Acquiring Tilt Series
The basic steps involved in acquiring a tilt series are quite simple: tilt by an increment, adjust image position to keep features of interest centered, adjust focus to maintain a desired level of defocus, and acquire and save an image. However, several factors make this process considerably more complicated. First, it is highly desirable to skip some of these steps as often as possible, because of the time and specimen dose required to perform them on every tilt, yet it is not trivial to skip steps without jeopardizing the quality of the tilt series. Second, the need to track image position with a lower magnification image when working at a high magnification, or with an image away from the area of interest when trying to minimize dose, makes this operation more complex. Third, things do go wrong that require manual intervention, and the program must have the flexibility to allow for a variety of interventions.
The part of SerialEM that manages tilt series acquisition is called the Tilt Series Controller (TSC). It deserves to be thought of as an entity in its own right because it takes control of many aspects of program operation. For example, you will not be able to tilt, copy images to certain buffers, or save images directly while the TSC is in control - you can do these things only through the TSC.
When the TSC is running, buffer usage is standardized as follows: images roll from buffer A through buffer C as new images are acquired; buffer D is the primary buffer for autoalignment, buffer E is a secondary buffer for aligning low magnification images or Trial images in low dose mode; and buffer F can be read into from a file. Copying to buffers E and F is disabled.
You can suspend a tilt series to change the image alignment or focus. See the help for the Tilt Series Resume dialog box for more details on what options are available when you do this. You can also back up to previous tilt angles; see the help for the Tilt Series Back Up dialog box for more details on this process.
Minimizing tracking errors
Running a tilt series continuously from one extreme tilt to the other can result in the area of interest not being well centered at zero tilt. This can happen either because of inaccurate tracking from high to low tilt, or because it was difficult to recognize the correct center position at the starting tilt angle. SerialEM has several features that can help prevent this problem. One is the 'Walk up' procedure, which allows one to start at zero tilt (or a higher tilt if desired) and tilt up to the starting angle in a series of coarse increments with tracking images that are successively aligned to each other. This could be particularly useful when it is hard to recognize the features of interest at high tilt. You can activate the 'Walk Up' procedure yourself (from the Tasks menu) before you start the TSC, or you can activate the TSC at a low tilt angle and have it walk up to the starting tilt angle.
The second feature that may help tracking is called a 'mid-tilt anchor'. This is an image taken at around 40 to 50 degrees that the TSC can use as a reference when the tilt series reaches that angle. The angle should be low enough so that tracking is likely to be accurate from that point downward, but high enough so that uncorrected Y-axis errors above that point are unlikely to cause data loss in the reconstruction. (Note that because the region of interest is foreshortened in the Y direction at high tilt, alignment errors can be quite high in Y without losing any of the region of interest.) Angles in the range of 40-50 degrees should satisfy these criteria. If you are starting the TSC at a low tilt angle and having it walk up to the starting angle, you can have it leave an anchor at an appropriate angle. If you are going to high tilt before starting the TSC, you can also specify an angle to leave an anchor at when you use the 'Walkup & Anchor' procedure.
The fallback feature for fixing a tracking error is manual intervention. Simply stop the series with the 'End' button. It will stop with a Record image in Buffer A. Shift this image to the desired alignment with the right mouse button. Then select the 'Resume' option, check the box 'Use image in buffer A as reference for alignments' in the Tilt Series Resume dialog box, and push the 'GO' button.
Bidirectional tilt series
A different strategy, running the tilt series in both directions from zero tilt, can be used to avoiding problems in reaching an initial high tilt angle and in tracking accurately when coming down from a high angle. This strategy is most useful in two situations, running multiple tilt series with cryo-specimens automatically, or taking tilt series at very high magnifications, but it may also be preferred even for single tilt series with some cryo-samples. Its main disadvantage is that it introduces a discontinuity in the tilt series at the starting angle. Almost certainly, tilt angles will not be evenly spaced across this discontinuity because of backlash. Plastic sections will shrink in the first half of the series, causing a substantial size change between the zero-degree image and the starting image of the second half of the series. On a few microscopes, stage imperfections cause an apparent change in X-axis tilt between the two halves of the series. All of these problems can be accounted for when aligning the tilt series in IMOD. In version 4.7, the eTomo user interface allows one to specify that a tilt series is bidirectional and sets some appropriate parameters automatically. The Scan Header function of Tomogram Setup also detects whether a series from SerialEM is bidirectional and sets the option appropriately, because the header of the output file has an entry in its second title line showing the starting angle after 'bidir = '.
Although bidirectional series avoid some tracking difficulties, they involve a different challenging step, returning to the starting angle for the second half of the series. The program takes a lower magnification image at the starting angle, the 'bidirectional anchor' image, and aligns to this image after returning to that angle for the second half of the series. However, this re-alignment occurs after a much bigger tilt angle change and longer time interval than any other alignment operation in a tilt series, so it may require an image with a larger field of view than in other operations. For example, in Low Dose mode, you may find that View images with a relatively small field of view work in all other aspects of the series but are inadequate for the anchor image. Because of this vulnerability, the Tilt Series Setup dialog shows the anchor's field of view in microns and highlights the value when a magnification is chosen that makes the value too small.
In order to produce output files in the proper order, SerialEM restacks the data from the first half of the series into a new file, in inverted order, before it adds images from the second half. This restacking is done in the background while the program is tilting back and realigning to the starting position. When those operations are done, it then finishes the restacking before tilting to the first angle of the second half. If a metadata autodoc (.mdoc) file is being written, a new file will also be produced in the proper order.
Control of specimen exposure
Because a sample becomes thicker in the direction of the beam when tilted, a smaller fraction of incident electrons reaches the camera at higher tilts. It is usually desirable to keep the signal-to-noise ratio in the acquired images roughly constant through the tilt series, which requires greater exposure to the specimen at higher tilts. The default (and original) method for increasing exposure at higher tilts is to increase the beam intensity by condensing the beam. The motivation for this approach is that beam-induced drift depends primarily on the total current hitting the sample rather than its local flux; condensing the beam thus tends not to increase this component of drift. However, it is also possible to increase exposure time instead of beam intensity, which is preferable in some cases. If you choose this approach, it is important to assess whether drift-free images can be obtained with the highest exposure time needed.
Whichever method you choose for increasing specimen exposure, there are two fundamentally different ways of controlling the change. One is to adjust the intensity or exposure time after each Record image so as to maintain a constant number of counts through the series. This method is generally suitable for plastic-embedded samples. The other method is to adjust intensity or exposure time by a formula based on the cosine of the tilt angle. This method is preferred for cryo-samples because the total dose and its relative distribution through the series can be determined accurately.
Starting a tilt series
When taking a tilt series in one direction, from one extreme angle to the other, the TSC can be started either at a low tilt angle or at the starting tilt angle. You can safely start at low tilt if you are fairly confident that:
If you have doubts about any of these factors, you should go to high tilt before starting the TSC, and make sure that images look good, that the specimen is focused, and that the beam or exposure time can be adjusted properly. These conditions may be satisfied if you have already done one tilt series on the specimen, in which case letting the TSC start from low tilt would save some effort.
Here is a sequence of steps for starting a tilt series:
After you start the tilt series, the TSC will perform a number of preliminary actions, which may include finding eucentricity, walking up, finding focus, getting a first reference image, and adjusting intensity. Also, the program will use the Reverse Tilt procedure to eliminate tilt backlash, so there is no need for you to do this yourself. This procedure involves tilting up and back by about 3 degrees, with lower magnification pictures taken before and after to track the potentially large shift in position that can occur. If there is 'pole touch' in this process, the program will try to compensate for this by taking lower magnification tracking images during the first few tilts, until the backlash is worked out.
How predictions work
The TSC minimizes the number of tracking and focusing positions by predicting where the specimen will be after each tilt. It predicts position separately in the X and Y directions (along and across the tilt axis) and uses image shift to compensate; it predicts the position in Z and changes focus to compensate. At the beginning of the tilt series, the information needed to make reliable predictions accumulates gradually. After two tilts, there is enough information to make simple extrapolations, but not to know how accurate they are. With data from three or more tilts, the program can assess the accuracy of a prediction from how well the positions fit a straight line. In addition, if predictions were made on the previous tilt, the program can determine how far off they were. Thus, at a given tilt angle, there are two kinds of prediction errors available: the computed error of the prediction for the next angle, and the actual error in the prediction for the current angle. Only when both of these errors are low enough will the TSC rely on a prediction and skip a tracking or focusing step.
The accuracy of the prediction generally increases as more data become available, but only as long as the positions change regularly. Because the specimen and stage do not behave ideally, it is necessary to restrict the positions used for the predictions to those from the most recent tilts. For each axis, there is a maximum tilt range over which positions will be fit (a parameter that can be set in the.) Moreover, the TSC may restrict the tilt range used for a particular fit even more if it substantially in increases the accuracy of the prediction. The number of points dropped from a prediction fit is reported as 'nDrop' in the line describing the prediction. These limitations are expressed in terms of tilt range rather than number of tilts because the deviations from ideal behavior that limit the quality of predictions probably occur over a certain range of motion independent of the number of steps. If you have smaller a tilt increment, the predictions should be good over a larger number of tilts.
Certain events are considered to disturb the predictability of the system and will cause the TSC to ignore some or all of the position data that are available. These events include resetting the image shift, tilting back to redo a Record image, and changing the alignment reference. If you stop then resume a tilt series, the TSC will try to determine whether you have done something that would jeopardize the predictions.