One of the approaches for elemental mapping by EELS makes use of the energy filtering in the transmission electron microscope imaging mode (EFTEM) (3, 4). Electrons that are transmitted through a thin specimen enter the magnetic field of the spectrometer and are then dispersed according to their kinetic energy or velocity. A slit placed at the dispersion plane of the spectrometer’s magnetic prism selects electrons of a particular energy loss, which then travel through a series of lenses to form a high resolution image of the specimen on a two-dimensional array detector. By inserting an energy-selecting slit, elastically scattered as well as unscattered electrons are allowed to pass, which result in improvements in contrast in some applications.

To obtain element-specific maps from sectioned cells, it is typically necessary to record two images, one at an energy loss just above the atomic core level excitation threshold, with image intensity I _post, and one at an energy loss just below the threshold, with image intensity I _pre. By carefully subtracting a fraction of I _pre intensity from I _post intensity, it is possible not only to determine the net core edge signal but also to quantify the number of atoms of a particular element in the specimen using the theoretical scattering cross section for the core level excitation (2). The number of atoms n _x of an element x in a pixel of dimension d can be determined from the net signal at energies above the core edge S _x (Δ, β) measured in a band of width Δ defined by the energy-selecting slit and for a collection angle β, which is defined by the objective aperture:
where I ₀ (Δ, β) is the measured zero-loss intensity in the transmitted electron beam, for the same energy band and objective aperture, and σ _x (Δ, β) is the inelastic scattering cross section for the core level excitation for the energy band Δ and collection semi-angle β (5). Usually the number of atoms of element x is much smaller than the total number of atoms in the volume of the specimen, so there is a strong background intensity underlying a relatively weak signal. Therefore, it becomes essential to subtract the background very carefully from the post-edge energy-selected image.

Several methods have been employed to take into account the background intensity (6, 7). The simplest approach is ratio mapping, for which post-edge I _post is divided by the pre-edge intensity I _pre. If the shape of the background remains constant across the entire image (which is rarely the case), the ratio value k _x gives a simple representation of the distribution of element x:

However, it is useful to obtain a modified ratio, by subtracting a constant k ₀ , which corresponds to the measured ratio value in a region of the specimen where the element x is known to be absent.

Multiplying this equation by I _pre gives the so-called two-window intensity S _x , which is an accurate measure of the core-edge signal, provided that the shape of the background remains constant over the field of interest.

When the specimen thickness is nonuniform, which often occurs, the shape of the spectral background varies significantly over the field of interest. Thus, the above mentioned method of calculating the core-edge signal becomes invalid. Previously, this change of spectral shape has been taken into account by collecting additional pre-edge images at different energy losses. For example, in the three-window method, two images are collected below the core-edge (8). By fitting the pre-edge background intensity as an inverse power function, its shape is allowed to vary across the image. However, for beam-sensitive biological specimens the electron dose is restricted and thus the three-window method drastically increases the statistical noise in the extrapolated background, which results in a substantial reduction in analytical sensitivity. For this reason, it is preferable to retain a two-window approach by modifying it to take into account subtle changes in the background shape as a function of the sample thickness t. In this case, it is convenient to measure the specimen thickness in units of inelastic mean free path λ, which can be readily determined from the fraction of transmitted electrons in the zero-loss peak:

Here, I _unflt is the total unfiltered image intensity. The inelastic mean free path for plastic embedding material is typically 150–200 nm for a microscope accelerating voltage of 120 kV and 300–400 nm for a microscope operating at 300 kV.

To incorporate the thickness variation into the two-window method, we compute the spectral intensity as a function of t/λ using logarithmic Fourier deconvolution method (7). It is then possible to calculate the ratio k ₀ as a function of t/λ, resulting in the k-curve or a fitted function, which provides a set of parameters that can be used to adjust the k-value at each imaged pixel. It is then straightforward to compute a thickness-corrected or modified two-window elemental map and write the following expression for the number of atoms in a pixel of dimension d.

This ability to correct for thickness effects in elemental maps by using a t/λ dependent k-ratio enables two-window elemental mapping to be performed in specimens that are as thick as one inelastic mean free path, without artifacts due to plural inelastic scattering.

Here we will mainly discuss the application of the modified two-window method on mapping phosphorus (P) in sectioned cells in 2D as well as in 3D. Since DNA contains one phosphate group for every base in the structure, EFTEM imaging of phosphorus provides an intrinsic molecular label for nucleic acid, and can be combined with nitrogen imaging, which provides a means of distinguishing between DNA and protein complexes (see Fig. 2). It is not feasible to map the distribution of phosphorus in frozen hydrated cells, except at low spatial resolution, because of their high sensitivity to damage. EFTEM can be used to image phosphorus in cells or tissues that have been prepared by high pressure freezing and freeze substitution or conventional chemical fixation, embedding and sectioning. In such preparations, it is important to avoid heavy metal fixatives and stains that could interfere with the phosphorus edge in the energy loss spectrum. Until recently EFTEM phosphorus mapping of cells in unstained plastic sections has involved the collection of two-dimensional maps. By combining EFTEM with tomography, it is now possible to obtain three-dimensional phosphorus distributions of cellular components (9).