In vivo continuous three-dimensional magnetic resonance microscopy: a study of metamorphosis in Carniolan worker honey bees (Apis mellifera carnica)

Magnetic resonance microscopy (MRM), in comparison to the conventional magnetic resonance imaging (MRI), has a higher spatial resolution and uses almost identical imaging routines. With MRM, higher gradient strengths permit better resolution and faster imaging, and the stronger the magnetic field, the greater the signal-to-noise ratio (Callaghan, 1991). Their optimal water/lipid content and their size make insects one of the ideal samples for the MRM studies (Hart et al., 2003). Previously, studies on the morphologies of locusts (Gassner and Lohman, 1987), ladybirds (Geoghegan et al., 2000), honey bees (Tomanek et al., 1996), diving beetles (Wecker et al., 2002), fruit flies (Meme et al., 2013) and, more recently, stag beetles (Seo, 2018) have been reported. Several studies were also focused on the development of insects, e.g. in wasps (Chudek et al., 1996) and tobacco hornworms (Conner et al., 1988; Rowland and Goodman, 2016), or on the metamorphosis of Lepidoptera (Goodman et al., 1995) and silk larvae (Mapelli et al., 1997). MRM efficiency has also been proven in the analysis of insect brains (Haddad et al., 2004; Michaelis et al., 2005; Watanabe et al., 2006) and the lipid content of the insects (Schilling et al., 2012). Another study reported on the parasite–host relationship in insects (Chudek et al., 1998).

In this feasibility study, the metamorphosis of the Carniolan worker honey bee (Apis mellifera carnica) from sealed larva up to the development of an adult was followed by the continuous three-dimensional (3D) MRM. The series of 3D images were analyzed to extract the information on transformation of the tracheal system, gastrointestinal tract and the muscles. Owing to its long duration and limited experimental time, the study was performed only on two specimens.

The experiment was performed in August 2018 and then in May 2020 on a single colony of Carniolan honey bees (Apis mellifera carnica Pollman 1879) in the Department of Biology, Biotechnical Faculty at the University of Ljubljana. On the day when the worker brood was capped, i.e. in mid-larval stage (day 0 in our study), a small section of it was removed from an experimental hive and immediately transported to the nearby MRI laboratory at the Jožef Stefan Institute. There, a single cell was carefully dissected from the brood, making sure that the cell remained capped and intact while its diameter was truncated to 10 mm, which was the diameter of the MRM probe in which the cell was then inserted (Fig. S1). Directly below the probe, an open container filled with humectant (saturated solution of NaNO₂ in water) was placed in order to maintain a constant humidity of 62±1% during the entire experiment (Greenspan, 1977), and the continuous MRM scanning of sample was started. During the scanning, which lasted for several days, i.e. until the formation of an adult bee, the sample was left uninterrupted in the MRM system and the temperature was kept constant at 34°C to within 0.5°C tolerance with the MRM probe conditioning system. The conditioning system had no effect on the performance of the MRM system. Owing to the restricted availability of the MRM system and its long experimental duration, the analysis was performed only on two worker bee specimens. To gain additional insight into the intra-specimen variations, a similar experiment was undertaken with the drones of the same hive, which were scanned simultaneously by having multiple drones in the same imaging field of view. A section of freshly capped drone brood was inserted into a 30 mm MRM probe and the scanning was initiated immediately. The drone sample was continuously scanned until the formation of the adult drones such that the temperature and humidity control factors remained the same as in the case of the worker bee experiment.

The study was performed on an MRM system consisting of a 9.4 T (400 MHz proton frequency) high-resolution superconducting vertical-bore magnet (Jastec, Tokyo, Japan), a Bruker Micro 2.5 gradient system (Bruker, Ettlingen, Germany) and a Tecmag Redstone spectrometer (Tecmag, Houston, TX, USA). Worker and drone bees were scanned continuously using the 3D T₁-weighted spin-echo imaging sequence (Callaghan, 1991; Vlaardingerbroek and Boer, 1996). Scanning parameters for the worker bees were: field of view (FOV) 20×10×10 mm³, imaging matrix 256×128×128 voxels, echo time (T_E) 2.2 ms, repetition time (T_R) 218 ms and number of signal averages 4. For the sample containing the drone bees, these parameters were: FOV 30×30×30 mm³, imaging matrix 256×256×256 voxels, T_E 2.0 ms, T_R 219 ms and number of signal averages 2. Imaging resolution was equal to 78 and 117 μm for the bee and the drone experiment, respectively, in all three spatial directions. In contrast, the temporal resolution of the experiments (scan time for a single 3D image) for the worker bee and the drone samples were 4 and 8 h, respectively.

Histological sections were prepared using one adult honey bee specimen that was previously subjected to MRM scanning. The entire body of the honey bee was fixed in 10% formaldehyde solution in phosphate buffered saline (PBS) in the decompression chamber at room temperature. After fixation and removal of the appendages, the specimen was dehydrated through an ascending series of ethanol and xylene, and was infiltrated with paraffin wax (Paraplast Plus, McCormick Scientific) at 60°C for 48 h. Dehydration and infiltration steps were performed in the decompression chamber. Afterwards, the specimen was embedded using a HistoCore Arcadia H paraffin embedding station (Leica, Wetzlar, Germany). Animal sagittal sections (7 µm) were prepared with a Leica RM2265 microtome, transferred to water on microscope slides and dried on a hot plate. After deparaffinization and rehydration, the sections were stained by a standard procedure of Hematoxylin & Eosin staining, using Weigert's iron hematoxylin, followed by dehydration and mounting. Imaging was performed with an AxioImager Z.1 light microscope (Zeiss, Oberkochen, Germany), equipped with an HRc Axiocam camera using Axiovision software.

Three-dimensional MR images of the bee were digitally processed by the Fiji distribution (Schindelin et al., 2012) of the ImageJ digital image processing software (National Institutes of Health, Bethesda, MD, USA), which includes plugins for image segmentation, volume rendering and processing of image stacks. Volume-rendered images of the bee during metamorphosis from sealed larva to adult were calculated from the original 3D image sets. The tracheal system, characterized by signal voids inside the bee's body, was segmented using the thresholding method. Similarly, the gastrointestinal tract was characterized by the brighter envelope. Because the signal intensity of the gastrointestinal tract did not differ significantly from other surrounding tissues, the segmentation was done manually. Segmented images of the whole animal and its organs, i.e. tracheal system and gastrointestinal tract, were also used to calculate the corresponding volumes as a function of time. The length of the segmented gastrointestinal tract was calculated using Skeletonize3D and AnalyzeSkeleton ImageJ plugins, while tortuosity of the tract was calculated as the ratio of its length to the distance between its ends. Volumes of the bee and its organs were determined as the voxels with a signal intensity above the specified threshold. The volumetric uncertainty was estimated as the square root of the volume of superficial voxels. Single voxel volume was approximated to 4.77×10⁻⁴ mm³ (calculated from FOV and imaging matrix).

The compositional analysis of the flight muscle as a function of time in two different ways was also performed. One method is the signal intensity histogram analysis of a 3D image segment of the muscle defined by a region of interest (ROI) with a size of 22×16×7 pixels. This segment was analyzed from the transformation of pupal stage to the emergence of the adult. For each time point, the mean and standard deviation of the ROI signal were calculated.

The second method utilizes the gray level co-occurrence matrix (GLCM) analysis (Haralick et al., 1973) of the same image segment (ROI). GLCM has elements p_i,j with values that correspond to the occurrences of pixels with intensity i having a pixel with intensity j in their neighborhood at a specified offset with respect to their position. GLCM is able to distinguish features in an image such as stripes or blocks from the noise even if they share the same average value and standard deviation. Each neighborhood-averaged GLCM matrix was represented by 11 second-order statistical texture features: (1) angular second moment (ASM), uniformity or energy; (2) entropy (ENT); (3) dissimilarity (DIS), (4) contrast (CON) or inertia; (5) inverse difference normalized (IDN); (6) correlation (COR); (7) inverse difference moment (IDM) or homogeneity; (8) variance (VAR); (9) cluster shade (SHA); (10) cluster prominence (PRO); and (11) maximum probability (MAX) (Yang et al., 2012). All of these statistical parameters were calculated for the ROI for most time points during development.

Metamorphosis from sealed larva to adult of the first worker bee specimen is depicted in a set of 3D T₁-weighted sequential images in the central sagittal plane for the selected times shown in Fig. 1A. In contrast, a complete set of both specimens is represented in Movie 1. Bright formations correspond to the fat-rich tissues that are used for the formation of other tissues during the first day of development. Prominent transformation of the specimen from larva to pupa occurred at day 3. After this transformation, all of the body parts and most of the internal organs are visible. During days 3 to 12, organs transform significantly. The gastrointestinal tract elongates up to day 5, then starts twisting up to day 7, becoming wider and growing towards the thorax section. A more detailed analysis of the thorax section shows myogenesis of flight muscles (Fig. S3). The muscles are initially brighter due to fatty inclusions which evolve into dark layered structure. Development of the brain along with the nervous system can also be seen. The size, orientation and shape of the head changes significantly during the development. Images at day 13 and later are blurred with both specimens because of an excessive motion of the bee trying to leave the cell. Three-dimensional images also enable volume rendering shown in Fig. 1B, Movie 1. The volume-rendered images enable the accurate visualization of the process showing all the changes of the external shape of the bee, such as development of extremities, head, thorax and abdomen.

Volume rendering of the segmented tracheal system enables the spatial visualization of the system (Fig. 1C, Movie 2). In the larval stage, rudiments of all the major airways exist which become wider and start branching further into narrow airways after the transformation into the pupal stage. Towards the end of the metamorphosis, the tracheal system becomes more complex and branched, and finer details can no longer be seen owing to the insufficient image resolution. The gastrointestinal tract (Fig. 1D, Movie 2) mainly shows rudiments of the tract in the larval state. Initially, a narrow and straight tube (up to day 2) becomes wider in course of time. During the transformation from larva to pupa (day 3 and 4 h), the tube curves and the segmented stomodeum and midgut section of the gastrointestinal tract can be seen on the corresponding image, while the later images show only the stomodeum or the midgut because of insufficient contrast for reliable segmentation. Stomodeum is then pushed from thorax into the abdomen, where it starts forming the honey stomach. These observations correspond well with the original description by Snodgrass (Snodgrass, 1956). No curvature of the stomodeum is observed until day 5. After that, the midgut starts U-shaping (day 7) and finally C-shaping.

Fig. 2 shows graphs of average volumes with standard errors of the worker bee and some of its organs during the metamorphosis from sealed larva to adult. The total volume of the bee (Fig. 2A) first decreases and then sharply increases, coinciding with the transition from larva to pupa. From then onward, a relatively steady decrease is observed. The final average volume of the bees at the end of the metamorphosis is equal to 118±18 mm³. The volume of the tracheal system (Fig. 2B) practically triples after the transformation from larva to pupa, following almost steady growth until the end. After day 9, a moderate decline in the volume of the tracheal system is observed. This correlates with the narrow growth of major airways. The volume of the gastrointestinal tract (Fig. 2C) increases rapidly just before the transition from larva to pupa. After the transition, the volume decreases and remains practically unchanged until day 9. From then onward, the growth is considerable again until the end of metamorphosis. Most of the growth of the honey stomach occurs in the first 5 days after the transition from larva to pupa. From the graph of the bee's gastrointestinal tract tortuosity as a function of time (Fig. 2D), it can be seen that the tortuosity was close to 1 until day 5, when it started increasing until day 10 and was stabilized at a value of 8 during the emergence of the adult.

The sensitivity of MRI to the structural and compositional tissue changes also enables the analysis of flight muscles. A segment of the flight muscle of the worker bee specimens (Fig. 3A) was analyzed using image signal intensity histogram analysis. Measured histograms are shown in Fig. S3A, while the graph of the flight muscle's mean signal intensity and its standard deviation as a function of time is shown in Fig. 3B. The histograms measured in 1 day intervals exhibit a normal distribution with changing means and widths. Initially, the high mean intensity (blue curve) decreases at day 4, recovers until day 8 and gradually decreases until day 12. The corresponding standard deviation (orange curve) reaches the highest value at day 4, decreases until day 6 and is relatively constant from then onwards. The high values of the mean signal suggest a higher accumulation of fat. The decrease indicates change in the muscle composition towards the low fat content. The increase in the signal's standard deviation found soon after the transformation from larva to pupa at day 4 suggests the beginning of muscle transformation in which fat is used to build the muscle. Results of the muscle structure analysis in Fig. 3C are shown by GLCM maps that correspond to T₁-weighted images of flight muscle ROIs in Fig. 3A at three different stages of metamorphosis. The GLCM maps have distinct shift of distribution peak and also marked dependence on time of some second-order statistical texture features derived from the GLCM maps: correlation (COR), cluster shade (SHA), cluster prominence (PRO) and angular second moment (ASM). Time dependencies of features ASM, COR and SHA are shown in Fig. 3C, while time dependencies of remaining and less distinctive features are shown in Fig S3B.

The 3D MRM method was found to be efficient in the assessment of morphological and compositional changes, and allowed the undisturbed analysis of the same specimen throughout the process of metamorphosis from sealed larva to an adult. Frequent scanning (more than 80 sequential 3D images) did not affect the course of metamorphosis.

The right ambient conditions are essential for the vitality of the bee and its undisturbed development. Metamorphosis from larva to adult took on average 12.5 days. The reduction of total volume by 13% (shown in Fig. 2A) can be explained by the use of nutrients for the organ build-up. Sub-optimal ambient conditions could also contribute to the deviations from the expected development dynamics, with variability also present between the specimens. The experiment with parallel imaging of drones in Fig. S2 indicated that variations of up to 10% in the volume of the insect body and its organs can be expected among the different specimens. These factors explain some anomalies in the development of organs, e.g. the decrease of the tracheal system volume after day 9. In the tracheal system, only the major sacks can be seen in the MR images owing to limited resolution. Major tracheal sacks additionally differentiate at the ends and could not be detected. Insufficient contrast between some sections of gastrointestinal tract and the surrounding tissues prevents a complete segmentation of the gastrointestinal tract and yields underestimated gastrointestinal tract volume.

Microscopic computed tomography (micro CT) is an alternative imaging method with higher spatial resolution, but it is not that efficient in distinguishing different soft tissues. However, it is also harmful owing to X-ray radiation. This method has been used previously for studying insect metamorphosis (Helm et al., 2018; Lowe et al., 2013; Martín-Vega et al., 2017) with parallel sample scans in a shorter time frame as compared with our method, which, however, has six times better temporal resolution. Problems such as blurring associated with the motion of the bee can be seen owing to longer 3D scanning time during the larval stage and some phases of the late pupal stage. However, most changes during development were sufficiently slow to not affect the image quality by motion artifacts. Absence of motion artifact in the image taken before and then after the transformation from larva to pupa indicates that the transformation was completed in less than a frame time, i.e. less than 4 h. According to some previous studies, as in case of Calliphora vicina (Hall et al., 2017), the transformation took less than 2 h.

Comparison of MRM with micro CT performed on the winter worker bee (Fig. S4A,B) and of MRM with microanatomy on the summer worker bee (second specimen, Fig. S4C,D) confirms that the structures visualized by MRM are identical to those seen by micro CT and microanatomy, which has superb contrast and resolution and can be considered as a reference standard. In microanatomical images, fine structures can be observed which can help to better identify the anatomical parts on the corresponding MRM image. Conversely, MRM images can give an insight into the immediate original state of structures, especially on the shape and content of the lumen in parts of the digestive system such as the rectum. MRM analysis is also free of possible artifacts that are due to organ deformations and washing out of the lumen content, which can occur during histological sample preparation. In the experiment on the winter bee, the abdomen section appears in the MRM image much brighter than the head and thorax sections. This is because T₁-weighted MRM is sensitive to the fat content and winter bees have more fats in the abdominal section. Sensitivity of this method to fat-rich tissues enables morphological visualization of different sections, and the compositional analysis, e.g. flight muscle analysis in Fig. 3, shows a transformation from fat-rich to low-fat containing tissues. Similar analysis could be performed on other organs that may undergo a substantial transformation, e.g. brains (Haddad et al., 2004).

There are several other possibilities of employing the presented 3D MRM method in other studies on the insects. For example, it would also be very interesting to investigate metamorphosis from egg to larval stage by MRM. In this study, we did not follow metamorphosis during first 9 days, i.e. from laid egg to sealed larva. Another interesting topic is insect–parasite interactions, for example between honey bees and Varroa mites (Rosenkranz et al., 2010). MRI is quite a capable modality for this type of study (Chudek et al., 1998). It enables long observation times, is harmless and would have very little or almost no effect on this interaction.

In conclusion, the present study shows that MRM is a powerful tool for the analysis of morphological and compositional changes during the metamorphosis in insects. Unlike many other imaging modalities, MRM is harmless, which enables continuous scanning on the same specimen for a longer period of time. Clearly, MRM is an ideal tool for studying insect development and other incremental biological processes on a similar spatial and temporal scale, provided the prolonged scanning time and restricted availability is not an issue.

The authors thank Mitja Štrukelj, M.D., for the research inspiration and providing some useful advice regarding planning and initiation of this study. We also thank Kanza Awais for proofreading an earlier version of the manuscript.