Development of the artificial intelligence assisted A-ESEM
A-ESEM is a novel, user-friendly method for complex in-situ physico-chemical analysis and macro- to nano-scale imaging of static or dynamically changing samples in their native state under environmentally compatible conditions, see Figure 1.
We used a low temperature approach for in‑situ handling and high‑resolution imaging of extremely sensitive structures of condensed mitotic metaphase chromosomes of barley. It was based on a low‑temperature method (LTM) and a newly developed artificial intelligence (AI) module for A-ESEM23.
An artificial intelligence algorithm was developed using the program ThermoDynamic Simulator (TDS), based on a combination of computational fluid dynamics (CFD) numerical simulations in ANSYS CFX and the support vector regression method (ε-SVR). The core of the algorithm was built on machine learning, specifically on a support vector machine, which works via statistical learning theory and Wolfe's dual programming theory24. Results from CFD analyses were used as training data for the algorithm. TDS was used for sophisticated estimation of thermodynamic parameters involved in the initial settings of the sample environment protocol. TDS calculates the integral values of temperature, pressure, and humidity in the vicinity of the sample depending on a working distance of the sample. First, two of the mentioned parameters are entered, then the last one is calculated. TDS considers the real geometry of the specimen chamber, gas pressure, the saturated water vapour flow rate and the real temperature of the sample in a drop of water with regard to heat losses of the material (Peltier cooling stage, Si-Cu pad, sample). Graphical outputs are visualized as contours placed in the vertical axis of the sample. The correct contour is selected using machine learning based on user specified input values.
Based on TDS results, measurements of thermodynamic parameters were performed to obtain data for simulation in the Ansys CFX software aiming at optimizing working parameters of the low temperature method (LTM). Transient analysis was carried out with a time step of 0.05 sec and total simulated time 240 sec. Heat transfer within the sample environment was simulated using a total energy model, which includes high-speed energy effects (in our case, a rapid drop in pressure during pumpdown of the microscope chamber). Laminar flow without turbulent models was used and radiation was not included. A solver scheme was set to high resolution and second order backward Euler with maximum 100 iterations per time step. Humid air with temperature-dependent thermophysical properties was used as the fluid material.
Parameters for transient analysis were monitored in-situ using a calibrated temperature sensor placed just below a replaceable Cu-Si pad (Figure 2a), humidity micro-sensors placed in the Cu base of the Cu-Si pad at the same level as the sample (Figure 2a), and two gas pressure capacitance gauges in the wall of the specimen chamber coupled. The sensor data were processed online using in-house software with a graphical interface to show the dependence of total pressure, water vapor partial pressure and saturated water vapor pressure on temperature and time at different points in the specimen chamber. The measurements obtained, in particular RH just above the sample surface, were used as boundary conditions for transient analysis of thermodynamic conditions during the LTM25. Significant changes in temperature when applying LTM can only be measured using the temperature sensor placed directly under the drop containing chromosomes (Figure 2a, c; red line vs. doted red line). Therefore, the data from the RH sensor can only be used as one of the boundary conditions for the simulations. Artificial intelligence-based simulations are the only way to confidently deduce humidity changes close to the drop (Figure 2c; green line vs. dashed green line).
Prior to the LTM, the specimen chamber was sufficiently hydrated through a series of purge flood cycles until the RH reached the required initial level of 65% (Figure 2c; green line) 26. Sample preparation for A‑ESEM observation was kept to a minimum, with only the application of an ionic liquid (IL). Since the sample was essentially in its native state (without harsh chemical treatments, post-fixation with osmium or drying) the IL could sufficiently permeate the sample. Chromosomes are very sensitive to the environment in which they are stored, therefore the IL exposure time was minimized and no pre-incubation was applied. The ability of ILs to remain liquid in the supercooled state is ideal for the LTM, and the increased electrical conductivity of the surface covered by a very thin IL layer allows imaging of non-coated chromosomes (Figure 2) under significantly lower gas pressure than with conventional ESEM. The increased thermal conductivity also allows for more rapid sample cooling, which inhibits ice crystal growth. The very thin IL layer also helps to scavenge free radicals and thus to reduce radiation damage to the sample during low-dose imaging, reduces water evaporation from the sample, and does not visibly distort the natural nanostructure of the surface.
Before the observation, chromosome samples were equilibrated in-situ following the LTM process directly in the A ESEM specimen chamber, on the Cu-Si pad of the Peltier cooling stage27. To ensure optimal velocity and direction of gas flow around the sample, including underneath, the sample was laid on Si with an etched checkerboard pattern. A matrix of 10 x 2 μm2 square holes with a spacing of 10 μm was etched onto the Si surface according to a pattern created by electron beam lithography. The shape of the pattern was designed based on experimental results and made it possible to image the very fine protrusion morphology on the chromosome surface (Figure 3). The shape and dimensions of the pattern, as well as its positive effect on the preservation of fine morphology was the outcome of a series of experiments. A series of regular depressions/thinning of the Si pad most probably causes it to be better cooled at the thinning points and disrupts the homogeneity of gas flow along the surface of the SI pad and around the sample. The sample is thus well surrounded by the gas and therefore its very fine morphology is preserved. The fine fibers on the sample surface are fluffed by the gas flow, they are not stuck to the surface.
The application of the LTM can be described in four phases (Figure 2c). In phase I (up to 69 sec), the chromosome samples were pre-cooled from 25°C to 0.5 °C (at a rate of 18.3 °C m-1 in) in the closed specimen chamber under atmospheric pressure. Preferential evaporation of water previously condensed on the base of the Peltier cooling stage increases RH (green lines) and protects the sample from drying during the continuous temperature decrease (red line) of the Cu-Si pad. The sample is still fully covered with a liquid and the volume change of the drop is minimal. When the sample temperature of 0.5 °C is reached, phase II starts and the specimen chamber is pumped (Figure 2c; 69 sec). The key factor in the successful use of the LTM is a synchronized decrease in sample temperature and gas pressure with respect to the initial humidity in the specimen chamber. The temperature decrease is accelerated both by the use of IL and by placing the sample near the region with the lowest temperature (down to -85 °C) in the specimen chamber (Figure 2b; Temperature - 97 sec). Here, the sample is intensively cooled by very cold gas flow (Figure 2b; Velocity - 97 sec) until the end of phase II. At this stage, gas is pumped out at a rate of 2000 Pa/s and the evaporation of water from the droplet is very intensive (Figure 2c; green dashed line) 28. Under these conditions, the supercooled drop immediately freezes at -14 °C (Figure 2c; 119 sec - start of phase III).
The higher the degree of supercooling, the higher the rate of ice nucleation and the faster the effective rate of freezing. This results in a high number of small ice crystals29. In combination with IL, crystallization damage risks are minimized. Sample freezing is accompanied by the release of the heat of crystallization (Figure 2c; red line, small peak at 119 sec). Sublimation of residual ice then occurs, indicated by the increase in simulated RH above the sample surface, up to 94%, and a decrease of sample temperature to -19 °C. Finally, the thickness of the layer on the sample is reduced, revealing the nanostructure of the sample covered only by an ultra-thin IL layer, not visible in the image. As a result of continuous gas pumping and the increase in sample temperature, RH decreases below 40% in 179 sec (start of the phase IV). By opening the valve of the hydration system (185 sec) to stabilize the gas pressure in the specimen chamber to 150 Pa (final value for imaging) RH increases to more than 60% at the end of the phase IV. The raise in pressure increases the probability of electron-induced ionization of gas molecules, which in turn increases the detection efficiency of the ISEDS. When phase IV is completed, the sample is transferred to the optical axis of the microscope and A‑ESEM observation may start. The observation parameters were: electron beam energy 10 keV, beam current 10 pA, dwell time 5 µs. Due to the increased temperature of gas in close vicinity to the pole piece of the objective (Figure 2a left, 2b in the middle) and thermal effects of the electron beam, any residual ice is sublimated from the sample surface and so the study of very fine nanomorphology of chromosome surfaces is possible4.
Chromosome observation
In contrast to previous SEM studies and conventional ESEM observations, our approach involves greatly simplified preparation of chromosome samples with the aim to preserve the natural nanostructure of the sample surface3. The only chemical treatment used was a mild formaldehyde fixation of root tips from which the chromosomes were released by mechanical homogenization and afterwards purified by flow cytometric sorting (Figure 3a).
As A-ESEM visualizes topography of chromosomes at native state, our observations should not suffer from preparatory and observation artefacts, which are inherent to conventional SEM and cryoSEM30,31. The average length of A-ESEM imaged barley chromosomes measured using Scandium software was 4.3 µm (n=95) (Figure 3b), which was 20% less than the length of 5.3 µm (n=200) as determined using fluorescence microscopy. The difference could be due to longitudinal extension of chromosomes during drying on a glass slide prior to microscopy. Importantly, A-ESEM revealed chromosome surface studded with numerous protrusions covering densely the chromosome surface (Figure 3c). The mean width of the protrusions was 32.9 nm (n=183) and their nature is at present not clear (Figure 3d). Such organization of chromosome surface has not been observed before and this observation contributes to the efforts aiming at unravelling the organization and function of the perichromosomal layer32. We also observed protrusions with a mean width of 64.3 nm (n=111), which could represent two structures fused together.
Our observations support the model of hierarchical organization of the perichromosomal layer and the role of pre-ribosomal RNAs in its establishment20. We have also confirmed the presence of regularly spaced bridges linking sister chromatids (Figure 3d) that were discovered using other method33. Other feature not observed before in condensed chromosomes were 11.6 nm features which may represent nucleosomes (Figure 3 f,g).
Post-fixation of flow-sorted chromosomes with 2% (v/v) formaldehyde for 20 min at 5 °C resulted in a dramatic decrease in the number of the 30 nm protrusions and increased the number of double sized protrusions. Moreover, the surface of post-fixed chromosomes was smoother as compared to non-treated ones (Figure 4a).
Prolonged exposure to electron beam resulted in the observation of chromatin cavities (Figure 4b). The presence of chromatin cavities was reported previously after the observation of mitotic chromosomes by SEM20,34. However, the cavities were not observed when ionic liquid preparatory method was used for SEM20. The absence of larger dark spots in non‑treated chromosomes representing void space below the chromosome surface in this work indicates that the earlier observations of cavities were most probably due to preparatory artefacts. Given that our approach does not rely on critical point drying and on ionic liquid preparatory steps, it is suited to clarify the presence of chromatin cavities. The observation using A-ESEM suggested that condensed mitotic chromosomes are free of hollow spaces.
Contrary to the published SEM images of barley chromosomes our observation using A-ESEM did not reveal parallel fibrils in centromeric regions12,35,36. The chromosome surface of this region was similar to the remaining parts of chromosome body, albeit with noticeable hinge and a decreased number of protrusions (Figure 4c). An interesting feature were the bridges linking sister chromatids (Figure 4d). The distance between the adjacent bridges was 390 nm, which compares well with 400 nm reported earlier as well as similar bridges in human and pig chromosomes33. However, given the high compaction and the ~2 µm size of the mitotic chromosome arms, we observed correspondingly smaller number of bridges. Further study of their molecular composition should reveal their role in sister chromatid cohesion. Our observations indicate that they are part of the perichromosomal layer and free of DNA as they disappeared after a mild RNase treatment (see below).
A model of hierarchical structure of perichromosomal layer suggests that intermediate layer comprising mainly pre-ribosomal RNAs, plays an important role in binding proteins on chromosome surface and the formation of chromosome cover20. A removal of the RNA layer caused loss of several nucleolar proteins from human chromosome surface20. To confirm the role of RNA, we treated barley chromosomes with 0.01 µg µl-1 RNase for 30 min at 37 °C. This treatment caused only minor swelling of chromosomes, but their overall shape remained unchanged (Figure 4e). However, the surface of RNase-treated chromosomes as observed by A-ESEM differed significantly from the untreated ones. The 30 and 60 nm protrusions were no longer visible and chromosome surface appeared levelled and with globular structures (Figure 4f). This type of chromosome surface was observed earlier by SEM12,36,37. The diameter of the globular structures ranged from 25.8 to 43.7 nm (n=105) (Figure 4g). Moreover, parallel fibers in centromeric region were exposed (Figure 4f) and the bridges between sister chromatids could no longer be detected. According to Booth and Earnshaw the chromosome periphery, which was most probably lost after the RNase treatment, has a thickness of 87‑150 nm18. However, we could not confirm this value due to chromosome swelling after RNase treatment.