interest.If the summed square of residual current at any applied potential of a CV (Q t ) exceeds the threshold value in a training set (Q α ) then a source of variation is not accounted for in the PCA model and the value should be retained 25,26 .An example of PCR can be observed in and 1.5E, respectively.PCR is a good method for removing spurious noise and interferents from color plot data but it is not a fully automated method for adenosine transient finding.Moreover, since the shape of adenosines CV changes with time PCR has difficulties discriminating the secondary peak.When dealing with large amounts of data a completely automated method, which accurately finds spontaneous adenosine transients is necessary.

Other peak identification techniques
The ability to automate identification of spontaneous adenosine transients is advantageous for two reasons: (1) experimental throughput, since data analysis requires considerable labor (2) accuracy of data analysis within data sets and between researchers.Principal components regression is useful in accurately determining concentrations and using residuals to remove non-relevant data from color plots but is not completely automated.Other groups have developed methods for automating identification of peaks in chromatography but normally use retention times in analyte recognition 28,29 .In one study retention times were expanded and contracted to fit a target chromatogram.Then the Pearson correlation coefficient was used to determine the degree in which the target and test chromatogram were linearly related 30 .However, spontaneous adenosine release does not produce consistent time markers that enable this type of automation.Furthermore, data from in vivo FSCV contains more noise and baseline drift than chromatographic techniques.Finally, since the brain is a complex matrix, unexpected analytes would further muddle these types of automated identification.In this thesis, I describe a computer program that automatically identifies and characterizes adenosine transients, which minimizes labor for data analysis and maximizes accuracy of resulting data.

Automatic identification of adenosine features
In this thesis a straightforward algorithm was designed to identify and to study other electroactive molecules like dopamine, since the algorithm is modular in design, with fast non-bias computer processing.In conclusion, this thesis will describe a new method for automatically identifying spontaneous adenosine transients, with minimal analyst input, and show how an algorithm can be developed for neurotransmitter identification in in vivo FSCV analysis.

References
Chapter 2: Building an algorithm to automatically identify spontaneous adenosine transients
However, there are 144,000 CVs collected in a four hour voltammetry experiment so they cannot be individually examined.Principal components regression (PCR) uses those cyclic voltammograms to identify compounds in mixture and remove noise from the data 17 .In particular, PCR has proven to be a powerful tool to separate dopamine from pH shifts.This method was used previously to identify adenosine and create concentration vs time traces that are analyzed by an analyst to identify adenosine transients.The problem with PCR for adenosine is that the cyclic voltammogram of adenosine changes over time, with a primary peak that is large in the first few cyclic voltammograms and a secondary peak that grows in over time.it is hard to select a representative training set, the residuals (i.e.noise) are large, and the residual noise This study demonstrates the reliability of the automated identification program for adenosine and the program is customizable to study other electroactive analytes in the future.Automated analysis of FSCV data will allow faster data analysis and less analyst bias for identifying and characterizing adenosine in vivo.

FSCV Transient
The adenosine feature detection algorithm, FSCV Transient, was written in

Incremental background subtraction
In the first part of the algorithm (Figure 2.1, Steps 1-3), incremental background subtraction is performed by choosing several times for the background and then subtracting the background charging current.The analyst sets the peak voltages for the primary and secondary peaks of adenosine, then i vs t data is searched for peaks.For example, an initial background subtraction occurs at t = 1.0 s and then the two i vs t .Analyst defined primary and secondary oxidation voltages are scanned for adenosine peaks (2).Lag time filter is applied to detected peaks to remove spurious peaks (3) and resultant peaks are background subtracted adjacent to the peak (4) then identified similarly to steps 2 and 3 (5).Signal-to-noise and ratio filter are applied to detected peaks to remove spurious peaks (6).Event time, concentration, and duration are written to file until all peaks are investigated (7).

Adjacent background subtraction
During the first part of the algorithm increment background subtraction is performed, which background subtracts the FSCV file at evenly spaced time steps.In the second part of the algorithm (Steps 4-7), adjacent background subtraction is accomplished by performing background subtraction approximately 10 s before, i.e.
adjacent, to each individual peak found during incremental background subtraction.
Adjacent background subtraction is standard procedure in FSCV because of baseline drift and leads to more accurate measures of adenosine peak characteristics.Some peaks found during the first part of the program are spurious and are rejected as adenosine during adjacent background subtraction because the two peaks are not identified or the primary peak doesn't precede the secondary peak.

Spurious peak filtering (in vivo)
Peaks are filtered in Step 5 (Step 5 is similar to Step 2 and 3) at analyst-defined thresholds for concentration, duration, and prominence.The minimum concentration that can be detected is usually around 40 nM and the minimum duration for an adenosine transient is 1.0 s.Prominence is the minimum peak height between two consecutive, possibly overlapping peaks.Thresholds are determined from running 5 FSCV data sets in the program and finding the minimum value for concentration that minimizes false negatives and positives.Duration and prominence thresholds are constant and are determined in the same way as concentration but do not need to be adjusted per experiment.Additionally, a signal-to-noise filter is applied to the data and only peaks that have a S/N > 3 are kept, with the noise defined as the SD of the baseline taken adjacent to the peak.Finally, another filter is applied which compares the ratio of the secondary peak max current to primary peak max current, S p,i / P p,i with an empirically determined value for adenosine.The minimum secondary to primary peak ratio for adenosine measured in vivo is 0.49, which was determined empirically from 100 in vivo adenosine transients.Thus, any peak with a ratio below the threshold of 0.49 is rejected as an adenosine peak.If peaks pass the S/N and ratio filter, they are accepted as adenosine peaks.Programmatic details of background subtraction, data filtering, threshold setting, and peak finding can be found in the Appendix at the end of this thesis.

Chemicals
The chemicals used to make phosphate buffered saline (PBS) were all purchased from Fisher Scientific (Fair Lawn, NJ, USA) unless otherwise stated.PBS buffer was used to test interferents using a flow-injection system 19  The interferent pH was tested by adjusting pH=7.4 PBS buffer to pH=7.3 or pH=7.5.
A Dagan ChemClamp potentiostat (Dagan Corporation; Minneapolis, MN, USA) was used to apply voltage to the CFME.All electrodes were scanned from a holding potential of -0.40 V and scanned to a switching potential of 1.45 V and back at 10 Hz versus a Ag/AgCl reference electrode, at a scan rate of 400 V/s.All data was background subtracted to remove any non-Faradic currents by taking the mean of 10 CVs and background subtracting that vector from the data set.All in vitro interferent tests were performed using flow-injection analysis by comparing 1.0 μM adenosine to 1.0 μM interferent in PBS buffer.

Data sets analyzed
In vivo data set (S1) was measured in the caudate putamen and data sets (S2 and S3) were measured in the hippocampus according to procedures previously described 5 .The brain slice data was measured in the prefrontal cortex according to procedures previously described 21 .

Error analysis
Sensitivity, precision, and accuracy were calculated from true positive (TP), false positive (FP), and false negative (FN) values determined from analyst validation of FSCV transient algorithm results (Results and discussion 3.2.3.).Sensitivity or recall is the fraction of relevant peaks that are returned by the algorithm from the data set.

!"
!"#$% Precision or positive predictive value is the fraction of peaks returned by the algorithm from the data set that are relevant peaks.

!"
!"#$" (2) Accuracy was calculated from the F 1 score, which is the harmonic mean of sensitivity and precision.The harmonic mean weights sensitivity and precision equally.

𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦 × 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑖𝑡𝑦+ 𝑝𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛
(3) The F 1 score was calculated because the amount of true negatives, or the number of times the algorithm missed a spurious peak, is unable to be calculated.Values calculated from equations (1-3) are between 0 and 1.0, with 1.0 being maximum sensitivity, precision, and accuracy for analyst validation.Data are presented as mean ± standard deviation.

Adenosine feature detection (algorithm)
To publish a paper for an in vivo experiment many animal experiments are needed and analyzing the resulting data sets takes much more time than collecting the data.Typically, researchers collect 4 hours of data from a single animal experiment but In the brain, spontaneous adenosine release is a random process so the time they will occur cannot be predicted.Electrochemical properties of adenosine must be exploited to identify adenosine transients.Adenosine is irreversibly oxidized 22 around 1.4 V and forms a primary product, which is subsequently irreversibly oxidized around 1.0 V to form a secondary product in FSCV experiments 4 .The peak oxidation voltage for both products is at a singular electrode and therefore these peaks are used to identify adenosine.The primary product is the precursor for the secondary product so there is a lag time between peak maximum, which is seen in current versus time traces in Figure 2.2A 23 .The peak maximums are marked as diamonds and the primary peak always occurs before the secondary.Additionally, as seen in the false color plot in Figure 2.2B, there is lag time between primary and secondary peak formation with the primary occurring before the secondary.
Lag time and peak oxidation voltage provides enough information for the algorithm to successfully identify adenosine transients.Primary and secondary max voltages, p max and s max , respectively, are analyst-defined and i vs t traces at these voltages are scanned for peaks above a threshold, illustrated by the white lines in the color plot in Figure 2.2B.For a peak to be identified as adenosine it must have a peak both at the primary and secondary voltages and the secondary peak must lag the primary peak by at least 0.1 s but be within 2.5 s of the primary peak.Additionally, another filter is applied which compares the ratio of adenosines secondary peak max current to primary peak max current, S p,i / P p,i .If peaks pass the threshold, lag time criteria, and secondary-to-primary peak ratio, then the signal-to-noise ratio (SNR) is calculated for each adenosine peak.If the resultant peak is above 3 × SD of baseline noise, the program saves the event time, peak concentration, duration, and SNR.This identification algorithm is the basis for automatically identifying and tabulating spontaneous adenosine transients, in large data sets with many peaks.

Background Subtraction
In FSCV experiments, stable non-faradaic currents occur due to background charging of the surface of the CFME and these background currents are subtracted to study Faradaic redox reactions.Because the location of adenosine transients is not known a priori, the program first picks several places for background subtraction, at defined increments, and then background subtracts each data set at these times.By examining the same data set background subtracted from different places, we correct for 2 fundamental problems.First, if we only picked one background subtraction time, it may inadvertently be during an adenosine peak and the results would not be interpretable.
Second, the background current drifts over time so background subtraction should be performed as near to a peak as possible to accurately identify and define all possible adenosine peak characteristics.Practically, the algorithm accomplishes this by first reading in a non-background subtracted data file into the program.Next, the program background subtracts at the start of the file, completely scans p max and s max for peaks, and then uses the identification algorithm to determine if any adenosine transients are present in the data set.After the initial background subtraction, the program iterates at the analyst-defined increment value (usually 10 s) (Figure 2.3A-C) repeating the procedure of doing a background subtraction and identifying adenosine transients using that background file.This iteration continues at the defined times until the end of the file.All potential adenosine transients are identified by event time and are saved for later use.Since the program is deterministically incrementing and adenosine transients are random, the program will detect spurious peaks, which need to be further explored.The purpose of this part of the identification algorithm is to tabulate all possible adenosine transient times for each color plot file.
During the incrementing part of the program the algorithm casts a wide net in order to obtain all possible adenosine events.This strategy works in gathering all peaks, real and spurious, but in order to determine if a peak is adenosine, it more standard to do background subtraction directly adjacent to the peak 5  (HDCV), a program developed in the Wightman lab 18 .The results are tabulated in Table 2.1.If the algorithm selects a peak that was not identified as adenosine by the analyst, it is counted as a FP.Moreover, if the analyst determines that the program missed an adenosine peak it is counted as FN.Each data set was measured in an independent animal experiment and the data sets were obtained from three independent experimenters.The first data set (S1), an in vivo measurement in the caudate putamen, an analyst determined 41 adenosine transients and the algorithm resulted in 5% FN and 2% FP.In data set S2, an in vivo measurement in the hippocampus, an analyst determined 397 adenosine transients and the algorithm resulted in 10% FN and 10% FP.
In data set S3, another in vivo measurement in the hippocampus, the algorithm resulted in 8% FN and 9% FP.Finally, in data set S4, a brain slice experiment from the prefrontal cortex, the algorithm resulted in 16% FN and 7% FP.The reason for S4 having a higher FN percentage than the other sets is due to thresholds being adjusted to minimize the selection of FP.Overall, analyst validation of the adenosine algorithm resulted in 10 ± 4% FN and 9 ± 3% FP in a total of 640 confirmed adenosine transients.
These results suggest that the adenosine algorithm is able to discern adenosine, with a high degree of certainty, from noise that is present during animal experiments.Borman | 33

Testing biologically relevant interferents (in vitro)
The brain is a complex organ with multiple electroactive molecules that could interfere with adenosine detection.During adjacent background subtraction, the algorithm checks if peaks exist at p max and s max voltages and that a lag time exists between these peaks.In order to test the robustness of the algorithm, adenosine and possible interferents were measured in a flow cell.Adenosine was measured in a flowinjection system to determine p max and s max , the voltages for the primary and secondary peaks.Since the oxidation voltage of adenosine remains constant during animal experiments, p max and s max are scanned for potential adenosine peaks.In order for an interferent to be counted as adenosine transient, the interferent must have oxidation potentials near the p max and s max of adenosine, a s p,i /p p,i ratio above adenosines threshold, and importantly the primary peak must occur before the secondary.
Adenosines s p,i /p p,i ratio, determined from in vitro analysis to be 0.34, was calculated from 20 injections of 1 μM adenosine at 5 electrodes, which was the minimum ratio calculated (mean=0.6±0.2, range=0.3%-1.02%,20 injections, 5 electrodes).The reason for the large standard deviation and upper range being above 1.0 is due to electrode noise.
To determine if the algorithm generates false positives (FP) (Figure 2.4) adenosine triphosphate, histamine, hydrogen peroxide, and pH were tested as possible adenosine interferents.First, 1 μM ATP (Figure 2.4B) was tested with the algorithm values for adenosine (Figure 2.4A) in order to try to generate FP and no data was omitted from analysis.ATP differs from adenosine by only three phosphate groups and has the same electrochemical moiety 23 .As seen in the i vs t trace for ATP, there are primary and secondary peaks.However, the max s p,i /p p,i ratio is 0.27 (mean=0.18±0.05,range=0.07-0.27,15 injections, 6 electrodes), which is below the threshold for adenosine and the secondary occurs before the primary maximum.Thus, ATP fails to be identified as adenosine.Furthermore, comparing adenosines cyclic voltammogram with ATP's, adenosine displays a more pronounced secondary peak than ATP.The algorithm also rejected histamine, a molecule whose cyclic voltammogram is similar to adenosine 24 .
The secondary peak for histamine is at 0.76V compared to 1.06V for adenosine.Thus, the i vs t trace for 1 μM histamine (Figure 2.4C) displays the s max occurring before p max and a max s p,i /p p,i ratio of 0.24 (mean=0.15±0.07,range=0.03-0.24,20 injections, 6 electrodes), which is below the threshold for adenosine of 0.34.Setting p max and s max constant exploits adenosines intrinsic oxidation potentials.Hydrogen peroxide (Figure

2.4D
) is another possible interferent of adenosine 21 , and is rejected as a transient from the computer algorithm because the max s p,i /p p,i ratio of 0.18 (mean=0.08±0.05,

Borman | 37
animal experiments does not detect histamine and it is concluded that during in vivo experiments detected transients are not histamine.This validation shows the algorithm is good at distinguishing histamine as an interferent both during in vitro calibration experiments and in vivo.

Conclusion
The ability to automate the identification of adenosine transient features will reduce the hours researchers spend on monotonous data analysis and normalize results between researchers.The first iteration of the algorithm was building a structure to acquire all possible adenosine peaks by incrementing background subtraction.Next, to maximize accuracy of adenosine feature detection adjacent background subtraction was added to the algorithm.Moreover, signal-to-noise, ratio filters, and analyst-defined thresholds can be adjusted to analyze independent data sets from multiple researchers.
In summary, this program can save more than two hundred hours of repetitive data analysis per publication.This accumulated time can be used to conduct more experiments and therefore increase laboratory throughput.

Future directions
The initial development of the algorithm generated promising result for identifying adenosine transients in FSCV data sets.Since CFMEs are manufactured in our lab and have different sensitivities, adjusting thresholds is expected for independent animal experiments.After a CFME is equilibrated in tissue the resulting noise is stable and is relativity constant during animal experiments.One way to automatically calculate threshold values is by calculating 3 × SD of the primary peak noise and set the peak concentration threshold with this value.Determining the concentration threshold is probably the most time consuming step for this algorithm to work and automatically setting this value would be advantageous.
This software is modular and has the ability to be programmed to identify numerous analytes detected in FSCV data other than adenosine.Moreover, the strategy for detecting adenosine transients could be extended to other analytes like oxygen and dopamine.For instance, oxygen and adenosine release are correlated, so the program could scan the reduction voltage for oxygen in a window after an adenosine peak is detected to scan for oxygen transients 25 .Additionally, the algorithm could scan oxidation and reduction voltages for dopamine, with the time-lag threshold set to zero, and automatically detect spontaneous dopamine transients.To make a dopamine algorithm more robust a reduction to oxidation ratio threshold would be empirically calculated.This ratio would help reduce possible FP from being accepted by the program and make the program more robust.Alternatively, dopamine could be preprocessed by PCR to remove noise from the color plot data and subsequently postprocessed by the dopamine transient algorithm.Thus, as long as there is enough information to make rules about detection from the electrochemical data, there are limitless possibilities for this modular spontaneous transient program in analyzing electroactive species.An automated analyte identification algorithm saves hundreds of hours of time in tedious peak feature detection and will normalize data between animals, researchers, and institutions.

References
Next, the final event time data, acquired from adjacent background subtraction, is stored in loc1 of dataFinal.

CODE: seed-dataFinal(1,i) <= 22;
The value of 2.2 seconds was used due to multiple peaks being present in the data.This value can be adjusted to suit experimental needs.

Figure 1 . 1 :
Figure 1.1: Mechanism for adenosine production.Intracellularly adenosine can be formed from ATP.Alternatively, adenosine can be formed from metabolism of ATP extracellularly.Modified by Pajeski et al.

(Figure 1 . 2 )
to traditional cyclic voltammetry but FSCV has faster temporal resolution.With FSCV, a carbon-fiber microelectrode (CFME) is scanned from a negative holding potential to a positive switching potential and immediately ramped back down.The total length of the scan is about 10 ms, and data is collected at 10 Hz, which gives 100 ms temporal resolution.Background charging currents are stable with CFMEs, which allows for accurate background subtraction.Charging currents result from the formation of a double layer of ions at the CFME interface that act similar to a capacitor 22 .Background subtracted cyclic voltammograms provide a selective fingerprint for each analyte of interest.The CFME has a diameter of 7 μm, in vivo spontaneous transient adenosine release (Figure 1.3A).During FSCV cyclic voltammograms are taken 10 times per second.A color plot displays cyclic voltammograms as a function of time with current represented in false color.As can be visualized the color plot, the start of the primary peak and secondary peak are not at the same time point.The dashed lines in the color plot represent times where CVs were taken for adenosine at three different time points (Figure 1.3B).The first CV, taken at

Figure 1 . 3 :
Figure 1.3: in vivo Spontaneous transient release of adenosine.(A) The top shows current versus time traces of adenosine at 1.4 V for the primary oxidation (orange) and 1.2 V for the secondary oxidation (black).The bottom is a false color plot with dashed lines showing where the CVs were taken.(B) Cyclic voltammograms taken at different time intervals.The top CV was taken at the start of the primary oxidation, the middle CV was taken at the primary peak maximum, and the bottom CV was taken at the secondary peak maximum.Figure taken from Nguyen et al.

Figure 1 . 4 :
Figure 1.4: Current versus time traces for primary peak, taken at 1.5 V, in triangles, and secondary peak, taken at 1.0 V, in circles of adenosine in flow injection analysis experiment.The rise of the primary peak occurs before the rise of secondary peak.Inset : At 3.5 seconds the rise of the primary peak starts and the rise of the secondary peak starts at 3.6 seconds.Therefore, the primary oxidation product must be produced before the secondary product is able to be formed.Figure taken from Swamy et al.

Figure taken from Swamy et al. Borman | 11 1. 3 .
Figure 1.4: Current versus time traces for primary peak, taken at 1.5 V, in triangles, and secondary peak, taken at 1.0 V, in circles of adenosine in flow injection analysis experiment.The rise of the primary peak occurs before the rise of secondary peak.Inset : At 3.5 seconds the rise of the primary peak starts and the rise of the secondary peak starts at 3.6 seconds.Therefore, the primary oxidation product must be produced before the secondary product is able to be formed.Figure taken from Swamy et al.

Figure 1 . 5 Figure 1 .
Figure 1.5B 27 .Essentially the CVs included in the training set are removed, which

Figure 1 . 5 :
Figure 1.5: Residual plots from in vivo stimulation of dopamine in freely moving rats.(A) in vivo color plot.(B) Residual plot when dopamine and pH are in training set.(C) Q trace of residual plots where the dashed line represents Qα at the 95% significance.All Q scores are below this threshold, therefore the training set accurately describes all relevent data.(D) Residual plot with only dopamine in training set.(E) pH is above the Qα threshold and is retained in the Q trace.Figure taken from Keithley et al characterize random adenosine transients from FSCV color plots.Typically, analysts tabulate adenosine transient features and this process is very tedious and time consuming, often taking 10-18 hours per experiment for a skilled analyst.In a single animal experiment 700 transients can be found and characterized and 24 experiments are needed in order to be statistically relevant and this amounts to 14,400 transients per publication.If an in vivo researcher performs 3 experiments per week the data analysis time will be approximately 30-54 hours, which is a significant amount of time.The developed adenosine transient program automatically reads, analyzes, and creates a report of adenosine features in 40 minutes per in vivo animal experiment.This thesis demonstrates the reliability of the automated identification program in determining adenosine features and speed of compiling these features.Automatically identifying analytes in FSCV data will allow researchers to analyze adenosine and is customizable measure adenosine is important in understanding the roles it plays in neuromodulation and homeostatic regulation.Recently, direct measurements of spontaneous transient adenosine release in vivo have been made by fast-scan cyclic voltammetry (FSCV) 5 .These events are spontaneous, rather than stimulated, and last only a few seconds.Several hundred transients can occur in the four hour data collection typical of an in vivo experiment.Current data analysis requires a human to pick the transients by hand; a human experienced in analyzing the data can identify a transient approximately every 1.5 minutes.Therefore, if a data set contains 700 transients, then it would take about 18 hours to analyze.Adenosine transient events are seemingly random and do not follow any redily identifible pattern so all the data must be painstakingly analyzed.In addition to being slow, identification by an analyst could be potentially biased.Automating identification of adenosine transients would save time and normalize data analysis between researchers.
denoting the training is not sufficient to predict the concentration of the neurochemical.The other major problem for finding adenosine transients is that they are random events, with no unique time markers.While many dopamine events are linked to behaviors or cues, finding adenosine transients requires an algorithm that does not use time as a rule for identification.In this study an algorithm was designed to identify and characterize random adenosine transients from FSCV data.Our program automatically reads, analyzes, and creates a report characterizing the duration, concentration, and event time for each adenosine transient.This automated analysis takes only about 40 minutes to analyze an in vivo data set.The program was validated with 4 data sets from 3 independent researchers and compared to the results of human analysts.The program resulting in 10 ± 4% false negatives (FN), due to multiple peaks and high thresholds, and 9 ± 3% false positives (FP), due to random noise that occurs in biological experiments.The algorithm was tested against ATP, histamine, hydrogen peroxide, and pH, known interferents in the brain, and only generated one false positive in 82 measurements.

Matlab 2014b (
The MathWorks Inc, Natick, MA, USA).First, non-background subtracted and non-filtered FSCV color plot data were exported from High Definition Cyclic Voltammetry (HDCV), a program developed in the Wightman lab18  .A typical in vivo FSCV experiment has files with 80-180s of data and the FSCV transient program individually reads each file for analysis.After the files are exported an analyst defines three user inputs: maximum (1) primary and (2) secondary oxidation voltage, p max and s max , respectively, and a background subtraction (3) increment value.Then FSCV data is read into the program and convoluted with a 2-D Gaussian filter (size=7, σ=7), which is a low-pass blurring filter.If the algorithm finds adenosine events, then features including event time, concentration, and duration are written to a comma separated value file (Figure 2.1, Step 7) until all files are read and analyzed.The program was run on a 3.4 GHz PC computer with Windows 10 for data analysis.

Figure 2 . 1 :
Figure 2.1: Algorithm for FSCV transient.Files are read into the program for incremental background subtracted(1).Analyst defined primary and secondary oxidation voltages are scanned for adenosine peaks(2).Lag time filter is applied to detected peaks to remove spurious peaks (3) and resultant peaks are background subtracted adjacent to the peak (4) then identified similarly to steps 2 and 3(5).Signal-to-noise and ratio filter are applied to detected peaks to remove spurious peaks(6).Event time, concentration, and duration are written to file until all peaks are investigated(7).

spend 10 -
18 hours identifying transients and calculating the event time, concentration and duration of the transients by hand.With FCSV cyclic voltammograms are taken 10 times per second, and multiple CVs are often viewed as a color plot, with data stacked as a function of time.The resulting color plot (Figure 2.2B) forms a three-dimensional plot of voltage and current as a function of time.Data files are typically 180s worth of data, 20-80 files per experiment.Thus, 36,000-144,000 individual cyclic voltammograms are collected and must be analyzed per experiment.Thus, an automated, unbiased method of analyzing thousands of cyclic voltammograms to identify hundreds of adenosine transients is needed.Building a computer program will automate this process, normalize data analysis between researchers and allow more time to conduct experiments and interpret experimental results.

Figure 2 . 2 :
Figure2.2: in vivo spontaneous adenosine transients from the hippocampus brain region.A) i vs t trace with 5 transients detected.The primary peak maximums (black trace) occur before secondary peak maximums (red trace) and is the basis for the lag time filter.The Sp,i / Pp,i ratio is > 0.50, thus this peaks would be accepted as adenosine transients.B) False color plot of adenosine transients with white lines representing the primary and secondary oxidation voltages.

Figure 2 . 3 : 3
Figure 2.3: Incremental background subtraction of in vivo adenosine transients from the hippocampus brain region.A-C) Incrementally background subtracted FSCV data every 50 seconds.The white vertical lines in the files show when the background was taken.
Any automated method for measurement and identification is subject to false negatives and false positives.Since many common interferents found in the brain have been rejected by flow-injection analysis experiments (vide infra), FP are mainly generated from random noise in the data.The main reason the adenosine transient algorithm will generate FN is due to multiple peak maximums occurring in a single peak.False negatives due to multiple peaks can corrected for in the program by adjusting the prominence threshold, which is the minimum peak height between two consecutive, possibly overlapping peaks.However, it will always be difficult to measure the concentration and duration of multiple peaks that do not go back to baseline in between.Alternatively, the transient program will reject peaks if the data is noisy and the threshold is set above the level of small adenosine transients.If thresholds are high during the incremental background subtraction, adenosine transients are rejected and cannot be measured during adjacent background subtraction.One strategy for setting up thresholds in both the incremental and adjacent background subtraction is to minimize the amount of FPs but this will ultimately increase FN, as seen in the brain slice experiment.Alternatively, setting thresholds properly in both background subtraction parts of the adenosine algorithm can achieve the minimization of both FP and FN.Overall, analyst validation of adenosine transient program had a mean precision of 0.91 ± 0.01, sensitivity of 0.90 ± 0.04, and accuracy of 0.90 ± 0.02.An accuracy of 0.90 is sufficient for the FSCV transient algorithm because analysts also fail to detect adenosine transients in data when counting.

Figure 2 . 4 :
Figure 2.4: in vitro testing of biologically relevant interferents.i vs t traces, cyclic voltammograms, and false color plots for A) adenosine, B) ATP, C) histamine, D) hydrogen peroxide, E) pH 7.3 shift, F) pH 7.5 shift.Interferents are rejected by the algorithm due to smax occurring before pmax and an interferent max sp,i /pp,i ratio below adenosines minimum sp,i /pp,i ratio.

Figure 2 . 5 :
Figure 2.5: in vivo stimulated histamine from the premammillary nucleus.A) Primary (black) and secondary (red) oxidation peak i vs t traces of stimulated histamine fail lag time filter because maximums occur at the same time.B) False color plot white lines are pmax and smax generated from adenosine transients in data set 2 (S2).

Table 2 . 1 :
Analyst validation of adenosine transient data sets.S1 is an in vivo measurement in the caudate putamen.S2 and S3 are in vivo measurements in the hippocampus.S4 is a brain slice experiment from the prefrontal cortex.Each data set is an independent experiment