Biomarker Development for Sudden Unexplained Death in Epileptic Patients: Roles of Respiration, Lactate, and Neuroinflammation

NHSJS Reports

Thomas Barbera

Peer Reviewer: Sanya Jain

Professional Reviewer: Dr. Misty D. Smith

Abstract

Sudden Unexplained Death in Epileptic Patients (SUDEP) is one of the leading causes of epilepsy related deaths.The risk of SUDEP in children with Dravet syndrome is estimated to be 15-fold greater than other childhood-onset epilepsies (Skluzacek JV, Watts KP, Parsy O, Wical B, Camfield P. Dravet syndrome and parent associations: The IDEA League experience with comorbid conditions, mortality, management, adaptation, and grief. Epilepsia. 2011;52(suppl 2):95–101.) Research in this field is focused on discovering a biomarker, or way to predict, SUDEP. Research is focused on three possible biomarkers: respiratory rates, serum lactate, and neuroinflammation. The following experiments explore each possible biomarker in separate experiments. The first experiment explores respiratory rates following seizures, and the hypothesis is that declining respiratory rates cause respiratory failure during sleep. This experiment used a non-invasive respiratory collar to measure baseline respiratory rates and respiratory rates following hyperthermia-induced seizures. The second experiment studies serum lactate, and the hypothesis is that the increase in serum lactate following a seizure causes cardiac or respiratory failure during sleep, thus causing SUDEP to occur. This experiment used a lactate kit to analyze the concentration of serum lactate in a sample of blood collected following euthanization of the mice used in the respiratory rates experiment. The third experiment studies neuroinflammation of astrocytes, a specific neurotransmitter that controls the Nav1.1 Sodium Channel in the hindbrain, and the hypothesis is that the inflammation of the astrocytes following a seizure causes brain dysfunction, eventually resulting in SUDEP. This experiment analyzed brain slices using color-based imaging following the staining of the brain slices for specific proteins. For all three of the experiments, only the serum lactate study yielded significant results. The p-value obtained for the difference in lactate following a seizure was 0.04978, which indicates a significant difference as the value is >0.05. However, all other p-values obtained were at least 0.3168, which indicates none of the measured respiratory rates yielded a significant increase following seizures. Antiepileptic Drug (AED) production should be focused on controlling increase in serum lactate, and if the rate of SUDEP decreases for patients on this AED, a correlation between SUDEP and serum lactate increase can be established.

Introduction

Sudden Unexplained Death in Epileptic Patients (SUDEP) is the cause of death for over 20,000 epilepsy patients each year in the United States, and accounts for over 15% of all epilepsy-related deaths1. All epilepsy patients are at risk of SUDEP, although it mainly occurs in patients between 20 to 45 years old2. SUDEP most commonly occurs in Dravet Syndrome epilepsy, a genetic strain of epilepsy diagnosed in infancy3. SUDEP is nearly impossible to predict, as there have been no biomarkers validated as accurate ways to predict SUDEP. However, SUDEP is always preceded by seizures, so researchers agree that there must be validatable biomarkers to prevent SUDEP4. Mice bred for Dravet Syndrome exhibit similar characteristics to those of human patients, making mice viable options for testing for biomarkers to validate their development5. This leads researchers to question the pre-ictal (before a seizure) and post-ictal (after a seizure) rates in Dravet Syndrome mice, and whether they are significant enough to be classified as possible biomarkers for Sudden Unexplained Death in Epileptic Patients.

Literature Review:

Sudden Unexplained Death in Epileptic Patients (SUDEP)

Dravet Syndrome leads to an increased chance of SUDEP 6. SUDEP is classified as the death of an epilepsy patient that is both sudden and unexpected from an unknown cause. There does not have to be evidence of a seizure, and all examinations after the death of the patient must reveal no direct instigator of death7. Despite extensive studies on the causes of SUDEP and its frequency in Dravet Syndrome patients, there have been no completely validated biomarkers of SUDEP, making the deaths unpreventable and unpredictable8

Some hypotheses as to what causes SUDEP are hindered respiratory, cardiac, and sensory functions following seizures; however, these causes are highly unpredictable, and there are no validated biomarkers for them9. Researchers discovered that generalized tonic-clonic seizures (GTCS), which involve the patient losing consciousness and experiencing severe muscle contractions, are a major risk factor for SUDEP10. One study found that SUDEP is 15 times more likely to occur in patients that have had more than two GTCS per month11

Many experiments, both those aforementioned and several others, are investigating biomarkers for the prediction and prevention of SUDEP12. Along with conducting the pre-mortem (prior to euthanization) studies to discover biomarkers, researchers are attempting to expand their knowledge of SUDEP by performing post-mortem (following euthanization) analyses and autopsies with SUDEP patients. These studies are then compared to post-mortem studies of animal models and the differences are analyzed to obtain more information about the effects of SUDEP, and their correlations with animal models13.

Dravet Syndrome

Dravet Syndrome causes cognitive impairments such as slowed brain development, and is commonly accompanied by psychological issues and physical defects14. These psychological issues include anxiety and depression; furthermore, the mental defects that accompany Dravet Syndrome affect patients physically, as patients are more likely to develop unhealthy coping mechanisms such as smoking or develop an unhealthy diet.15. There are also specific characteristics that differentiate Dravet Syndrome from other forms of epilepsy16. Dravet Syndrome is a pharmacoresistant form of epilepsy, meaning AEDs do not affect the patient’s seizure frequency or amplitude. The syndrome is unique, as it is diagnosed when the patients are infants17

Dravet Syndrome patients’ genetic mutation is commonly referred to as the SCN1A mutation, which is a mutation on the GABAergic neurons in the brain. GABAergic neurons are neurons that produce GABAergic receptors, which are inhibitory neurotransmitters (Ribak, C., 1987)). GABAergic receptors constrain the functions of sodium channels, which usually allow for communication between neurons through the transfer of sodium ions18. When these neurons produce damaged receptors, they can suddenly inhibit specific brain functions that cause the patient to seize19. Dravet Syndrome impairs the Nav 1.1 sodium channel, which directly affects the efficiency and strength of certain features of the brain in both the pre-ictal and post-ictal phases20.

Pre-Mortem Studies

Pre-mortem studies help researchers further understand SUDEP for more accurate predictions. Studies addressing pre-mortem biomarkers for SUDEP usually include investigating respiratory rates. Seizures in Dravet Syndrome are mainly caused by hyperthermia, commonly known as a fever. Hyperthermia-induced seizure tests begin by increasing the internal body temperature of the mouse, and simulating the thermal effect of a hyperthermia seizure21. The tests were previously administered to several age groups, and their conclusion supported the hypothesis that adults are more prone to SUDEP than children22

Other pre-mortem studies focus on measuring respiratory rates, along with other breathing-related measurements such as oxygen saturation through both mouse assays and human testing. Sleep apnea, a sudden cessation of breathing during sleep, appears to have a connection to SUDEP23. Furthermore, several researchers hypothesize that tachycardia, an abnormally fast heart rate, occurs due to a decline in respiratory rates following seizures, thus causing SUDEP24. By studying the difference in pre-ictal and post-ictal respiratory rates, researchers may discover that hindered respiratory rates could be a possible biomarker of SUDEP. 

Post-Mortem Studies

Some post-mortem studies explore peripheral lactate, a type of lactate produced when a body must perform a physically demanding action25. One study which analyzed the increase of lactate supported the conclusion that the increase of the UCH-L1 branch of lactate and plasma both correlate with age; however, the sex of the patient did not affect the levels of UCH-L1 or plasma, which fits with existing literature26. In another study exploring connections between general tonic-clonic seizures and SUDEP, the researchers discovered that serum lactate significantly increased after a GTCS compared to other forms of seizures. This significant difference in lactate between GTCS and other types of seizures presents the possibility of lactate as a biomarker to predict SUDEP27

Some post-mortem tests examine the brains of Dravet Syndrome mice following their seizures. Astrocytes, nervous system sensors primarily found in the medulla that control breathing and heart rate, may send abnormal signals following seizures28. Astrocytes can also sense movements within the brain, specifically the changes in glucose and lactate, which activate the astrocytes and begin neuroinflammation29. Measuring astrocyte inflammation involves staining predetermined sections of the brain and imaging them30. The pictures are then analyzed for differences with saline-controlled slices through the staining and imaging systems to quantify differences in the pictures31.

One study focused on the correlation between epileptic seizures and non-epileptic seizures by monitoring the NaV1.1 sodium channel following seizures32. Researchers discovered that the greatest levels of dysfunction occurred following hyperthermia seizures33. This further suggested neuroinflammation as a possible biomarker for SUDEP. Another study discovered that over 50% of human patients tested had miniscule brain damage, and that approximately 25% of brains showed signs of swelling following SUDEP34. This study also found that there are no significant differences in brains due to age or sex in SUDEP cases, which helps researchers by allowing them to test a wider variety of brains without having to meet specific parameters35

Methods:

Justification for Experiments

The literature on SUDEP and its possible biomarkers presents substantial evidence that respiratory, lactate, and neuroinflammatory analyses may reveal possible biomarkers for SUDEP. Due to the decrease in respiration following seizures, respiratory rates following seizures may be a possible biomarker for SUDEP36. The correlation between seizures and a positive increase in serum lactate following the seizures warrants an analysis of the levels of serum lactate in the body directly following the seizure, which may indicate an increased risk for SUDEP37. Because lactate directly affects neuroinflammation, inflammation of astrocytes in the medulla of the brain may also be a viable biomarker for SUDEP38. All of these potential biomarkers may lead researchers to further understand SUDEP and prevent it. This evidence and examination of literature suggests the hypotheses that differences between pre-ictal and post-ictal measurements in respiratory rates, serum lactate, and in Dravet Syndrome mice are substantial enough to provide accurate biomarkers for SUDEP.

Risk Assessment and Funding

All of the research was conducted in a University of Utah College of Pharmacy research lab and was supervised by qualified scientists. Qualified scientists supervised the use of all hazardous chemicals and all animals both during their uses and during their storage. All procedures were approved by the University’s Institutional Animal Care and Use Committee (IACUC) prior to experimentation. All researchers also underwent animal handler training prior to all experimentation and access to the animals. This training was sanctioned by the IACUC, provided by the University, and allowed for safe use of chemicals and animals during experimentation. Funding for the experiments comes directly from the University of Utah College of Pharmacy.

Subjects

The following experiments use mice diagnosed with Dravet Syndrome, as experiments that use these mice provide data that correlates almost indistinguishably with human Dravet Syndrome patients39. This provides reasons for the use of mice in these experiments instead of other smaller organisms, as other animals would not provide as accurate data as mice diagnosed with Dravet Syndrome. The experiments used the same 12 mice, 6 with the SCN1A mutation (with Dravet Syndrome) and 6 from the same parents but without the mutation (Wild Type). These values were chosen to ensure that data would be conclusive without extending the research beyond the length of time available. The mice were between the ages of 2-4 weeks old and were obtained from a University breeding center for Dravet Syndrome mice. During experimentation, the mice were given arbitrary numbers by a third party to prevent bias throughout the study.

Respiratory Study:

Hyperthermia-Induced Seizure Induction

Hyperthermia seizures were induced before collecting post-ictal measurements of respiratory rates. A thermometer was inserted rectally into the mouse to allow for measurement of the internal body temperature of the mouse while it was heated. The seizure was then induced by placing each mouse into individual containers with a heat lamp placed above each container. Each mouse was monitored until it either seized or reached a temperature of 42.5?, as previous studies show that Dravet Syndrome mice must seize before reaching this temperature40. If the mouse seized, the temperature at which this happened was recorded and the mouse was immediately removed from the container and moved back to continue respiratory data collection.

Respiratory Testing

A respiratory collar was used to record all respiratory rates to analyze during data analysis. The mice had their necks shaved prior to the collar testing, and were then habituated to a new area for at least 30 minutes. The first mouse was placed into a testing container, and a practice collar (figure 1) was placed on the mouse for five minutes to acclimate the mouse to the collar before beginning the test.

The test collar (figure 2) was then placed around the neck of the mouse. Data was then recorded using the program MouseOX, which is a non-invasive monitoring system designed specifically to record respiratory functions of mice using a collar placed around the neck of the mouse41. After 5 minutes of collecting data, the collar was then taken off of the mouse, and the mouse was moved to another section of the lab to have a hyperthermia seizure induced. After the hyperthermia seizure was induced, the mouse was immediately returned to the test collar to have its post-ictal respiratory rates measured. After pre-ictal and post-ictal measurements were taken, the data was saved onto the computer and exported.


Serum Lactate Study:

Serum lactate was collected and run through a lactate kit to determine if there was an increase of lactate following a seizure. Each mouse was perfused and had their trunk blood, cerebral blood, and brain collected to obtain the materials for the lactate and neuroinflammation studies. Each of these blood samples were taken and mixed with a lactate kit to allow for spectrophotometric analysis: analysis done by shining a light through a solution to determine concentration (Lundgaard, I et. al, 2017). To allow for a standard curve during data analysis, 10 µL of Lactate Standard (LS) was mixed into 990 µL of Lactate Assay Buffer (LAB) and stirred together. Each well was then filled with 10 µL of the solution, and more LAB was added to obtain molarities ranging from 2-10 µM of the Lactate Standard. Samples of lactate taken from both Dravet Syndrome and wild type mice were then placed into each of the wells containing the solution. 

Test samples were created to convert from absorption to lactate using the standard curve to analyze the difference in lactate following a seizure. In a separate well container, each lactate test sample was placed in a well with 50 microliters (50 µL) of a solution containing 46 µL of LAB, 2 µL of Lactate Substrate Mix (LSM), and 2 µL of Lactate Enzyme Mix (LEM). After the serum was diluted in the solution for two hours, samples from the wells were then moved to different wells containing the original 50 µL solution to obtain dilution values closer to the standard curve. The solutions were then incubated in the new dilution for 30 minutes before spectrophotometric analysis began.

Figure 3: ELIZA Spectrophotometer

Solutions then had to be run through a spectrophotometer to collect data on the amount of lactate in each sample. Both the standard curve solutions and the test solutions were then placed into an ELISA spectrophotometer (figure 3). This specific spectrophotometer is designed to measure levels of lactate in blood samples42. The values collected from the spectrophotometer were then exported from the computer and later used for data analysis.

Neuroinflammation Study:

Brain Slicing

Figure 4: Microtome

A brain sample was taken from each mouse during their perfusion, and the hind portions were sliced to allow for staining of astrocytes. The slicing process began by placing an Optimal Cutting Temperature (O.C.T.) compound on a microtome (figure 4) at -20.0 °C. Immediately after the compound was applied to the surface, the brain was placed into the compound. A 40 micrometer (40 µm) slice was then taken from the brain. The slice was then taken off of the blade and was placed into a well filled with 1X PBS (phosphate-buffered saline) (see figure 5) if it contained the medulla of the brain. The well was then marked with the designated section of the brain that the slice was from. This process was continued for each brain, and then the slices were mounted.

Figure 5: Wells for Brain Slices

Brain Mounting

Brain slices were then moved to image slides to allow for staining, photography, and quantitative analysis. After all of the brain slices were put into wells, the slices were mounted onto Superfrost Plus slides (figure 6). Each slice was mounted using 1X PBS and a paintbrush, and each slide contained 3-4 slices. Gel from an Immuno Edge pen was then applied around each slice to prevent the solutions from sliding off of the slides. After the gel was applied, the gel was allowed to dry until it was a translucent ring. The slides were then washed for five minutes three times and then moved into a humidified black slide box in preparation for their staining.

Figure 6: Imaging slide

A solution composed of 1% Sodium Deoxycholate, 0.2% Triton X-100 in PBS was placed on each of the slices, and the slices were kept in the humidified box at room temperature while the solution permeated the slices43. The solution was then suctioned off of each slice. After the solution was suctioned off of the slices, the glial fibrillary acidic protein (GFAP) antibody, which was conjugated to Cy3 (Sigma) and diluted to 1:1000 in Cyto Q and Sodium Deoxycholate, was placed onto each of the slices and they were incubated in the antibody for two hours. The GFAP antibody stains for GFAP, which appears on astrocytes. The antibody was then removed via the suction method, and the slices were washed44. A 4',6-diamidino-2-phenylindole (DAPI) solution diluted in 1X PBS was then placed onto each of the slices and they were incubated for 10 minutes in the solution. The sections were washed again before coverslips were applied using a Prolong Gold (Invitrogen) solution and placed into non-humidified slide boxes stored at 4? until they were viewed and photographed using the imaging microscope.

Photography

Figure 7: Camera used for color-based imaging

Each of the slides were taken and placed individually under the microscope (figure 7). GFAP staining appears under red light and DAPI staining appears under blue light, so both a red and a blue light were shined onto each of the slides separately to scan for levels of GFAP and DAPI respectively. Photographs were taken of several sections on each slide under each color45. Each photograph was also marked with the arbitrary number given by the third party that correlated with the mouse whose brain slices were on each slide.

Data Analysis/Results:

Respiratory Study

A significant difference between post-ictal and pre-ictal respiratory rates may further validate the use of respiratory rates as a biomarker for SUDEP, as respiratory complications may cause SUDEP to occur. A p-value obtained from a t-test quantifies a difference between two data sets as either significant or insignificant based on the p-value obtained from the test. The p-value obtained must be below or equal to 0.05 to be classified as a significant difference between the data sets46. The standard deviation of the data sets shows whether the range of the data set varies widely from the mean of the data. This helps classify the data as accurate or inaccurate, as a low standard deviation compared to the mean value means the data points are all similar in value.


Figure 8a: Tables of pre-ictal and post-ictal measurements of oxygen saturation, heart rate, breath rate, and pulse  distension of the Dravet Syndrome  (P 1-4) and Wild Type (R 1-5) mice

Figure 8b: (Beginning in upper left corner) Graphs of respiration, oxygen saturation, (SpO2) heart rate, and pulse distension of wild type (wt) and heterozygous (het) mice. The standard deviation of each mean is shown with a line above the mean value (apex of the grey region). All t-tests show no significance in the difference between respiratory rates following seizures compared to pre-seizure rates (p> 0.05)

The average of the differences for each of the four measurements of the five Dravet Syndrome and wild type mice were obtained from the MouseOX program (figures 8a & 8b). The standard deviation of the data from each of the measurements is shown using a line beginning at the mean value to the positive standard deviation (figure 8b).  The averages for oxygen saturation of the wild type and Dravet Syndrome mice are 95.80±6.09% and 96.44±6.79% of the baseline oxygen saturation respectively with a p-value of 0.8878. The averages for the heart rate of the wild type and Dravet Syndrome mice are 92.82±9.56% and 106.2±23.1% of the baseline respectively with a p-value of 0.3168. The averages for respiration (breath rate) of the wild type and Dravet Syndrome mice are 128.1±31.3% and 107.9±39.3% of the baseline respectively with a p-value of 0.4324. The averages for pulse distension of the wild type and Dravet Syndrome mice are 100.8±21.2% and 86.14±27.74% respectively with a p-value of 0.5482. None of these p-values proved the difference in ranges significant, as none of the values are below 0.05. Furthermore, none of the standard deviations vary widely from the mean, so the data is statistically precise. 

Serum Lactate Study

Lactate is created following a seizure due to muscle contractions in the heart, and it is hypothesized that a significant increase in lactate following a seizure could be a possible biomarker for SUDEP. The data obtained from this experiment was used to create a standard curve of lactate with respect to the optical density (O.D.) to convert the absorptions of the test samples to micromolarity (figures 11 a&b).

SampleConcentrationFound Conc.WellsValuesMeanValueStd.Dev.CV%
Sta0122.691C11.11.100








Sta0222.691D11.11.100








Sta0344.651E11.5081.50800








Sta0444.055F11.3841.38400








Sta050-0.552G10.4250.42500








Sta060-0.605H10.4140.41400








Sta0764.92A21.5641.56400








Sta0866.149B21.821.8200








Sta098MaskedC2MaskedMaskedMaskedMasked








Sta108MaskedD2MaskedMaskedMaskedMasked


Figure 9: Concentration of each well obtained by the spectrophotometer with their respective means, standard deviations, and coefficient of variation (%CV)

hetsnmoldilution factor
calc (uM)
1R0.046355229971004.6355229970.4635522997
2R1.007085385100100.708538510.07085385
4R0.320163324110032.016332413.201633241
5R0.644409751410064.440975146.444097514





wt



1P0.039149753811003.9149753810.3914975381
2P0.358592530310035.859253033.585925303
3P-0.22024738810000
4P-0.253872943410000
5P-0.172210880310000

Figure 10: The nanomoles, dilution factor, and molarity for the Heterozygous (1R-5R) and Wild Type (1P-5P) mice

The absorptions obtained from the ELISA spectrophotometer for the test samples were then converted to nanomolarity using the line of best fit on the standard curve. The r2 value on the line of best fit is 0.95, indicating a linear relationship between molarity and wavelength. This value statistically indicates the line of best fit is an accurate way to convert wavelength to molarity, the amount of lactate per liter of blood. All of the wavelengths obtained for the test solutions were then converted to molarity by using the equation for the line of best fit (figure 11a). The mean micromolarity (µM) of the lactate for the wild type mice was 0.1500±0.2057 µM. The mean micromolar lactate for the mice with Dravet Syndrome was 5.045±4.147 µM. A t-test analysis on the difference between the micromolarity of the samples from the Dravet Syndrome and wild type mice yields a p-value of 0.04978, indicating a significant difference between the lactate created by the Dravet Syndrome mice and the lactate created by the wild type mice.


Figure 11a (Top): The standard curve created from the base solutions of LS and LAB. The optical density (O.D.) of 450 nm was specifically selected, as it allows for minimum absorption of chemicals from the lactate kit. Equation in upper right-hand corner was used to convert test samples from absorption to molarity.

Figure 11b (Bottom): Graphs of micromolar lactate of the wild type (wt) and heterozygous (het) mice following the conversion from O.D. to micromolarity. P-value showed significance (p?0.05) between post- and pre-ictal rates.




Neuroinflammation Study

Astrocytes inflame following a seizure, and it is hypothesized that this inflammation inhibits their ability to produce proteins. If this production is hindered, this will cause complications in the Nav1.1 Sodium Channel, which relies on these proteins, making neuroinflammation a possible biomarker for SUDEP warrant SUDEP.

The quantitative differences between the photographs taken of the DAPI solutions were negligible. 3-Dimensional quantitative analysis was unable to determine any difference between the Dravet Syndrome and wild type slices. Because DAPI production is not hindered following inflammation, the protein must be contained rather than produced by astrocytes. 


Figure 12: Photograph of the GFAP staining on the hind brain of  a Dravet Syndrome mouse. Astrocytes are outlined in red, with the inflamed astrocytes appearing much larger with enlarged shapes. Brain shows much more inflammation than the Wild Type brain.

Figure 13: Photograph of the GFAP staining on the hind brain of a Wild Type mouse. Astrocytes are outlined in red, with the inflamed astrocytes appearing much larger with enlarged shapes. Brain shows much less inflammation than the Dravet Syndrome brain.

The images for the GFAP-stained slices showed more promising results. Inflamed astrocytes appear in jagged lines or are more enlarged shapes. The Dravet Syndrome image (figure 12) shows much more inflammation than the wild type image (figure 13). These images are potential indicators of the difference in inflamed astrocytes following a seizure; however, these images provide no quantitative data. 3-Dimensional quantification must be done to allow for analysis of the number of inflamed astrocytes, as astrocytes may appear on other layers of the slices or may be vertical in the slice and therefore are not captured by the image.

Conclusions and Discussion

Respiratory Rate Study:

The hypothesis for this study was not supported by the data. All data from this experiment yields no significant difference between any of the respiratory rates when comparing the post-ictal and pre-ictal measurements. The data was highly varied, as both respiration and pulse distension were lowered following seizures, whereas oxygen saturation and heart rate were increased following seizures, further supporting the conclusion that respiration is not an accurate biomarker for SUDEP. When compared, mean values for wild type mice versus Dravet Syndrome mice showed no significant difference, as p-values were at least 0.3168, indicating all differences were insignificant. Furthermore, standard deviations do not widely vary from the mean values, as the values were on average ±20.635%, which further supports the insignificant results obtained from the t-tests. This data effectively closes the avenue of research investigating respiratory rates as a biomarker for SUDEP.

Limitations for this portion of the experiment mainly appear during the hyperthermia-seizure induction. When mice were moved between the respiratory collar and the hyperthermia container, handling-induced seizures would sometimes occur. To minimize handling time, the container was moved directly next to the collar. Another limitation to this experiment was the respiratory collar data measurements. If the mouse moved during the data collection, the MouseOX program would fail to record specific data values. To minimize the effect on the data, the program gave error code. Every data point with an error code was omitted from the data analysis to prevent any inaccurate readings from the data analysis to provide the most accurate analysis from the data collected.

Serum Lactate Study:

This data from this experiment yields the most significant difference after analysis, and therefore correlates with the hypothesis that serum lactate significantly increases following a seizure. The molarities for the test samples are accurate, as the r2 value obtained from the line of best fit provides an accurate linear equation to convert the measurement obtained from the spectrophotometer to molarity. The p-value for the difference between the post-ictal rates for the wild type and Dravet Syndrome mice is 0.04978, proving the difference significant. This statistically significant difference offers substantial evidence to further this avenue of research. The hypothesis that lactate would significantly increase following seizures was supported by the data. Therefore, the large increase of lactate following seizures has the greatest potential of being a biomarker for SUDEP. As this experiment solely proves lactate increase following seizures, further experiments must determine whether there is a connection between this increase in lactate and SUDEP.

However, as the significance value is close to 0.05, limitations are greatly important, as they may cause the difference to become insignificant. Furthermore, the standard deviation of the experiments is at least ±82.20% of the mean value, indicating varied data, which further emphasizes the significance of the limitations.  The main limitation to this experiment is the spectrophotometric analysis of other forms of lactate. Lactate is mainly produced when muscles contract, and during a seizure, specifically a tonic-clonic seizure, the patients must have extensive muscle contractions. Serum lactate is only produced by muscles in the heart during a muscle contraction. Other forms of lactate may be incorrectly analyzed as serum lactate during spectrophotometric analysis, and therefore obfuscate the final results. To minimize this error, each solution was diluted twice using the lactate kits. Because the kits are specifically designed to only allow pigments of serum lactate, this minimized the amount of other forms of lactate collected by the spectrophotometer.

Neuroinflammation Study:

The images produced from this experiment offer promising results for a 3-dimensional quantification of GFAP production following neuroinflammation. However, the lack of data to analyze from this experiment is due to the main limitation of the experiment itself: the 3-dimensional quantification of GFAP. Technology is available to quantify proteins such as the DAPI protein. However, there is still no technology available to quantify the amount of GFAP produced by inflamed astrocytes. This prevents a plausible correlation between the hypothesis for this experiment and the slices obtained from the experimental process. Technology, such as the 

instrument used to analyze DAPI staining, is currently being refined to allow for GFAP quantification. Once this technology is refined, the quantification will be possible, and therefore the hypothesis can be tested against the data obtained from the quantification of the slices and their images.

Another limitation for this experiment is the limited data due to a limited sample size. Several hind brains were deformed or defiled during brain slicing, and therefore could not be used for staining. This was minimized by freezing the brains several ways before slicing began, and freezing them to both the O.C.T. compound and the Microtome prior to slicing. During photography, several slices appeared to have only regions stained rather than the entire slice, so these slices were omitted from data analysis. These limitations do not alter the data collected from the remaining slices; however, they further limit the amount of data that can be used to compare the inflammation in the Dravet Syndrome brains versus the wild type brains.

Further Directions

The initial research question focused on furthering potential biomarkers for predicting SUDEP to allow for the creation of an AED that could control the biomarker and therefore decrease the rate of SUDEP. Serum lactate had a significant difference following a seizure, thus answering the initial research question, as serum lactate increase proved to be the only potential biomarker according to currently analyzed data. Therefore, serum lactate is the only potential biomarker that warrants further directions. The first direction that must be pursued in this avenue of research is the expansion of the test group. An exact or extremely similar study should be performed with a larger test group, as the data from the original experiment provided a borderline significance. Further research must also substantiate that any seizure phenotype not occuring due to Dravet Syndrome would not cause similar changes in lactate. If the data maintains its significance following an increased number of subjects, and can determine the increase in lactate only increases following Dravet Syndrome seizures, the research can be used for other studies involving SUDEP. 

This research presents data that narrows the creation of antiepileptic drugs (AEDs) to focus on preventing the increase in lactate following seizures. As the increase in lactate may cause cardiovascular or respiratory complications, maintaining a similar level of serum lactate post-seizure will allow for a larger understanding of SUDEP. If SUDEP rates decrease when patients are taking the AED, then a correlation can be established between an increase of serum lactate and the occurrence of SUDEP. If the rate of SUDEP does not decline after an AED has been created to decrease serum lactate following seizures, other hypotheses as to what may cause SUDEP to occur should be studied. However, as the other two avenues that were explored did not provide promising results, focusing AED production on controlling the increase in serum lactate following a seizure appears to be the most promising avenue of research to establish a biomarker of SUDEP.

Currently, further experiments are being conducted to attempt a correlation between the increase in lactate and SUDEP. As the study previously conducted solely discovered a significant increase in serum lactate, further experiments must prove the correlation between this increase in serum lactate and SUDEP. The current hypothesis for further research based off of the serum lactate data presented in this paper is that lactate following seizures may change more dramatically if the patient is susceptible to SUDEP. Researchers are currently monitoring lactate increase following a single seizure induced using hyperthermia-induction. The researchers are monitoring the occurrence of SUDEP in Dravet Syndrome mice while simultaneously having their serum lactate levels measured to determine whether there is a correlation between a lasting increase in serum lactate following seizures and SUDEP. If there is a significant correlation between a lasting increase in serum lactate and SUDEP, a correlation can be established between serum lactate levels following seizures and SUDEP, therefore making serum lactate a biomarker for Sudden Unexplained Death in Epileptic Patients.

Works Cited

Bittolo, T., Raminelli, C. A., Deiana, C., Baj, G., Vaghi, V., Ferrazzo, S., . . . Tongiorgi, E. (2016). 

Pharmacological treatment with mirtazapine rescues cortical atrophy and respiratory deficits in MeCP2 null mice. Scientific Reports, 6(1). doi:10.1038/srep19796

Bozorgi, A., & Lhatoo, S. D. (2013). Seizures, Cerebral Shutdown, and SUDEP. Epilepsy Currents, 

13(5), 236-240.

Brackenbury, W., Yuan, Y., O'Malley, H., Parent, J., & Isom, L. (2013). Abnormal neuronal patterning 

occurs during early postnatal brain development of Scn1b-null mice and precedes hyperexcitability. Proceedings of the National Academy of Sciences of the United States of America, 110(3), 1089-1094.

Cheah, C., Yu, F., Westenbroek, R., Kalume, F., Oakley, J., Potter, G., . . . Catterall, W. (2012). Specific 

deletion of Na v 1.1 sodium channels in inhibitory interneurons causes seizures and premature death in a mouse model of Dravet syndrome. Proceedings of the National Academy of Sciences of the United States of America, 109(36), 14646-14651. Retrieved from http://www.jstor.org/stable/41706261

Couzin, J. (2008). When Death Strikes without Warning. Science,321(5885), 31-33. Retrieved from 

http://www.jstor.org/stable/20054397

D., Accorsi Mendonca, B, Zoccal. D., & G, Bonagamba. L. (2013). Glial cells modulate the synaptic 

transmission of NTS neurons sending projections to ventral medulla of Wistar rats. Physiol Rep, 1:e00080.

Devinsky, Orrin, et al. “Sudden Unexpected Death in Epilepsy: Epidemiology, Mechanisms, and 

Prevention.” The Lancet Neurology, vol. 15, no. 10, 2016, pp. 1075–1088., doi:10.1016/s1474-4422(16)30158-2.

Fuller S, Steele M, Munch G. Activated astroglia during chronic inflammation in Alzheimer’s 

disease-- do they neglect their neurosupportive roles? Mutat Res 2010;690:40-49

Gano, L. B., & Grabenstatter, H. L. (2017). Modulation of Abnormal Sodium Channel Currents in 

Heart and Brain: Hope for SUDEP Prevention and Seizure Reduction. Epilepsy Currents, 17(5), 306-310.

Harden, C., Thomson, T., Gloss, D., Buchhalter, J., Cross, J. H., Donner, E., & ... Ryvlin, P. (2017). 

Practice Guideline Summary: Sudden Unexpected Death in Epilepsy Incidence Rates and Risk Factors: Report of the Guideline Development, Dissemination, and Implementation Subcommittee of the American Academy of Neurology and the American Epilepsy Society. Epilepsy Currents, 17(3), 180-187.

Hawkins, N. A., Anderson, L. L., Gertler, T. S., Laux, L., George, A. L., & Kearney, J. A. (2017). 

Screening of conventional anticonvulsants in a genetic mouse model of epilepsy. Annals of Clinical and Translational Neurology, 4(5), 326-339. doi:10.1002/acn3.413

Jha MK, Lee IK, Suk K,. Metabolic reprogramming by the pyruvate dehydrogenase kinase lactic acid 

axis: Linking metabolism and diverse neuropathophysiologies. Neurosci Biobehav Rev 2016; 68: 1-19

Kim, Y., Bravo, E., Thirnbeck, C. K., Smith-Mellecker, L. A., Kim, S. H., Gehlbach, B. K., . . . 

Kobau, R., Zahran, H., Thurman, D., Zack, M., Henry, T., Schachter, S., & Price, P. (2008). Epilepsy 

Surveillance Among Adults — 19 States, Behavioral Risk Factor Surveillance System, 2005. Morbidity and Mortality Weekly Report: Surveillance Summaries, 57(SS-6), 1-20. Retrieved from http://www.jstor.org/stable/42568258

Lhatoo, Samden, et al. “Sudden Unexpected Death in Epilepsy: Identifying Risk and Preventing 

Mortality.” Epilepsia, vol. 56, no. 11, 2015, pp. 1700–1706., doi:10.1111/epi.13134.

Lundgaard I, Lu ML, Yang E, et al. Glymphatic clearance control state-dependent changes in brain 

lactate concentration. J Cereb Blood Flow Metab 2017;37:2112-2124

Matz, O., et al. “Lactate as a Diagnostic Marker in Transient Loss of Consciousness.” Seizure, vol. 

40, 2016, pp. 71–75., doi:10.1016/j.seizure.2016.06.014.

Metcalf, C. S., & Trandafir, C.,. (2018). Biomarker development for sudden death in Dravet 

Syndrome: Role of neuroinflammation and lactate. Neuroscience Initiative Research Proposal, 1-3.

Mondello, S., Palmio, J., Streeter, J., Hayes, R. L., Peltola, J., & Jeromin, A. (2012). Ubiquitin 

Carboxy-Terminal Hydrolase L1 (UCH-L1) is increased in cerebrospinal fluid and plasma of patients after epileptic seizure. BMC Neurology, 12(1), 85-91. doi:10.1186/1471-2377-12-85

Nurminen, M. (1997). Statistical significance — a misconstrued notion in medical research. 

Scandinavian Journal of Work, Environment & Health, 23(3), 232-235. Retrieved from http://www.jstor.org/stable/40966639

NT Kadima, R Kobau, MM Zack, S Helmers Comorbidity in Adults with Epilepsy — United States, 

2010. (2013). Morbidity and Mortality Weekly Report, 62(43), 849-853. Retrieved from http://www.jstor.org/stable/24854737

Oakley, J., Kalume, F., Yu, F., Scheuer, T., & Catterall, W. (2009). Temperature- and Age-Dependent 

Seizures in a Mouse Model of Severe Myoclonic Epilepsy in Infancy. Proceedings of the National Academy of Sciences of the United States of America,106(10), 3994-3999. Retrieved from http://www.jstor.org/stable/40428488

Parihar R, Ganesh S. The SCN1A gene variants and epileptic encephalopathies. Journal Of Human 

Genetics [serial online]. September 2013;58(9):573-580. Available from: Academic Search Premier, Ipswich, MA. Accessed September 7, 2018.A

Ribak, C. (1987). GABAergic Abnormalities Occur in Experimental Models of Focal and Genetic 

Epilepsy. The Journal of Mind and Behavior, 8(4), 605-617. Retrieved from http://www.jstor.org/stable/43853435

Richerson, G. B. (2018). Severe peri-ictal respiratory dysfunction is common in Dravet syndrome. 

Journal of Clinical Investigation, 128(3), 1141-1153. doi:10.1172/jci94999

Skluzacek JV, Watts KP, Parsy O, Wical B, Camfield P. Dravet syndrome and parent associations: 

The IDEA League experience with comorbid conditions, mortality, management, adaptation, and grief. Epilepsia. 2011;52(suppl 2):95–101.

Smith, Misty, et al. “Discovery of Antiepileptic Drugs.” Neurotherapeutics, vol. 4, no. 1, Jan. 2007, 

pp. 12–17., doi:10.1016/j.nurt.2006.11.009.

Sugawara, T., Tsurubuchi, Y., Agarwala, K., Ito, M., Fukuma, G., Mazaki-Miyazaki, E., . . . 

Takahashi et al, Neurobiology of Disease 2010; 40: 573-85

Thom, M., Michalak, Z., Wright, G., Dawson, T., Hilton, D., Joshi, A., & ... Sisodiya, S. M. (2016). 

Audit of practice in sudden unexpected death in epilepsy ( SUDEP) post mortems and neuropathological findings. Neuropathology & Applied Neurobiology, 42(5), 463-476. doi:10.1111/nan.12265

Velagapudi, P., Turagam, M., Laurence, T., & Kocherii, A. (2012). Cardiac Arrhythmias and Sudden 

Unexpected Death in Epilepsy (SUDEP). Pacing & Clinical Electrophysiology, 35(3), 363-370. doi:10.1111/j.1540-8159.2011.03276.x

Volkers, L., Kahlig, K. M., Verbeek, N. E., Das, J. G., van Kempen, M. A., Stroink, H., & ... Rook, M. 

B. (2011). Nav1.1 dysfunction in genetic epilepsy with febrile seizures-plus or Dravet syndrome. European Journal Of Neuroscience, 34(8), 1268-1275. doi:10.1111/j.1460-9568.2011.07826.x

What are Tonic-Clonic Seizures? (n.d.). Retrieved from  What are Tonic-Clonic Seizures? (n.d.). 

Retrieved from https://www.epilepsy.com/learn/types-seizures/tonic-clonic-seizures

What is Dravet Syndrome? (n.d.). Retrieved from 

What is a seizure? https://www.epilepsy.com/learn/about-epilepsy-basics/what-seizure (n.d.)

Yamakawa, K. (2001). A Missense Mutation of the Na Channel ?II Subunit Gene Nav1.2 in a 

Patient with Febrile and Afebrile Seizures Causes Channel Dysfunction. Proceedings of the National Academy of Sciences of the United States of America, 98(11), 6384-6389. Retrieved from http://www.jstor.org/stable/3055814

  1. Gano & Grabenstatter, 2017 and Devinsky, Orrin, et al., 2016 []
  2. United States Morbidity and Mortality Weekly Report, 2010, Devinsky, Orrin, et al., 2016, and Gano & Grabenstatter, 2017 []
  3. Skluzacek JV, Watts KP, Parsy O, Wical B, Camfield P. Dravet syndrome and parent associations: The IDEA League experience with comorbid conditions, mortality, management, adaptation, and grief. Epilepsia. 2011;52(suppl 2):95–101. []
  4. Lhatoo, Samden, et al. 2015, and Metcalf, Cameron et al. 2018 []
  5. Metcalf, Cameron et al. 2018 []
  6. Cheah, et al, 2012 []
  7. Devinsky, Orrin, et al., 2016 []
  8. Lhatoo, Samden, et al., 2015 []
  9. Gano, L. B., & Grabenstatter, H. L, 2017 and Bozorgi, A., & Lhatoo, S. D. 2013 []
  10. Harden, C., et al. 2017 []
  11. Harden, C., et al. 2017 and Bozorgi, A., & Lhatoo, S. D. 2013 []
  12. Bozorgi, A., & Lhatoo, S. D. 2013 []
  13. Velagapudi, P. et al, 2012 []
  14. Volkers, L. et al, 2011 []
  15. Kobau, R. et al, 2008 []
  16. Volkers, L. et al, 2011 []
  17. Ceah, C. et al, 2012 []
  18. Sugawara, T. et al, 2001 []
  19. Ribak, C., 1987 and Oakley, J et al. 2009 []
  20. Parihar, R. and Ganesh, S., 2013 []
  21. Oakley, J. et al, 2009 []
  22. Oakley, J. et al, 2009 []
  23. Kim, Y., et al. 2018 []
  24. Metcalf, Cameron et al. 2018 and Kim, Y., et al. 2018 []
  25. Metcalf, C. S., et al. 2018 []
  26. Mondello, S., et al. 2012, Oakley, J. et al, 2009 []
  27. Matz, O., et al. 2016 []
  28. Fuller S. et al, 2012, Metcalf, C. S., et al. 2018, and D., Accorsi, et al. 2013 []
  29. MK, Jha, et al. 2016 []
  30. Brackenbury, W., et al. 2013 []
  31. Brackenbury, W., et al. 2013 []
  32. Yamakawa, K., 2001 []
  33. Yamakawa, K., 2001 []
  34. Thom, M., et al. 2016 []
  35. Thom, M., et al. 2016 []
  36. Metcalf, Cameron et al. 2018 and Kim, Y., et al. 2018 []
  37. Matz, O., et al. 2016 []
  38. Thom, M., et al. 2016 []
  39. Metcalf, Cameron et al. 2018 and Cheah, C. et al, 2012 []
  40. Volkers, L., et al, 2011 []
  41. Bittolo et al., 2016 []
  42. Lundgaard, I et. al, 2017 []
  43. Takahashi et al, 2016 []
  44. Takahashi et al, 2016 []
  45. Brackenbury, W., et al. 2013 []
  46. Nurminen, M., 1997 []

The Effect of Types of Organic Materials on the Production of Ethanol

Featured

Author: Allyson Wang
Peer Reviewer: Tanya Singh
Professional Reviewer: Levente Pap

Abstract

Currently, burning fossil fuels and their released greenhouse gases are contributing to global warming. Because fossil fuels are running out quickly, alternative energy sources such as biofuels are emerging because of their renewability and abundance. The purpose of this experiment was to observe the effects of different organic materials on the production of ethanol. It was hypothesized that if orange peels were fermented for eight days, then they would have the highest potential of producing ethanol. There was no control because there is no “typical” organic material. To start the experiment, the same amount of each organic material (orange peels, pistachio shells, cornstalk, and Switchgrass) was boiled for an hour. The four organic materials were each placed into twenty jars, and after fermenting for two, four, and eight days, refractometers are used to measure the refraction index/specific gravity. The results determined that orange peels had an average specific gravity of 1.011 on Day 8, higher than the values for pistachio shells, cornstalk, and Switchgrass: 1.002, 1.004, and 1.003 respectively. A t-test was performed which showed that all data was significant, rejecting the null hypotheses. The results did not support the research hypothesis (peels produced more ethanol on Day 4 than Day 8), but supported that orange peels overall produce more ethanol than other organic materials on Day 8. Because orange peels produced the most ethanol, it was found that the peels have a high content of cellulose which can be converted into ethanol. Further research can include longer boiling time and using a variety of other organic materials.

Introduction

Technology is advancing rapidly, and the world’s population is constantly growing. Worldwide, 78% of the energy comes from fossil fuels, 3% is from nuclear energy, and 19% is from renewable energy sources like biomass (Balan, 2014). The current energy source is dependent on fossil fuels like petroleum and coal which are nonrenewable and causing global warming with their greenhouse gases. Because of their high production of air pollution and countries’ increasing demand for oil, researchers and scientists around the world are looking for alternatives like renewable biomass fuels (Islam et al., 2015). 

Using biomass for fuels is a stable option because of its renewability, abundance, and benefits to society (Demiral et al., 2009). In addition, biofuels bring energy security to countries dependent on oil and increase job opportunities for fermentation specialists and scientists (Balan, 2014). These agricultural wastes can help the world by reducing greenhouse gases to slow global warming. It is predicted that by 2050, around 27% of the total transportation fuel consumption will be biofuels (Islam et al., 2015). Fuels are used to heat homes, for transportation, and other essential uses, making it a huge necessity (Demiral et al., 2009).

Organic material, also known as biomass, can be converted into energy sources and be used alternatively in industries. Biomass includes agricultural wastes like corn stalks and nutshells; wood materials such as sawdust and bark; aquatic plants and algae; municipal waste consisting of wastepaper and yard clippings; and energy crops like Switchgrass and willows. Biomass contains a complex pattern of molecules of mostly carbohydrates and lignin. The carbohydrates are primarily composed of cellulose, or hemicellulose fibers that strengthen the plant’s structure. The cellulose are homopolymers of glucose; this is what most living organisms use as a main source of sugar for biochemical energy.

Around 140 billion metric tons of biomass are produced every year. The combustion used to convert carbon dioxide into biomass is called the carbon cycle. Biomass also fixes carbon dioxide balance in the atmosphere using photosynthesis. Currently, 7% of energy comes from biomass each year. Biomass can potentially create 2.89 x 104 exajoules of energy, around eight times the world’s energy consumption from all sources (Tao, 2002). 

Orange peels, pistachio shells, cornstalk, and Switchgrass are being thrown away in landfills as food and agricultural wastes, creating unwanted greenhouse gas emissions. Orange peels from Florida alone can generate an estimated 200 million gallons of ethanol annually (Casey, 2010). Fifteen point six million tons of orange peels globally could be used to make other beneficial materials like fuels. According to the Orange Peel Exploitation Company (OPEC), the cellulose in the orange peels can be made into biofuels (Oranges.com, n.d.). Orange peels can also produce large amounts of citric acid, suggesting its high content of cellulose (Torrado et al., 2011).

Pistachio shells are one of the world’s favorite nuts and, annually, more than 65,000 tons of pistachios are produced. Turkey has begun to use pistachio shells as fuel. Few studies have shown that there is cellulose in the shells (Demiral et al., 2009). Additionally, corn stalks are agricultural wastes, and it has been hypothesized that they contain cellulose to produce ethanol (Steil, 2013). Moreover, energy crops, specifically Switchgrass, are labeled as cellulosic biomass and will greatly increase in production in the next 30 years (400 to 700 million dry tons). Research has revealed that Switchgrass produces more ethanol than corn (Biello, 2008).

Ethanol is inexpensive and easily produced by fermenting organic matter (Watts, 2012). Bio-ethanol is a biofuel used mainly as a substitute for gasoline as well as for transportation and heaters (Markov, 2012a). Edible foods produce first generation bio-ethanol which contains high sugar content and can be easily converted into fuel. Because first generation feedstock comes from edible foods; however, it could come into competition with food industries, land, and water. Second generation biofuels are based on non-edible lignocellulosic biomass like agricultural and municipal wastes. The residues contain high sugars of polysaccharides which can be processed into second-generation biofuels. Some major advantages of the non-edible feedstock include its low cost, high availability, and is noncompetitive with land, water, and food sources (Islam et al., 2015).

To obtain fuels, pretreatment is performed where the sugars are extracted from the materials. Before pretreatment, the lignin encloses and organizes the cellulose within the cell. After the material undergoes high temperature and pressures, the lignin in the cells ruptures and scatters the sugars inside (Liu & Fei, 2013). The biomass sources then convert into fuels under the processes of fermentation, pyrolysis, or chemical modification. Industrial fermentation is the most efficient and popular way biofuels are produced by sugars. The sugar is extracted by enzymes, and the yeast cells convert the sugar into ethanol and carbon dioxide (C6H2OH ? 2C2H5OH + 2CO2). The ethanol is then separated from fermentation using distillation (Markov, 2012b).

This experiment will determine if different durations of fermentation and organic materials will affect the amount of ethanol produced and their potential to convert into biofuels. The independent variables in this experiment are different organic materials: orange peels, pistachio shells, cornstalk, and Switchgrass; and different durations of fermentation of two days, four days, and eight days. There is no control because there is no typical organic material. The dependent variable is the production of ethanol which is determined by the specific gravity (no unit). Based on previous studies and research, it is held that if orange peels were fermented for eight days, then it would produce the most ethanol. In a previous experiment, it was shown that orange peels produced high amounts of citric acid from its high cellulose content. Because it contains lots of cellulose, there is more sugar to ferment which potentially means that peels could produce lots of ethanol (Torrado et al., 2011). 

EDD

Title: The Effects of Different Organic Materials on the Production of Ethanol

Hypothesis: If orange peels are fermented for 8 days, then they would have the highest potential of producing ethanol.

Durations of Fermentation: 2, 4, and 8 days

Independent Variable: Organic Materials
Orange PeelsCornstalkSwitchgrassPistachio Shells
20202020

Dependent Variable: Specific Gravity (no unit) (production of ethanol)

Constants: type of environment (lab temperature), type of water added to cooking pot, amount of grinded material (400 grams), same boiling time (1 hour), type of cooking pot, amount of pressure/boiling temperature, timer, type of jar, type of refractometer, type and amount of yeast, pipette brand

Methods and Materials

Four hundred grams of each organic material (orange peels, pistachio shells, cornstalk, and Switchgrass) were gathered and measured using a gram balance scale. The materials were then crushed, blended, and cut into small pieces. One of the organic materials was separated into one high pressure cooking pot filled with three liters of water. The material was boiled at high pressure for one hour. After an hour, the liquid in the pot was strained to separate from the solid matter. They were cooled down to room temperature (23 degrees Celsius), and measured using a thermometer. A measuring cup was then used to pour 200 milliliters of the boiled liquid in each of the twenty 568 mL (1-pint) jars. Next, 20 grams of the boiled material and five grams of yeast were distributed into each jar. The jars were then sealed with lids and set aside to be fermented at room temperature. The same procedure was repeated for each of the organic materials: orange peels, pistachio shells, cornstalk, and Switchgrass. On days 2, 4, and 8, the specific gravity was measured for all IV levels with a refractometer. A pipette was used to place three drops of the fermented solution on the refractometer slide. White paper towels were used to clean the refractometer. After the data was collected, the supervisor disposed the homemade alcohol by using a 500 mL beaker and cloth to separate the liquid and solid matter. The solid matter was placed in an autoclave bag for safety purposes while the liquid was poured down the drain. The pots and jars were rinsed thoroughly. Additionally, a fire extinguisher was kept in the room in case of a fire from boiling materials, and an apron and gloves were used for safety. A permission form was signed by a parent that indicated that they had read and understood the risks and possible dangers involved in the research and they consented to their child participating in this research.

Results


Table 1. Statistics of the Effect of Types of Organic Materials on the Production of Ethanol

Graph 1. The Effect of Types of Organic Materials on the Production of Ethanol

Graph 2. The Effect of Types of Organic Materials on the Production of Ethanol

Table 2. Raw Data of Orange Peels of the Effect of Types of Organic Materials on the Production of Ethanol
*Specific Gravity has no unit


Table 3. Raw Data of Pistachio Shells of the Effect of Types of Organic Materials on the Production of Ethanol
*Specific Gravity has no unit


Table 4. Raw Data of Cornstalk of the Effect of Types of Organic Materials on the Production of Ethanol
*Specific Gravity has no unit

Table 5. Raw Data of Switchgrass of the Effect of Types of Organic Materials on the Production of Ethanol
*Specific Gravity has no unit
C:\Users\hcps-wangah\Documents\Science Project\Pictures for Sci Proj\IMG_4378.jpg
Picture 1: Fermented Orange Peels on Refractometer Slide

The effects of different organic materials on the production of ethanol was observed, and the results are shown in Tables 1, 2, 3, 4, 5 and Graph 1 and 2. It was hypothesized that if the orange peels were fermented for eight days, then they would have the greatest potential of producing ethanol. The means were calculated for each of the independent variables. The central tendencies between orange peels (1.011) and pistachio shells (1.002) shows that orange peels have a higher density than pistachio shells, meaning that the peels contain more cellulose that can be made into ethanol and the shells contain the least. The means for pistachio shells (1.002), cornstalk (1.004), and Switchgrass (1.003) were low and remained consistent at all three intervals, which supports that the three materials have very low density, and do not contain as much cellulose compared to orange peels. This also indicates that if the materials were to be fermented for longer than eight days, the production of ethanol would remain consistent. If orange peels were fermented for a longer duration, the data may vary according to the results which can be tested in additional experiments.

From these results, the data supported the research hypothesis because orange peels had the most amount of cellulose which leads to their high production of ethanol. However, orange peels produced more cellulose on Day 4 compared to Day 8. The variance and standard deviation were determined for each independent variable. Overall, the standard deviations were very low, meaning that the raw data collected for all levels were very clustered and precise. There were outliers for pistachio shells (1.004) since it was outside the SD 2 range (1.0006-1.0034), and there was an outlier for Switchgrass (1.004) that was outside the SD 2 range (1.0024-1.0036). These points could have been caused by error or other factors. Specify these factors.  

A t-test was done at a level of significance of 0.05 with the degrees of freedom of 38. The null hypotheses were that there would be no difference between each of the organic materials (orange peels, pistachio shells, cornstalk, and Switchgrass). All calculated t-values (32.863; 28.577; 34.268; 10.691; 5.822; and 8.305) were greater than the table t-value of 2.024. This implies that the null hypotheses were rejected, and there are significant differences between each organic material. The data collected for this experiment was most likely affected by the independent variables and not by chance. The probability of the results happening by chance is less than 0.05 based on the level of significance. The data for organic materials having an effect on how much ethanol they produce is significant.

Discussion and Conclusions

This experiment was to determine the effects of different organic materials (orange peels, pistachio shells, cornstalk, and Switchgrass) on the production of ethanol. It was hypothesized that if orange peels were fermented for eight days, then they would have the highest potential of producing ethanol. It was found that orange peels had the highest potential of producing ethanol at a specific gravity of 1.012 on Day 4 out of all organic materials. Based on the results, the research hypothesis was not supported, but the data did show that the orange peels produced the most ethanol (of 1.011) overall compared to other materials. A t-test was done for this experiment to observe if the data collected was significant. The data for all the organic materials was statistically significant which shows that the data was due to the independent variable, implying that different types of organic materials have an effect on the production of ethanol.

Other studies and research have shown that orange peels produce a large amount of ethanol and cellulose. According to an article, orange peels can produce almost 200 million gallons of ethanol every year (Casey, 2010). This reveals its high cellulose content which is also supported by a study on citric acid (Torrado et al., 2011). Another study shows that the OPEC have already started to convert orange peels into usable biofuel (Oranges.com, n.d.). A research team in Sweden has successfully produced ethanol from different materials and is diverting their attention towards orange peels (Meade, 2009). As a non-edible feedstock that produces a huge amount of ethanol, it provides many benefits to our society such as transportation fuel and heat for homes (Islam et al., 2015; Demiral et al., 2009).

Florida is taking a huge step towards using orange peels for ethanol. A professor at the University of Central Florida has been working on converting the peels for secondary fuel. Other scientists believe their findings and experiment may turn out satisfactory (Kotala, 2010). In addition, an energy ethanol plant in Florida is planning to convert peels into ethanol that can be sold to Floridians at gas stations (Meade, 2009). Meanwhile, another department in Florida who partnered with other companies are investigating the peels’ potential for ethanol production. The industry has been researching citric energy and performing basic steps in the laboratory (Dunford, 2008).

Overall, there was no significant difference between the data for different durations of fermentation. However, for orange peels, Day 4 had an increase of 0.001 from Day 2, implying that orange peels may produce more ethanol in shorter durations of fermentation. This data also may have been a misreading (specifically because of the minimal increase) since all other materials had a constant value for the three days. For pistachio shells, cornstalk, and Switchgrass, there were no differences between the durations of fermentation in specific gravity which explains that different durations of fermentation do not greatly affect the production of ethanol. 

Based on previous research, orange peels can be converted into biofuel under high temperature. The more cellulose there initially is, the more biofuel is created (Tao, 2002). This research supports Torrado’s study, stating that orange peels have a high cellulose content to produce a high amount of ethanol (Torrado et al., 2011). Studies still support that materials like pistachio shells, cornstalk, and Switchgrass are capable of producing ethanol, but it would take lots of heat and material in order to extract a large amount of ethanol (Tao, 2002). All the results were significant because the amount of ethanol for each organic material had a significant difference which rejects the null hypotheses. The independent variables affected the results which were statistically apparent.

In the experiment, there were a few sources of error. Longer boiling time and a larger amount of materials could be used in order to receive better and more accurate results.                                                    Additionally, more specific measurements of the measuring cup, refractometer, and thermometer could have been used in order for more precise data. When taking measurements for each ferment day, some gas may have leaked into the air which could have prevented the production of ethanol on Days 4 and 8 because of the introduction of oxygen. Also, the refractometer itself may have been the source of error. The standard instruments GC-FID or GC-MS monitor the fermentation and detect a more exact ethanol concentration. However, because the instrument was not available, a refractometer was used instead to track the ethanol concentration. For further experimentation, other organic materials could be tested to observe if they contain cellulose to produce ethanol. An additional experiment could be performed on orange peels specifically with different factors such as amount of peels, boiling time, and amount of water in order to expand the knowledge on this topic.

Literature Cited

Peer-Reviewed

Balan, V. (2014, May 5). Current Challenges in Commercially Producing Biofuels from Lignocellulosic Biomass. ISRN Biotechnology, 2014, 1-31. Retrieved October 30, 2017, from doi: 10.1155/2014/463074

Demiral, I., Atilgan, N., and Sensoz, S. (2009, October 20). Production of Biofuel from Soft Shell of Pistachio. Chemical Engineering Communications, 196 (1-2), 104-115. Retrieved on October 5, 2017, from doi: 10.1080/00986440802300984

Islam, Z., Zhisheng, Y., Hassan el B., Dongdong, C., and Hongxun, Z. (October 3, 2015). Microbial conversion of pyrolytic products to biofuels: a novel and sustainable approach toward second-generation biofuels. Journal of Industrial Microbiology & Biotechnology, 42(12), 1557-1579. Retrieved November 4, 2017, from doi:10.1007/s10295-015-1687-5

Liu, Z. and Fei, B. (2013, May 15). Characteristics of Moso Bamboo with Chemical Pretreatment. InTech, 1, 3-14. Retrieved November 10, 2017, from doi:10.5772/55379

Markov, S. (2012a). Biofuels and Synthetic Fuels. Applied Science (1, 199-203). Hackensack: Salem Press.

Markov, S. (2012b). Industrial Fermentation. Applied Science (3, 1041). Hackensack: Salem Press.

Torrado, A. M, Cortés, S., Salgado, J.M., Max, B., Rodríguez, N., Bibbins, B., Converti, A., and Domínguez, J.M. (2011, November 3). Citric Acid Production from Orange Peel Wastes by Solid-State Fermentation. Brazilian Journal of Microbiology, 42, 394-409. Retrieved October 22, 2017, from doi: 10.1590/S1517-83822011000100049

Watts, C. (2012). Bioenergy Technologies. Applied Science (1, 185). Hackensack: Salem Press.

Non-Peer Reviewed

Biello, David. (2008, January 8). Grass Makes Better Ethanol than Corn Does. Retrieved November 17, 2017, from https://www.scientificamerican.com/article/grass-makes- better-ethanol-than-corn/

Casey, Tina. (2010, February 25). A Sustainable Recipe for Biofuel: Ethanol from Orange Peels and Tobacco. Retrieved September 30, 2017, from https://cleantechnica.com/2010/02/25/a-sustainble-recipe-for-biofuel-ethanol-from- orange-peels-and-tobacco/

Dunford, Nurhan. (2008). Converting Orange Peel to Ethanol. Retrieved December 31, 2017, from http://fapc.biz/files/converting_orangepeel_toethanol.pdf/

Kotala, Zenaida. (2010, February 18). Orange Peels, Newspapers may lead to cheaper, cleaner ethanol fuel. Retrieved December 31, 2017, from http://www.research.ucf.edu/News/2010/CleanerEthanolFuel.html

Meade, Jenna. (2009, December 6). Four ways to turn an orange peel green. Retrieved December 31, 2017, from https://newatlas.com/orange-peel-ethanol-biogas- recycling/13521/

Oranges.com. Orange Biofuel. n.d. Retrieved September 30, 2017, from http://oranges.com/orange-biofuel

Steil, Mark. (2013, December 17). New ethanol plants to make fuel from 'biomass'. Retrieved November 17, 2017, from https://www.mprnews.org/story/2013/12/16/new-plants-to- make-ethanol-from-biomass

Tao, B. (2002). Biomass. McGraw-Hill Encyclopedia of Science & Technology (3, 69-70). New York: McGraw Hill.


Synthesis of a Biodegradable Bioplastic Alternative to Polypropylene Utilizing a Blend of D-Mannose and Acemannan from Aloe vera

Featured

Authors: Brian Lu and Haripriya Dukkipati
Peer Reviewer: Ashley Yoon
Professional Reviewer: Sidra tul Muntaha
Saint Francis High School 

Abstract

The vast majority of commercial plastic that a general consumer encounters on store shelves, uses on a daily basis, and casually tosses into garbage bins is petroleum-based [1]. However, these materials degrade with much difficulty and originate from less-than-eco-friendly manufacturing processes; as a result, harmful plastic waste accumulates with negative ramifications for living beings in every conceivable ecosystem [3, 4]. In order to address the conundrum of plastic waste buildup, the authors of this study formulated a biodegradable bioplastic using D-mannose and powderized acemannan from dehydrated Aloe vera. This new material was created through acid-catalyzed hydrolysis and dehydration synthesis reactions in the presence of a glycerol plasticizer and a hydrochloric acid catalyst. A Decomposition Test projected that the D-mannose and acemannan bioplastic would fully biodegrade into its smaller molecular components within an average of 1.6 months, as opposed to petroleum-based plastics which would not visibly degrade at all within that timeframe [5]. A subsequent Force Test which simulated wear and tear through routine bending corroborated the hypothesis that the bioplastic at least rivalled commercial polypropylene in durability. Finally, a Water Resistance Test indicated that the new material was significantly more efficient at hydro-degradation than polypropylene. Thus, this study concludes that the D-mannose and acemannan bioplastic is a viable replacement for petroleum-based plastics, with likely applications in a diverse array of products from bottling and packaging to medical instruments.

Introduction

Plastics

Plastics are a category of man-made polymers now nearly ubiquitous among consumer products [1]. From wiring and insulation to dishware and packaging, plastics differ widely in their chemical composition but share a few core properties: their flexibility, non-hazardous nature, chemical inertness, and cost-effectiveness [1]. However, most plastic items obtainable in stores today are made of polypropylene, polyethylene, or other fossil-fuel derivatives and are therefore largely non-biodegradable [1].

Certain chemicals can be mixed into the plastic to tweak its properties toward a desired ideal. One important category of additive is the plasticizer, a substance that increases the flexibility of the plastic by acting analogously to a molecular lubricant and keeping individual polymer chains from entangling or locking together [2]. In this experiment, glycerol functioned as the plasticizer. 

Plastic Waste

Petroleum-based plastics tend to cause twofold environmental harm because of their environmentally unfriendly sourcing and their slow degradation. The extraction of plastic precursors from raw petroleum requires harsh chemicals, and the industrial processes and reactions that turn them into mass-produced plastics often deposit harmful pollutants into the atmosphere and the ocean [3]. Secondly, the nonpolar structure of the hydrocarbon chains that compose most plastics hinders their decomposition in the natural environment. These hydrocarbon chains are insoluble in common polar solutions such as water or the acids in the stomachs of animals, and all the C-H bonds in petroleum-based plastics have the same bond strength, so there are no weaker links for enzymatic or chemical reactions to easily break [4]. Due to these difficulties in degradation, petroleum plastics tend to accumulate as pollutants regardless of their method of disposal, thus leaching into drinking water systems, killing wildlife, and cluttering terrestrial and aquatic ecosystems [5].


Figure 1: Chemical Structure of an Example of a Petroleum-Based Plastic: Polypropylene [6].

Bioplastics and Biodegradability

There are two key improvements to conventional plastic that this study is concerned with: bioplastics and biodegradable plastics. Each of these more eco-friendly types of plastic attempts to solve one of the dual problems of harmful sourcing and slow degradation as explained above. All bioplastics are derived from organic materials, thus allowing for more eco-friendly synthesis, but they do not necessarily have to be biodegradable [7]. For example, corn-based polyethylene is organically sourced, but decomposes extremely slowly because it is chemically identical to existing petroleum-based polyethylene [8]. On the other hand, biodegradable plastics must wholly decompose within 180 days into smaller non-harmful molecules such as water or usable nutrients [9, 10]. Unfortunately, there also exist potential problems with biodegradable plastics—because they are optimized to degrade quickly in natural conditions, toxic or otherwise harmful chemicals might have been used during their production to prolong their shelf-life or to diminish costs [10]. Thus, this study attempts to solve both problems by combining the properties of the two categories of plastics and creating a biodegradable bioplastic.

D-mannose and Acemannan


Figure 2: Chemical Structure of D-mannose [11, 12].


D-mannose is a monosaccharide that is often sold in the form of a white granular powder for use as an over-the-counter diet additive or supplement [12]. It is present in certain types of algae and seaweed, fruits such as apples, and common household plants [12, 13]. According to Zhang, Pan, Quian, and Chen, researchers at the Zhejiang University in China, its extraction can be  cost-effective [13]. D-mannose represents the largest component of this study’s biodegradable bioplastic as measured by percent mass.


Figure 3: Chemical Structure of Acetylated Mannose [14, 15].

Acemannan is a polysaccharide and a polymer of the monosaccharide acetylated mannose, which is itself a version of D-mannose with extra CH3 groups added to its chemical structure [14, 15]. These CH3 groups act as nonpolar regions surrounding an otherwise polar molecule, decreasing acemannan’s interactions with polar molecules such as water [14, 15]. In this study, small amounts of acemannan in the form of dried Aloe vera powder were incorporated into the majority D-mannose bioplastic to increase its water resistance. Aloe vera was selected for this experiment because according to Pinghuai Liu, Deli Chen, and Jie Shi, researchers at Hainan University in China, it contains one of the highest concentrations of acemannan and very little of other unwanted solid compounds, which contributes to ease of extraction [16]. 


Figure 4: Chemical Structure of Acemannan [14, 15].

Table 1: Chemical Composition of Aloe vera Gel [16].

While the extraction of D-mannose and acemannan have been established to be comparatively hassle-free and cost-effective, the D-mannose and acemannan used in this experiment were purchased in powder form because of lack of access to the necessary laboratory equipment for extraction.

Acid Catalyzed Hydrolysis

Hydrolysis refers to the decomposition reaction in which a larger molecule is cleaved into two smaller molecules with the addition of water [17]. When this hydrolysis is catalyzed by a dissociated hydrogen ion from a protic acid, it is called acid catalyzed hydrolysis [18]. For this experiment, acid catalyzed hydrolysis is utilized when the polysaccharide acemannan is broken down into units of acetylated mannose in preparation for polymerization and plasticization. Hydrochloric acid was chosen to catalyze the hydrolysis of acemannan because it always completely dissociates into a hydrogen ion (H+) and a chloride ion (Cl-), thus producing a consistent supply of H+ ions for the acid catalyzed hydrolysis of acemannan [18].


Figure 5: A Proposed Mechanism of the Acid Catalyzed Hydrolysis of Acemannan  [14, 15].

Dehydration Synthesis

Dehydration synthesis—the reverse reaction of acid catalyzed hydrolysis—is a subtype of polymerization that links smaller molecules together in a chain-like structure to form a larger molecule and results in the generation of one water molecule per linkage formed [17]. During the synthesis of the biodegradable bioplastic, dehydration synthesis occurred when the blend of D-mannose and acetylated mannose was polymerized into plastic in the presence of the glycerol plasticizer. In this case, the water generated from the polymerization reaction did not dilute the components of the plastic because it completely evaporated into the surrounding air, as a result of the high temperatures at which the dehydration synthesis was performed.


Figure 6: Polymerization of D-mannose and Acetylated Mannose During the Synthesis of Biodegradable Bioplastic [11, 14].

Objective

The objective of this research project is to synthesize a biodegradable bioplastic from D-mannose and acemannan, to compare its strength against that of polypropylene, and to confirm its viability for everyday use.

Materials for Synthesis of Bioplastic

  • D-mannose
  • Glycerol
  • Aloe vera powder
  • 0.5 M Hydrochloric acid (HCl)
  • 0.5 M Sodium Hydroxide (NaOH)
  • 18 Plastic Petri dishes (6 cm diameter x 1 cm height)
  • Aluminum foil
  • Gram balance
  • Hot plate
  • 2 100 mL Graduated cylinders
  • 1 500 mL Glass beaker
  • 2 Plastic weigh boats
  • Thermometer
  • Magnetic stir bar
  • Forceps
  • Acetone (for cleaning glassware)
  • Permanent marker

Materials for Preparation of Polypropylene Control

  • Polypropylene bottle caps
  • 1 500 mL glass beaker
  • Fume hood
  • 3 glass Petri dishes (6 cm diameter x 1 cm height)
  • 12 Plastic Petri dishes (6 cm diameter x 1 cm height)
  • Gram balance
  • Hot plate
  • Thermometer
  • Forceps
  • Acetone (for cleaning glassware)
  • Permanent marker

Materials for Decomposition Test

  • 6 Gardening pots (15 cm diameter x 16 cm height)
  • Potting soil
  • Ruler
  • Camera

Materials for Force Test

  • String (0.3 cm thick)
  • Bench vise
  • Flat surface
  • Plastic bucket
  • Assorted weights
  • Gram scale
  • Ruler

Materials for Water Resistance Test

  • Stopwatch
  • 6 Plastic cups (8 cm diameter x 12 cm height)
  • Distilled water
  • Gram scale

Methods

Synthesis of Bioplastic

Each of the 18 plastic Petri dishes was labeled from P1 to P18 with the permanent marker. The interior of each Petri dish was lined with aluminum foil to prevent any molten bioplastic from fusing with or deforming the plastic Petri dishes. The biodegradable bioplastic was synthesized in six batches of three samples each to reduce the variability between individual samples. Only three foil-lined Petri dishes were used for each batch of plastic; the remaining 15 were set aside temporarily.

To begin the synthesis of one batch of plastic (three samples), the gram balance was used to measure out 33.01 grams of D-mannose in a plastic weigh boat, 0.83 grams of Aloe vera gel powder in a second plastic weigh boat, 0.70 grams of 0.5 M HCl in a graduated cylinder, 0.70 grams of 0.5 M NaOH in a second graduated cylinder, and 4.30 grams of glycerol in a 500-mL glass beaker. The beaker of glycerol was set on the hot plate, and the magnetic stir bar and thermometer were placed into it.

Next, the hot plate was turned on to a heat setting of 5/10 and a stirring speed of 5/10 (the hot plate was capable of reaching a temperature of 210 degrees Celsius at its maximum heat setting). The D-mannose was gradually mixed into the glycerol and heated until the mixture resembled a thick clear-white paste. The HCl and Aloe vera powder were then added, immediately resulting in bubbling. In this step, the HCl catalyzed the hydrolysis of the acemannan within the Aloe vera powder into acetylated mannose [18]. Simultaneously, the heat energy provided by the hot plate initiated the dehydration synthesis polymerization of the D-mannose and the acetylated mannose [17]. In the presence of the plasticizer glycerol, the long polymer chains of D-mannose and acemannan became a plastic [2]. This mixture was continuously heated and stirred until it reached a temperature of 105 degrees Celsius. When the mixture began to froth vigorously, the NaOH was added to neutralize the still present H+ ions and prevent the creation of an acidic bioplastic, causing a thinning of the paste and more frothing.

Once the bubbles finally subsided, demonstrating that all excess water produced by dehydration synthesis polymerization and acid-base neutralization had been evaporated from the molten bioplastic mixture, the glass beaker was removed from the hot plate with heat-resistant gloves and forceps. The magnetic stir bar was retrieved from the beaker with the forceps. The contents of the beaker were swirled manually with the forceps for 30 seconds to cool, and distributed as equally as possible among the three foil-lined Petri dishes. The molten bioplastic was then moved to a cool, dry location to set for several days. Once the bioplastic samples solidified, the foil was peeled away and discarded.

The glassware, the magnetic stir bar, the thermometer, and the forceps were all cleaned with acetone and distilled water to remove all bioplastic residues and then dried. The above procedure was then repeated to produce additional batches of three bioplastic samples at a time until a total of 18 samples of bioplastic were synthesized.

Preparation of Polypropylene Control

Before the polypropylene was melted and remolded into cylindrical samples that were the same size and shape as the bioplastic samples, the hot plate and three glass Petri dishes were placed in a well-ventilated fume hood because polypropylene produces noxious fumes when heated [1]. Each of 12 plastic Petri dishes was labeled from C1 to C12 with the permanent marker and set aside. 

The polypropylene was prepared in four batches of three pieces. 26.76 grams of polypropylene bottle caps were measured out for each batch of three; caps were broken by hand to approximate this amount as closely as possible. This required amount of polypropylene was determined through a stoichiometric calculation, solving for the mass of polypropylene given its density and the desired volume of synthesized bioplastic (so that the bioplastic samples and polypropylene controls could be identical in diameter and thickness). The appropriate amount of polypropylene was placed into a 500-mL glass beaker, along with a thermometer. The beaker was placed on a hot plate, and the fume hood was powered on to ensure proper ventilation of the harmful fumes that would be produced by molten polypropylene. The hot plate was then adjusted to a heat setting of 9/10. No stirring was necessary. The polypropylene was heated at approximately 200 degrees Celsius for 25 minutes until it fully melted.


Figure 7: Setup of Fume Hood, Including Hot Plate, Beaker, and Thermometer.

Once the polypropylene completely liquified, the beaker was swirled manually with the forceps for 30 seconds to cool its contents. The molten polypropylene was distributed as equally as possible among the three glass Petri dishes and was allowed to set for several days. When each polypropylene sample cooled and solidified, it was transferred into the corresponding plastic Petri dish. A knife or other sharp tool was necessary to pry the polypropylene samples out of the glass Petri dishes.

         The glass Petri dishes, glass beaker, thermometer, and forceps were all cleaned with acetone and distilled water to remove polypropylene residues, then dried. The above procedure was then repeated to produce additional batches of three polypropylene samples until a total of 12 samples of polypropylene were prepared.

Random Assignment

Out of the 18 bioplastic samples and 12 polypropylene samples, three samples of bioplastic and three samples of polypropylene were randomly assigned using a random number generator to each of the three tests: Decomposition Test, Force Test, and Water Resistance Test. This random assignment minimizes the skewing effects of any hidden or confounding variables. Moreover, any samples with visible cracks or other major physical flaws were excluded from testing to prevent incorrect conclusions due to variations in structural integrity, exposed surface area, etc. In total, nine of the 18 bioplastic samples and nine of the 12 polypropylene samples were chosen for the tests; the unused samples were set aside and not tested.

The random assignments used in this study are depicted below:



Table 2: Random Assignment of D-Mannose Bioplastic Samples and Polypropylene Samples to the Decomposition Test, Force Test, and Water Resistance Test.

Decomposition Test

Three samples of bioplastic (P8, P16, and P18) and three samples of polypropylene (C3, C4, and C11) were selected, as assigned in Table 2 above. Six gardening pots, each with a 15 cm diameter and 16 cm height, were filled halfway (eight centimeters high) with potting soil. Each pot was labelled according to the sample to be contained inside. Each sample’s initial diameter and thickness were measured and recorded. A camera was used to take pictures of each sample in order to record its shape prior to decomposition. To start the Decomposition Test, each sample was buried in its corresponding pot about two centimeters below the soil’s surface; it was ensured that none of the samples were directly visible. The pots were then placed in a row in a room-temperature room.

         Each sample was dug up and inspected daily. Any excess dirt clinging to the sample was lightly brushed off with a gloved finger as much as possible. The diameter and thickness of each plastic was measured with a ruler to record the day-to-day progression of the decomposition. Then, each sample was re-buried in the correct pot of soil in its original position (two cm below the soil’s surface). This procedure was repeated each day for a total of 35 days.

Force Test

The diameter and thickness of each plastic sample was first measured with a ruler to allow for later calculation of the force-bearing cross-sectional area. A bench vise was set up on the edge of a flat surface such as a table. Three samples of bioplastic (P10, P11, and P15) and three samples of polypropylene (C1, C5, and C10) were selected for the Force Test, as assigned in Table 2 above. The gap in the bench vise was adjusted to a width of two centimeters and the plastic sample was placed across it. Each end of a 100 cm length of string was then tied to a plastic bucket. The string was hung across the middle of the cylindrical sample so that the vertical cross-section containing the diameter of the sample supported the entire weight of the string and bucket. In Figure 8 below, the Petri dish models where a sample was placed during a real trial of the Force Test.


Figure 8: Setup of Force Test.

Assorted weights were then placed into the bucket, applying stress in a plane perpendicular to the circular base of each sample. The samples were supported on opposite ends by the edges of the bench vise and a force was applied across the sample’s midsection, simulating the forces experienced by plastic when bent through everyday use. More and more weights were added until the sample snapped into two pieces. When it did, the total mass of the string, bucket, and weights required to break the sample was measured with the gram balance and recorded. The above procedure was repeated until all three chosen samples of bioplastic and polypropylene had been tested. Then, the average mass supported per square centimeter of cross-sectional area was computed for each plastic type and compared.

Water Resistance Test

         Three samples of bioplastic (P5, P6, and P14) and three samples of polypropylene (C6, C7, and C9) were selected for the Water Resistance Test, as assigned in Table 2 above. Each sample’s initial mass was measured with a gram balance and recorded. Six plastic cups (all with an eight centimeter diameter and 12 cm height) were prepared by filling each with distilled water until the water was two centimeters deep. Each cup was labelled according to the sample to be contained inside. Next, the samples of bioplastic and polypropylene were submerged in their respective cups. A stopwatch was immediately started.

After five minutes had elapsed, each plastic sample was retrieved from the water, excess water was removed by gently dabbing the sample with an absorbent paper towel, and the sample’s mass was measured again with the gram balance and recorded. Then the sample was re-submerged. The above procedure was repeated again 15, 25, 35, 60, 120, 180, 240, 360, and 480 minutes after the six samples were initially submerged.  These specific times were chosen in order to develop a more accurate model of water’s effects on the bioplastic and the polypropylene samples over both the short-term and the long-term.

Results

Measurements of Plastic Samples

After all the bioplastic samples had set (with the aluminum foil peeled off) and all the polypropylene samples had solidified, the mass of each sample was measured with a gram balance and recorded to provide a baseline for comparison with the results of the Decomposition Test, Force Test, and Water Resistance Test. A picture of each sample was also taken with a camera to record initial observations. All of the synthesized bioplastic samples were noted to be translucent, hazel in color, and malleable enough to be bent or shaped without shattering; additionally, they clung to the aluminum foil even after hardening. In contrast, the polypropylene samples were opaque, black (perhaps due to a dye already present in the bottle caps that were melted down), rigid, and smooth.


Figure 9: The 12 Control Polypropylene Samples.


Figure 10: The 18 D-Mannose and Acemannan Bioplastic Samples.

Table 3: Initial Mass Measurements Obtained After the Synthesis of Bioplastic and Preparation of Polypropylene Control.

Results of Decomposition Test – Polypropylene Control


Figure 11: Polypropylene Diameter Over 35 Days of Decomposition.

Figure 12: Polypropylene Thickness Over 35 Days of Decomposition.

Figure 13: Polypropylene Volume Over 35 Days of Decomposition.

This Decomposition Test was intended to measure and compare the decomposition of the samples of controlled polypropylene and the samples of synthesized bioplastic. The controlled polypropylene samples (C3, C4, and C11) did not significantly decompose to any degree over the 35-day testing period of the Decomposition Test. The measured diameter of sample C4 actually decreased on day 17 as shown in Figure 11. However, it was deduced that this reduction in diameter was not due to decomposition of the polypropylene, but rather cracking and chipping that was initially present in the sample and which worsened with frequent handling. The diameter, thickness, and volume of each sample remained approximately constant throughout the entire test as displayed in Figures 11, 12, and 13.

Using Microsoft Excel, the degree-5 polynomial of best fit was generated and plotted for each controlled polypropylene sample in the graph of polypropylene volume (Figure 13). These polynomials can be extrapolated to predict that C3 would require 428505 days (1173.98 years), C4 would require 198975 days (545.136 years), and C11 would require 399750 days (1095.20 years) to completely decompose and reach a volume of zero. These projections that were generated from the collected data and the calculated volume of each cylindrical sample serve to corroborate the known fact that petroleum-based plastics like polypropylene are non-biodegradable and often require exorbitant lengths of time to fully decompose [5].

Results of Decomposition Test – Bioplastic


Figure 14: Bioplastic Diameter Over 35 Days of Decomposition.


Figure 15: Bioplastic Thickness Over 35 Days of Decomposition.


Figure 16: Bioplastic Volume Over 35 Days of Decomposition.

In contrast to the polypropylene control, the diameter and thickness of each of the three bioplastic samples (P8, P16, and P18) decreased steadily until Day 10, then gradually levelled off after around Day 20. Some exceptions to this trend existed; for example, the measured thickness of sample P18 actually increased on the eighth day due to an unusually thick coating of dirt which was difficult to fully remove without damaging the sample itself. The diameter of the samples decreased at an average rate of 0.075 cm/day (calculated from Figure 14) and their thickness decreased at an average rate of 0.002 cm/day (calculated from Figure 15). As a result, the samples’ calculated volumes decreased at an average rate of 0.060 cm3/day (calculated from Figure 16). Within ten days, the tested bioplastic samples had decomposed to less than one-half of their initial volume as calculated from the initial measurements of diameter and thickness. By the conclusion of the Decomposition Test, the bioplastic samples had each decomposed to less than one-eighth of their original volume.

Using Microsoft Excel, the degree-5 polynomial of best fit was also generated and plotted for the volume graph of each bioplastic sample in Figure 16 above. While none of the bioplastic samples completely decomposed during the 35 days of the Decomposition Test, the extrapolations generated in Microsoft Excel predicted that P8 would completely decompose within 43.56 days, P16 would completely decompose within 57.92 days, and P18 would completely decompose within 48.16 days. Because these three bioplastic samples are projected to fully degrade in an average of approximately 50 days (well within the time limit of 180 days), the D-mannose bioplastic synthesized in this study qualifies as biodegradable [9, 10].

Results of Force Test


Table 4: Raw Data of Force Test - Polypropylene Samples.

Table 5: Raw Data of Force Test - Bioplastic Samples.

This Force Test simulated the forces arising from bending through everyday use and compared polypropylene and the synthesized bioplastic based on the amount of stress that each could sustain before snapping in two. While the thickness of each polypropylene sample correlated strongly with the amount of force it could withstand, all of the bioplastic samples in this test tolerated roughly the same amount of force. After controlling for the differently sized cross-sectional areas (diameter x thickness), it was found that the bioplastic samples supported an average of 14.20 Newtons of force per square centimeter of force-bearing area more than the polypropylene control did. 

In other words, if a bioplastic sample and a polypropylene sample of exactly the same diameter and thickness were simultaneously bent, data suggests that the polypropylene would snap first. These results demonstrate that the synthesized bioplastic is capable of withstanding a quantity of stress at least comparable to the amount that polypropylene can handle. This bioplastic therefore has the potential to be an effective polypropylene substitute and could be used for commercial purposes such as creating durable tubing, rods, and bottlecaps.

Results of Water Resistance Test


Table 6: Raw Data of Water Resistance Test.

Figure 17: Comparing Hydro-degradation of D-mannose Bioplastic and Polypropylene.

This test compared the water resistance of the synthesized bioplastic and the polypropylene control, which is necessary because plastics commonly come into contact with or are used as containers for water and other liquids. The bioplastic samples used in this test (P5, P6, and P14) experienced a precipitous decrease in mass during the first 15 minutes of submersion in distilled water, but then settled at slightly less than half of their initial mass. The polypropylene samples were not significantly affected by the water at all. 

This large difference in hydro-degradability can be attributed to differences in the chemical structures of the monomers that make up the bioplastic and the polypropylene. The D-mannose subunits are polar and hydrophilic (the acetylated mannose is less so as explained in the Introduction, but still hydrophilic), while the propylene subunits are entirely non-polar and hydrophobic [6, 11, 14, 15]. Moreover, as sugars, all forms of mannose are highly soluble in water; when the synthesized bioplastic is initially immersed in an aqueous environment, it is likely to decrease in mass as the D-mannose and acetylated mannose rapidly enter solution [11, 12]. However, this dissolution reaction reaches equilibrium after some time, so the rate of reduction in mass decreases as the solution nears saturation. On the other hand, polypropylene is much less soluble in water than mannose, so its solid mass at equilibrium would be much larger than the solid mass of the synthesized bioplastic at equilibrium (in other words, much less of it dissolves) [6].

Discussion

Conventional petroleum-based plastics such as polypropylene require almost a millennium to decompose, precipitating the accumulation of harmful waste products and contaminants across the planet and its oceans [5]. The increasingly widespread reliance on disposable plastics in an assortment of fields, such as industrial manufacturing and medicine, intensifies humanity’s need for more biodegradable alternatives [1].

The objective of this study was to develop a biodegradable bioplastic alternative to petroleum plastics and assess its viability for common uses. In order to accomplish this, a blend of D-mannose and water-resistant acemannan from Aloe vera powder was plasticized in the presence of glycerol, a non-toxic and all-natural food preservative and sweetener. The results of the Decomposition Test and Force Test demonstrated that the synthesized bioplastic would decompose within an average of approximately 50 days and was capable of withstanding at least as much stress from bending as commercial polypropylene. Moreover, the water resistance test indicated that the bioplastic was also more hydro-degradable than polypropylene. Because the synthesized bioplastic is at least as durable as polypropylene, yet is both biodegradable and hydro-degradable, it can be considered a practical alternative to commercial plastics that is viable for everyday use.

Despite humanity’s best intentions and increased efforts to be more “green” in recent years, plastics still commonly pollute oceans and other ecosystems to this day [5]. Even in the worst case scenario where the bioplastic developed in this study becomes pollution or waste, it will still have a less negative impact on the environment than polypropylene would. If buried underground in a landfill, as simulated by the Decomposition Test, the bioplastic would likely decompose in a mere 50 days (as opposed to centuries for polypropylene) and return to its non-toxic components of D-mannose and acemannan, thus posing much less risk to terrestrial ecosystems and organisms [5, 9, 10]. If disposed in an aqueous environment, the synthesized bioplastic would also decompose rapidly into its non-toxic constituents due to its high hydro-degradability, causing minimal harm to marine life [10, 19]. On the other hand, petroleum-based plastic does not easily degrade within any part of the natural environment, so it has a higher probability of negatively impacting wildlife [1, 19]. For example, animals in terrestrial and aquatic ecosystems often confuse plastic for nutritional sustenance and ingest bits of it, mistakenly thinking it is a food source [19]. Worse, if these harmful materials are slow to degrade like polypropylene, they can build up in the digestive tracts of organisms and get in the way of nutrient absorption, causing drawn-out and painful deaths [19].

Limitations to this study include a lack of access to equipment for the Force Test. Tensile testing, a standard procedure within materials science that analyzes the levels of tension a material could withstand before structural failure, could not be performed without expensive industry instruments. An alternative setup for the Force Test had to be utilized, instead operatively defining strength as the quantity of stress withstood. In addition, there was also some inevitable lab error. During the Decomposition Test and Water Resistance Test, human error was compounded by the impossibility of completely removing all excess soil or water from the samples, thus resulting in incorrectly increased measurements of the dimensions and the mass for all samples involved in these tests. When pouring the molten bioplastic into the Petri dish molds, some of the liquid mixture inevitably clung to the side of the beaker due to the molten bioplastic’s viscosity. This phenomenon resulted in samples that were thinner than expected and thus might have caused the bioplastic samples to snap more easily or dissolve more easily in water due to the larger surface area to volume ratio.

Based on the results of this study, the hydro-degradability of the synthesized bioplastic should be more carefully modulated in the future because the three bioplastic samples that were analyzed lost more than half of their initial mass within the eight hours of the Water Resistance Test. Moreover, D-mannose bioplastics could be synthesized with varying levels of flexibility, durability, and strength for different applications from wiring insulation to medical instruments to environmentally friendly packaging. Due to their remarkable properties, D-mannose bioplastics have the potential to revolutionize the materials industry without compromising the planet we share. 

Acknowledgements

We would like to thank Ms. Jennifer Thomas at Saint Francis High School for providing the use of her lab space, lab equipment, and fume hood. This study would not have been possible without her encouragement, guidance, and resources.

References

[1] Helmenstine, A.M. What is Plastic? Definition and Examples in Chemistry. ThoughtCo. 2018 May. [https://www.thoughtco.com/plastic-chemical-composition-608930]

[2] Iowa State University. Bioplastics Lab. Biorenewables Education Laboratory. Pp. 15. 2015. [http://www.engineering.iastate.edu/brl/files/2011/10/brl_bioplastics_instructor.pdf]

[3] Geyer R, Jambeck JR, Law KL. Production, Use, and Fate of All Plastics Ever Made. Sci Adv. Vol. 3, Num. 7. 2017 Jul. Doi: 10.1126/sciadv.1700782. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5517107/]

[4] Yates D. Researchers Describe ‘Implausible’ Chemistry That Produces Herbicidal Compound. Illinois News Bureau. 2009 Jun. [https://news.illinois.edu/view/6367/205911]

[5] Parker, L. A Whopping 91% of Plastic Isn’t Recycled. National Geographic. 2017 Jul. [https://news.nationalgeographic.com/2017/07/plastic-produced-recycling-waste-ocean-trash-debris-environment/]

[6] Maddah, H. Polypropylene as a Promising Plastic: A Review. Scientific and Academic Publishing. 2016. [http://article.sapub.org/10.5923.j.ajps.20160601.01.html]

[7] European Bioplastics. What Are Bioplastics? European Bioplastics. 2016 Jan.  [http://docs.european-bioplastics.org/2016/publications/fs/EUBP_fs_what_are_bioplastics.pdf]

[8] International Renewable Energy Agency. Production of Bio-Ethylene. Energy Technology System Analysis Programme. 2013 Jan. [https://iea-etsap.org/E-TechDS/PDF/I13IR_Bioethy_MB_Jan2013_final_GSOK.pdf

[9] Californians Against Waste. Bioplastic Enforcement Campaign. Californians Against Waste. [https://www.cawrecycles.org/bioplastic-enforcement-campaign]

[10] Atlantic Poly, Inc. Biodegradable Plastic. Atlantic Poly. 2016 Oct. [http://atlanticpoly.com/biodegradable-plastics]

[11] NCBI. D-Mannose. U.S. National Library of Medicine. [https://pubchem.ncbi.nlm.nih.gov/compound/mannose]

[12] Hu X, et al. D-Mannose: Properties, Production, and Applications: An Overview. Comprehensive Reviews in Food Science and Food Safety. Vol. 15, Num. 4. Pp. 773-785. 2016 May. Doi: 10.1111/1541-4337.12211. [https://onlinelibrary.wiley.com/doi/full/10.1111/1541-4337.12211]

[13] Zhang T, et al. Isolation and Purification of D-Mannose from Palm Kernel. Carbohydrate Research. Vol. 344, Num. 13. Pp. 1687-1689. 2009 Sep. Doi: 10.1016/j.carres.2009.06.018. [https://www.ncbi.nlm.nih.gov/pubmed/19631314]

[14] NCBI. (4S,5S,6R,7R)-4,5,6,7,8-Pentahydroxyoctane-2,3-dione. U.S. National Library of Medicine. [https://pubchem.ncbi.nlm.nih.gov/compound/22840310]

[15] Ray A, Ghosh S. Aloe vera Gel: Biochemical Composition, Processing and Nutraceutical Applications. Recent Progress in Medicinal Plants. Vol. 41. Pp. 1-22. 2014 Dec. Doi: 10.13140/RG.2.1.5056.6564. [https://www.researchgate.net/figure/Basic-molecular-Structure-of-acetylated-mannan-or-acemannan-a-predominant-component_fig1_275338434]

[16] Liu P, et al. Chemical Constituents, Biological Activity and Agricultural Cultivation of Aloe vera. Asian Journal of Chemistry. 2013 May. [http://www.asianjournalofchemistry.co.in/User/ViewFreeArticle.aspx?ArticleID=25_13_1]

[17] Synthesis of Biological Macromolecules. Boundless Biology. [https://courses.lumenlearning.com/boundless-biology/chapter/synthesis-of-biological-macromolecules/]

[18] Clark, J. Acid Catalyzed Hydrolysis of Esters. Chemistry LibreTexts. 2019 June. [https://chem.libretexts.org/Bookshelves/Organic_Chemistry/Supplemental_Modules_(Organic_Chemistry)/Esters/Reactivity_of_Esters/Acid_Catalyzed_Hydrolysis_of_Esters]

[19] Parker, L. Animals Eat Ocean Plastic Because It Smells Like Food. National Geographic. 2016 Nov. [https://www.nationalgeographic.com/news/2016/11/animals-eat-ocean-plastic-because-of-smell-dms-algae-seabirds-fish/]

Works Consulted

Gade R, Tulasi M, Bhai V. Seaweeds: A Novel Biomaterial. International Journal of Pharmacy and Pharmaceutical Sciences. Vol. 5, Num. 2. Pp. 40-44. 2013 Apr. [http://www.ijppsjournal.com/Vol5Suppl2/6880.pdf]

Dehydration Synthesis - Definition and Examples. Biology Dictionary. [https://biologydictionary.net/dehydration-synthesis/]

Arutchelvi J, et al. Biodegradation of Polyethylene and Polypropylene. Indian Journal of Biotechnology. Vol. 7. Pp. 9-22. 2008 Jan. [https://pdfs.semanticscholar.org/2450/5625e6d7b99dff9081815785ad5196defe5d.pdf]

Wolchover, N. Why Doesn’t Plastic Biodegrade? LiveScience. 2011 Mar. [https://www.livescience.com/33085-petroleum-derived-plastic-non-biodegradable.html]

Song JH, et al. Biodegradable and Compostable Alternatives to Conventional Plastics. Philosophical Transactions: Biological Sciences. Vol. 364, Num. 1526. Pp. 2127-2139. 2009 Jul.

[https://www.jstor.org/stable/pdf/40485986.pdf?refreqid=search%3A9a5cfd4e82a382dd5187882c04265c21&seq=1#page_scan_tab_contents]

Araki K. A Potential Usefulness of Agar for Packaging and More. Agar Plasticity. 2016. [https://www.kosuke-araki.com/agarplasticity]

Gupta B, et al. Polylactic Acid Fiber: An Overview. Progress in Polymer Science. Vol. 32, Num. 4. Pp. 455-482. 2007 Apr. Doi: 10.1016/j.progpolymsci.2007.01.005. [https://www.sciencedirect.com/science/article/pii/S007967000700007X]

Tullo A. Making Wood Last Forever with Acetylation. Chemical and Engineering News ACS. Vol. 90, Num. 32. 2012 Aug. [https://cen.acs.org/articles/90/i32/Making-Wood-Last-Forever-Acetylation.html]

Fringant C, et al. A Biodegradable Starch Based Coating to Waterproof Hydrophilic Materials. Starch. Vol. 50, Num. 7. Pp. 292-296. 1998 Dec.

[https://onlinelibrary.wiley.com/doi/abs/10.1002/%28SICI%291521-379X%28199807%2950%3A7%3C292%3A%3AAID-STAR292%3E3.0.CO%3B2-%23]

Fritz JS, Schenk GH. Acid-Catalyzed Acetylation of Organic Hydroxyl Groups. Analytical Chemistry. Vol. 31, Num. 11. Pp. 1808-1812. 1959 Nov. Doi: 10.1021/ac60155a034. [https://pubs.acs.org/doi/pdf/10.1021/ac60155a034]

Hemsri S, et al. Improvement of Toughness and Water Resistance of Bioplastic Based on Wheat Gluten Using Epoxidized Natural Rubber. IOP Conference Series: Materials Science and Engineering. 2015. Doi: 10.1088/1757-899X/87/1/012049. [http://iopscience.iop.org/article/10.1088/1757-899X/87/1/012049/pdf]

North EJ, Halden RU. Plastics and Environmental Health: The Road Ahead. Rev Environ Health. Vol. 28, Num. 1. Pp. 1-8. 2013 Jan. Doi: 10.1515/reveh-2012-0030. [https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3791860/]

A Feasibility Study for Q-Learning Applied to a Dynamic Ant-Foraging Model

Featured

Written by: Malika Shah

Peer Reviewer: Akhila Gundavelli

Professional Reviewer: David Witman, PhD.

Abstract:

Reinforcement learning algorithms, such as Q-Learning, have been shown in certain situations to mimic the behavior of biological systems.  One example is ant-foraging models, where the goal of each individual ant is to explore an environment with the purpose of providing food for its colony.  Although traditional ant-foraging models have not included explicit reinforcement learning algorithms, they are well suited for this type of approach. In this work, we propose an ant-foraging model using Q-Learning to explore a constantly changing terrain environment that obeys the rules of a general Abelian Sandpile Model (ASM).  We consider a stochastic version of the ASM in order to show learned behavior is not dependent on deterministic conditions. Additionally, we will show that the agents (ants), given enough time, will eventually learn the most optimal path to the food; however, in some cases due to the terrain, the agents were not always able to take it. Our research provides an approach for understanding how Q-Learning algorithms explore a constantly changing, random environment and its associated complexities.

Introduction:

General Ant-Foraging Models

Ant-foraging models, included in the broader topic of optimal foraging theory, describe how ants locate and gather food (Pyke, 2000). Note that the ants will be referred to agents in our simulation.  One goal of optimal foraging theory is to understand multi-agent scenarios, where a set of agents has the ability to interact with an environment (Beckers, Holland, & Deneubourg, 1994) (Deneubourg, et al., 1991).  In typical ant-foraging models, pheromones are used to communicate information among the agents in the environment (Holldobler & Wilson, 1990).  Generally speaking, there are two types of pheromones that have been used in past research. The first pheromone represents the smell released by the food in the predefined environment which the agents use to locate the food source.  While the second pheromone is released by each individual agent after it has successfully located the food; this serves as a communication mechanism between agents (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004) (Panait & Luke, 2004) (Wolf & Wehner, 2000).  This work models the typical characteristics of Cataglyphis fortis, more generally known as Desert Ants, which can smell food up to 3 meters away (Wolf & Wehner, 2000).  (Panait & Luke, Learning ant foraging behaviors, 2004) showed through a preliminary ant-foraging model that given only pheromones as a guide and basic encoded logic, the agents could collectively determine an optimal path (Panait & Luke, Learning ant foraging behaviors, 2004).  Additionally, they included various obstacles in their environment including multiple blocks and a rotating “clock” that needed to be avoided in order to locate the food source, showing that the majority of agents were able to find the most optimal path (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004).

1 Pheromones are chemical signals that many species use to communicate a variety of signals including territorial,
alarming, and food trailing (Silva-Junior, et al., 2018). Although all of the chemicals in pheromones are not known,
the main one in ants is pyrazine (Silva-Junior, et al., 2018).

Abelian Sandpile Model

The Abelian Sandpile Model, also known as the Bak-Tang-Wiesenfeld Model, has a few general practical applications.  We use this model, however, because it provides us a simple random model on which we can build on. The model represents a theoretical simulation of a dynamic environment where grains of sand are continuously added to a defined two-dimensional map, either randomly or systematically (Bak, Tang, & Wiesenfeld, 1987). After a certain amount of time, local “avalanches” occur when a site has exceeded its capacity and the grains topple onto adjacent sites. Eventually, these local avalanches will cascade into a global “avalanche” affecting the entire region.  

The ASM we make use of requires a finite, nonnegative square lattice, Z where represent the i horizontal and j vertical cell indices respectively, a time-step, k, with where represents the duration of the simulation, and a critical value, K, which denotes a critical number of sand grains (Bak, Tang, & Wiesenfeld, 1987).  The model proposed by Bak, Tang, and Wiesenfeld has three steps (Bak, Tang, & Wiesenfeld, 1987):

If any , then the site topples and the following occurs:

  1. The site gives a grain to each of its four neighbors

2. The site is left with zero grains.

3. If any of the other sites is equal to or exceeds the K value, then the steps are repeated. If there is more than one “avalanche”, the order in which the “avalanches” occur does not
matter as well as the order in which the sand grains are distributed (Bak, Tang, & Wiesenfeld,
1987) . In our simulation, we do not restrict the sand grains on Z, so if a sand grain is supposed to
be distributed to a site that is not on Z , it leaves the system.

Q-Learning:

Q-Learning is a reinforcement-learning algorithm used in many current applications today like robotics, modeling and simulation, and gaming (Watkins C. J., 1989) (Chen, Li, & Dong, 2008) (Al-Tamimi, Lewis, & Abu-Khalaf, 2007) (Mnih, et al., 2013) . For example, in robotics, (Chen, Li, & Dong, 2008) were able to show how a robot can execute biped balance using Q-learning. Another notable use of Q-Learning is in teaching an agent playing the Atari game, Space
Invaders (Mnih, et al., 2013) . The agent was able to learn how to play the game as well as earn a high score (Mnih, et al., 2013) .

Q Learning utilizes Bellman’s Equation, which relies on parameters that quantify: states, actions, rewards and two parameters that define learning rates (Watkins C. J., 1989) . The state (s) is a numerical quantization of the environment at a given point in time a ind space depending on the scenario (Watkins C. J., 1989) . In our simulation, the state with respect to each agent is defined by its location on the grid and the corresponding number of sand grains present at that site. At every single discrete state, there is a set of actions (a) available to the agent (Watkins C. J. 1989) . The actions available to each individual agent are the movements to an adjacent cell (diagonally, left, right, up or down). Given a state-action pair, a reward (r) (scalar or function value) metric provides feedback to the agent quantifying how beneficial that action at that given state was (Watkins C. J., 1989) . For our study, the agent’s reward can be represented as the number of pheromones in the site and whether or not it reached the food in our simulation. The learning rate is a scalar between 0 and 1 and allows for a configurable parameter for the feedback to the agent (Watkins C. J., 1989) . Theoretically, the greater the learning rate, the faster the agent will learn; however, this is not guaranteed. The discount rate also set between 0 and 1, allows for future rewards to be worth less than the immediate rewards (Watkins C. J., 1989) . If the discount rate is 0.9, the agent will care more about its long-term rewards. Contrastingly, if the discount rate is 0, the agent will only care about its short- term rewards; this is said to be “greedy” behavior because the agent is trying to maximize its immediate rewards (Russell & Norving, 2010) . The Bellman equation, named after Richard E. Bellman, is an optimization method used in dynamic programming (Dixit, 1990) . The Q- Learning algorithm utilizes the following Bellman equation (Watkins C. J., 1989) :

Although all Q-Learning processes use the same equation, Q-Learning can be implemented in various forms, such as a tabular approach as well as a function approximation via a Neural Network (Watkins & Dayan, 1992) (Mnih, et al., 2013) .

A tabular approach, the approach being used in this work, implements a Q table or an array containing real numbers storing the value of Q based on the state and set of available actions (Watkins C. J., 1989) . As an agent explores an environment, it uses the Q table to determine an optimal state action pair (Watkins C. J., 1989) . Given an action procured by the agent, a resultant reward and new state information are used to update the Q table (Watkins C. J., 1989).

Using a function approximation (like a neural network) in lieu of the Q table approach is generally known as deep q learning. This approach uses a neural network to map the state-action relationship instead of storing values in a Q-table. (Mnih, et al., 2013) . As the agent explores the environment, the respective state-action pairs are stored, updated and used to train the Neural Network (Mnih, et al., 2013) . The neural network then has the ability to approximate the Q-Table in a more condensed form (Mnih, et al., 2013) . These approximations are revised as necessary based on future experiences in the environment and its ability to approximate the given space (Mnih, et al., 2013) .

Method:

The main goal of our research is to determine how a Q-learning algorithm applied to an ant-foraging model behaves in a constantly changing, random environment.  Previous ant-foraging research has not focused on the aspect of a random environment. The following sections correspond to various components of our simulation.   The first section describes the environment in which the agents traverse. Then we will discuss the constraints imposed upon the agents in the simulation. Finally, we will discuss the specific parameters associated with the implemented Q-Learning Algorithm.

Environment Configuration

Our model builds on Luke’s and Panait’s model with the rotating “clock” for obstacles (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004).  In our model, we use the ASM to represent the terrain that the agents must traverse. For this work, a stochastic definition of the ASM is used, meaning the grains are placed randomly at different locations throughout time.  This creates a “mosaic” type of effect where “hills” of different heights form over time creating the effect of constantly changing terrain.  

For this simulation we allow the ASM to create an initial environment before the agents are released onto the domain.  We denote the number of grains that the ASM places before the agents begin to move towards the food as   The nest location is in the top left corner of the domain, while the food location is in the bottom-right corner, where and are the coordinates of the center of the food.  The food is not restricted to a single cell so, a radius is defined for the food placement such that multiple cells are contained. The number of sand grains at a site and time is denoted as   The number of pheromones released is denoted as Np and how long the pheromones last on a site is denoted by .  Finally, the pheromones associated with the food are denoted by .

Ant Logic

To define the agent behavior and logic we utilize an ant-foraging model, similar to that of (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004).   However, our work implements a Q-Learning tabular approach, which completely determines the agent’s actions instead of the decision being based on the pheromones available at a given state.  Generally, ants have an impressive memory as it relates to foraging; in fact, about 50% of their memory is solely dedicated to locating food sources (Hourcade, Muenz, Rossler, & Devaud, 2010).  Using this information, we make two assumptions in regards to the agent logic. First, each individual agent has its own version of which represents its learned knowledge.  Second, each agent has full memory of its initial starting point, i.e. the agent nest.  This second assumption allows the agent to immediately return to the nest, via the most direct path, after collecting food.

Rewards: 

The rules that we impose on the agents concerning their movement on the terrain are described as follows:

The first rule creates obstacles for the agent, since the agents can become blocked by the grains of sand on their way to and from the food.  Our implementation of ASM represents a stochastic implementation seeing as the grains are placed randomly. Thus, the obstacle configuration cannot be one hundred percent known by the agents, no matter how much they explore the environment.  The agents must have a holistic approach and be adaptable to different situations.

Once an agent collects food, it releases pheromones on its way back to the nest.  Since we have assumed perfect knowledge of the original nest location each agent makes its way back to the nest by calculating the minimum straight-line distance between neighboring sites and the nest.  This assumption is made by (Collett, Dillmann, Giger, & Wehner, 1992), who found that an ant’s memory could accurately pinpoint nest location during the foraging process. This process allows the agents to take the theoretically optimal path back to the nest; however, the agents still have to follow the rules determined by the terrain. 

Simulation Configuration

The parameters of the Q-Learning algorithm and the simulation, as a whole, are set with the values as follows:


Table 1 shows our chosen values for the various parameters used in our simulation.

We completed 50 trials of each experiment to get a 95% confidence interval.  The value was found through analysis of Monte Carlo sampling trials (Driels & Shin, 2004).  The standard deviation, s, the mean, Y, and the estimated error, , were approximated using preliminary trials (Driels & Shin, 2004).  We decided to complete 50 trials in order to account for any error that may have gone into calculating the minimum value of iterations.

Our simulation parameters follow a similar setup to that of (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004) and our  values were chosen from the recommendation from (Even-Dar & Mansour, 2003).  

The pseudo code for the simulation:


Results:

Figure 5: The configurations of the board at each of the specified time-steps. 

Baseline Experiment: Preliminary observations

For our first experiment, we measured the average number of moves it took a single agent to get to the food for all trips (see Figure 6).  If the number of steps decreases with respect to the overall number trips, then we would imply “learned” behavior.  

From Figure 6 we see that after 15 trips, the agents have decreased the total number of steps to less than 100 steps.  This is an improvement from the more than 1500 steps that the agents took when they ventured out for the first time.


Figure 6: The mean numbers of steps the agents take each time they go to the food (trip number).  The overall trend shows that the number of steps overtime that the agents take decreases

After 15 experiences of getting food, the agents were overall roughly 80% more efficient at getting food than their first experience; however, not all of the agents were taking the optimal path.  Although most of the agents were able to learn the environment, some of the agents could not if they got stuck because the sand grains blocked their path and took thousands of steps to get to the food. That is why the average is considerably higher than it should be; however, one can see in the next experiment that the minimum value is about 50, which is consistent with the expected results.

Experiment 2: Dynamic and static environments

For our second experiment, we studied the effect of the dynamic environment on the minimum number of steps it took for an agent to locate the food. Two cases were analyzed: a static environment where the terrain is fixed and a dynamic environment where sand is constantly added to the terrain. Our static environment was pre-populated with 2000 grains of sand, and then fixed for the remainder of the scenario. The average minimum number of steps from the static environment was 47 steps, while the dynamic environment required 85 steps. The average minimum number of steps is also a measure of how fast it took one agent to reach the food. In Figure 7, the comparison of the static environment and changing environment is further observed. The transition to fewer steps to the nest is seen as the agents progress through the time-steps.


Table 2: The comparison between changing environment and static environment.  


Figure 7: The distributions over 2500 time steps in a static environment (blue) and changing environment (orange). The red and the green show the medians in their respective environments.

We went a step further in this experiment by letting the changing environment run for more time steps.  At 10000 time steps the minimum number of steps the agents took in the changing environment was 45 steps.


Figure 8: the distribution for the minimum number of steps the agents took in a changing environment for 10000 time steps.

Given enough time steps (10000), the changing environment is able to achieve the level of the static environment at 45 steps as seen in Figure 8.

Experiment 3: Number of ants on board

For our third experiment, we examined how the number of agents in the environment affects how well the agents are able to learn the environment.  We observed the percentage of agents that have food at any given time-step with respect to the total number of agents in the environment.  Figure 9 shows a violin plot of the resultant distributions for environments with 20, 60, and 100 agents.  As the number of agents increases, the maximum number of agents that have food at any given time-step also increases (Table 3).  However, the percentage of agents that have food at any given time-step stayed roughly the same between 20 and 25% in the simulation. The maximum percentage of agents that have food at any given time-step theoretically assuming perfect learned behavior should be 50%.  This would imply that the agents spend half their time traveling to the food source and half dropping the food off at the nest.  Figure 9 is created using data from 2500 time-steps run with 50 repetitions for 100 agents. 


Figure 9: A comparison graphically of the effect of the number of agents in the environment on the dynamics in the environment.   The number of agents that have food at a given time-step is shown. The violin plot shows the distribution of all of the time-steps as well as the range of the percentages.  


Table 3: A comparison of the number of agents on the board to the maximum number of agents that had food at any given time-step

Experiment 4: Variable starting sand grains

For our fourth experiment, we observed how , the number of sand grains that are added to the board before the agents are placed, affects the minimum number of steps the agents need to take to the food.  From Figure 10, the general trend shows that as more grains are placed on the board, the more steps the agents needed to take to get the food.  This trend is seen up until a point at which a plateau is reached.  This means that, at least initially, more sand grains implies longer travel times to initially find the food.  At around 3500 sand grains, the minimum number of steps plateaus at roughly 130 steps, this can be considered a steady state terrain for this scenario.  This result is expected because the ASM eventually becomes “organized” and there are not any more obstacles than there were before.


Figure 10: The numbers of sand grains on the board before the agents transverse it affects how fast the agents get to the food.

Conclusion:

Although the agents were placed in a dynamic environment with random configurations, generally speaking, the RL ant-foraging model was still able to locate the food source.  In certain cases, the routes developed by the agents were suboptimal due to the dynamic nature of the terrain.

The Baseline Experiment reinforces the idea that the Q-Learning algorithm can help optimize the path of the agents in a changing environment quickly, although it takes many more time steps than initially thought for an agent with Q-Learning to learn to navigate a changing, random environment.  

Additionally, when comparing dynamic and static environments, we conclude that agents in a changing environment take longer than agents in a static environment.  This may be due to randomness, as well as the changing nature of the environment. However, given enough time-steps, the agents in the changing environment were able to learn the environment at a similar accuracy to the agents in the static environment evidenced in Figure 8.  

By analyzing the effect of the number of ants on board, we showed that adding more agents to the environment increased the overall complexity of the scenario.  It also reveals how not all of the agents can have food at the same time. All the agents do not have food at the same time, but go in cycles of picking up and dropping food.  The maximum number of agents that had food at any given time is at about 25% of the number of total agents in the environment. So, our agents did not reach our theoretical efficiency of 50%.  This shows how in a changing, random environment, even though the agents know the optimal path, they are hindered from taking it because of the unexpected sand grains.  

Finally, by varying the initial starting sand grains, we were able to observe the effects of the dynamic nature of the terrain.  The more sand grains, the longer the agents took to get to the food because the board had more obstacles, and the more obstacles there are, the more complex the terrain gets.  At around 3500 sand grains, the number of steps the agents took plateaus because the complexity reaches a steady state. The grid configurations do not become more complex. If the environment is even more complex, the agents may not do as well or may take a considerably long time to explore the environment and reach the goal. 

Comparing our results with the past work of (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004), we note a few similarities as well as differences when compared to this work.  Both of our agents were able to find the most optimal path; however, our agents were not always able to take it due to the terrain (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004).  Additionally, pheromones played a large role in our simulations. They allowed us to see emergent properties in the behavior of the agents. Our agents took longer than (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004) model to find the food initially; however, over time it became shorter (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004).  Using a Q-Learning algorithm allows us to explore other types of environments, ones that cannot be explored only using pheromones. If Q-Learning was applied to (Panait & Luke, A pheromone-based utility model for collaborative foraging, 2004) model, the agents would have learned much faster than basing their entire logic on just using pheromones. For example, incorporating Q-Learning into our simulation allows us to observe different macro behaviors of the agents (Hourcade, Muenz, Rossler, & Devaud, 2010). 

Our research had the goal of observing how the Q-Learning algorithm with the tabular approach behaves in a constantly changing, random environment.  We investigated whether or not it is feasible and useful to apply to future work, which involves a random, changing environment. Through our results, we would recommend using a tabular approach in future work if the environment is comparatively small and there are no real-time limitations.  Our work points to implementing a functional approximation via a neural network for large environmental problems. As seen from our results, the agents were able to explore the environment, but they could not fully adapt to the changes in the environment. A functional approximation via a neural network may be able to resolve the mentioned limitations; however, more experimental evidence is needed.

References

Al-Tamimi, A., Lewis, F. L., & Abu-Khalaf, M. (2007). Model-free Q-learning designs for linear discrete-time zero-sum games with application to H-infinity control. Automatica, 373-481.

Bak, P., Tang, C., & Wiesenfeld, K. (1987). Self-organized criticality: An explanation of 1/f noise. Physical Review Letters, 381-384.

Beckers, R., Holland, O. E., & Deneubourg, J. (1994). From local actions to global tasks: Stigmergy and collective robotics. Artificial Life IV: Proceedings of the International Workshop on the Synthesis and Simulation of Living Systems.

Chen, C., Li, H. X., & Dong, D. (2008). Hybrid control for robot navigation-a hierarchical Q -learning algorithm. IEEE Robotics and Automation Magazine.

Collett, T. S., Dillmann, E., Giger, A., & Wehner, R. (1992). Visual landmarks and route following in desert ants. Journal of Comparative Physiology A, 435-442.

Deneubourg, J. L., Goss, S., Franks, N., Sendova-Franks, A., Detrain, C., & Chretian, L. (1991). The dynamics of collective sorting: robot-like ants and ant-like robots. Animals to Animals: Proceedings of the First International Conference on Simulation of Adaptive Behavior, 356-363.

Dixit, A. K. (1990). Optimization in economic theory. Oxford University Press on Demand.

Driels, M. R., & Shin, Y. S. (2004). Determining the number of iterations for Monte Carlo simulations of weapon effectiveness.

Even-Dar, E., & Mansour, Y. (2003). Learning rates for Q-learning. Journal of Machine Learning Research, 1-25.

Holldobler, B., & Wilson, E. (1990). The ants. Harvard University Press.

Hourcade, B., Muenz, T. S., Rossler, W., & Devaud, J. M. (2010). Long-term memory leads to synaptic reorganization in the mushroom bodies: a memory trace in the insect brain. Journal of Neuroscience, 6461-6465.

Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., & Riedmiller, M. (2013). Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602.

Oberle, W. (2015). Monte Carlo Simulations: Number of Iterations and Accuracy. US Army Research Laboratory.

Panait, L. A., & Luke, S. (2004). Learning ant foraging behaviors. Artificial Life XI Ninth International Conference on the Simulation and Synthesis of Living Systems, 575-580.

Panait, L., & Luke, S. (2004). A pheromone-based utility model for collaborative foraging. Autonomous Agents and Multiagent Systems, 36-43.

Panait, L., & Luke, S. (2004). Learning ant foraging behaviors. Artificial Like XI Ninth International Conference on the Simulation and Synthesis of Living Systems.

Pyke, G. (2000). Optimal foraging theory: a critical review. Annual Reviews Ecological Systems, 523-575.

Russell, S., & Norving, P. (2010). Artificial Intelligence: A Modern Approach. Prentice Hall.

Silva-Junior, E. A., Ruzzini, A. C., Paludo, C. R., Nascimento, F. S., Currie, C. R., Clardy, J., & Pupo, M. T. (2018). Pyrazines from bacteria and ants: convergent chemistry within an ecological niche. Scientific Reports.

Sornete, A., & Sornete, D. (1989). Self-organized criticality and earthquakes. Europhysics Letters, 197.

Watkins, C. J. (1989). Learning from delayed rewards. 

Watkins, C., & Dayan, P. (1992). Q-learning. Machine Learning, 279-292.

Wolf, H., & Wehner, R. (2000). Pinpointing food sources: olfactory and anemotactic orientation in desert ants, cataglyphis fortis. The Journal of Experimental Biology, 857-868.

Sound Localization and Speech Detection to Assist the Hearing Impaired

Yeji Cho

Anthony Kim

Valmik Ranparia

Sky Shia

Young Kim
Chadwick High School/Peninsula High School

Peer Reviewer: Annasimone Andrawis and Lily Ge

Professional Reviewer: Mark Nimmer

Sound Localization and Speech Detection to Assist the Hearing Impaired

ABSTRACT

Auditory assistive devices are becoming increasingly ubiquitous, and audio visualization devices represent an innovative method of helping the hearing impaired. The purpose of this project is to evaluate a method of creating an auditory visualization device that displays the real- time location of a sound source on a 3D grid interface. In addition to determining the location of the sound source, a functioning prototype requires the differentiation between audio frequencies for background noise reduction and identifying sound types such as human voices or passing cars. The hypothesis was that if audio frequency selection is used in cross-correlation, then sound source localization of each frequency source position could be achieved. Audio data was recorded and compared using the cross-correlation method. This method yielded a single value representing the calculated sound location. This calculated location was then compared to the actual location. This process was repeated in multiple trials with sounds of both high and low frequencies. The FFT-IFFT technique was used to remove one of the frequencies. The source position of the other frequency was then calculated. Analysis of the one-source results showed that most positions were found within a 40-degree angle of error. Analysis of the frequency- removed data supported the hypothesis and indicated that determining each sound source position can be achieved from one mixed sound signal. The results supported the hypothesis, and the sound localization technique developed in the present study can be further developed and implemented in future research to create a practical, wearable device.

INTRODUCTION

Development and improvement of auditory assistive devices is critical, as 11% of Americans are categorically hearing impaired (Hear-It Organization, 2009). 28.8 million American adults stand to benefit from the use of hearing aids (U.S. Department of Health and Human Services, 2018). Only 28.5% of Americans who are hearing impaired employ the use of hearing aids (Hear-It Organization, 2009). In the period of 2005-2008, the number of hearing impaired Americans increased by 9%, doubling the rate of the population growth of 4.5% in the same period of time (Hear-It Organization, 2009). Sales of hearing aids have been on the rise accordingly, with sales increasing by 5.3% in 2018, exceeding what market experts deem in the “normal” range of 2-4% annual growth (Hearing Review, 2019). Globally, the hearing aid and auditory assistive device retail market is expected to expand with a significant compound annual growth rate of 7.2% during 2019-2028 (Market Watch, 2019).

Audio-visualization devices are a key innovation in how auditory assistive devices improve the lives of the hearing impaired. Visualization devices can show locations on a display from which sound originates in their field of vision, show icons that represent familiar sounds (e.g. ambulance sirens, cars honking, etc), or convert speech to text. Users of audio visualization devices would be able to rely heavily upon the devices to provide cues that are key to navigating their daily environments.

Sound localization refers to the ability to detect the origin of a sound source. Human sound localization is primarily based on two factors: interaural time and intensity differences (Wightman & Kistler, 1992). Sound intensity is more commonly understood, and refers to the “loudness” or power carried by sound waves. This factor is used to understand the distance from which noise is being produced. Interaural time difference refers to the lag time between sound arrival to each ear. This subtle difference can be used to subconsciously calculate the angle from which sound is arriving.

These two factors -- sound intensity and interaural time difference -- are utilized in the development of audio visualization devices. However, other factors, such as varying frequencies of environmental sounds, also play a role in the creation of a usable device, making this technology even more difficult to make precise. For example, users need to be able to distinguish sounds of different frequencies for various reasons; some examples include the need to identify specific voices and the intensity of the screech of car tires. Multiple sounds are often produced simultaneously, and ambient noise is present in nearly every real-life setting. A device that cannot distinguish between multiple sound sources would render itself unusable, producing an inaccurate location for what it assumes is only one sound.

Despite the clear importance of audio visualization devices, development continues to be very limited. Current prototypes cannot provide a comprehensive analysis of noise and provide little aid in real-world applications. For example, a now-abandoned industry prototype visualized sound using LED lights placed around a pair of glasses to indicate the direction of sound. The simplistic nature of this device prevents the user from understanding other characteristics of sound, including the intensity or source type. The lack of a screen or display interface prevents the device from communicating to its user more specific spatial dimensions of environmental sounds. Other prototypes included spectrograph and “positional ripples” visualization of sound source locations (Ho-Ching, Mankoff, & Landay, 2003). The “positional ripples” display was among the most user-friendly, in that it portrayed sound location, source type, and other suchnecessary factors. However, each of these prototypes proved to be ineffective in spaces with ambient noise due to difficulty in distinguishing the target sound.

The purpose of this project is to find ways to distinguish between multiple sounds occurring at the same time. Methods that currently exist to distinguish between sounds include use of amplitude, frequency, and other physical properties of sound. Devices that can measure the amplitude of sound waves in order to evaluate sounds are already available. Thus, this project focused on frequency as an identifying factor for sounds. It was hypothesized that if frequency selection was used in audio recordings, then the location of each frequency-typed sound source could be identified.

PRINCIPLES & THEORY

Cross-correlation is a method of measuring the similarity of two series by evaluating the displacement of one series relative to the other. If two series x(i) and y(i) are evaluated and = 0, 1, 2, ... N-1, the cross-correlation at delay can be defined where mx and my are the means of the series and (see Equation 1). If is evaluated for all delays, = 0, 1, 2, ... N-1, then it results in a cross-correlation series twice the length of the original series.

Equation 1. Cross-correlation r at delay d.

The output is a number which can be used to determine the difference in the distances between two microphones and the sound source.

Another technique used was the FFT-IFFT technique, which converts time domain audio recordings into frequency domain data, and then converts the data back into the original time domain data.

FFT is an algorithm that computes the discrete Fourier transform (DFT) of a sequence, or its inverse (IDFT) (Heideman, Johnson, & Burrus, 1984). The DFT is obtained by decomposing a sequence of values into components of different frequencies (see Equation 2). An FFT rapidly computes such transformations by factorizing the DFT matrix into a product of sparse (mostly zero) factors (Heideman, Johnson, & Burrus, 1984).

Equation 2. Formula for discrete Fourier transform (DFT).

MATERIALS & METHODS

Method 1: Single Source Localization

The first set of data was collected through single source localization methodology. Four microphones were arranged into a tetrahedral formation to ensure that recordings could be used to determine a specific point in the 3D grid (see Figure 1A & 1B). A single sound file was played at nine different locations (30, 60, and 90 cm horizontally from mics 1, 2, and 3) and captured using the four microphones.

Fig. 1A. Recording equipment: experiment setup with microphone locations labelled. Four microphones were placed in formation of the diagram. Three microphones were placed 120 degrees apart from each other. The final microphone was placed at the center.

Fig. 1B. Picture of recording equipment (note: recording environment not pictured).

From recordings captured by each microphone, the difference in time (?time) of when the sound was captured by each microphone was calculated using cross-correlation function on Matlab, similar to the use of interaural time difference in mammals. This ?time was converted into ?distance by multiplying it with sound velocity. The ?distance was calculated between each set of 2 microphones (mics 1 & 2, mics 1 & 3, mics 1 & 4, mics 2 & 3, mics 2 & 4, and mics 3 & 4), giving a total of six combinations of microphones.

The recording environment was scanned in 10 centimeter increments (see Figure 1C). At each 10 centimeter increment, distance between that point and each of the 4 microphones was calculated and compared to the 6 ?distance values. If the values were equal, this meant that the point was a match for the sound source. This method shall be referred to as “Position Search”.

Fig. 1C. Position search algorithm: visualization of recording environment scanning.

Angle of error was determined from two vectors: first was the position of known sound source, and second was the position of calculated sound source. Angles of error were charted for each of the 9 tested sound sources (see Figure 2A & 2B).

Fig. 2A. Single sound source localization. Y-axis represents angle of difference between actual sound source and calculated sound source. X-axis shows each of the nine tested sound sources.

Fig. 2B. Visualization of actual sound sources. Each number shows the location of the source, and the value of each number shows the calculated angle of error.

Method 2: Frequency Selection Localization

The second set of data involved distinguishing between two sounds occurring simultaneously. A low frequency sound (male voice) and a high frequency sound (female voice) were recorded from different locations at the same time, following the same microphone setup as the Single Source Localization method. Position search was done with the same method as the Single Source Localization case, but with mixed voice sound files.

Using FFT technique, mixed voice recordings were converted into frequency domain data. Initially, the high frequency sound portion was removed and converted back to time domain sound file using the IFFT technique, producing a recording with a low frequency isolated sound file. This same single voice extraction process was followed to produce a high frequency isolated sound file. Using the cross-correlation, the angle of error between the exact and calculated sound source positions were determined for (“Quick Statistics,” 2018) the original mixed voice sound file, (Wightman & Kistler, 1992) the low frequency only sound file (see Figure 3A), and (Kim, Choi, & Kim, 2013) the high frequency only sound file (see Figure 3B).

Fig. 3A. Low frequency sound source localization.

Fig. 3B. High frequency sound source localization.

RESULTS

The results of single sound source localization gave a maximum error value of 46.4 degrees and a minimum error value of 15.8 degrees (see Table 1 & Figure 2A). This angle of error was calculated by determining the angle of difference between the three-dimensional coordinates for the actual versus the calculated sound source. This angle, measured in degrees, can be visualized in a three-dimensional space (see Figure 4A).

Fig. 4A. Depiction of the range given by any error value ?.

Table 1. Single sound source localization. Angle represents the angle of difference between the actual sound source and the calculated sound source.

The low frequency and high frequency sounds were evaluated in two distinct sets of data, as each sound came from a different location at the same time. First, the coordinates of each of the sound sources was compared to the calculated coordinate when the two sounds were evaluated together with no frequency limitations. This provided the angle of error titled “original sound.” These error values ranged from 32.5 degrees to 73.8 degrees, with one outlier of 17.5 degrees (see Table 2 & 3). Then limitations were placed on the frequency to run the cross- correlation using only low or high frequency sounds. Again, the angle of error was evaluated, and titled “low frequency sound” or “high frequency sound.” These error values were far smaller, and ranged from 8.9 degrees to 20.1 degrees (see Table 2 & 3). A notable improvement was seen when the frequency was limited, showing that one of the sounds was successfully eliminated from evaluation. This meant that limitations on frequency values allowed cross- correlation to be taken of either the low or high frequency sound.

Table 2. Low frequency sound source localization.

Table 3. High frequency sound source localization.

DISCUSSION

The major finding of the present study is that single source localization and frequency selection localization methods did not produce significantly accurate results, but were able to greatly increase accuracy by isolating one frequency range prior to cross correlation and FFT- IFFT method.

In the set of trials titled “Single Source Localization,” the majority of the sound source positions were found with less than 40-degree error (see Figure 4A). Given that the recording was done in a closed room with the potential of sound reflecting off the walls and creating a high level of background noise in the recordings, the error value indicates that while accuracy is a continued prospect for future research, localization is possible. Though more efforts should be taken in future research to reduce the error (i.e. using a recording booth with near-zero audio rebound), current results show that even rudimentary audio cleaning is possible using FFT-IFFT technique.

As expected for the Frequency Selection Localization trials, calculating sound source position with mixed voice sound files did not yield accurate results. The angles of error were generally larger than the single sound source case. This indicates that the sounds occurring simultaneously caused errors when interpreted as one source.

However, upon removing one of the voices (isolating low frequency or high frequency sound), angles of error were greatly reduced. These results suggest that positions of sound sources can be accurately determined using the cross-correlation technique. Frequency-isolation method, calculated using FFT-IFFT technique, yielded much more accurate sound source position calculations and higher resolution of sound. This increase in accuracy may be a result of isolating one voice from the original recordings. The removal of one of the frequencies may have reduced data noise (background static, sound reflecting off of walls, etc.), which may have interfered with sound position calculations in the first Single Source Localization method.

The results support the hypothesis and suggest that frequency selection was successfully achieved in cross-correlation methodologies tested. Evaluating the frequency of audio recordings allowed sound source localization and differentiation between simultaneous sounds of varied frequencies.

A novel method of sound source localization was determined through isolation of sounds of different frequencies. Results indicate that future research could involve the use of this program to create devices for use as an auditory aid. Work has been done on speech-to-text, which can be used alongside sound localization software in order to create a usable device. Future research should focus on implementation of this methodology in commercially applicable devices with good visualization of sound localization.

REFERENCES

“35 million Americans suffering from hearing loss.” (n.d.). Hear-It Organization. Retrieved from https://www.hear-it.org/35-million-Americans-suffering-from-hearing-loss

“Global Hearing Aid Retail Market To Expand with Significant CAGR of 7.2% During 2019- 2028.” (2019, January 10). Market Watch. Retrieved from https://www.marketwatch.com/press- release/global-hearing-aid-retail-market-to-expand-with-significant-cagr-of-72-during-2019- 2028-2019-01-10

Gorman, B. M. (2014). VisAural:. Proceedings of the 16th International ACM SIGACCESS Conference on Computers & Accessibility - ASSETS 14. doi:10.1145/2661334.2661410

Hearing Aids Market Size, Share | Industry Analysis Report, 2019-2025. (n.d.). Retrieved from https://www.grandviewresearch.com/industry-analysis/hearing-aids-market

Heideman, M., Johnson, D., & Burrus, C. (1984). Gauss and the history of the fast fourier transform. IEEE ASSP Magazine, 1(4), 14-21. doi:10.1109/massp.1984.1162257

Ho-Ching, F. W., Mankoff, J., & Landay, J. A. (2003). Can you see what I hear? Proceedings of the Conference on Human Factors in Computing Systems - CHI 03. doi:10.1145/642611.642641

Kim, K.W., Choi, J.W., & Kim, Y.H. (2013). An assistive device for direction estimation of a sound source. Assistive technology: The official journal of RESNA, 25, 216-21. 10.1080/10400435.2013.768718.

“Quick Statistics About Hearing.” (2018, October 05). Department of Health and Human Services. Retrieved from http://www.nidcd.nih.gov/health/statistics/quick-statistics-hearing

“US Hearing Aid Sales Increase by 5.3% in 2018.” (n.d.). Hearing Review. Retrieved from http://www.hearingreview.com/2019/01/us-hearing-aid-sales-increase-5-3-2018-approaches-4- million-unit-mark/

Wightman, F. L., & Kistler, D. J. (1992). The dominant role of low frequency interaural time differences in sound localization. The Journal of the Acoustical Society of America, 91(3), 1648- 1661. doi:10.1121/1.402445

ACKNOWLEDGEMENTS

We would like to acknowledge Dr. Kim for his technical guidance throughout the experimentation and Dr. Nimmer for his role in the development and background information for this project.