TY - JOUR
AU1 - Sofela,, Samuel
AU2 - Sahloul,, Sarah
AU3 - Bhattacharjee,, Sukanta
AU4 - Bose,, Ambar
AU5 - Usman,, Ushna
AU6 - Song,, Yong-Ak
AB - Abstract Type 2 diabetes is the most common metabolic disease, and insulin resistance plays a role in the pathogenesis of the disease. Because completely functional mitochondria are necessary to obtain glucose-stimulated insulin from pancreatic beta cells, dysfunction of mitochondrial oxidative pathway could be involved in the development of diabetes. As a simple animal model, Caenorhabditis elegans renders itself to investigate such metabolic mechanisms because it possesses insulin/insulin-like growth factor-1 signaling pathway similar to that in humans. Currently, the widely spread agarose pad-based immobilization technique for fluorescence imaging of the mitochondria in C. elegans is laborious, batchwise, and does not allow for facile handling of the worm. To overcome these technical challenges, we have developed a single-channel microfluidic device that can trap a C. elegans and allow to image the mitochondria in body wall muscles accurately and in higher throughput than the traditional approach. In specific, our microfluidic device took advantage of the proprioception of the worm to rotate its body in a microfluidic channel with an aspect ratio above one to gain more space for its undulation motion that was favorable for quantitative fluorescence imaging of mitochondria in the body wall muscles. Exploiting this unique feature of the microfluidic chip-based immobilization and fluorescence imaging, we observed a significant decrease in the mitochondrial fluorescence intensity under hyperglycemic conditions, whereas the agarose pad-based approach did not show any significant change under the same conditions. A machine learning model trained with these fluorescence images from the microfluidic device could classify healthy and hyperglycemic worms at high accuracy. Given this significant technological advantage, its easiness of use and low cost, our microfluidic imaging chip could become a useful immobilization tool for quantitative fluorescence imaging of the body wall muscles in C. elegans. C. elegans, microfluidics, fluorescent imaging, mitochondria, hyperglycemia Insight As a simple model, Caenorhabditis elegans possesses insulin/insulin-like growth factor-1 signaling pathway similar to humans and is an attractive model for diabetes study. In this study, we report a simple microfluidic chip for imaging of its mitochondria under hyperglycemic conditions. By exploiting its innate behavior, called proprioception, to rotate the body in a microchannel for body undulation, we quantified subtle changes of the mitochondria in the body muscle cells using fluorescence images that were unobservable using traditional agarose pad-based imaging. Our microfluidic channel-based immobilization and imaging approach lends itself to machine learning-based image classification for potential high-throughput image analysis. INTRODUCTION Mitochondria are dynamic organelles that play diverse roles in signaling, physiology and metabolism [1]. Some of these roles include the regulation of several vital cellular processes for the skeletal muscle physiology [1, 2]. As such, the mitochondria are crucially involved in muscle cell metabolism, energy supply, reactive oxygen species (ROS) production and regulation of energy-sensitive signaling pathways [3]. Mitochondrial dysfunction has been attributed to a number of muscle-related diseases such as disuse-induced muscle atrophy [4], Duchenne muscle dystrophy [5, 6], aging-related muscle mass loss [7–11], ventilator-induced diaphragmatic dysfunction [12], and development of insulin resistance (IR) [13] as a result of defect in insulin secretion by pancreatic beta cells and subsequent progression to hyperglycemia. Mitochondrial DNA damage has also been reported to be involved in development of type 2 diabetes [14]. Caenorhabditis elegans has been utilized as a relevant disease model in this context with its metabolic regulation mechanism at the molecular level that closely matches the human model [15]. One of the key similarities between C. elegans and mammals include signaling pathways such as the insulin/insulin-like growth factor-1 signaling pathway, thereby making C. elegans a useful system to further understand complex diseases like diabetes and impact of high-glucose stress [16]. A study by Schulz et al. [17] used a model of impaired glucose metabolism to show that an in increase in ROS leads to secondary hermetic increase in stress defense leading to reduced net stress levels. In contrast, high-glucose culture conditions have been reported to limit the life span of C. elegans by increasing ROS formation and AGE (advanced glycation end product) modification of mitochondrial proteins in a daf-2-independent manner [18]. High-glucose diet has been reported [19] to cause swelling of the mitochondria and accumulation of damaged mitochondria in the muscle cells of C. elegans. The study also observed disorganization of myofilaments, myofibrils and the Z line in the muscle cells. As a transparent model organism, C. elegans allows for in vivo visualization of fluorescently labeled proteins which are transgenetically expressed, thereby revealing insights of several subcellular functions [20] such as mitochondrial dysfunction. However, traditional handling methods of C. elegans are laborious and time-consuming. These methods typically require each worm to be placed on an agarose pad on a glass slide and use of anesthetic for immobilization which can impact developmental processes for long-term studies. Especially in 2D quantitative imaging, due to randomness of the C. elegans body orientation, a repeatable and reliable comparison of 2D quantitative image data between different worms, even from the same population, could therefore be challenging. Based on the same rationale, subtle but significant changes in response to external stimuli could simply be overlooked when randomly orientated. In the last decade, microfluidics has become a powerful platform in the handling and study of C. elegans due to its suitable dimensions relative to the worm’s size, incorporation of active components (such as valves) and multiplexing for high-throughput studies [21]. As such, several channel and droplet-based microfluidic chips have been developed for C. elegans culture [22], sorting [23], immobilization [24], microsurgery [25] and imaging [26, 27]. Lee et al. [28] used a densely packed multichannel microfluidic device for worm immobilization and high-throughput neuronal imaging. The device had 140 single-worm microtraps, at an inclined angle, positioned along a serpentine microchannel. To gain insight into muscle-related diseases, Cornaglia et al. [29] designed a microfluidic platform to monitor protein aggregation in worm models for both amyotrophic lateral sclerosis and Huntington’s diseases. Mondal et al. [30] further probed C. elegans polyglutamine aggregation to model Huntington’s disease at commendable high throughput. To understand metabolic disorders relevant to diabetes in C. elegans, Zhu et al. [31] integrated microfluidic device for long-term culture, immobilization and imaging of C. elegans under hyperglycemic conditions. From all these microfluidic chip-based studies, it is clear that immobilization avails precise control of orientation, and this possibility offers a great technical advantage over the traditional agarose pad-based immobilization when quantitatively imaging body wall muscle cells of C. elegans. Despite its obvious technical advantage, a broad usage of microfluidic devices has still not been fully taken place in the C. elegans community. Among several potential reasons, we speculate that the low acceptance may be related to the device complexity and operation, resulting in higher cost and required time for training that all pose serious barriers to entry. In addition, the above-mentioned microfluidic devices with a typical channel aspect ratio of height to width well below one maintained the worms in a lateral orientation implicitly which is suitable for non-fluorescence imaging. However, for studies involving fluorescence intensity measurement of body wall muscles as in this study, a dorso-ventral worm orientation is required, and to the best of our knowledge, there has been no microfluidic device that allows controlling the orientation from lateral to dorso-ventral by changing the channel aspect ratio and exploiting the proprioception of C. elegans as our device demonstrated. In this study, we propose a simple-to-build and facile-to-use microfluidic device for trapping and imaging C. elegans, operated completely manually without any prior training and using complex instrumentation other than a fluorescence microscope with a CCD camera for imaging. A straight channel with a width similar to the worm diameter, at an aspect ratio of height to width over one, connected the inlet and outlet reservoirs. A notch with an aspect ratio of ~1:3 (width to height) was incorporated into the straight channel which impeded the flow of a worm for imaging, however, deep enough to allow the worm to squeeze through the gap without any obvious damage. Before fully trapping the head or tail of worm in the notch region of the constriction channel, each worm was given a time to reorient itself in the given space confinement with a channel aspect ratio of ~1.3. Inducing a reorientation of its body to gain the maximum space for its body undulation required for propulsion in the microchannel, it was possible to achieve the dorso-ventral body orientation of up to ~85% consistency. This body orientation was more favorable for the quantification of the fluorescent mitochondria in the body wall muscles avoiding an overlap of the muscle quadrants when imaging from the bottom of the microchannel. Using this approach, we quantified subtle fluorescence intensity signal changes in the mitochondria under hyperglycemic conditions that were not detectable using the traditional agarose pad-based imaging approach. Confocal imaging confirmed significant changes in the mitochondria upon exposure to the hyperglycemic conditions. The fertility and locomotory assays showed no obvious behavioral and fertility changes of worms after passing through the narrow notch region after imaging. Finally, this microfluidic channel-based imaging enabled subsequent image analysis using a machine learning-based image classification approach. By using a pretrained deep convolutional neural network (CNN) architecture, we demonstrated a case study where the CNN model could classify between healthy and hyperglycemic C. elegans relatively accurately despite a limited number of the available fluorescent images of the mitochondria, opening up new opportunities for automatic quantitative imaging and image analysis. MATERIALS AND METHODS Microfluidic chip fabrication The imaging device was fabricated by conventional soft lithography technique using a single layer of polydimethylsiloxane (PDMS) peeled from the surface of a 4″ silicon master wafer. The designed pattern was developed using a layout editing software (Klayout®) and printed on a chrome plate by a mask writing machine (Heidelberg Instruments DWL 66+). Using photolithography, the pattern was transferred onto a silicon wafer using a negative photoresist, SU 8 2025. The photoresist was spun on a silicon at 2000 rpm resulting to a thickness of ~50 μm. The photoresist was patterned using a mask aligner (Karl Suss MA8) with exposure dose of 200 mJ/cm2 and developed using standard SU 8 developer. The master was then silanized for 5 h. The PDMS polymer was obtained by mixing 10:1 weight ration of base (Sylgard® 184 silicone elastomer) and curing agent using a degassing mixer (Think® ARE 250). The mixture was cast over the master and further degassed in a vacuum jar for 30 min. Before it was transferred to an oven to cure for ~4 h at 70°C. The cured PDMS was peeled off from the master mold and treated with O2 plasma for 2 min and bonded onto a glass slide. Worm culture One transgenic model strain SJ4103 (zcls14[myo-3::GFP(mit)] with GFP expressed in the mitochondria of body wall muscle cells was used in this study. It was purchased from Caenorhabditis Genetics Center (University of Minnesota, Minnesota, MN). Worms were cultured on nematode growth medium (NGM) with a lawn of Escherichia coli (OP50) [32]. Sodium hypochlorite treatment was carried out on gravid adult animals to get embryos, and then eggs were allowed to hatch at room temperature. L1 worms were cultured at 17.5°C for ~70 h until young adult stage. For the hyperglycemic study, L1 stage worms were cultured to young adult stage on an ultraviolet (UV)-treated E. coli (OP50) in addition to 1 mL of two glucose concentrations 200 mM and 400 mM. For drug treatment, metformin (1, 1-Dimethylbiguanide hydrochloride (Sigma Aldrich, D150959)), a common diabetes drug was used. To treat the worms, metformin was added to the NGM and subsequently used for hyperglycemic worm culture. The final concentration of metformin was 25 mM. This study was carried out under the following four conditions: control, 200 mM of glucose, 400 mM of glucose, and 25 mM of Metformin with 400 mM of glucose. Experimental setup and image acquisition The experiment was independently carried out using two platforms: microfluidic chip and agarose pad for comparison. The worms from the same batch were used in both platforms. The agarose pads were prepared as described in the Wormatlas. Generally, 1–6% could be used to immobilize worms or embryos with different degrees of compression against the cover slip. In our study, we used 3% agarose pads [33, 34]. Around 15 worms in 4 μL of M9 were placed on a cover slip, and then an equivalent amount of 20 mM of levamisole was added resulting in a final concentration of 10 mM of levamisole. The cover slip was placed on top of the agarose and then imaged using an inverted microscope (Nikon® Eclipse Ti-U) equipped with a CCD camera (Andor® Clara E) under fluorescein isothiocyanate (FITC) filter with 100 ms exposure time. To study the morphology of the mitochondria, the same procedure was used with a final concentration of 15 mM levamisole. The mitochondrial morphology in body wall muscles of the tail was examined using a confocal microscope Leica TCS SP8 with 63× objective (with 1.5× zoom). For the microfluidic-based imaging, each imaging device (Fig. 1) was inspected under a stereoscope to ensure the channel was not clogged. Two hundred microliters of worm suspension was pipetted on a clean petri dish. Using a stereoscope (Leica® M125), 5 μL containing ~30 worms were picked with pipette and loaded into the inlet reservoir of the device. This loading procedure was necessary in order to ensure that no debris or bacteria residue was loaded into the microfluidic device which could potentially clog the constriction channel. After loading, the inlet reservoir was sealed with a tubing connected to a syringe that was preloaded with M9 buffer solution. The outlet reservoir was connected to a falcon tube for waste collection via a tubing. With a gentle finger tap of the plunger of the syringe, a single worm was pushed into the constriction channel for imaging. The constriction channel was 52 μm deep with a width of 40 μm and 50 μm at the bottom and roof of the channel, respectively (see Fig. 1b). The fluorescence signal of the mitochondria in the body wall muscles of the worm was captured using an inverted microscope (Nikon® Eclipse Ti-E) equipped with a CCD camera (Andor® iXon Ultra 897 EMCCD). The images were taken at 10× magnification with a FITC filter. The exposure time was kept constant throughout the experiment at 100 ms. Figure 1 Open in new tabDownload slide Microfluidic imaging chip for C. elegans. (a) Schematic diagram and optical image (right) of the microfluidic device showing a notch integrated into the constriction channel to stop a worm in order to take images. (b) Pictorial image (left) and scanning electron micrograph of the device showing the dimensions of the notch and constriction channel sections. The cross-section was slightly trapezoidal due the high thickness of SU-8 layer. The distance between the entrance of the constriction channel and the notch was ~1100 μm allowing only for a single worm to enter into the constriction channel at a time. Figure 1 Open in new tabDownload slide Microfluidic imaging chip for C. elegans. (a) Schematic diagram and optical image (right) of the microfluidic device showing a notch integrated into the constriction channel to stop a worm in order to take images. (b) Pictorial image (left) and scanning electron micrograph of the device showing the dimensions of the notch and constriction channel sections. The cross-section was slightly trapezoidal due the high thickness of SU-8 layer. The distance between the entrance of the constriction channel and the notch was ~1100 μm allowing only for a single worm to enter into the constriction channel at a time. Image analysis Fluorescence intensity of the images obtained from both the agarose pad and the microfluidic device was quantified using ImageJ (Fiji®). Due to size variation in the worms, the measured intensities were normalized with the area of the worm. This was obtained by dividing the intensity by the size of the worm. This normalization eliminated the possibility of the subtle size changes affecting the final result. Thrashing force measurement The thrashing forces of the worms were obtained using elastomeric micropillars. The deflection of each micropillar, due to the thrashing behavior of the worm, was optically recorded and analyzed. Due to the non-linear deflection of the micropillars, the conversion of the micropillar displacements to thrashing force was performed using a custom finite element model which we have previously reported [35]. The maximum deflection of each micropillar was obtained and the average taken as per the number of worm samples used. For this study, young adult animals were used, similar to those used for fluorescence intensity quantification. More details of the methodology used in the thrashing force can be found in our previous paper [35]. Machine learning-based image classification Over the past few years, convolution neural networks (CNNs) have achieved remarkable success in various application domains such as computer vision, medical image analysis, speech recognition and autonomous driving. In traditional machine learning techniques, a domain expert needs to identify the features to make patterns more visible for learning algorithms. The major advantage of CNN lies in its automatic extraction (i.e. feature learning) of discriminating features at multiple levels of abstraction from raw inputs. This eliminates the need of domain expertise and complicated feature extraction. The training of a CNN from scratch is very challenging as it requires tons of labeled images, large computational and memory resources and expertise to define the CNN architecture and hyper-parameters tuning. An alternative to building a CNN model from scratch is to use pretrained models and fine tune to fit it for another purpose by using transfer learning. In medical image analysis, Tajbakhsh et al. [36] showed that fine tuning of pretrained deep CNN can avoid the extensive training of CNNs from scratch. Gopakumar et al. [37] leveraged transfer learning to predict cell class using ~30 cell images using a CNN, pretrained on the ImageNet database [38]. In transfer learning, an existing model is trained and subsequently used for classification. The use of convolutional neural network (CNN) offers the advantage that it can automatically extract multiple important features for image classification rather than the singular feature of fluorescence intensity. As such, it could use morphological features, such as size and distribution, for more accurate classification. In this paper, we use a pretrained deep CNN architecture (MobileNetV2) to classify between a hyperglycemic and non-hyperglycemic (healthy control) C. elegans worm from its image with transfer learning. The MobileNetV2 has two modules: feature extraction and classification (see Supplementary Fig. S1). The detailed description of each module can be found in [38]. The feature extraction module was used as a black box for obtaining features which were used by the classification module. The classification module was altered and two classes were added for the classification. After the classification, we trained the model (supervised learning) using 90% of the total data in the dataset (342 hyperglycemic and 342 non-hyperglycemic worms, i.e. a total of 684 images). For evaluating the accuracy of the trained model, we used the remaining 10% of the data for blind testing. The training data were further divided into training and validation sets. We have performed 3000 training steps with the training data by setting the hyper-parameter learning rate η = 0.01. Supplementary Figure S2 shows the plot of training and validation accuracy with respect to the training step. After the retraining process, we have achieved 87.2% accuracy on the blind test set. Viability, reproductive fitness and locomotory assay All of the assays were carried on young adult worms under two conditions on the control (worms washed off the NGM plate without being loaded into the chip), and the worms collected from the microfluidic device after imaging. In the viability assay, worm growth was observed on NGM plates seeded with OP50 for 3 days. The reproductive fitness (egg count after 48 hours) and locomotory behavior (velocity, number of bends, and amplitude) were recorded according to standard protocol of the WormBook [39]. Pictures were taken using a CCD camera (LEICA® MC170 HD) mounted on a stereoscope (LEICA® M125). RESULTS AND DISCUSSION Manual device operation The use of agarose pad for C. elegans imaging allowed limited number of worms, ~10–15 per pad, thereby necessitating for multiple agarose pads for imaging a significant number of worms. This batchwise imaging process in addition to having to prepare fresh pads for each experiment made this technique laborious and inefficient for imaging studies requiring large number of worms. Our microfluidic device offered improvement in throughput while enabling ease of use by using a straight channel and constriction notch (Fig. 1b). The design criteria for the notch were based on our previous study [35]. Several worms (~50) were manually loaded into the inlet reservoir and pushed one by one into the straight constriction channel by a gentle tap of the syringe plunger to load only one worm into the constriction channel at a time. However, since several worms were loaded into the inlet reservoir at the same time, there was a possibility that the entrance of the constriction channel could be clogged with several worms trying to enter the constriction channel simultaneously. In such cases, the syringe plunger was simply released, causing dispersion of worms from the entrance. This sequence of pushing and releasing of the plunger was repeated for 2–3 times to clear the clogged constriction channel entrance and to allow a single worm entering the constriction channel. Once entered into the constriction channel, each worm was significantly restricted in its free undulating motion due to the channel size that was on the same order in width similar compared to the diameter of the young adult worm (~45 μm). At the same time, the entire body was elongated in the constriction channel which was beneficial for fluorescence body imaging. An additional narrow stop in the form of a notch in the middle of the constriction channel allowed to maintain the same location of imaging (Fig. 1b). Once imaging was completed, the worm imaged was further pushed through the notch region of the constriction channel to the collection reservoir on the other side while allowing simultaneous loading the next worm to the imaging zone. The continuous imaging of worms using our microfluidic device resulted in a throughput of ~10 worms/min. Also, the easiness of device operation and complete elimination of a syringe pump were major advantages of our microfluidic imaging chip. Figure 2 Open in new tabDownload slide Quantitative fluorescence imaging of mitochondria in body wall muscles in two different body orientations. (a) In the lateral orientation, fluorescent mitochondria in the body wall muscles can only be viewed along the perimeter of the worm body due to the vertical alignment of the muscle quadrants relative to the substrate allowing only partial quantification of the fluorescence intensity of the muscle cells. (b) In the dorso-ventral orientation, however, the muscle quadrants are orientated parallel to the substrate allowing quantitative measurement of fluorescent mitochondria in the two neighboring muscle quadrants at the same time along the entire length of the worm. Images were captured by focusing the objective at the bottom of the channel at 10× (center) and 40× (right) magnification. (c) For laterally oriented worms, there was no significant difference in the fluorescence intensity of the mitochondria between worms cultured in the absence and presence of 400 mM glucose. In the dorso-ventral orientation, however, the fluorescence intensity of the mitochondria showed a significant difference between worms cultured in the absence (control) and presence of 400 mM glucose. N = 30; P < 0.0001. Significant differences were analyzed using student t-test. N = 28; P > 0.05. Figure 2 Open in new tabDownload slide Quantitative fluorescence imaging of mitochondria in body wall muscles in two different body orientations. (a) In the lateral orientation, fluorescent mitochondria in the body wall muscles can only be viewed along the perimeter of the worm body due to the vertical alignment of the muscle quadrants relative to the substrate allowing only partial quantification of the fluorescence intensity of the muscle cells. (b) In the dorso-ventral orientation, however, the muscle quadrants are orientated parallel to the substrate allowing quantitative measurement of fluorescent mitochondria in the two neighboring muscle quadrants at the same time along the entire length of the worm. Images were captured by focusing the objective at the bottom of the channel at 10× (center) and 40× (right) magnification. (c) For laterally oriented worms, there was no significant difference in the fluorescence intensity of the mitochondria between worms cultured in the absence and presence of 400 mM glucose. In the dorso-ventral orientation, however, the fluorescence intensity of the mitochondria showed a significant difference between worms cultured in the absence (control) and presence of 400 mM glucose. N = 30; P < 0.0001. Significant differences were analyzed using student t-test. N = 28; P > 0.05. Impact of orientation Caenorhabditis elegans has 95 body wall muscle cells longitudinally organized in pairs in four quadrants of the worm body [40]. The quadrants are situated subventrally and subdorsally. This arrangement of muscles cells in C. elegans makes it necessary to control the body orientation for accurate 2D optical evaluation of the body wall muscles. C. elegans moving on an agarose plate have a lateral body orientation (Fig. 2a) where both the ventral and dorsal nerve cords can be seen on either side of the worm [41, 42]. This orientation is mostly used in analysis of neuronal processes that travel along the anterior–posterior axis [41]. On the other hand, worms could be oriented dorso-ventrally such that either the ventral or dorsal side of the worm faces up and only one of the nerve cords can be observed (Fig. 2b). From the imaging point of view, this orientation was more preferred for evaluation of body wall muscle cells because the dorsal and ventral quadrants could be imaged without significant overlap when viewing from underneath. To the best of our knowledge, only one study [43] has reported the ability to reorient the worm to a dorso-ventral orientation. In the study, worms were captured pneumatically using a glass capillary integrated into a microfluidic device. The other end of the capillary was connected to a 3D-printed fixture used to rotate the capillary and captured worm simultaneously. However, this technique obviously required more complex manipulation system to build and operate. It could also mechanically damage the worm as it requires 30% of the worm to be trapped in the glass capillary. In addition, the device had a relatively low throughput (20–30 worms/h) compared to our device (~600 worms/h). To our advantage, this orientation was mostly obtained in ~85% of cases when a worm was loaded into the constriction channel. A potential explanation for taking the dorso-ventral orientation inside a narrow microfluidic channel was hypothesized as follows. When entering the constriction channel, it “sensed” the tight clearance in the lateral direction which does not allow continuing its undulating motion in the lateral direction for propulsion as was the case on agarose pad. In order to propel its body by undulating motion in the narrow microchannel, it was most likely forced to rotate its body by 90° (Fig. 3; Supplementary Video S1) so it could undulate its body in the vertical direction along the channel depth where there was more space between its body and the top of the channel due to the higher aspect ratio (AR = width/height) of ~1.3 (= 52/40 μm) compared to ~0.8 (= 40/52 μm)in the lateral position (Fig. 2b). To keep worms in the lateral orientation inside microfluidic channels with aspect ratio of ~1, some studies have reported incorporation of geometric constraints to the channel such as the U-shaped curvature [42] and stepped taper [30] in the height. We believe that this behavioral adaptation in a narrow confinement biasing its body orientation toward the vertical direction with more space than in the lateral/horizontal direction with AR > 1 probably caused a change of the lateral to the dorso-ventral orientation of the C. elegans worm body. For the quantitative fluorescence imaging of the mitochondria in whole body wall muscles, this body adaptation turned out to be more advantageous. This finding could open up a new exciting approach in the future to control the worm body orientation based on the geometrical constraints of the microchannel in use. Figure 3 Open in new tabDownload slide Time-lapse images showing orientation adaption of a single worm in the microfluidic channel. The worm entered the channel in the lateral orientation, as shown at t = 0 s. Within ~2 s in the channel, it started to reorient itself from the lateral to the dorso-ventral orientation. After 2 s, the worm had completely switched to the dorso-ventral orientation using its proprioception. The worm remained in this new orientation. The corresponding video is provided as Supplementary Video S1. Figure 3 Open in new tabDownload slide Time-lapse images showing orientation adaption of a single worm in the microfluidic channel. The worm entered the channel in the lateral orientation, as shown at t = 0 s. Within ~2 s in the channel, it started to reorient itself from the lateral to the dorso-ventral orientation. After 2 s, the worm had completely switched to the dorso-ventral orientation using its proprioception. The worm remained in this new orientation. The corresponding video is provided as Supplementary Video S1. Using the microfluidic imaging chip, we compared the intensities for worms cultured in the absence (control) and presence of 400 mM glucose concentration. For worms imaged with the dorso-ventral orientation, as was the case in the microfluidic channel, the result showed a significant decrease by 26.7% in fluorescence intensity from the mitochondria of the body wall muscles cultured with 400 mM glucose as compared to the negative control (Fig. 2c). However, when imaged in the lateral orientation on agarose pad, no significant change could be measured in the fluorescence intensities between the both culture conditions. This finding implied that the dorso-ventral orientation, the forced body orientation inside a narrow microchannel, was more suitable for image-based evaluation of body wall muscles. It provided a view of the two dorsal or ventral quadrants of the muscle cells without significant overlap when viewing them from the bottom as was the case in the lateral orientation. Effect of hyperglycemia Using both the microfluidic chip and agarose pad, we quantified the fluorescence intensity of the mitochondria in the body wall muscles of worms in the absence (control) and presence of 200 mM and 400 mM glucose (Fig. 4a) mimicking the glucose concentration of type 2 diabetes in humans [18]. Figure 4 Open in new tabDownload slide Analysis of fluorescence intensity of mitochondria in body wall muscle cells using ImageJ. (a) Fluorescence images from agarose pad and microfluidic device for worms cultured with and without glucose. (b) Fluorescence intensity for control and hyperglycemic worms with and without drug treatment. On agarose pad, no significant difference was observed across the samples. However, images obtained using microfluidic device showed up to 31% decrease in the fluorescence intensity of the mitochondria. Significant differences were analyzed using student t-test. N = 60; P < 0.0001 error bars indicate SEM. Figure 4 Open in new tabDownload slide Analysis of fluorescence intensity of mitochondria in body wall muscle cells using ImageJ. (a) Fluorescence images from agarose pad and microfluidic device for worms cultured with and without glucose. (b) Fluorescence intensity for control and hyperglycemic worms with and without drug treatment. On agarose pad, no significant difference was observed across the samples. However, images obtained using microfluidic device showed up to 31% decrease in the fluorescence intensity of the mitochondria. Significant differences were analyzed using student t-test. N = 60; P < 0.0001 error bars indicate SEM. Using the microfluidic device, the normalized fluorescence intensity decreased by 28.5% and 31.1% for 200 and 400 mM glucose-seeded worms, respectively (Fig. 4b). This result implied that hyperglycemic conditions impacted the mitochondria integrity in the worm muscles. A recent study [19] has shown that exposure of C. elegans to high-glucose concentration results in swelling of mitochondria and endoplasmic reticulum in germ and muscle cells and an accumulation of damaged mitochondria. The study also reported that high-glucose concentration upregulated mitophagy genes, DCT-1 and PINK-1 [44]. These findings seemed to suggest that the change of fluorescence intensity of GFP that is expressed in mitochondria of body wall muscles of the transgenic strain SJ4103 (zcls14[myo-3::GFP(mit)] at hyperglycemic conditions might be related to mitophagy. However, normalized intensities recorded from the agarose pads showed no significant change in fluorescence intensity. This result could be attributed to the orientation of worms measured on agarose pad which were mostly (~90%) oriented in the lateral position. This orientation did not allow for accurate quantification of fluorescence intensity of body wall muscles as discussed. In the microfluidic chip on the other hand, ~85% of the worms showed the dorso-ventral position. This result further evidenced that our microfluidic device would be suitable especially for quantification of fluorescence intensity of the mitochondria in body wall muscles. Using 25 mM metformin, the fluorescence intensity of the worm did not decline on the microfluidic platform. Similar fluorescence intensity was recorded to that of the control worms. This result indicated that metformin reduced the impact of hyperglycemia on the integrity of the mitochondria in the body wall muscles. Prior studies have shown that metformin increases the accumulation of andesine monophosphate (AMP), resulting in activation of 5′ AMP-activated protein kinase (AMPK) pathway [45–47]. Subsequently, AMPK promotes general metabolism and mitochondrial respiration. Our result was in agreement with the previous studies that fully functional mitochondria were required to obtain glucose-stimulated insulin and the dysfunction of the same could lead to the development of type 2 diabetes [14]. In our previous work [35], we demonstrated that the thrashing force of C. elegans degraded with increasing glucose concentration. To investigate the relationship between the decrease of fluorescence signal intensity of mitochondria and the thrashing force as its functional output, we quantified the thrashing force of worms cultured with and without high-glucose concentration (Fig. 5). Our result showed a ~23.2% decrease in the average thrashing force of worms seeded with 400 mM glucose (~13.1 μN) as compared to those seeded in the absence of glucose (~17 μN). When treated with diabetes drug, metformin, the thrashing force was restored, and there was no significant difference compared to control worms. Our biophysical functional assay result corroborated the fluorescence imaging result that hyperglycemic conditions degraded the mitochondria of body wall muscles of C. elegans impacting its thrashing force as functional output. In addition to this functional assay data, the confocal microscopy images, as shown in Figure 6, also showed a morphological change of the mitochondria under 400 mM glucose indicating a mitochondrial dysfunction in hyperglycemic conditions. Figure 5 Open in new tabDownload slide Effect of hyperglycemia on thrashing force using young adult worms. Compared to worms cultured with no glucose (control), hyperglycemic worms exhibited a thrashing force decrease of at least ∼23.2%. There was a significant difference in the measured thrashing forces between healthy and glucose-seeded worms (P < 0.0001). No significant difference was observed between worms seeded with 400 mM glucose + 25 mM metformin and healthy worms (P > 0.1). Significant differences were analyzed using two-way ANOVA with Tukey’s multiple comparison; error bars indicate SEM, N = 27 for all samples. Figure 5 Open in new tabDownload slide Effect of hyperglycemia on thrashing force using young adult worms. Compared to worms cultured with no glucose (control), hyperglycemic worms exhibited a thrashing force decrease of at least ∼23.2%. There was a significant difference in the measured thrashing forces between healthy and glucose-seeded worms (P < 0.0001). No significant difference was observed between worms seeded with 400 mM glucose + 25 mM metformin and healthy worms (P > 0.1). Significant differences were analyzed using two-way ANOVA with Tukey’s multiple comparison; error bars indicate SEM, N = 27 for all samples. Figure 6 Open in new tabDownload slide Change of mitochondria morphology from control to the hyperglycemic condition observed using confocal microscopy: (a) control sample with score 1 showed healthy morphology with highly abundant networked mitochondria, and (b) hyperglycemic sample with score 2 show damaged mitochondria morphology with less abundant and fewer networked mitochondria when exposed to 400 mM glucose. Figure 6 Open in new tabDownload slide Change of mitochondria morphology from control to the hyperglycemic condition observed using confocal microscopy: (a) control sample with score 1 showed healthy morphology with highly abundant networked mitochondria, and (b) hyperglycemic sample with score 2 show damaged mitochondria morphology with less abundant and fewer networked mitochondria when exposed to 400 mM glucose. Study of mitochondria morphology under hyperglycemic conditions using confocal microscopy Confocal microscopy allowed for detailed characterization of the mitochondria morphology to observe changes under hyperglycemic conditions and corroborate the results from fluorescence intensity quantification. As previously reported in reference [48], we also observed a change in mitochondria morphology and its abundance as well. Furthermore, 3D imaging also showed that the traditional agarose pad-based imaging cannot control the body orientation precisely. The confocal images taken from two different worms on the agarose pad showed the lateral orientation with two muscle quadrants (indicated with a white arrow) stacking on each other (see Supplementary Fig. S3a and b) and the dorso-ventral orientation for comparison with two muscle quadrants lying parallel next to each other in the plane (see Supplementary Fig. S3c). To evaluate mitochondria health qualitatively, two scores have been used; score 1 indicated healthy, highly abundant and networked mitochondria, while score 2 showed less abundant, fewer networked and fragmented mitochondria [48]. Seventy nine percent of control worms showed healthy mitochondrial structure (Fig. 6a). Under hyperglycemic conditions, however, only 67% of the worms showed less abundant and fewer networked mitochondria (score 2) (Fig. 6b). These confocal microscopy images clearly showed a morphological change of the mitochondria under 400 mM glucose indicating a mitochondrial dysfunction in hyperglycemic conditions and, thus, explained the muscle force loss as shown in the biophysical functional assay (see Fig. 5). Machine learning-based image classification result Image classification using machine learning is becoming increasingly widespread in biology [49, 50]. Our microfluidic-based imaging platform with its precisely controlled environment with a sequential loading and immobilization scheme of worms instead of the batchwise scheme on agarose pad that enables a reproducible body orientation could be an ideal source of images that can be fed to the automatic image classification program for high-throughput image analysis in future. To demonstrate this applicability, we used the images from the microfluidic–imaging platform and assessed its accuracy even though only a limited number of images were available for proper training of the CNN model. The accuracies of our CNN classification are shown Table 1. We obtained an accuracy of 56 and 79% for classification of images with a test sample size of 16 and 76 obtained from agarose pad and microfluidic device, respectively. The low number tested on agarose pad was due to the throughput limitation of the technique. Also, the classification also showed better precision (78%), specificity (77%) and sensitivity (81%) with the microfluidic device compared to the agarose pad. It is clear that the images obtained from our microfluidic device are more suitable for machine learning classification in terms of the image quality and the number. This result in addition to the fluorescence intensity analysis discussed in the previous section which further underlines the advantage of our microfluidic for imaging of the mitochondria in body wall muscle cells of C. elegans with subsequent image classification based on machine learning. Table 1 Comparison of CNN model classification accuracy of fluorescence images obtained from agarose pads and microfluidic device. Parameters . Agarose pad-based imaging . Microfluidic-based imaging . Accuracy 0.56 0.79 Misclassification rate/Error rate 0.44 0.21 True positive rate/sensitivity 0.62 0.81 False positive rate 0.50 0.23 True negative rate/specificity 0.50 0.77 Precision 0.55 0.78 Prevalence 0.50 0.50 Number of samples for training 129 684 Number of samples for blind test 16 76 Parameters . Agarose pad-based imaging . Microfluidic-based imaging . Accuracy 0.56 0.79 Misclassification rate/Error rate 0.44 0.21 True positive rate/sensitivity 0.62 0.81 False positive rate 0.50 0.23 True negative rate/specificity 0.50 0.77 Precision 0.55 0.78 Prevalence 0.50 0.50 Number of samples for training 129 684 Number of samples for blind test 16 76 Open in new tab Table 1 Comparison of CNN model classification accuracy of fluorescence images obtained from agarose pads and microfluidic device. Parameters . Agarose pad-based imaging . Microfluidic-based imaging . Accuracy 0.56 0.79 Misclassification rate/Error rate 0.44 0.21 True positive rate/sensitivity 0.62 0.81 False positive rate 0.50 0.23 True negative rate/specificity 0.50 0.77 Precision 0.55 0.78 Prevalence 0.50 0.50 Number of samples for training 129 684 Number of samples for blind test 16 76 Parameters . Agarose pad-based imaging . Microfluidic-based imaging . Accuracy 0.56 0.79 Misclassification rate/Error rate 0.44 0.21 True positive rate/sensitivity 0.62 0.81 False positive rate 0.50 0.23 True negative rate/specificity 0.50 0.77 Precision 0.55 0.78 Prevalence 0.50 0.50 Number of samples for training 129 684 Number of samples for blind test 16 76 Open in new tab Furthermore, we evaluated our model with 121 images of metformin-treated worms. The model classified 74.3% of the images as healthy worms which was close to the accuracy obtained with the classification of healthy worms. This result implies that our model can be used to classify the healthiness of worms after drug treatment as well. The classification result agreed with our fluorescence intensity study which showed no significant difference between control and metformin-treated worms as shown in the case of microfluidic chip in Figure 4. Viability, reproductive fitness and locomotory assay The viability assay presented the ability of the worms to grow to the adult stage and reproduce normally as presented (Supplementary Fig. S4a) over 3 days in both groups. The reproductive fitness assay showed that both the control worms and worms used in the microfluidic device had similar number of progeny 55 and 57, respectively. The locomotory assay included investigation for three parameters: amplitude, body bends and velocity (see Supplementary Fig. S4b–d). There was no significant difference between the body amplitude of the control group and worms used in the microfluidic device with amplitudes of 110 and 105 μm, respectively. Our result also showed no significant difference in number of body bends per 3 min (66 for control and 72 for worms used in the chip) and worm velocity (6005 μm/min for control and 5497 μm/min for worms used in the chip). The assay has validated that the imaging device have no effect on the worm locomotory. Based on the visual assessment for viability assay and statistical analysis for reproductive fitness and locomotory assay, the worms did not exhibit any morphological changes, decrease in their fertility or a change in their motility behavior. However, there might be changes in gene expression for which further experiments will be necessary for validation. CONCLUSION We have developed a simple-to-use microfluidic device to quantitatively evaluate the effect of high-glucose concentration on the mitochondria of C. elegans. Our device utilized a constriction notch embedded within a straight channel to temporarily immobilize the worm for imaging while allowing for viability and locomotory integrity of worms forced through the narrow notch zone for collection. The design of the microfluidic device favored the innate behavior of the worm within a straight channel whereby the worm was oriented more dorso-ventrally than laterally. The narrow channel in the lateral direction was forcing worms to change their body orientation by rotation and to assume the dorso-ventral rather than lateral orientation for propulsion by undulation. We have shown that this forced orientation was particularly advantageous for accurate quantification of the mitochondrial fluorescence intensity in the body wall muscle cells of C. elegans. On agarose pad, where worms were laterally oriented for propulsion by lateral undulation motion, no difference in the fluorescence intensity was measured for C. elegans cultured in the absence and presence of high concentration (200 and 400 mM) of glucose. However, when the dorso-ventral orientation was adapted as a behavioral response to the narrow constriction channel, where there was more space with AR > 1 in the height than in the width, we could measure a decrease in mitochondrial fluorescence intensity up to ~31% under hyperglycemic conditions. Using the same device, we quantified the efficacy of metformin, a common drug for treating type 2 diabetes, by measuring changes in the mitochondrial fluorescence intensity upon treatment, whereas on agarose pad we did not measure any significant changes in fluorescence intensity upon drug treatment. The confocal images confirmed a significant change in the mitochondria of body wall muscles up on exposure to hyperglycemic conditions supporting the findings on the microfluidic chip. Using pretrained deep CNN architecture, the images acquired from the microfluidic channel could further automatically be classified into healthy and hyperglycemic worms at an accuracy of 79%, even with the limited number of images available for training. Our PDMS-based microfluidic device was easy to fabricate and low-cost as well as operated completely manually using just a syringe at relatively higher throughput compared to agarose pad based on the sequential loading and immobilization scheme. Most importantly, it allows to induce a preferred orientation for quantitative fluorescence imaging of the mitochondria where subtle changes are difficult to quantify because of the innate lateral orientation with significant overlap of the body wall muscles when imaging from the top or bottom. Our device has the potential to be multiplexed to have multiple channels thereby further improving throughput. With all these technological advantages, it has a potential to become a useful immobilization tool for imaging of C. elegans. ACKNOWLEDGEMENTS The authors would like to acknowledge support from Dr. Hala Fahs, Suma Gopinathan and Fathima Refai of the Center for Genomics and Systems Biology, NYU Abu Dhabi, in sourcing the C. elegans strains used in this work. We acknowledge Navajit Baban’s support in drawing the 3D schematic images. We are also thankful for the support of NYUAD microfabrication core facility for the device fabrication and Rachid Rezgui for training on confocal microscopy. Conflict of interest statement No conflict of interest. FUNDING This work was supported by the Al Jalila Foundation (AJF201633). S.S. was supported by the NYUAD Global PhD Fellowship program. References 1. Gouspillou G , Hepple RT. Editorial: mitochondria in skeletal muscle health, aging and diseases . Front Physiol 2016 ; 7 : 446 . Google Scholar Crossref Search ADS PubMed WorldCat 2. Putti R , Migliaccio V, Sica R et al. . Skeletal muscle mitochondrial bioenergetics and morphology in high fat diet induced obesity and insulin resistance: focus on dietary fat source . Front Physiol 2016 ; 6 : 426 . Google Scholar Crossref Search ADS PubMed WorldCat 3. Brookes PS , Yoon Y, Robotham JL et al. . Calcium, ATP, and ROS: a mitochondrial love-hate triangle . Am J Physiol Physiol 2004 ; 287 : C817 – 33 . Google Scholar Crossref Search ADS WorldCat 4. Min K , Smuder AJ, Kwon O et al. . Mitochondrial-targeted antioxidants protect skeletal muscle against immobilization-induced muscle atrophy . J Appl Physiol 2011 ; 111 : 1459 – 66 . Google Scholar Crossref Search ADS PubMed WorldCat 5. Bernardi P , Bonaldo P. Mitochondrial dysfunction and defective autophagy in the pathogenesis of collagen VI muscular dystrophies . Cold Spring Harb Perspect Biol 2013 ; 5 : a011387 . Google Scholar Crossref Search ADS PubMed WorldCat 6. Godin R , Daussin F, Matecki S et al. . Peroxisome proliferator-activated receptor γ coactivator 1-α gene transfer restores mitochondrial biomass and improves mitochondrial calcium handling in post-necrotic mdx mouse skeletal muscle . J Physiol 2012 ; 590 : 5487 – 502 . Google Scholar Crossref Search ADS PubMed WorldCat 7. Hepple RT . Mitochondrial involvement and impact in aging skeletal muscle. Vol. 6, Frontiers in aging . Neuroscience 2014 ; 6 : 211 . OpenURL Placeholder Text WorldCat 8. Picard M , Ritchie D, Wright KJ et al. . Mitochondrial functional impairment with aging is exaggerated in isolated mitochondria compared to permeabilized myofibers . Aging Cell 2010 ; 9 : 1032 – 46 . Google Scholar Crossref Search ADS PubMed WorldCat 9. Chabi B , Ljubicic V, Menzies KJ et al. . Mitochondrial function and apoptotic susceptibility in aging skeletal muscle . Aging Cell 2008 ; 7 : 2 – 12 . Google Scholar Crossref Search ADS PubMed WorldCat 10. Short KR , Bigelow ML, Kahl J et al. . Decline in skeletal muscle mitochondrial function with aging in humans . Proc Natl Acad Sci U S A 2005 ; 102 : 5618 – 23 . Google Scholar Crossref Search ADS PubMed WorldCat 11. Dirks AJ , Leeuwenburgh C. Aging and lifelong calorie restriction result in adaptations of skeletal muscle apoptosis repressor, apoptosis-inducing factor, X-linked inhibitor of apoptosis, caspase-3, and caspase-12 . Free Radic Biol Med 2004 ; 36 : 27 – 39 . Google Scholar Crossref Search ADS PubMed WorldCat 12. Picard M , Azuelos I, Jung B et al. . Mechanical ventilation triggers abnormal mitochondrial dynamics and morphology in the diaphragm . J Appl Physiol 2015 ; 118 : 1161 – 71 . Google Scholar Crossref Search ADS PubMed WorldCat 13. Goodpaster BH . Mitochondrial deficiency is associated with insulin resistance . Diabetes 2013 ; 62 : 1032 – 5 . Google Scholar Crossref Search ADS PubMed WorldCat 14. Newsholme P , Gaudel C, Krause M. Mitochondria and diabetes. An intriguing pathogenetic role . Adv Exp Med Biol 2012 ; 942 : 235 – 47 . Google Scholar Crossref Search ADS PubMed WorldCat 15. Markaki M , Tavernarakis N. Modeling human diseases in Caenorhabditis elegans . Biotechnol J 2010 ; 5 : 1261 – 76 . Google Scholar Crossref Search ADS PubMed WorldCat 16. Kimura KD , Tissenbaum HA, Liu Y et al. . Daf-2, an insulin receptor-like gene that regulates longevity and diapause in Caenorhabditis elegans . Science 1997 ; 277 : 942 – 6 . Google Scholar Crossref Search ADS PubMed WorldCat 17. Schulz TJ , Zarse K, Voigt A et al. . Glucose restriction extends Caenorhabditis elegans life span by inducing mitochondrial respiration and increasing oxidative stress . Cell Metab 2007 ; 6 : 280 – 93 . Google Scholar Crossref Search ADS PubMed WorldCat 18. Schlotterer A , Kukudov G, Bozorgmehr F et al. . C. elegans as model for the study of high glucose-mediated life span reduction . Diabetes 2009 ; 58 : 2450 – 6 . Google Scholar Crossref Search ADS PubMed WorldCat 19. Alcántar-Fernández J , González-Maciel A, Reynoso-Robles R et al. . High-glucose diets induce mitochondrial dysfunction in Caenorhabditis elegans . PLoS One 2019 ; 14 : e0226652 . Google Scholar Crossref Search ADS PubMed WorldCat 20. Cornaglia M , Lehnert T, Gijs MAM. Microfluidic systems for high-throughput and high-content screening using the nematode Caenorhabditis elegans . Lab Chip 2017 ; 17 : 3736 – 59 . Google Scholar Crossref Search ADS PubMed WorldCat 21. Kamili F , Lu H. Recent advances and trends in microfluidic platforms for C. elegans biological assays . Annu Rev Anal Chem 2018 ; 11 : 245 – 64 . Google Scholar Crossref Search ADS WorldCat 22. Zhuo W , Lu H, McGrath PT. Microfluidic platform with spatiotemporally controlled micro-environment for studying long-term: C. elegans developmental arrests . Lab Chip 2017 ; 17 : 1826 – 33 . Google Scholar Crossref Search ADS PubMed WorldCat 23. Sofela S , Sahloul S, Rafeie M et al. . High-throughput sorting of eggs for synchronization of: C. elegans in a microfluidic spiral chip . Lab Chip 2018 ; 18 : 679 – 87 . Google Scholar Crossref Search ADS PubMed WorldCat 24. Chokshi TV , Ben-Yakar A, Chronis N. CO2 and compressive immobilization of C. elegans on-chip . Lab Chip 2009 ; 9 : 151 – 7 . Google Scholar Crossref Search ADS PubMed WorldCat 25. Yanik MF , Cinar H, Cinar HN et al. . Functional regeneration after laser axotomy . Nature 2004 ; 432 : 822 . Google Scholar Crossref Search ADS PubMed WorldCat 26. Aubry G , Zhan M, Lu H. Hydrogel-droplet microfluidic platform for high-resolution imaging and sorting of early larval Caenorhabditis elegans . Lab Chip 2015 ; 15 : 1424 – 31 . Google Scholar Crossref Search ADS PubMed WorldCat 27. Hulme SE , Shevkoplyas SS, Apfeld J et al. . A microfabricated array of clamps for immobilizing and imaging C. elegans . Lab Chip 2007 ; 7 : 1515 – 23 . Google Scholar Crossref Search ADS PubMed WorldCat 28. Lee H , Kim SA, Coakley S et al. . A multi-channel device for high-density target-selective stimulation and long-term monitoring of cells and subcellular features in C. elegans . Lab Chip 2014 ; 14 : 4513 – 22 . Google Scholar Crossref Search ADS PubMed WorldCat 29. Cornaglia M , Krishnamani G, Mouchiroud L et al. . Automated longitudinal monitoring of in vivo protein aggregation in neurodegenerative disease C. elegans models . Mol Neurodegener 2016 ; 11 : 17 . Google Scholar Crossref Search ADS PubMed WorldCat 30. Mondal S , Hegarty E, Martin C et al. . Large-scale microfluidics providing high-resolution and high-throughput screening of Caenorhabditis elegans poly-glutamine aggregation model . Nat Commun 2016 ; 7 : 13023 . Google Scholar Crossref Search ADS PubMed WorldCat 31. Zhu G , Yin F, Wang L et al. . Modeling type 2 diabetes-like hyperglycemia in C. elegans on a microdevice . Integr Biol (United Kingdom) 2016 ; 8 : 30 – 8 . Google Scholar Crossref Search ADS WorldCat 32. Stiernagle T . Maintenance of C. elegans. WormBook : the online review of C. elegans biology . 2006 ; 1 – 11 . 33. Driscoll M . Mounting animals for observation with Nomarski DIC optics . Wormatlas 2008 . https://www.wormatlas.org/agarpad.htm OpenURL Placeholder Text WorldCat 34. Rothman J and Andrew S. Caenorhabditis elegans: Cell Biology and Physiology , Volume 107, 2nd Edition, Academic Press, 2012 . OpenURL Placeholder Text WorldCat 35. Sofela S , Sahloul S, Stubbs C et al. . Phenotyping of thrashing forces exerted by partially Immobilized C. elegans using elastomeric micropillar arrays . Lab Chip 2019 ; 19 : 3685 – 96 . Google Scholar Crossref Search ADS PubMed WorldCat 36. Tajbakhsh N , Shin JY, Gurudu SR et al. . Convolutional neural networks for medical image analysis: full training or fine tuning? IEEE Trans Med Imaging 2016 ; 35 : 1299 – 312 . Google Scholar Crossref Search ADS PubMed WorldCat 37. Gopakumar G , Hari Babu K, Mishra D. Cytopathological image analysis using deep-learning networks in microfluidic microscopy . J Opt Soc Am A 2017 ; 34 : 111 . Howard AG, Zhu M, Chen B et al. MobileNets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861 2017. https://arxiv.org/abs/1704.04861v1 Google Scholar Crossref Search ADS WorldCat 38. Howard AG , Zhu M, Chen B et al. . MobileNets: efficient convolutional neural networks for mobile vision applications . arXiv preprint arXiv : 1704.04861 2017 . https://arxiv.org/abs/1704.04861v1 OpenURL Placeholder Text WorldCat 39. Korb J . Locomotive activity as an indicator of acute chlorpyrifos toxicity in C. Elegans . WormBook 2010 ; 0 – 1 . http://www.wormbook.org/chapters/www_behavior/behavior.html. Girard LR, Fiedler TJ, Harris TW et al. (2007). WormBook: the online review of Caenorhabditis elegans biology. Nucleic acids research, 35(suppl_1), D472-D475. OpenURL Placeholder Text WorldCat 40. Girard LR , Fiedler TJ, Harris TW et al. . WormBook: the online review of Caenorhabditis elegans biology . Nucleic acids research 2007 ; 35 (suppl_1): D472 – D475 . Google Scholar Crossref Search ADS PubMed WorldCat 41. De Carlos CI , Porto DA, Gallotta I et al. . Automated screening of: C. elegans neurodegeneration mutants enabled by microfluidics and image analysis algorithms . Integr Biol (United Kingdom) 2018 ; 10 : 539 – 48 . Google Scholar Crossref Search ADS WorldCat 42. Cáceres I d C , Valmas N, Hilliard MA et al. . Laterally orienting C. elegans using geometry at microscale for high-throughput visual screens in neurodegeneration and neuronal development studies . PLoS One 2012 ; 7 : e35037 . Google Scholar Crossref Search ADS PubMed WorldCat 43. Ardeshiri R , Mulcahy B, Zhen M. et al. , A hybrid microfluidic device for on-demand orientation and multidirectional imaging of C. elegans organs and neurons . Biomicrofluidics 2016 ; 10 : 064111 . Google Scholar Crossref Search ADS PubMed WorldCat 44. Palikaras K , Lionaki E, Tavernarakis N. Coordination of mitophagy and mitochondrial biogenesis during ageing in C. elegans . Nature 2015 ; 521 : 525 – 8 . Google Scholar Crossref Search ADS PubMed WorldCat 45. Ahmad W , Ebert PR. Metformin attenuates Aβ pathology mediated through Levamisole sensitive nicotinic acetylcholine receptors in a C. elegans model of Alzheimer’s disease . Mol Neurobiol 2017 ; 54 : 5427 – 39 . Google Scholar Crossref Search ADS PubMed WorldCat 46. Onken B , Driscoll M. Metformin induces a dietary restriction–like state and the oxidative stress response to extend C. elegans healthspan via AMPK, LKB1, and SKN-1 . PLoS One 2010 ; 5 : e8758 . Google Scholar Crossref Search ADS PubMed WorldCat 47. De Haes W , Frooninckx L, Van Assche R et al. . Metformin promotes lifespan through mitohormesis via the peroxiredoxin PRDX-2 . Proc Natl Acad Sci U S A 2014 ; 111 : E2501 – 9 . Google Scholar Crossref Search ADS PubMed WorldCat 48. Hartman JH , Smith LL, Gordon KL et al. . Swimming exercise and transient food deprivation in Caenorhabditis elegans promote mitochondrial maintenance and protect against chemical-induced mitotoxicity . Sci Rep 2018 ; 8 : 1 – 16 . Google Scholar Crossref Search ADS PubMed WorldCat 49. Tang B , Pan Z, Yin K. et al. . Computational biology. Front Genet 2019 ; 10 : 214 . Crossref Search ADS PubMed 50. Webb S . Deep learning for biology . Nature 2018 ; 554 : 555 – 7 . Google Scholar Crossref Search ADS WorldCat © The Author(s) 2020. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: journals.permission@oup.com. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)
TI - Quantitative fluorescence imaging of mitochondria in body wall muscles of Caenorhabditis elegans under hyperglycemic conditions using a microfluidic chip
JF - Integrative Biology
DO - 10.1093/intbio/zyaa011
DA - 2020-06-19
UR - https://www.deepdyve.com/lp/oxford-university-press/quantitative-fluorescence-imaging-of-mitochondria-in-body-wall-muscles-a06xvHAyL7
SP - 150
VL - 12
IS - 6
DP - DeepDyve
ER -