Tools and Resources

Cell Biology

Crosshair, semi-automated targeting for electron microscopy with a motorised ultramicrotome

Cell Biology and Biophysics Unit, European Molecular Biology Laboratory (EMBL), Germany
Collaboration for Joint PhD Degree Between EMBL and Heidelberg University, Faculty of Biosciences, Germany
Francis Crick Institute, United Kingdom
Electronic Workshop, European Molecular Biology Laboratory (EMBL), Germany
Mechanical Workshop, European Molecular Biology Laboratory (EMBL), Germany
Electron Microscopy Core Facility, European Molecular Biology Laboratory (EMBL), Germany

Nov 15, 2022

Open access
Copyright information

Abstract
Editor's evaluation
Introduction
Results
Discussion
Materials and methods
Data availability
References
Article and author information
Metrics

Abstract

Volume electron microscopy (EM) is a time-consuming process – often requiring weeks or months of continuous acquisition for large samples. In order to compare the ultrastructure of a number of individuals or conditions, acquisition times must therefore be reduced. For resin-embedded samples, one solution is to selectively target smaller regions of interest by trimming with an ultramicrotome. This is a difficult and labour-intensive process, requiring manual positioning of the diamond knife and sample, and much time and training to master. Here, we have developed a semi-automated workflow for targeting with a modified ultramicrotome. We adapted two recent commercial systems to add motors for each rotational axis (and also each translational axis for one system), allowing precise and automated movement. We also developed a user-friendly software to convert X-ray images of resin-embedded samples into angles and cutting depths for the ultramicrotome. This is provided as an open-source Fiji plugin called Crosshair. This workflow is demonstrated by targeting regions of interest in a series of Platynereis dumerilii samples.

Editor's evaluation

Meechan et al. present a systematic approach to semi-automate an ultramicrotome operation for targeting a specific plane aided by X-ray tomography measurements, and it is a fundamental work of great interest to any users of electron microscopy (EM), particularly when targeting the imaging of thin sections in a select region of interest by ultramicrotomy, or when targeting volume EM of select sample regions. The article documents with exceptional detail a workflow including both microtome modifications and software adaptations for semi-automated targeting of structures with micrometre precision, resulting in a faster and more accurate orientation of the image acquisition planes for volume electron microscopy, a task that has traditionally been difficult and time-consuming. Therefore, this work will reduce sample preparation labour and, critically, facilitate the comparison of the ultrastructure of multiple samples. The method is based on X-ray imaging acquisition prior to any sectioning and proposes a solution for the two instruments commercially available in the field, and by transparently sharing all the data, hardware, and software, and describing every detail of the workflow, this fundamental method can be readily adopted by any practitioner, enabling its wide application – it is a key step in the field regarding speed-up, accuracy, and reproducibility in electron microscopy.

https://doi.org/10.7554/eLife.80899.sa0

Introduction

Imaging samples with electron microscopy (EM) is a time-consuming process – often requiring weeks or months of continuous acquisition for large samples (Scheffer et al., 2020; Zheng et al., 2018; Vergara et al., 2021). This means that it is rarely feasible to acquire image volumes from entire specimens, and we must instead target specific regions of interest. For resin-embedded samples, this process is usually done with an ultramicrotome that allows manual trimming of the block using a razor blade and/or diamond knife, followed by cutting of thin sections at precise locations.

Targeting with an ultramicrotome is a difficult and manual process. The operator must rely on surface features that are visible through the ultramicrotome’s binocular microscope. This makes it challenging to target regions deep within a sample, especially when heavy metals make the sample opaque. In addition, the orientation of cutting must be set manually by adjusting three different ultramicrotome axes. This kind of 3D thinking is challenging, taking much time and training to master.

Other studies have improved the precision and ease of targeting by leveraging external 3D maps from light or X-ray microscopy (Karreman et al., 2016; Karreman et al., 2017; Xu et al., 2021; Musser et al., 2021; Bushong et al., 2015; Ronchi et al., 2021). X-ray offers many advantages for EM targeting – for example, it is readily compatible with standard EM sample preparation methods (Bosch et al., 2022; Kuan et al., 2020; Bushong et al., 2015; Karreman et al., 2017) and provides images that highlight similar structures to volume EM (although at a lower resolution) (Bosch et al., 2022; Kuan et al., 2020; Bushong et al., 2015; Musser et al., 2021). Laboratory micro-CT systems are becoming ever more popular and can provide an isotropic resolution of about 1 micron, with scan times in the range of a few hours (Withers et al., 2021; Karreman et al., 2016). In addition, there is increasing access to synchrotrons for X-ray imaging, which can provide resolutions in the range of hundreds or even tens of nanometres, and scan times in the range of minutes (depending on the resolution required) (Withers et al., 2021; Bosch et al., 2022; Kuan et al., 2020). X-ray imaging therefore offers a fast, non-destructive method for obtaining 3D maps of the internal features of a sample for targeting.

Most methods that use light or X-ray microscopy for targeting rely on measuring the depth of regions of interest from the resin block’s surface, followed by trimming with an ultramicrotome. As these methods do not compensate for the exact position and orientation of the sample in the ultramicrotome, progress must be checked at regular intervals by iterative rounds of X-ray or light microscopy and trimming, slowing this process down. In addition, these methods focus on the depth of the region of interest, but still require the orientation of cutting to be set entirely manually.

Other studies have sought to bring more automation to the ultramicrotome to make it easier to use (Baena et al., 2019; Templier, 2019; Lee et al., 2018). These efforts have mostly focused on the collection of serial sections, with little automation of the initial trimming and targeting process. An exception is Brama et al., 2016, in which a miniature fluorescence microscope was integrated into the ultramicrotome, allowing automated tracking of regions of interest during cutting. However, this requires a fluorescent signal to be present within the sample embedded in the resin block and is therefore not compatible with standard heavy-metal stained specimens. Also, while this technique allows detection and collection of sections from regions of interest, it does not automate the depth or orientation of ultramicrotome cutting.

Here, we created a workflow for semi-automated targeting of regions of interest with a modified ultramicrotome. We introduced automation to two recent commercial systems in the form of motorisation for each of the ultramicrotome axes. This allowed precise and automated angular movement for both systems, as well as translational movement for one system. We also created new software that allows selection of a plane of interest from X-ray images, and automatic conversion into angles and cutting depths for the ultramicrotome. This is provided as an open-source, user-friendly Fiji (Schindelin et al., 2012) plugin called Crosshair.

Results

Automation of ultramicrotome

We modified two of the most recent commercial ultramicrotomes (Leica EM UC7 and RMC PowerTome PC [PTPC]) to add automation of movement (Figure 1, Figure 1—figure supplement 1). For the UC7, this consisted of three extra motors (one for each rotational axis) controlled by a small Raspberry Pi computer (Figure 1B and D). This system can be controlled via a simple touchscreen interface (developed with Node-RED), allowing the angles of each axis to be precisely set and measured (see section ‘Ultramicrotome systems and calibration’). All translational movements of the knife (north–south [NS] and east–west [EW]) were controlled manually via the existing Leica stage motors and Leica user interface.

Figure 1 with 4 supplements see all

Download asset Open asset

Motorised ultramicrotome.

(A) Summary of the main ultramicrotome axes of rotation (sample tilt, sample rotation, and knife tilt), and movement (north–south [NS]/east–west [EW]), which are common to both Leica and RMC systems. (B) Overview of motorised Leica system. (C) Overview of motorised RMC system. (D) Zoom of (B), showing motors attached to each axis. (E) Zoom of (C), showing motors attached to each axis.

For the RMC, motors were added to the same rotational axes, with additional automation of the two translational stage movements (Figure 1C and E). These additional NS and EW motors allowed the position of the knife to be precisely controlled within the same user interface as the orientation. This is in contrast to the Leica system, where currently the translational and rotational movements are controlled separately. The five motors on the RMC system were controlled by a Trinamic TMCM-6110 6-axis stepper motor driver board. The graphical user interface (GUI) was developed with LabVIEW and installed onto a touchscreen PC, allowing users to control the motions via keyboard/mouse, touchscreen, or a joystick (see section ‘Ultramicrotome systems and calibration’).

Both systems were calibrated with special adaptors made to convert a rotational movement into the rotation of the dial of a Thorlabs CRM1PT/M rotation mount, whose Vernier scale provides 5 arcmin (0.083°) resolution (see section ‘Ultramicrotome systems and calibration).

Targeting calculations

Our goal was to allow automatic calculation of the ultramicrotome moves required to cut to a specific target plane. The target plane defined the final position and orientation of the desired block surface (Figure 2A). Thin sections could then be taken for transmission electron microscopy (TEM) or array tomography or the surface could be imaged directly with volume scanning electron microscopy (SEM).

Figure 2

Download asset Open asset

Targeting problem.

(A) Left: diagram of a resin block with a *Platynereis* sample inside. The top surface has been trimmed flat with a diamond knife. Right: same resin block with labelled block plane (i.e. the plane parallel to the flat trimmed surface) and target plane. (B) Diagram of the block in the ultramicrotome before (left) and after (right) the orientation solution is applied. A top view (with diamond knife at the bottom, and sample holder at the top) and side view (with diamond knife on the left, and sample holder on the right) are shown. $θ_{K}$ is the knife tilt angle, $θ_{T}$ is the sample tilt angle, and $θ_{R}$ is the sample rotation angle. (C) Diagram of the block in the ultramicrotome before (left) and after (right) the distance solution is applied. Top/side view are the same as in (B). $D_{N S}$ is the NS cutting distance.

This targeting problem was broken down into two steps: orientation and distance. For the orientation problem, we found a combination of three angles (knife tilt, sample tilt, sample rotation; Figure 1A) that brought the target plane to be vertical and parallel to the knife (Figure 2B). For the distance problem, we found the depth to cut from the sample surface to reach the target plane (Figure 2C).

The orientation problem required a mathematical description of how the target plane orientation changed as the knife tilt, sample tilt, and sample rotation were varied. This was a similar problem to that which must be solved to, for example, orient the ‘end-effector’ of a robotic arm by varying the angle of its individual joints. In robotics, this is solved by construction of a ‘forward kinematics’ equation that calculates the end-effector position, given the angles of the joints, and an ‘inverse kinematics’ equation, which calculates the joint angles, given the desired end-effector position. We therefore applied similar principles, constructing a forward and inverse kinematics equation for the ultramicrotome.

We first constructed the forward kinematics equation to describe the orientation of the target plane in terms of the three input angles. To do so, we defined orthogonal coordinate frames for each of the ultramicrotome parts (e.g. the knife holder, the sample holder, etc.) (Figure 3A and B, Figure 3—figure supplement 1) and calculated the rotations that relate each piece. This is summarised in the pose diagram in Figure 3C. The relation between the sample holder and sample block surface is unknown in this pose diagram. This is because the sample holder does not hold the sample in a fixed orientation, meaning that the resin block has a slightly different orientation every time it is placed inside.

Figure 3 with 2 supplements see all

Download asset Open asset

Coordinate frames, pose diagrams, and solutions.

(A) Diagram of all coordinate frames. World: x is parallel to EW, y is parallel to NS, z is vertical. Knife: z points out of the page, and is parallel to the world frame z. Arc: x is parallel to the world frame x. Holder: y points into the page and is parallel to the arc frame y. Block: x is parallel to the bottom edge of the block face. Two variations are shown: a rectangular block face (left) and a trapezium block face (right). Target: y is perpendicular to the target plane (shown in grey). (B) Diagram of relation between block and target coordinate frames. Left: 3D view of the block, showing the block frame (red) and target plane (grey). Middle: 2D view of the block frame’s xy plane (i.e. looking down on the block from above). The target plane intersects along the shown grey line, and $θ_{t o}$ is the angle between the red and blue x axes, that is, the rotation angle about the block frame z. Right: 2D view of the blue axes’ zy plane (i.e. looking down the line of intersection from the middle panel). $θ_{t r}$ is the angle between the blue and green z axes, that is, the rotation angle about the blue frame x. The green axes are the final target coordinate frame (as shown in A). (C) Pose diagram showing relations between coordinate frames. $R_{x}$ , $R_{y}$ , $R_{z}$ : rotation matrices about the x, y, and z axes, respectively; $θ_{T}$ : sample tilt angle; $θ_{R}$ : sample rotation angle; $θ_{K}$ : knife angle; $θ_{t o}$ : target offset angle; $θ_{t r}$ : target rotation angle. Dashed arrows are unknown relations. (D) Pose diagram when the knife is aligned to the block face. Here, the knife frame is in the same orientation as the block (red arrow), and therefore we can infer the Holder to Block relation that was unknown in (C). $θ_{I T}$ is the initial sample tilt angle and $θ_{I K}$ is the initial knife tilt angle, when the knife and block are aligned. We define $θ_{R}$ to be zero at this aligned orientation. (E) Graphs of knife and sample tilt solution for all sample rotation angles. This used values of $θ_{I K}$ = 10, $θ_{I T}$ = 10, $θ_{t o}$ = –3.3, and $θ_{t r}$ = 5.4. (F) Solution graphs with more extreme initial angles. $θ_{I K}$ = 10, $θ_{I T}$ = 10, $θ_{t o}$ = –60, and $θ_{t r}$ = 5.4.

To calculate this relation, we therefore required an initial alignment step where the knife is manually aligned to the block face. This is standard procedure for cutting with an ultramicrotome, and so should be familiar to anyone who has used a ultramicrotome previously. In this aligned position, the orientation of the knife and block coordinate frames was the same, allowing the holder to block relation to be calculated (Figure 3D).

This meant that the full forward kinematics equation, describing the rotation from the World coordinate frame to the target plane, was

F = R_{x} (θ_{T}) R_{y} (θ_{R}) R_{x} (- θ_{I T}) R_{z} (θ_{I K}) R_{z} (θ_{t o}) R_{x} (θ_{t r})

where $R_{x}$ , $R_{y}$ , and $R_{z}$ are rotation matrices about the x, y, and z axes, respectively. $θ_{T}$ is the sample tilt angle and $θ_{R}$ is the sample rotation angle. $θ_{I T}$ , $θ_{I K}$ , $θ_{t o}$ , and $θ_{t r}$ are constants for a particular targeting run representing the initial sample tilt angle (on alignment), the initial knife tilt angle (on alignment), the offset angle between the block face and the target plane (measured from X-ray), and the rotation angle between the block face and the target plane (measured from X-ray). See Figure 3B for details of how $θ_{t o}$ and $θ_{t r}$ are calculated.

With the forward kinematics equation complete, we then constructed and solved an inverse kinematics equation to determine the required angles for each solution (see section ‘Orientation calculation’). This gave the solution tilt and knife angles as

θ_{T} = \arctan (C_{1} \cos (θ_{R}) + C_{2} \sin (θ_{R}))

θ_{K} = \arctan (\frac{C_{3} (C_{4} \cos (θ_{R}) + C_{5} \sin (θ_{R}))}{(\sqrt[2]{{C_{3}}^{2} + (C_{4} \sin (θ_{R}) - C_{5} \cos (θ_{R}))^{2}}) | C_{3} |})

where $C_{1 - 5}$ are constants for a particular targeting run (see ‘Materials and methods’ for details). This means that, as expected, there are many possible solutions for each targeting setup. In fact, any point on these lines (Figure 3E) is a possible solution. Note that there will be some situations where no solution is possible due to the constraints of the ultramicrotome axes, for example, in Figure 3F, the sample tilt of some solutions exceeds the maximum sample tilt of 20°. In these situations, the sample would need to be remounted or re-trimmed to aim for an angular difference within the ultramicrotome constraints.

For the distance problem, we calculated the distance between the block surface and target plane (from the X-ray images), and compensated for the current knife angle to give the true cutting distance:

D_{N S} = \frac{D_{P}}{\cos (θ_{K})}

where $D_{N S}$ is the north–south (NS) distance (cutting distance), $D_{P}$ is the perpendicular distance between the target plane and the furthest surface point, and $θ_{K}$ is the knife angle (see section ‘Distance calculation’ and Figure 3—figure supplement 2).

Crosshair targeting workflow

The entire workflow is summarised in Figure 4, with details given in the section ‘Targeting workflow’. In brief, the resin block was first trimmed manually close to the sample, with a rectangular face in its surface. This block face (which can be any shape with four corners and straight lines in between) provided the reference to define the block coordinate frame (Figure 3A).

Figure 4

Download asset Open asset

Targeting workflow.

Summary of targeting workflow steps with representative images/3D renders.

Next, the sample was imaged with X-ray at the highest resolution possible (around 1 micron isotropic voxel size is feasible on lab-based micro-CT systems). From this X-ray image, two planes were defined: one on the flat block surface, and one on the target plane using the Crosshair plugin (see section ‘Crosshair’ and Figure 2A). Then, the corners of the block surface were marked in the software as these were used to determine where cutting would begin and the distance required. Finally, the orientation was decided, that is, which edge of the block would face upwards in the ultramicrotome.

Next, we moved to the ultramicrotome. The zero position of the sample tilt and knife tilt axes was checked (see ‘Materials and methods’) to ensure all angular measures began from the correct position. Then, the resin-embedded sample was inserted, and the knife and block face manually aligned. This aligned position was defined as the zero position for the sample rotation to simplify the targeting calculations.

The aligned sample and knife tilt angles ( $θ_{I T}$ and $θ_{I K})$ were then precisely read from the motorised ultramicrotome system and entered into the targeting calculations. This allowed all solutions to be calculated and one selected by the ultramicrotome user. For example, the user may want to minimise the knife and sample tilt angles to make it easier to collect thin sections after targeting. Finally, the ultramicrotome was precisely set to the three solution angles using the motors and the solution distance cut.

At this point, thin sections could be taken for TEM, or array tomography, or the block could be transferred for volume imaging in a focused ion beam scanning electron microscope (FIB-SEM) or serial block face scanning electron microscope (SBF-SEM).

Crosshair

To make this workflow accessible to all, we developed the open-source Fiji (Schindelin et al., 2012) plugin Crosshair (https://github.com/automated-ultramicrotomy/crosshair; Meechan, 2022b), which allows user-friendly measurement of X-ray images, and access to the calculations described above (Figure 5, Figure 5—figure supplement 1). It uses BigDataViewer (Pietzsch et al., 2015) to allow browsing of large images in arbitrary 2D slices, and the ImageJ 3D Viewer (Schmid et al., 2010) to allow browsing in 3D.

Figure 5 with 1 supplement see all

Download asset Open asset

Crosshair.

(A) Crosshair plugin user interface, with an example *Platynereis dumerilii* X-ray image. Left: controls; middle: BigDataViewer window showing 2D cross-section. Right: ImageJ 3D viewer showing volume rendering with block and target planes (blue and green, respectively). The green target plane is the same as the 2D slice shown. (B) Left: microtome mode controls and solution display. Right: corresponding 3D representation of ultramicrotome and sample.

Crosshair allows the planes and points required for the targeting workflow (Figure 4) to be created and viewed interactively. It also allows the orientation of the block to be marked and visualised in 2D and 3D.

To make selection of solutions easier, Crosshair also has a ‘microtome mode’ which allows a simple representation of the ultramicrotome to be manipulated in 3D (Figure 5B). This has the constraints of the ultramicrotome built in and displays a volume rendering of the X-ray imaged sample at the appropriate orientation. In this way, users can interactively move sliders in the user interface and view the orientation of the ultramicrotome and sample in real time. This also outputs the corresponding angles and cutting distance for each solution.

Finally, Crosshair also has a ‘cutting mode’ which allows the predicted cutting progression to be viewed in 2D and 3D (Figure 5—figure supplement 1C). In this way, users can choose a solution and then easily view which structures will be visible at which cutting distances. This makes it simple to choose an appropriate solution for their targeting problem.

Accuracy measures

To test the accuracy of this workflow, we targeted a series of 6-day-old Platynereis dumerilii samples. Platynereis is a marine annelid, which is an established model system for development, evolution, and neurobiology (Özpolat et al., 2021; Fischer and Dorresteijn, 2004; Williams and Jékely, 2016). As they can be bred quickly and easily in the lab, and have well-defined tissue morphology, we decided to use them as a test case for targeting regions of interest. We set our target plane on each sample to cut through both of the anterior dorsal cirrus that protrude from either side of the Platynereis head.

All samples were targeted with the workflow described above: five on the motorised Leica system (samples a–e) and five on the RMC system (samples f–j). After targeting, samples were X-ray imaged again and registered to the pre-targeting scan. This registration was completed with another Fiji plugin we created – RegistrationTree – which is also freely available on GitHub (https://github.com/K-Meech/RegistrationTree; Meechan, 2022c). This plugin is a wrapper around two existing registration software – BigWarp (Bogovic et al., 2016) and elastix (Klein et al., 2010; Shamonin et al., 2013) – allowing easy passing of images from one software to the other, and smooth use of very large images with elastix. It is standalone, so it can also be used for general registration of large images, independent of the Crosshair workflow.

Once the images were registered, we calculated a series of angle and distance accuracy measures using the built-in features in the Crosshair plugin (Figures 6 and 7; see section ‘Accuracy tests’).

Figure 6 with 2 supplements see all

Download asset Open asset

Accuracy test results.

(A) Graph of point-to-plane distance error vs. angle error for all 10 samples (**a–j**). (B) Graph of absolute solution distance error vs. angle error. (C) For each sample (**a–j**), two images are shown. Top: registered before and after X-ray images. Bottom: snapshot of the 3D viewer in Crosshair. In the 3D view, the block surface is shown in blue, the target in green, and the plane reached in yellow. In most cases, the green and yellow planes are too close together to distinguish well. The axes shown in some of these images are the 3D viewer coordinate system.

Figure 7 with 1 supplement see all

Download asset Open asset

Comparison of target plane (from the before scan) to the block surface (after scan).

Each sample (**a–j**) has a column with three images. Top row: the target plane from the before targeting scan. Middle row: the block surface from the after targeting scan. Bottom row: cross-section parallel to block surface, but 3 µm inside the block. The middle row images can be quite dark/noisy as they slice right along the surface of the block, so the bottom row is provided so the cross-section can be seen more easily.

Comparison of the target plane and trimmed block surface showed that all samples hit their target of cutting through both anterior dorsal cirrus, and provided similar Platynereis cross-sections to those predicted from the X-ray (Figure 7). The accuracy measures gave a mean angular error of 0.9°, a mean absolute solution distance error of 3.1 microns, and a mean point-to-plane distance error of 3.1 microns (see section ‘Accuracy tests’ for details of these different measures). Note that there will be some error in these distance measures due to slight inaccuracies in the registration process – estimated at around 2 microns error (see section ‘Accuracy tests’ and Figure 6—figure supplement 1). Given that our X-ray voxel size is only 1 micron, there is a limit to how precisely such small distances can be measured. This could be improved in future by moving to higher-resolution X-ray imaging from synchrotrons, which can provide resolutions in the range of hundreds of nanometres.

Note that there were also some differences between the mean targeting accuracy reached with the Leica and RMC systems. The Leica system had a mean angular error of 1.1°, mean absolute solution distance error of 4.5 microns, and a mean point-to-plane distance error of 4.8 microns. The RMC system had a mean angular error of 0.7°, mean absolute solution distance error of 1.6 microns, and a mean point-to-plane distance error of 1.3 microns. There are a number of factors that could have contributed to this difference. Firstly, the samples tested with the Leica system were X-ray imaged with a different micro-CT system to those tested on the RMC system (see section ‘X-ray imaging’). While all were scanned with the same voxel size (1 micron), the amount of detail that could be seen inside the Platynereis was quite different (Figure 7—figure supplement 1). Therefore, the higher quality of the X-ray scans for samples f–j (tested with RMC) may have contributed to the improved average targeting accuracy. Secondly, there are still a number of manual steps in this workflow (such as the manual alignment of knife and block face), which can allow human error to contribute to the accuracy. Samples a–e and f–j were targeted by different researchers, so this may also contribute to the differing average accuracy. Finally, the Leica and RMC systems use different hardware (e.g. motors) which will also contribute to their differing average accuracy.

Discussion

The motorised ultramicrotome systems and targeting workflow presented here provide a user-friendly solution for targeted EM. The workflow allows X-ray images of resin-embedded samples to be converted into a set of three angles (sample tilt, sample rotation, and knife tilt) and a cutting distance for the ultramicrotome. These moves can then be precisely executed with the two motorised ultramicrotome systems we developed to reach a target plane. All measurement steps are integrated into the open-source Fiji plugin Crosshair, making it easily accessible to all. Across 10 targeting runs, the workflow provided a mean angular error of 0.9° and a mean distance error of 3.1 microns. Once the X-ray images have been acquired, the Crosshair measurements and trimming can be done in 1–2 hr, depending on the sample. This is a significant speed up where, in our experience, non-assisted targeting from X-ray may take 1–2 days for complex samples.

There are a number of factors that limit the accuracy of the current system – the first is the resolution of the X-ray images. All X-ray imaging in this article was done with lab-based micro-CT systems, which offer a maximum resolution of around 1 micron isotropic voxel size. As all measurements are based on these images, this voxel size limits the achievable accuracy and contributes to our registration uncertainty. A future direction would be to combine Crosshair with higher-resolution X-ray imaging from a synchrotron. Synchrotrons are extremely powerful sources of X-rays and can provide resolutions in the range of hundreds or even tens of nanometres. In recent years, there have been a number of studies that combine X-ray imaging from a synchrotron with EM (Bosch et al., 2022; Kuan et al., 2020; Musser et al., 2021), demonstrating the power of combining the large field of view and fast scan times of X-ray, with the high resolution of targeted EM acquisition. As this combination becomes ever more popular, it is more important than ever to have user-friendly targeting solutions like Crosshair.

Another accuracy limitation are the manual steps that remain in this targeting workflow. While we automate the motorised movements and angle/distance calculations, a number of steps still require manual use of the ultramicrotome, and will therefore vary in accuracy depending on how experienced the user is. For example, the setting of the zero positions of the knife and sample tilt, alignment of the knife and block face, and approach of the knife to the block surface are all done manually. This will therefore introduce some variation due to human error.

Each of these manual steps should be minimised or eliminated in future work to improve accuracy. For example, the zero for the sample tilt could be set by precise measurement of the arc piece, rather than by eye. Also, it would be useful to find a more automated way to set the knife zero (although this is more complex as the knife is removed and replaced each time, and different knives are used for different purposes).

The manual alignment of the knife and block face could be removed if another method was found to determine the relation between the block surface and the sample holder. For example, the sample could be X-ray imaged directly inside the sample holder, although this is challenging in many micro-CT systems due to its large size. Alternately, a thin pin could be designed that would fit in the micro-CT and also fit tightly inside the sample holder at a fixed orientation. This is very similar to how samples f–j were mounted for targeting (see section ‘Initial sample trimming and mounting’ and Figure 6—figure supplement 2). Each of these samples was mounted to a thin aluminium pin at the start of the targeting process, and all trimming steps completed with a Gatan 3View Rivet holder that allows the pin to slot tightly inside. To eliminate the manual alignment step, this pin/holder would have to be adapted so the pin only slots inside in a fixed orientation. For example, by adding a groove to the side of the pin, that fits a corresponding projection on the sample holder. This would allow the relation between the block surface and sample holder to be calculated directly from the X-ray images alone.

The final manual step is the approach of the knife to the block surface. Currently, the user must move the knife forward and assess whether it has reached the block surface visually by examining the block through the ultramicrotome’s binocular microscope. If cutting starts from one corner, it is easy to cut some distance before debris are visually detected on the knife (especially for less experienced ultramicrotome users). This results in errors in the final distance cut. This error increases each time the ultramicrotome is reset, meaning the greatest accuracy is achieved when the cutting distance is less than one feed length (200 microns). This being said, we did still achieve good accuracy with samples h, i, and j in this study that had cutting distances greater than one feed length (requiring one reset mid-run) (Figure 6). Even so, the requirement to manually reset and approach every 200 microns means that larger samples require more manual user intervention, and therefore more time and effort. Automation of this step is challenging and would require hardware improvements that extend the feed length of the ultramicrotome to minimise resets or a method to allow automatic resets. An automatic reset method would require a means to know the absolute distance between the knife and block surface, which is currently complex to achieve.

Moving away from improving the workflow’s accuracy, it would also be interesting to expand the workflow’s scope in future work. For example, Crosshair’s calculations should be applicable to any 3D map (not only X-ray), and so could be expanded to targeting from 3D light microscopy. This would be useful for targeting fluorescently labelled structures during CLEM (correlative light and electron microscopy) as in Ronchi et al., 2021. Also, Crosshair could be expanded to allow multiple modalities to be used together; for example, displaying registered light microscopy and X-ray images of the same sample.

In addition, Crosshair could be expanded to automate hitting not only a target plane, but a specific location within that plane. For example, for the Platynereis, we may cut to a target plane that gives a cross-section through the entire head, and then automatically trim the block sides to leave only a certain sub-structure of interest. This is useful for methods such as FIB-SEM that cannot image large block surfaces. This would require precise, automated east–west movement of the knife (as already implemented on the RMC system), and expansion of the targeting calculations and Crosshair plugin.

In summary, this targeting workflow offers a first step towards fully automated targeting of structures within resin blocks for EM. By making it simpler, and faster, to target a specific plane of interest, we can achieve higher throughput and shorter overall acquisition times. Also, by providing a user-friendly software for this process, we lower the barrier for newcomers to get started with 3D targeting. By providing the motorised ultramicrotome plans and software fully open-source, we hope that this work will enable more EM labs to rapidly target regions of interest for EM.

Constant	Full expression
$A$	$\cos (θ_{I K} + θ_{t o})$
$B$	$\sin (θ_{t r}) \sin (θ_{I K} + θ_{t o})$
$C$	$\sin (θ_{I T}) \sin (θ_{I K} + θ_{t o})$
$D$	$\cos (θ_{I T}) \sin (θ_{I K} + θ_{t o})$
$E$	$\cos (θ_{t r}) \sin (θ_{I K} + θ_{t o})$
$F$	$\sin (θ_{I T}) \cos (θ_{t r})$
$G$	$\sin (θ_{t r}) \cos (θ_{I T})$
$H$	$\sin (θ_{I T}) \sin (θ_{t r})$
$I$	$\cos (θ_{I T}) \cos (θ_{t r})$

Step no.	Step	Time	Temperature	Microwave (W)	Vacuum
1	2% paraformaldehyde, 2.5% glutaraldehyde in seawater	2 min	21°C	100	On
2	2% paraformaldehyde, 2.5% glutaraldehyde in 0.1 M cacodylate	2 × 14 min (2 min on/off cycles)	21°C	100	On
3	0.1 M cacodylate	1 immediate Then 2 × 40 s.	21°C	100	On
4	2% OsO₄ in 0.1 M cacodylate	2 × 14 min (2 min on/off cycles)	21°C	100	On
5	2.5% K₄[Fe(CN)₆].3H₂O in 0.1 M cacodylate	2 × 14 min (2 min on/off cycles)	21°C	100	On
6	Water wash	1 immediate Then 2 × 40 s	21°C	100	On
7	1% TCH unbuffered	2 × 14 min (2 min on/off cycles)	40°C	100	On
8	Water wash	1 immediate Then 2 × 40 s	40°C	100	On
9	2% OsO₄ aqueous	2 × 14 min (2 min on/off cycles)	21°C	100	On
10	Water wash	1 immediate Then 2 × 40 s	21°C	100	On
11	Dehydration series in ethanol (25%, 50%, 75%, 3 × 100%)	40 s each	10°C	250	Off
12	Infiltration series in Durcupan (25%, 50%, 75%, 90%, 3 × 100%)	3 min each	21°C	150	On
13	100% Durcupan	Overnight	21°C	N/A	N/A
14	Polymerisation in oven	48 hr	60°C	N/A	N/A

Share this article

Cite this article

Motorised ultramicrotome.

Targeting problem.

Coordinate frames, pose diagrams, and solutions.

Targeting workflow.

Crosshair.

Accuracy test results.

Comparison of target plane (from the before scan) to the block surface (after scan).

Symbols for orientation calculation.

Constants for orientation calculation.

Table of Platynereis preparation steps.

Author details

Kimberly Meechan

Contribution

For correspondence

Competing interests

Wei Guan

Contribution

Competing interests

Alfons Riedinger

Contribution

Competing interests

Vera Stankova

Contribution

Competing interests

Azumi Yoshimura

Contribution

Competing interests

Rosa Pipitone

Contribution

Competing interests

Arthur Milberger

Contribution

Competing interests

Helmuth Schaar

Contribution

Competing interests

Inés Romero-Brey

Contribution

Competing interests

Rachel Templin

Present address

Contribution

Competing interests

Christopher J Peddie

Contribution

Competing interests

Nicole L Schieber

Present address

Contribution

Competing interests

Martin L Jones

Contribution

Competing interests

Lucy Collinson

Contribution

Competing interests

Yannick Schwab

Contribution

For correspondence

Competing interests

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

Categories and tags

Research organism

Further reading