Sage Journals: Discover world-class research

Abstract

This article presents the design and development of a new hands-free ultrasonication robot for filamentous fungi homogenization. The platform was constructed with a modified inexpensive 3D printer, equipped with an upward-facing camera, a custom-designed wash station, and an add-on sonicator. While machine vision accomplished sample well screening based on image subtraction and color thresholding, it also determined the level of fungi homogeneity using color variance. Model fitting reveals that the process of filamentous fungi homogenization using ultrasonication included a period of significant exponential decay. Therefore, this procedure allowed for the rapid homogenization of the fungal samples during the initial stages of ultrasonication treatment followed by a deceleration in homogenization. Furthermore, a factorial experiment showed that higher sonicator power and temperature accelerated the homogenization process, while the cultivation time exhibited no effect on homogenization. In addition, the model parameters were varied between the wells, even when subjected to the same settings, meaning that the system cannot use the same asymptote of the homogeneity level to establish the termination time for different wells. Therefore, we used the standard deviation of the four most recent homogeneity level values to determine the termination time. This method was used for feedback control, forming a fully automated robot that did not require manual intervention during the experiment. A validation test on filamentous fungi demonstrated that the system was able to provide target quality of samples efficiently.

Keywords

laboratory automation robotics ultrasonication fungi homogenization measurement fungi homogenization modeling

Introduction

Microorganisms broadly serve as cell factories for the production of a variety of important high- and low-value products, such as lipids, biosurfactants, antibiotics, enzymes, therapeutic agents, exopolysaccharides, and bioplastics. It is possible to easily isolate some of these products following extracellular storage and excretion into the cultivation medium. However, others, such as lipids and polyhydroxyalkanoates, depend on intracellular production, necessitating their extraction for analysis via a process usually initialized by cell homogenization.¹ In addition, biochemical fingerprinting of microorganisms by high-throughput Fourier transform infrared (FTIR) spectroscopy, particularly in the case of filamentous fungi,^2,3 uses homogeneous cell suspensions that require the application of ultrasonic homogenization.

Generally, it is possible to achieve cell homogenization by using chemical or mechanical/physical methods,⁴ and recently, the development of a variety of cell homogenization procedures facilitated the efficient, low-cost, and effective release of intracellular products.⁵ The effectiveness of these different homogenization techniques depends on the properties of the microbial cell, such as the physical strength of the cell wall and the location of the desired intracellular product.⁵ Ultrasound disintegration—ultrasonication—is a common mechanical cell homogenization method based on the high shear force created by a high-frequency ultrasound (above 16 kHz),⁵ which can achieve complete disruption of the microbial cell.^5–7 Additional benefits of ultrasonication are that it reduces cell disruption processing time and energy consumption.⁸ Unfortunately, however, the process also presents three main disadvantages. First, manual use of the ultrasonication system requires highly skilled technicians to hold, mix, identify, and monitor during processing.⁹ Second, the process requires lengthy and repetitive operations, making it unsuitable for screening studies in which a large number of samples need to be sonicated within a short period of time. Finally, high-sound intensity has negative effects on human health, causing headaches, discomfort, and irritation.¹⁰ It is, therefore, important that a platform be developed for the automation of manual ultrasonication operations.

Mancia and Love¹¹ presented a robotic ultrasound disintegrator with an electronically controlled sonicator probe mounted on a five-axis industrial robotic arm. Similarly, Almo et al.¹² developed an ultrasonication robot equipped with an EPSON robotic arm. In addition, previous work by Li et al.⁹ used an ultrasonic module with a FESTO three-axis arm to prepare samples for analysis by FTIR spectroscopy. These systems simply used industrial robotic arms for ultrasonication manipulation without closed-loop control strategies for monitoring or automated control. The open-loop feature might, however, result in insufficient or excessive ultrasonication and, more important, might not always provide well-homogenized samples. Previous studies have provided neither automatic measurement methods for determining the degree of fungi homogeneity nor any mathematical models to quantify the phenomena of filamentous fungi homogenization by ultrasonication. Without appropriate sensing abilities, these machines continue to rely on the manual input of sample locations and technicians to monitor the process, adjust ultrasonication time, and so on.

In the laboratory automation field, many platforms with sensing systems have been developed to automate experiments and tests ranging from robotic sample preparation systems¹³ to a machine vision-based system for sorting zebrafish embryos¹⁴ and an automated system to test irrigation sprinklers.¹⁵ Recently, the well-developed, open-source software for 3D printing technology allowed for rapid and low-cost prototyping. Based on a 3D printer, Nejatimoharrami et al.¹⁶ presented a liquid-handling robot named EvoBot while also developing a 3D printer-based platform together with an imaging system for screening microscopy.¹⁷

The main contributions presented in this article are the following:

We developed a novel, fully automated ultrasonication robot for the homogenization of fungal cells. The robot used machine vision to distinguish sample wells and blank wells.

We quantified the fungi homogenization process using model fitting, suggesting that homogeneity level to ultrasonication time of all the tested sample wells can be well fitted with exponential decay equations. A further factorial experiment showed temperature, sonicator power, fungal varieties, and well replicates have a significant influence on homogeneity level.

We developed a feedback control strategy that used the standard deviation of local homogeneity values to determine the ultrasonication termination time.

The proposed system was validated by the homogenization of filamentous fungi but might also be applicable to other types of microorganisms, such as yeasts, bacteria, and algae. The developed ultrasonication robot could be of particular use in processes that prepare filamentous fungi samples for biochemical phenotyping by FTIR spectroscopy, extraction of intracellular metabolites, and DNA from microbial cells.

Materials and Methods

System Design and Hardware Description

Figure 1 shows the hardware assembly of the developed platform. It comprises five functional modules: (1) an XYZ motion stage; (2) a control and vision module, including a controller and an upward-facing camera; (3) a washing module consisting of a pump and a newly designed wash station; (4) a sonicator module; and (5) the power supply. The overall size of the ultrasonication platform, excluding the water bottles, is 59 × 45 × 80 cm (length × breadth × height).

Figure 1.

The ultrasonication robot: (a) robot setup and (b) schematic diagram of components and their connections.

XYZ motion stage

Figure 1a shows that the modification of a low-cost 3D printer formed the basis for the robot actuation module (Anet A8, Anet, China). This provided three degrees of motion for the sonicator, with an original working space of 220 × 220 × 240 mm. A 3D-printed sonicator holder with which to mount the sonicator on the robot arm replaced the extruder module used for 3D printing ( Suppl. Fig. S1 ). Furthermore, a custom-designed mixer with 96 pins was mounted on the same slider next to the holder. Before taking pictures, the mixer decreases sediment formation that might affect image processing. When removing mixer pins from the well, the robot moves the pins into contact with the well wall to remove droplets. After processing a microtiter plate of samples, the mixer can be washed in either a well plate with fresh water or a general container.

A laser-cut wooden board attached to a 3D-printed plate holder replaced the hotbed of the printer. Modifications to the original 3D printer controller facilitated motion control while being flashed with open-source firmware (0.92.5; Repetier, Willich, Germany) with a robust motion control algorithm that promoted external control through the G-code protocol. One disadvantage of the firmware is that it does not provide feedback upon completion of motion. However, this feedback is important in minimizing cycle time, since its utilization can trigger precisely sequential motions between the motion stage, pump, and sonicator. Therefore, we added a new function to the firmware to monitor when motions are complete and update the information to the upper controller. Since the robot arm is open-loop controlled, it requires calibration after each restart of the robot, as it is in 3D printing.

Control and vision module

The control board can be found on the left side of the platform. Selecting a Raspberry Pi 3 (Model B, Raspberry Pi Foundation, Cambridge, UK) as the main controller was due to its suitability for both image processing and electronics control. As shown in Figure 1b , the Pi controller has a serial port to communicate with the original 3D printer controller using G-code protocol for motion control. A centrally mounted, upward-facing camera (Raspberry Pi V2; Raspberry Pi Foundation) located underneath the plate holder provided a view of sample plates for image processing. The camera has a maximum resolution of 3280 × 2464 pixels; however, a lower resolution setting of 544 × 640 pixels facilitated fast processing. An LED panel consisting of several RGB LED units (5050; Golden Sun, Shenzhen, China), fixed on the center of the platform provided, a homogeneous light source for image processing. Customization of the LED panel shell allowed for easy attachment to the robot, and the lighting area (132 × 125 mm) was, therefore, large enough to pass light through the 96-well microtiter plate (128 × 86 mm) (CR1496; EnzyScreen, Heemstede, Netherlands).

On the top-left corner of the platform is a user-friendly control panel with one power switch and two buttons (start and pause), which connected directly to the GPIO pins of the Pi controller. An intentionally positioned low-cost four-channel relay module (SainSmart, Lenexa, KS) together with a transistor array (ULN2003A; Texas Instruments, Dallas, TX), segregated the Pi controller from the 3D printer main board. As shown in Figure 1b , the relay module is used to control the 3D printer controller, pump, LED panel, and sonicator controller for on/off operations. The Pi controller commands the relay module with current amplification through the transistor.

Sonicator module, washing module, and power supply

A sonicator (Q125; Qsonica, Newtown, CT) combined with a standard probe facilitated the sonication of fungal cells. The washing module consists of a peristaltic pump (WPL 810; Williamson, Southwick, UK), a newly designed wash station (middle right of Fig. 1a and Suppl. Fig. S2 ), a fresh water bottle, a waste water bottle, and several silicone tubes. To avoid contamination during washing, the design of the wash station allows for the rinsing of the sonicator probe tip rather than bathing it, as was the case in the previous platform.⁹ The input voltage is 5V DC for the Pi controller, 12V DC for the motion stage, and 24V DC for both the pump and the LED panel. Hence, in addition to the 12V DC power supply module of the printer, a dual-output embedded power (RD-65B; Mouser, Shanghai, China) supplies all other voltages.

Image Processing

Measurement of fungi homogeneity level

Accurate measurement of the fungi homogeneity level is essential to close the loop of ultrasonication control. Since the level of fungi homogenization in a well plate varies from well to well, analysis transpired via single-well-based image processing. This necessitated the segmentation of the wells and the identification of their centroid locations in the image pixel coordinates. One simple method was first to calculate the real-world coordinates (X, Y, Z) of the well centroids using geometry, which were then transferred into pixel coordinates (u, v) by means of camera calibration. Using the MATLAB camera calibrator toolbox and based on the classic Zhengyou Zhang calibration theory,¹⁸ the camera position changed 15 times, capturing images of a standard calibration chessboard from each position to obtain the camera’s intrinsic parameters as well as its distortion coefficients. These results allowed for the identification of the relationship between the pixel coordinates and the camera coordinates (x_c, y_c, z_c). Attaching the calibration chessboard to the plate holder with its rows or columns parallel to the holder edges for camera calibration permitted the acquisition of the extrinsic coordinate transformation from the camera coordinates to the real-world coordinates. With additional geometry dimensions, the well centroids in the real world were translated to the pixel coordinate system. Supplemental Figure S3 shows the region of interest (ROI) pixels for image processing in a camera view. The enlarged figure (lower right) illustrates the location of the ROI pixels in the center of the well and that they are smaller than the well to avoid edge effects. To reduce computation resources in the Pi controller, the transformation was conducted offline in MATLAB, and the pixel coordinates of the well centroids were imported directly into the Pi controller for real-time processing. A few months of laboratory tests indicated that the continuous use of the calibration results was possible until the camera or the plate holder position changed.

Filamentous fungi have a multicellular structure, where hyphae cells form a complex structure called mycelium. When filamentous fungi were homogenized by ultrasound sonication, we observed disruption of mycelium structure, and fungal cells were homogenized. With time, the ultrasonication treatment reduced the size of single or several large pieces of fungal mycelia in each well. Hence, the first measurement method attempted to calculate the fungal area for each well based on color thresholding. However, because of differences in the fungal biomass quality, amount, and possible fungal varieties in each well, the thresholds varied from well to well. Furthermore, during the ultrasonication, in some cases, the whole pieces of fungal biomass split into several small pieces, making area calculation difficult. Homogenization of the filamentous fungal mycelia disrupted the mycelia into smaller pieces, causing the sample suspension to become more homogeneous. Homogeneity level, on the opposite side, also means variance level. Therefore, this study proposed a robust method based on the color variance within each single well, which calculated and combined all of the three basic color channels, R (red), G (green), and B (blue), ranging from 0 to 255. The total variance $V a r_{R G B}$ in a well can be expressed as

V a r_{R G B} = [\begin{array}{l} \sum_{i = 1}^{N} {(R_{i} - μ_{R})}^{2} + \sum_{i = 1}^{N} {(G_{i} - μ_{G})}^{2} \\ + \sum_{i = 1}^{N} {(B_{i} - μ_{B})}^{2} \end{array}] / N,

(1)

where $μ_{R / G / B} = (\sum_{i = 1}^{N} R_{i} / G_{i} / B_{i}) / N$ . Generally, within a well, the smaller the $V a r_{R G B}$ , the more homogeneous the samples.

Sample screening

The purpose of sample screening was to extract the sample wells (filled with filamentous fungal mycelia) from the well plate so that the treatment could skip the blank wells (empty wells) without any preinput information, a feature that is essential for hands-free operation. Figure 2a shows the original picture of a well plate where a few sample wells are randomly located. To identify the sample wells, hue, saturation, and value (HSV) color thresholding was considered, since it is widely used for object segmentation.¹⁹ As mentioned above, due to the different angles at which the camera photographs the wells, the light saturation detected by the camera is uneven, and there are also shadows caused by the wells. This means that fixed global color thresholds cannot work well on different wells in the original picture, as verified by these tests. In addition, different fungal species of samples have different colors and thus require different color thresholds. To make the vision system robust to different wells and types of samples, we used a picture $B (i, j)$ that captured from a blank well plate to subtract the corresponding pixel values from the target original picture $O (i, j)$ (e.g., Fig. 2a ), thereby obtaining a subtracted picture $S (i, j)$ :

S (i, j) = B (i, j) - O (i, j),

(2)

where i and j represent the row and column pixel values, respectively. After this operation, as is shown in Figure 2b , the subtracted image $S (i, j)$ clearly shows the changes between the target picture $O (i, j)$ and the blank well plate picture $B (i, j)$ , thus indicating that the fungal mycelia in the sample wells already underwent segmentation. Figure 2c shows that employing color thresholding to extract the sample wells and converting the color image into a binary image successfully removed noise. During this process, pixels that were within the fine-tuned threshold range $(T h r e_{H S V})$ were set as white pixels (value 1), while pixels out of $T h r e_{H S V}$ were set as black pixels (value 0). Since the white pixels could have been disconnected within a well, rather than an area calculation, the quantity of white pixels ( $Q_{w h i t e}$ ) served as the indicator to quantify the likelihood that a well either contained a sample or was empty. This can be expressed as

Q_{w h i t e} = \sum P i x e l_{T r u e}, P i x e l_{T r u e} = {\begin{matrix} 0, P i x e l_{H S V} \notin T h r e_{H S V} \\ 1, P i x e l_{H S V} \in T h r e_{H S V} \end{matrix},

(3)

where $T h r e_{H S V}$ denotes whether or not the pixel color $T h r e_{H S V}$ is within the threshold range $T h r e_{H S V} .$ After thresholding, the blank wells were skipped, as illustrated in Figure 2d .

Figure 2.

Sample screening approach: (a) original picture; (b) a subtracted image that was obtained by image subtraction operation between a blank well plate picture and the original picture (a), in which wells with samples were extracted; (c) a binary picture shows the samples in white pixels obtained by use of color thresholding; (d) as a result, the blank wells were skipped correctly; (e) experiment results for determining the extraction threshold of $Q_{w h i t e} :$ generally, sample wells (orange) and blank wells (blue) show a significant difference of $Q_{w h i t e}$ .

Furthermore, an experiment performed on fourteen 96-well plates (1344 wells) with different sample well distribution and different types of filamentous fungi determined the extraction threshold of $Q_{w h i t e}$ for regular plates. As shown in Figure 2e , the orange square points represent $Q_{w h i t e}$ of sample wells while the blue circular points represent $Q_{w h i t e}$ of blank wells. Generally, the $Q_{w h i t e}$ values of sample wells and blank wells were significantly different. As a result, a segmentation line of 183 pixels (centerline between the minimum point of sample well and the maximum point of blank well) separated blank wells and sample wells with a perfect classification result (100%).

Filamentous Fungi Samples for the Ultrasonication Experiment

The modeling and control tests involved two varieties of filamentous fungi—namely, Mucor circinelloides VI 04473 (Norwegian School of Veterinary Science, Norway) and Mucor hiemalis UBOCC-A-101359 (Université de Bretagne Occidentale Culture Collection, France). The selection of these two varieties was due to the well-known fact that Mucor fungi have a thick cell wall containing different polymers (chitin/chitosan/polyphosphate) that make the homogenization process difficult.

Fungi were cultivated in 800-µL liquid medium malt extract broth (Oxoid, Basingstoke, UK) in 96-square polypropylene deep-well plates using the Duetz-MTPS (EnzyScreen) microtiter plate system for 2 to 3 days at 25 °C. After cultivation, the filamentous fungal mycelia were washed with distilled water in three centrifugation steps. After each centrifugation step, the mycelia formed a pellet at the bottom of wells of the microtiter plate, and supernatant above the mycelia was removed. Thereafter, the mycelia were transferred into new 96-square polypropylene deep-well plates filled with 2 mL of deionized water to homogenize fungi. We used distilled water according to the microbiology protocol for washing mycelia from the growth medium. Cells of mycelia already appeared in swallowed form before washing, since the fungi grew in a liquid medium.

Modeling and Control

Modeling of the Filamentous Fungi Homogenization Process

The method used to measure the fungi homogeneity level was also the technique employed to investigate fungi homogenization changes during ultrasonication treatment. Knowledge of the exact relationship between the fungi homogeneity level and the ultrasonication treatment time could reveal the characteristics behind ultrasonication homogenization of filamentous fungi, which denoted a possible approach to precisely control ultrasonication procedures.

The first experiment intended to observe the $V a r_{R G B}$ changes against ultrasonication time by monitoring sample temperatures and liquid density. Four M. circinelloides wells with 2 days’ cultivation were processed at room temperature (22.7 °C) with an ultrasonication duration of 15 s for each well in each interval. Each well was treated 40 times for a total of 600 s. During the experiment, the homogeneity level was measured after each 15-s treatment. We used a rapid noncontact infrared thermometer (IRT260; Biltema, Drøbak, Norway) to measure the temperature of the well bottom before and after each ultrasonication treatment. To track the sample density changes, this study used an optical density spectrophotometer (UV-T500PRO; GENESYS, Waltham, MA) to measure the sample liquid, excluding the undisrupted mycelium.

Figure 3 displays the three parameter changes during ultrasonication of a M. circinelloides well, and Supplemental Figure S4 shows the corresponding visual variation of this well captured by the robot camera. As shown in Figure 3a , generally, except for the first point (big black point at 15 s), the $V a r_{R G B}$ decreased rapidly at the beginning of the treatment and then steadily approached an asymptotic value. The same trend is evident in Supplemental Figure S4 , in which the size of the black mycelium in the well became smaller and smaller and the decreasing speed seemed faster at the beginning yet slower at the end. In Figure 3a , the $V a r_{R G B}$ value of the first point is lower than those of the second and third points, because the mycelium at 15 s occupies most of the ROI region ( Suppl. Fig. S4 ), thus reducing the color variance of the ROI region. The same phenomenon can also be found in the original picture of the well ( Suppl. Fig. S4 ). This means that the $V a r_{R G B}$ may increase at the beginning of the ultrasonication due to the inherent properties of the measurement method. Without considering the original $V a r_{R G B}$ and also removing the first point, the remaining points in Figure 3a tend to be smooth. They can be modeled well by an exponential decay equation, as follows:

V a r_{R G B} = f (x) = a e^{- b x} + c,

(4)

where a, b, and c are the free model parameters. As can be seen in Figure 3a , the fitted model has a high R² value. This model indicates that homogenization of the filamentous fungal samples occurs rapidly using ultrasonication at the beginning of the treatment, after which the homogenization becomes slow. The decreased size of the mycelium might play an important role in this phenomenon as there was less material to be disrupted. Also, Figure 3b shows that the density of the sample liquid grew gradually. This might be a second reason for the exponential decay phenomenon since the higher the density, the higher the viscosity. Higher viscosity can reduce the mycelium movement during ultrasonication so that the mycelium has less chance to be moved to the high disruption effect area around the sonicator probe. The temperature of the sample rapidly increased at the beginning, fluctuating thereafter by around 27 °C ( Fig. 3c ). The well cools naturally while the sonicator probe processes other wells or during the washing procedures.

Figure 3.

$V a r_{R G B},$ sample density, and temperature changes during ultrasonication of a Mucor circinelloides well: (a) $V a r_{R G B}$ —ultrasonication time curve; (b) absorbance—ultrasonication time curve, showing the density of the sample liquid increased gradually; (c) temperature—ultrasonication time curve.

To identify whether factors such as temperature, sonicator power, fungi varieties, cultivation time, or even the well positions affect changes in the $V a r_{R G B}$ , a factorial experiment was performed to assess these influences. In the first subtest, the temperature was set to either 26 °C or 50 °C. The well plate was preheated to the target degree before each treatment using a heating circulator (TC120; Grant Instrument, Shepreth, UK). The ultrasonication was performed with 50% or 60% of the sonicator’s rating power (effective power 62.5 or 75 watts [W]) in a pulse-regime (pause or power on, can be externally controlled) mode. For each setting, there were at least three replicates of wells. Supplemental Figure S5 shows the results of some wells in different settings, and Supplemental Figure S6 displays the corresponding pictures of these wells. Generally, all of them fit well with equation (4), but the model parameters (a, b, and c) vary in the different settings, as well as in the same setting with different well replicates. Two varieties of filamentous fungi (M. hiemalis and M. circinelloides) were tested with different cultivation times—2 days and 3 days. As shown in Supplemental Figure S7 , although the settings were different, the $V a r_{R G B}$ changes in these wells still followed equation (4). This led us to conclude that filamentous fungi homogenization by ultrasonication might be an exponential decay process and is similar to many other natural decay phenomena, such as radioactive decay and temperature decrease.

This study used a statistical software (JMP 13, SAS, Medmenham, UK) to perform factor effect analysis on all the data using one-way analysis of variance (ANOVA). In this analysis, the response was $V a r_{R G B},$ while all other settings were the factors. As shown in Supplemental Table S1 , with the exception of the cultivation time, all the factors had a statistically significant influence on $V a r_{R G B},$ using α = 0.05. Particularly, the p-value of wells verified the view that even when all the settings are the same, the well replicates affect the response $V a r_{R G B} .$ It may be due to the variation, related to biomass quality and chemical composition obtained, in cultivation procedure. The ANOVA also provided a linear model using these factors, as shown in Supplemental Figure S8 . In addition to ultrasonication time, both sonicator power and temperature are predicted to have negative influences on $V a r_{R G B},$ indicating that the higher sonicator power and temperature would accelerate the homogenization process. Furthermore, the prediction indicates that M. circinelloides is more difficult to homogenize than M. hiemalis, which supports the point that it normally occurs in biology. The cell wall of M. circinelloides is richer in polymers chitin/chitosan than that of M. hiemalis, and because of that, theoretically it should be more difficult to homogenize it. Also, as shown in Supplemental Figure S7b , these curves were consolidated to pass through the same point (52, 700) by using x-axis translation for better comparison. It is evident that the $V a r_{R G B}$ decreasing speed of M. hiemalis (solid and dashed lines) tends to be faster than for M. circinelloides (dash-dotted and dotted lines).

System Control

Termination time

Since the timeous termination of ultrasonication ensures efficiency, it is essential to be familiar with the target homogeneity level regardless of the control method employed. However, according to the results shown in Supplemental Table S1 , the parameters of equation (4) vary for different well replicates even when the other factors remained the same. Therefore, it is challenging to subject all the wells to the same cutoff value (asymptote). Although the wells display different asymptotes for $V a r_{R G B},$ the derivative $V a r_{R G B}^{'}$ has the same asymptote, 0. Therefore, $V a r_{R G B}^{'}$ was considered to determine whether proper homogenization of the wells has occurred. Supplemental Figure S5 and Supplemental Figure S6 represent the four wells at 26 °C and 75 W. The four circled points/wells (termination points) in these images indicate the required homogeneity levels based on the manual estimation. Therefore, it is evident that the $V a r_{R G B}$ changes were minimal when closing the target points. To calculate $V a r_{R G B}^{'}$ at each point, four local points (three former closest points and the current point) were used to fit a linear model and obtain the slope, $S l o p e_{l o c a l} .$ Supplemental Figure S9 shows the $S l o p e_{l o c a l}$ –ultrasonication time curves for the four wells in Supplemental Figure S5 (26°C and 75 W). In all the wells, the $S l o p e_{l o c a l}$ increased rapidly during the initial stages, while decelerating when approaching 0. Similarly, the circled points represent the time that the wells reached full homogenization. Except for C4, employing a fixed value for the determination of proper homogenization points in the other three wells remained a challenge, since the slopes fluctuated around 0 before the termination points.

Then we used the standard deviation $S t d_{l o c a l}$ of the four local points (three former closest points and the current point) to determine the termination time, and it can be expressed as

S t d_{l o c a l} = \sqrt{\sum_{i = 1}^{4} {(V a r_{R G B i} - \bar{V a r_{R G B}})}^{2} / 4} .

(5)

Figure 4 shows the $S t d_{l o c a l}$ –ultrasonication time curves for the four wells in Supplemental Figure S5 (26°C and 75 W), where the circled points indicate the termination time. Generally, the curves in Figure 4 appear smoother than those in Supplemental Figure S9 . Most important, the $S t d_{l o c a l}$ values exhibited a gradual decrease before the termination points, and their $S t d_{l o c a l}$ values at termination time were 5.7, 4.9, 3.8, and 4.3, respectively. To eliminate fluctuations, the $S t d_{l o c a l}$ value was reviewed three times before reaching a final decision. Hence, the system only terminates the treatment when the three most recent $S t d_{l o c a l}$ values are smaller than the required value. The three points following the termination point for each curve in Figure 4 were all smaller than 3.3. Therefore, 3.3 represented the termination value during the automation process.

Figure 4.

$S t d_{l o c a l}$ –ultrasonication time curves for the four wells in Supplemental Figure S5 (26 °C and 75 W): the circled points (termination points) represent the time that the wells were fully homogenized; these curves show more significant changes before the termination points than those in Supplemental Figure S5 and Supplemental Figure S9 .

Sequential control

Equation (4) shows promise in model-based control for the ultrasonication process. The idea is that, after an initial period of ultrasonication (e.g., 30 s), the system uses a model from equation (4) to estimate the remaining treatment time based on the difference between the target and current $V a r_{R G B}$ values. Theoretically, the model-based control method is the fastest way to achieve fungi homogenization as it processes wells continuously within the estimated remaining time. Moreover, the sonicator tip requires only one washing per well. However, continuous ultrasonication will heat up the samples, thus damaging the sample proteins and other chemical structures. Thus, laboratory experiments usually use 15- to 30-s ultrasonication duration for 62.5-W sonicator power, with a cool-down period for the samples between each treatment. In addition, according to the manual, the sonicator cannot be powered continuously for more than 60 s. Therefore, model-based control is not suitable for filamentous fungi as used in this article, but it might be applicable for samples requiring only short periods of ultrasonication, such as bacteria.

Due to the limitations of ultrasonication, this system used a screening-based control method for the automation. Similar to the modeling process, the robot treated each well with 15 s of ultrasonication at each interval and monitored the termination time for all wells. When reaching the termination time, the well was passed. Figure 5 shows the entire control sequence of the robot. The codes in the Pi controller were written in Python 2.7, and the image processing was based on the Picamera and OpenCV libraries. Three main steps were evident in the ultrasonication sequence following placement of the sample well plate on the robot:

Robot initialization: The Pi controller ran a server script automatically after booting. The server program ran in a loop waiting for the activation of a key followed by the initialization of the XYZ motion stage. The main program initialized once the operator pressed the start button.

Sample screening: After calibration of the arm, the vision system distinguished between blank wells and sample wells. The locations of the sample wells were recorded for ultrasonication, while the blank wells were disregarded.

Feedback control: Once the sample wells had been screened out, the feedback control step was initiated, starting with ultrasonication treatment for a duration of 15 s. During ultrasonication, the sonicator tip in the well moved in a compound manner (circling and up-down) to encourage more homogeneous fungi homogenization, which resembled the way human operators would apply the device. Furthermore, between processing any two wells, the sonicator tip was cleaned thoroughly in the wash station to avoid contamination. As shown in Figure 6c , fresh water passes through the outer hollow space of the wash station to clean the sonicator tip and then flows through the hollow inner space into the waste bottle. To avoid droplets from contaminating other wells, the sonicator was powered on for 1 s to remove the droplet from the tip of the sonicator probe after withdrawing it from the well ( Fig. 6b ). In addition, when extracting the sonicator probe following the washing process, the robot touched the inner top of the wash station to remove the droplet ( Fig. 6d ). The top of the wash station has a conical surface for better scraping of the droplets. Subsequently, measurement of the fungi homogeneity levels of all sample wells occurred online using the image-processing system. After completion of the first four treatments, the system began to calculate the $S t d_{l o c a l}$ value for determining the termination time. Wells that did not meet the termination requirements (reaching target homogeneity level three times continuously) were subjected to the ultrasonication process again. Once a well reached the proper termination time, no further treatment was required. Contrary to the open-loop control approach subjecting all wells to the same ultrasonication treatment time,⁹ the feedback control method reduced the ultrasonication time for an entire well plate since disregarding the homogenized wells. The maximum ultrasonication time was set to 450 s, which could be readily changed depending on fungi species and target homogeneity level.

Figure 5.

Workflow of the robot control sequence: the ultrasonication sequence can be divided into three main steps: robot initialization, sample screening, and feedback control. The system measured the fungi homogeneity levels of all sample wells after each 15-s ultrasonication and screened them again for a new cycle of ultrasonication until the termination requirements were satisfied or the wells had completed a 450-s processing cycle.

Figure 6.

Schematic of ultrasonication and washing steps: (a) ultrasonication of the fungal sample; (b) the sonicator is powered on for 1 s to remove the droplet when the probe is taken out from the well; (c) washing the sonicator probe in the wash station; (d) the sonicator probe is touching the inner top of the wash station to remove the droplet.

This screening-based control is simple, is robust, and ensures high homogenization quality. Consequently, the vision-based feedback control system allowed for the full automation of the machine without requiring manual intervention during the experiment.

Results and Discussion

Measurement Precision and Accuracy

To evaluate the precision of the fungi homogenization measurement, 11 wells with different homogeneity levels were measured continuously without any treatment and repeated 25 times. The results, illustrated in Supplemental Figure S10a , show horizontal lines of $V a r_{R G B}$ that were relatively stable, suggesting the acceptable repetition precision of the proposed measurement method. An average percentage deviation of 2.29%, obtained from $100 % * \sum_{i = 1}^{N} s_{i} / (\bar{s} * N),$ was evident, where $s_{i}$ was the standard deviation of each line in Suppl. Fig. S10a .

To evaluate the measurement accuracy, this study utilized fully homogenized samples to fill the well plate. As shown in Supplemental Figure S10b , wells located at the edges of the microtiter plate (low or high row number) presented higher $V a r_{R G B}$ , while the wells in the middle row exhibited relatively low $V a r_{R G B}$ . This could be attributed to the changes in the camera angle when observing each well. Due to different angles, the camera detected the various light intensities and shadows that were present in the wells. The standard deviation of these points was 1.9, but the average percentage deviation was 26.3%. Furthermore, evaluation of the accuracy occurred by randomly placing the six physically heterogeneous samples into different wells. Although not significant, various fluctuations were evident in $V a r_{R G B}$ of the different wells, with a standard deviation of 10.2 and an average percentage deviation of 11.4%, as shown in Supplemental Figure S10c . The relative low accuracy indicated that it was advisable to control the wells independently when using $V a r_{R G B}$ .

System Control Result

To validate the control method, an experiment was performed with six M. circinelloides sample wells using the same setting as during the modeling process (2 days’ cultivation, 26 °C, and 75 W). The vision system successfully screened out the six wells. During the feedback control step, the wells underwent ultrasonic treatment until the system detected the termination time. Figure 7a shows images of the entire procedure, where the pictures circled in black indicate the final states. Figure 7b shows the corresponding $V a r_{R G B}$ and $S t d_{l o c a l}$ changes of all wells, clearly indicating that the ultrasonication duration of the six wells varied between 150 s and 435 s. Compared to the open-loop control approach that treated all wells with the same ultrasonication time,⁹ the screening-based control reduced the ultrasonication time by 870 s (33.3%) for the six wells if the open-loop control treatment time was set to 435 s.

Figure 7.

Validation test results: (a) images of six wells during all stages of ultrasonication: all the filamentous fungi samples were ultimately homogenized (black circled pictures), and the treatment time varies from 150 s to 435 s; (b) $V a r_{R G B}$ and $S t d_{l o c a l}$ to ultrasonication time curves of the six wells in (a).

Consequently, sufficient homogenization of all the fungi samples transpired and reached similar states comparable to the modeling process (refer to the circled wells in Suppl. Fig. S6 ). However, as shown in Figure 7a , wells C5, C6, C7, and D7 were entirely homogeneous in the final states, while D5 and D6 still exhibited a small black dot of mycelium that did not appear to display significant changes in size before termination. This result was especially evident in D5, where the size of the mycelium remained the same from the 225-s mark onward, which was also verified by the variance curve in Figure 7b . Unless higher sonicator power was used during the manual tests, the black mycelium dots were extremely difficult to disrupt in some wells. The application of this control method can prevent wasting time to homogenize these challenging mycelium dots if no change is evident in the homogeneity levels after three repetitions.

Limitations and Suggestions

Since the homogeneity measurement method depended on an upward-facing camera, the bottom mycelia primarily determined the color variance calculation. Therefore, the system was not sensitive to the mycelia that remained suspended in the well. On some occasions, the mycelia did not fall to the bottom before the images were captured, which resulted in the acquisition of a smaller $V a r_{R G B}$ . For example, in Figure 7a , well C5 seems homogeneous at 120 s and 135 s and the $V a r_{R G B}$ values are small in Figure 7b . Owing to the termination method by which $S t d_{l o c a l}$ was checked three times before terminating, the robot continued treating the sample and the mycelium returned at 150 s.

However, based on our tests, for some filamentous fungi, such as Penicillium expansum, it seems that the density of the mycelia was lower and close to water, so the mycelia were always suspended in the water. This made homogeneity levels challenging for the imaging system to measure due to the upward-facing camera. A possible improvement might be to use a camera that faces the sides of the wells. However, this requires placing the sample in a separate well for ultrasonication rather than a well plate. Another solution might be to aspirate the sample after each treatment and measure it from the side in a separate well. However, this process would unnecessarily complicate the system since the separate well requires thorough cleaning between samples.

In addition, the P. expansum samples were easily homogenized. The tests in this study indicated that the P. expansum samples could be fully homogenized within 15 s using the same setting as for M. circinelloides (2 days’ cultivation, 26 °C, and 75 W). However, the control system only terminated after obtaining three small, continuous $S t d_{l o c a l}$ values, suggesting that the treatment for each well was only terminated after 105 s. This feature wasted the valuable processing time due to overtreatment but did not affect the final results. To minimize the time required, sample wells that are easily homogenized could be preloaded into the system.

We tested the system with filamentous fungi and found that filamentous fungi homogenization by ultrasonication might be an exponential decay process. However, this theory has not been verified by testing various fungi. Further work should include testing of other types of fungi or even other microorganisms.

In conclusion, this study developed a novel inexpensive platform to automate filamentous fungi homogenization using ultrasonication and constructed the platform hardware by modifying an inexpensive 3D printer with various custom-designed parts and additional electronics. While machine vision distinguished between sample wells and blank wells based on image subtraction and color thresholding, it also measured the level of fungi homogeneity using color variance. The utilization of model fitting indicated that all the tested fungal homogenization using ultrasonication caused significant exponential decay. This process allowed for the rapid homogenization of fungal samples during the initial stages of ultrasonication treatment followed by gradual homogenization. In addition, one-way ANOVA analysis denoted that higher sonicator power and temperature accelerated the homogenization process, while the cultivation time exhibited no effect on homogenization. Moreover, the model parameters varied between the wells, even when subjected to the same settings, meaning that the system cannot use the same asymptote of the homogeneity level to establish the termination time for different wells. Therefore, we used the standard deviation of the four closest homogeneity level values to determine the termination time. This method allowed for feedback control, forming a fully automated robot that did not require manual intervention during the experiment. A validation test on filamentous fungi demonstrated that the system was efficient and able to provide target quality of samples.

Supplemental Material

DS_TECH861361 – Supplemental material for A Laboratory-Built Fully Automated Ultrasonication Robot for Filamentous Fungi Homogenization

Supplemental material, DS_TECH861361 for A Laboratory-Built Fully Automated Ultrasonication Robot for Filamentous Fungi Homogenization by Ya Xiong, Volha Shapaval, Achim Kohler and Pål Johan From in SLAS Technology

Footnotes

Acknowledgements

We thank Julie Eymard for helping to prepare filamentous fungi samples.

Supplemental material is available online with this article.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by Norwegian University of Life Sciences.

ORCID iD

Ya Xiong

References

Shapaval

Møretrø

Wold Åsli

, et al. A Novel Library-Independent Approach Based on High-Throughput Cultivation in Bioscreen and Fingerprinting by FTIR Spectroscopy for Microbial Source Tracking in Food Industry. Lett. Appl. Microbiol. 2017, 64, 335–342.

Shapaval

Schmitt

Møretrø

, et al. Characterization of Food Spoilage Fungi by FTIR Spectroscopy. J. Appl. Microbiol. 2013, 114, 778–796.

Shapaval

Møretrø

Suso

H. P.

, et al. A High-Throughput Microcultivation Protocol for FTIR Spectroscopic Characterization and Identification of Filamentous Fungi. J. Biophontonics 2010, 3, 512–521.

Burden

Guide to the Disruption of Biological Samples. Random Prim. 2012, 25, 1–25.

Klimek-Ochab

Brzezińska-Rodak

Zymańczyk-Duda

, et al. Comparative Study of Fungal Cell Disruption-Scope and Limitations of the Methods. Folia Microbiol. (Praha). 2011, 56, 469–475.

Borthwick

K. A. J.

Coakley

W. T.

McDonnell

M. B.

, et al. Development of a Novel Compact Sonicator for Cell Disruption. J. Microbiol. Methods 2005, 60, 207–216.

Gogate

P. R.

Kabadi

A. M.

A Review of Applications of Cavitation in Biochemical Engineering/Biotechnology. Biochem. Eng. J. 2009, 44, 60–72.

Tiwari

B. K.

O’Donnell

C. P.

Cullen

P. J.

Effect of Sonication on Retention of Anthocyanins in Blackberry Juice. J. Food Eng. 2009, 93, 166–171.

Shapaval

Kohler

, et al. A modular liquid sample handling robot for high-throughput Fourier transform infrared spectroscopy. In Advances in Reconfigurable Mechanisms and Robots II, 2016, 36, 769–778.

10.

Smagowska

Pawlaczyk-Łuszczyńska

Effects of Ultrasonic Noise on the Human Body: A Bibliographic Review. Int. J. Occup. Saf. Ergon. 2013, 19, 195–202.

11.

Mancia

Love

High-Throughput Expression and Purification of Membrane Proteins. J. Struct. Biol. 2010, 172, 85–93.

12.

Almo

S. C.

Garforth

S. J.

Hillerich

B. S.

, et al. Protein Production from the Structural Genomics Perspective: Achievements and Future Needs. Curr. Opin. Struct. Biol. 2013, 23, 335–344.

13.

Wei

Feng

Rehmani

I. J.

, et al. A High-Throughput Robotic Sample Preparation System and HPLC-MS/MS for Measuring Urinary Anatabine, Anabasine, Nicotine and Major Nicotine Metabolites. Clin. Chim. Acta 2014, 436, 290–297.

14.

Breitwieser

Dickmeis

Vogt

, et al. Fully Automated Pipetting Sorting System for Different Morphological Phenotypes of Zebrafish Embryos. SLAS Technol. 2018, 23, 128–133.

15.

Saretta

de Camargo

A. P.

Botrel

T. A.

, et al. Test Methods for Characterising the Water Distribution from Irrigation Sprinklers: Design, Evaluation and Uncertainty Analysis of an Automated System. Biosyst. Eng. 2018, 169, 42–56.

16.

Nejatimoharrami

Faina

Stoy

New Capabilities of EvoBot: A Modular, Open-Source Liquid-Handling Robot. SLAS Technol. 2017, 22, 500–506.

17.

Schneidereit

Kraus

Meier

J. C.

, et al. Step-by-Step Guide to Building an Inexpensive 3D Printed Motorized Positioning Stage for Automated High-Content Screening Microscopy. Biosens. Bioelectron. 2017, 92, 472–481.

18.

Zhang

A Flexible New Technique for Camera Calibration (Technical Report). IEEE Trans. Pattern Anal. Mach. Intell. 2002, 22, 1330–1334.

19.

Xiong

Liang

, et al. Development of a Prototype Robot and Fast Path-Planning Algorithm for Static Laser Weeding. Comput. Electron. Agric. 2017, 142, 494–503.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.55 MB