NanoSIMS for biological applications: Current practices and analyses

Secondary ion mass spectrometry (SIMS) has now enjoyed more than a century of development since Sir J. J. Thomson first produced and described secondary ions.¹ Honig wrote an excellent treatise on the early history of SIMS for the interested reader.² Still, as of this writing, the SIMS field is continuing to grow, where every aspect of SIMS instrumentation, experimentation, and data analysis are being intensively developed. Arguably, one of the current pinnacles of development in SIMS instrumentation is the commercially available NanoSIMS 50/50L (CAMECA, Gennevilliers Cedex, France). Implementation of the NanoSIMS has had a significant impact in the analysis of biomaterials and biological samples, which will be the focus of this review. There are a number of other reviews written about this technique in multimodal investigations, which we recommend to the interested reader.^3–8 In early 2009, Boxer et al. published a review of SIMS for biological applications, which heavily focused on work reported using the NanoSIMS up to that point in time.⁹ We intend for this review to be similar in focus, but exclusively NanoSIMS-centric, and primarily to cover work that has been reported since ∼2009 to early 2017. It is first worth exploring, briefly, the theory and history of SIMS to serve as a primer to understanding the origin of the NanoSIMS and its current practices and applications, especially within the biological field. While this review aims to be as comprehensive as possible, there are reports we might have missed.

In a SIMS experiment, material under high-vacuum (typically 10⁻⁵–10⁻¹⁰Torr) is removed by a primary ion beam in a process referred to as “sputtering.” Primary ion impact energies vary but are generally in the hundreds of electron-volt to the tens of kilo-electron-volt range. The sputtered material consists of neutral and ionized atoms and molecules, as well as secondary electrons, which are ejected from the top monolayers of the sample's surface. The ionized atomic and molecular species are then accelerated into a mass spectrometer (MS) for analysis. As a result, SIMS is capable of creating ion maps of different species across a sample, and in a multiplexed fashion (i.e., can visualize more than one ion at once), which reveals the samples composition.

Quadrupole, time-of-flight (ToF), and magnetic sector instruments have historically been the most common types of mass spectrometers in use for SIMS applications, but new Fourier transform-based mass spectrometry (FTMS) methods are being pioneered.^10,11 Each of these types of mass analyzers have their own advantages and limitations (e.g., mass resolution, transmission, duty cycle, and quantitativeness). Further, there exists a distinction between “static” and “dynamic” SIMS modes. In static SIMS, the primary ion dose is generally below 10¹³ ions cm⁻², which corresponds to impacting less than ∼1% of the atomic sites on the sample surface.¹² Thus, molecular fragments containing significant chemical bond information exist. In dynamic SIMS, relatively large amounts of material are removed, as primary ion doses exceed the so-called “static limit.” In general, NanoSIMS is useful as a dynamic mode technique, where molecular information is generally not sought. In part, this is because the reactive primary ion beam utilized in these experiments, in conjunction with the beam's orthogonal alignment to the surface, fragments surface molecules quite significantly, and therefore NanoSIMS is operated in dynamic mode to maximize signal.

For secondary ion imaging, two broad classes of instruments exist: ion microscopes and ion microprobes.¹³ Ion microscopes preserve the spatial orientation of ions sputtered from the sample surface throughout the mass spectrometer to the detector, analogous to photons traveling through a light microscope and have the advantage that spatial resolution is independent of primary ion beam size, thus permitting the use of large ionizing primary probes.¹⁴ However, spatial resolution is improved by removing secondary beam aberrations with physical apertures in the mass spectrometer, which also limits transmission and, therefore, sensitivity. Another restriction of ion microscopy is the spatially resolved detectors for imaging (e.g., microchannel plate)¹⁴ that are inherently limited in their accuracy and precision. In contrast to ion microscopes, ion microprobes raster the primary beam over the sample surface and the secondary ion beam is collected and focused onto the mass spectrometer entrance aperture using transfer optics within the mass analyzer. Primary ion beam position and secondary ion signal are correlated under software control to reconstruct an ion image. Here, image resolution is directly related to beam diameter and pixel density, and thus a main limitation of ion microprobes is the requirement for finely focusable primary ion beams. This limits the choice of the primary beam and increases cost due to the need for improved primary column optics. While the NanoSIMS is designed as a dedicated ion microprobe, there are instruments that can operate in either microprobe or microscope mode (e.g., the CAMECA IMS series and early versions of the PHI TRIFT series).^13,14

The history of SIMS and mass spectrometry imaging (MSI) arguably started at the dawn of mass spectrometry itself, as J. J. Thompson visualized positive rays onto photographic plates.¹ In the reporting of these experiments, he passively mentioned that material was sputtered off the cathodes, and a fraction of this material was charged (although there was no mention of how this was observed). However, modern SIMS started in the early 1950s, when the first attempts were made to determine if secondary ions produced by primary ion bombardment could be used to characterize the surface composition materials (e.g., oxide-coated cathodes).² The 1960s brought on an explosion of SIMS growth and development. Here, ion microscopes became the first imaging mass spectrometers to be developed and, eventually, commercialized. During the early years of modern SIMS, this technique became indispensable for characterizing inorganic materials, where, for example, its involvement in the characterization of dopant distributions within semiconductors cannot be overstated. While previous exploratory studies into characterizing organic materials had been reported, it was in 1968 when Benninghoven described the static SIMS approach that a tipping point for ion bombardment methods for analysis of “soft” materials occurred.¹⁵ With the development and implementation of charge neutralization methods in the early 1980s, static SIMS became significantly more utilized in the analysis of organic macromolecules (including biomolecules). Eventually, as ToF mass analyzers were further developed, they began replacing quadrupole MS in static SIMS instrumentation.¹⁶ 

Static ToF-SIMS had become a widespread tool in surface characterization of biologically relevant material by the early 1990s. During this period, Slodzian and coworkers described and developed a new ion beam configuration, which has become especially useful for high-lateral resolution dynamic SIMS measurements.^17,18 Here, the primary ion beam is highly focused, where a very minimal working distance between the last primary focusing optic and the sample exist, and orthogonally impacts the sample surface. This orthogonal primary ion beam, which acts coaxially with the secondary ion beam extraction, provides improved spatial resolution compared to instruments using oblique beams that have ovoid spot shapes, but it requires opposite polarity for primary and secondary ions. This configuration has now been commercialized by CAMECA as the NanoSIMS 50, and it utilizes a magnetic sector mass spectrometer to measure secondary ions as they are generated from the sample and extracted toward the MS. Figure 1(a) illustrates the optic configuration of the current NanoSIMS 50L system.

The NanoSIMS employs reactive primary beams, where Cs⁺ is used for negative ion analysis and O⁻ (or rarely O₂⁻) for positive secondary ion analysis. The limit of offering only these two sources is minimal, as positive ion yields are increased under O⁻ bombardment and negative ion yields are increased under Cs⁺ bombardment (see Sec. II A). The development of the microcesium source by CAMECA, which is coupled to the NanoSIMS, provides the ability to focus the primary ion beam to 50 nm. These conditions require a significant reduction in beam current, though, and more commonly the instrument is operated with a beam current of 1–2 pA (beam size ∼100 nm). It should be noted that Levi-Setti et al. demonstrated in the mid-1980s the use of a ∼20 nm Ga⁺ source for ion microprobe analysis.¹⁹ However, this primary ion source was operated at 47 keV, resulting in high lateral resolution at the expense of depth resolution and with lower ionization efficiencies than Cs⁺ primary ion beams.¹³ The standard duoplasmatron oxygen source on most NanoSIMS instruments has a limited lateral resolution in comparison to the cesium source (100 s of nanometer) and has had little improvement since its development for the CAMECA f-series. A new radio frequency (RF)-based oxygen source is now available as discussed in Sec. V A. Since its inception, the primary column has been improved over the years to increase primary current at the sample stage.

For secondary ion collection and detection, the NanoSIMS utilizes a double-focusing design, incorporating both an electrostatic sector and a magnetic sector, and has the advantage of focusing along a focal plane, eliminating the waste of sample due to peak switching. This design achieves a good compromise between high transmission and high mass resolution conditions. The early prototype version of the NanoSIMS contained four moveable electron multipliers (EMs), which were used to detect preselected secondary ions, and a stationary Faraday cup (FC) used for diagnostic purposes.^20,21 The first production NanoSIMS 50 instruments were delivered with five miniature EMs (four of which were moveable), an additional large, fixed EM located past the miniature EMs, and a single FC accompanying an EM on the first detector trolley. Although it was possible to obtain higher isotopic precision compared to EM/EM analyses on some spot measurements by using the FC for an abundant isotope (e.g., ¹²C, ¹⁶O, or ³²S), low noise FC amplifiers and secondary electron repellers were not added until later, allowing for true high-precision measurements. The most current version is the NanoSIMS 50L [Fig. 1(a)]. This instrument has six movable EM/FC trolleys and one stationary EM/FC, where monitoring of up to ten signals in a single imaging experiment has been reported.²² The NanoSIMS 50 has a mass range (M_max/M_min in parallel acquisition) of ∼13. In practice, this translates to a mass range that can accommodate analysis of, for example, m/z 12 to 156, where collection of adjacent masses up to m/z 30 is possible, allowing for multicollector isotope measurements below this mass. The NanoSIMS 50L has a mass range of ∼21, allowing for detection of mass intervals between adjacent detectors up to m/z 58 (thus, making simultaneous collection of Fe isotopes possible). However, magnetic-peak switching is possible on both instruments to overcome mass range issues within a given experiment. This method employs alternating magnetic field strengths between imaging frames to detect various ions of interest. The type of species detected, importance of transmission, mass resolution, and simultaneous ion species collection will be discussed in Sec. II.

Since its introduction, the NanoSIMS 50/50L has arguably driven the state-of-the field in high lateral-resolution MSI. At the time of this publication, there are 42 instruments installed worldwide. While subcellular imaging is possible using ToF-SIMS,²³ NanoSIMS continues to be a popular method for high lateral-resolution analysis of biologically relevant samples.

There are a variety of factors that must be carefully considered for successful utilization of NanoSIMS to analyze samples of interest. First and foremost are the basic principles of analytical chemistry and instrumental analysis, but other factors, such as data representation, are perhaps equally important. Universal to nearly all SIMS methodologies is to ensure that samples are high vacuum compatible and exhibit minimal topography. Conductive samples are also preferable, but insulating samples can be analyzed via charge neutralization (i.e., conductive coating and electron gun bombardment). Additional considerations when performing NanoSIMS analysis include sample labeling and predetermination of signals of interest. This section will address these topics and more in detail.

Prior to running an experiment, consideration must be given to what is being measured. Since NanoSIMS is a dynamic SIMS method, which generates highly fragmented secondary ions with limited molecular information, results from biological samples are essentially inferred from detection of monoatomic or diatomic species. Table S1 lists examples of ions detected within the reports reviewed here.³⁰⁷ There are several studies, like those discussing elucidating metal accumulation within biological specimens^24,25 or searching for evidence of biological activity as indicative of isotope fractionation,^26,27 where samples were able to be used “as is.” However, in most NanoSIMS analysis endeavors, light isotope and/or rare element labeling schemes are utilized to enable detection of processes or species of interest within biological samples. Further, incorporation of specific probes, like that developed by Vreja and coworkers,^28,29 can increase precision in biomolecular analysis by removing background-related signals. Whatever labeling scheme is decided upon, researchers must be cognizant that only five or seven species can be detected within a single imaging scan (depending on model 50 or 50L, respectively), and the finite physical footprint of the magnetic sector mass analyzer dictates the balance between mass range and isotope resolving power.

Stable isotope-labeled species (e.g., nutrients) are often used as probes for revealing metabolic processes within biological systems. For example, this methodology is often employed to illuminate how carbon and/or nitrogen are fixed within microbial and plant systems, where samples can be incubated in an enriched ¹³CO₂ or ¹⁵N₂ atmosphere or an enriched ¹³CO₃⁻ or ¹⁵NO₃⁻ growth media. Metabolic activity can be visualized using NanoSIMS by identifying where isotopic enrichment is localized. Typically, this is based on measuring the ¹³C⁻/¹²C⁻ (or ¹³C¹²C⁻/¹²C₂⁻) or ¹²C¹⁴N⁻/¹²C¹⁵N⁻ value and comparing it to a standard sample and/or value. Note, due to the extreme instability of N⁻, among other mass measurement factors, nitrogen is typically measured using the CN⁻ ion.³⁰ The combination of NanoSIMS and fluorescence in situ hybridization (FISH) in these types of studies has been especially useful, as it provides the ability to link metabolic activity to taxonomy,^31,32 though users are warned combining these two methods could impact results (e.g., isotope enrichment dilution).³³ FISH methodology has been expanded to incorporate elements into the nucleic acid probes (e.g., F⁻, Br⁻, I⁻), where NanoSIMS detection of these elements provides the phylogenetic identification of species of interest that can be localized with metabolic activity, thus removing the need for fluorescence microscopy.^34,35

Within eukaryotic studies, localization of cell-specific molecules, cellular components, and cellular active species can be achieved by a variety of methods. For example, incorporation of isotopes into various lipid species have been successfully performed via metabolic labeling strategies.^36–39 The Kraft group has done this successfully by incorporated ¹⁵N-sphingolipid precursors into the cellular growth media as a way to map sphingolipid distribution within cells.^36,37 These types of labeling methodologies require (1) a comprehensive understanding of the metabolic pathways that can be exploited and (2) careful optimization of cell culturing to ensure labeling is sufficiently high for species of interest with little nonspecific incorporation into other species. The latter requirement might require the use of complementary bulk methodologies (e.g., liquid chromatography tandem MS) to ensure labeling is optimized.⁴⁰ Similar strategies, where cells are cultured in media containing ¹³C-fatty acids, have been used to visualize cellular components (e.g., plasma membranes and lipid droplets).^38,41

Imaging macromolecular structures within cells is relatively straightforward, as there are numerous commercially available complementary molecules for specific labeling (e.g., proteins-antibodies). For example, immunogold labeling, where antibodies are labeled with gold (Au) nanoparticles, are quite prevalent in electron microscopy because their high electron density produces a pronounced dark spot in images. These labels can also be used in NanoSIMS, where detection of ¹⁹⁷Au indirectly measures protein location.³¹ Alternatively, Au nanoparticles provide a large surface area that can be functionalized with fluorinated labels. These so-called “fluorinated colloidal gold immunolabels” afford the ability to image light isotopes (e.g., ¹³C⁻/¹²C⁻ or ¹²C¹⁴N⁻) in parallel with the labeled antibody without having to perform magnetic-peak switching between imaging scans.^42,43 An antibody probe that incorporates both a ¹⁹F-enriched label and a fluorophore to simultaneously visualize protein distribution and cellular metabolism has also been reported.²⁹ Recently, Angelo et al. demonstrated the ability to measure ten labels simultaneously within breast cancer tissue using the same immunohistochemistry methodology, where antibodies were labeled with metals that included lanthanides.²² 

Labeling nanoparticles and drugs with stable isotopes can permit the ability to reveal their intercellular mode of action. For example, Wedlock et al. used ³⁴S to localize anticancer Au nanoparticle complexes in the nuclear and cytoplasmic regions of the cell.⁴⁴ There are a few reports using metal-based anticancer drugs that are functionalized with stable isotope molecules in order to elucidate their behavior within a cell.^45–47 Most recently, Proetto et al. developed a nanocarrier containing cytotoxic plutonium, which incorporated both a fluorescent and ¹⁵N-label in its architecture.⁴⁸ Accordingly, the authors were able to use NanoSIMS to image how the nanocarrier entered the cell using high resolution fluorescence microscopy, and if the nanocarriers remained intact once they entered the cell.

Arguably, the universal rule across all MSI analyses is that sample preparation is the most critical step of the whole process. Goodwin comprehensively described elsewhere how carelessness in this step may create image artifacts.⁴⁹ For NanoSIMS analysis, samples must be vacuum compatible and free of topography. Further, most samples need to be free of salts or other compounds that could interfere with the ionization process. Mapping prepared samples with complementary imaging modalities (e.g., optical and electron microscopy) prior to NanoSIMS analysis can enhance the efficiency of the analysis, while simultaneously providing correlative information about the sample. The NanoSIMS is equipped with an optical microscope, which allows ease in sample navigation to locate areas of interest. Often these areas are identified by fiduciary markers that can be observed by both optical and electron microscopy.

Preparing samples for in vacuo MSI analysis is nontrivial. Ideally, samples should be preserved as close to their native states as possible to maintain the asymmetric distribution of biomolecules. Compared to MSI modalities capable of visualizing molecular species (e.g., matrix assisted laser desorption/ionization), sample preparation for NanoSIMS is relatively simple. Since elemental/isotopic distributions are being measured, samples can often be chemically fixed and/or embedded to make them vacuum compatible. Chemical fixation with glutaraldehyde is the gold standard for crosslinking proteins, but it is nonideal for fluorescence microscopy due to autofluorescence. Alternatively, fixation with paraformaldehyde is also possible, but is less effective at preserving cellular ultrastructure (typically unresolvable by NanoSIMS). Osmium tetroxide can be used to crosslink lipids and preserve their cellular localization.^38,39 After chemical crosslinking, samples are typically washed with water to remove salts, followed by ethanol dehydration using ethanol:water solutions of increasing ethanol concentration. There are several different embedding media that can be utilized depending on the type of sample being analyzed. Commercially available polymers used in electron microscopy measurements (e.g., acrylic, epoxy) permit room temperature microtome sectioning.⁵⁰ Flash freezing samples can assist in maintaining the distribution of biomolecules without chemical fixatives. These samples are often embedded in optimal cutting temperature media and carboxymethylcellulose, as well as water/ice to some varying degree of success, and then may be sectioned using a cryomicrotome. Note that depending on the sample composition, both chemical fixation and embedding processes can dilute isotopic signals of interest in a sample. For example, most embedding materials contain a significant amount of carbon that can considerably decrease measured ¹³C-enrichment.^33,51 Postfixation functionalization methods for complementary image analysis, like FISH, can also reduce isotopic enrichment signatures even further,³³ and maybe disruptive to biological features of interest.

Ideally, samples should be conductive as to minimize charging during NanoSIMS analysis, but charge neutralization can be performed during analysis. Silicon chips are commonly used because of their conductivity properties, affordability, and ability to be diced from large wafers to desirable sizes. Transmission electron microscopy (TEM) grids are another sample support option. However, ultrathin sample thicknesses preferable for TEM might not generate ideal SIMS images, whereas thicker sections (up to 500 nm) that can give more desirable NanoSIMS images over larger areas produce lower quality TEM images.³¹ Indium tin oxide (ITO)-coated glass slides and coverslips provide a conductive substrate that is optically transparent. ITO-coated glass substrates are ideal for correlative transmission light optical microscopy methods. Metal-coated polymer filters used for isolating cells from their environment have also been used as substrates.^52,53 There are numerous examples of biological systems that can be directly grown on these conductive substrates^38,48,54 though, in most cases, samples must be transferred onto substrates. Coating the substrate surface with poly-l-lysine, vector bond, or other compounds might be required to enhance adhesion of sample to the substrate. Sample topography should be minimized, as it induces varying ionization trajectories across the sample and can complicate quantification. Often the final step in sample preparation is to conductively coat with Au, Ir, or C (to name a few) using a sputter coater or evaporator. This step enables sample charging to dissipate to ground more readily.

Without the addition of a postablation ionization mechanism,⁵⁵ energy for ionization comes exclusively from primary ion impact. As a result, the sensitivity of SIMS spans more than 5-orders of magnitude, depending on the species, where it is correlated to first ionization potential in the positive secondary ion spectrum and electron affinity in the negative spectrum.⁵⁶ Accordingly, one of the first considerations when designing a SIMS experiment is simply which secondary ion polarity will provide better signal. Figure 2 illustrates the best ionization source for the detection of different elements. For example, if the goal is to determine the location of calcium or potassium deposits within a microfossil, then ionization with the duoplasmatron (O⁻) primary ion source is necessary for detection of positive secondary ions.

The NanoSIMS has very high mass specificity. Explicitly, it has the ability to distinguish between adjacent masses, as defined by mass resolving power (M/ΔM), where M is the nominal mass being measured and ΔM is the mass difference resolvable between neighboring species. As designed, the NanoSIMS is capable of M/ΔM ≈ 10 000.¹⁷ It is worth noting that M/ΔM of this dynamic SIMS, magnetic sector-based instrument is not directly comparable, based upon the fashion in that ions are collected and measured, to the M/ΔM of ToF-MS and FTMS-based instruments, which are capable of M/ΔM > 10 000 and > 1 000 000, respectively.^9,57

Within the NanoSIMS, the secondary ion beam must be tightly focused to allow collection of effectively 100% of ions of interest into the detectors in order to attain high precision mass measurements (accuracy and precision are discussed in Sec. II D). Proper tuning should ideally allow a margin for small variations in the magnetic field or other potential shifts in the mass line (e.g., due to room temperature fluctuations). In practice, this means optimizing the transmission of secondary ions through a narrow energy window at the entrance slit to the mass spectrometer and then optimizing deflector conditions in front of the detector trollies, which were prepositioned to collect desired secondary ions. As the mass line is scanned (e.g., voltage scanning the detector deflectors), proper tuning would result in a step-sided, flat-topped peak. The sharpness of this flat-top peak is its M/ΔM, which is effectively applicable across all masses. Note, the NanoSIMS software estimates M/ΔM based on the steepness of the side slopes as

where R_eff is the effective radius of the detector position, and L_10/90 is the mean lateral distance between the 10% and 90% heights of the side slopes. A result of this estimation is that the M/ΔM_NS50 is ∼150% greater than M/ΔM based on its standard notation. Some groups report M/ΔM based upon this correction.³¹ 

Reporting of isotopic ratio measurements differs somewhat by discipline and experimental relevance. We present here an abbreviated primer on isotope notation and use the stable C-isotope system as an example. Agricultural and other environmental studies that use isotopic labeling have often used atom% (AP) or atom% excess (APE) to express isotope enrichment. Here, AP is given by

where, in the case of a NanoSIMS analysis, ¹²C and ¹³C are estimates of the total ion counts for a given measurement. APE is a formulation that is convenient for some mathematical manipulations and is derived by subtracting a background ¹³C value from AP¹³C. Thus, for ¹³C measurements, a possible formulation is

In this case, we chose the AP¹³C for the internationally accepted standard Vienna Pee Dee Belemnite (VPDB, see Table I) to convert AP¹³C to APE¹³C,⁵⁸ but any convenient background value could be substituted for a particular experiment. For the remainder of this section, we constrain our consideration to relative isotope ratios using δ notation as this is the most common notation reported in the literature.

The concepts presented here are applicable to other formulations such as R, AP, and APE. The most fundamental isotope ratio measurement is the relative abundance or simply isotope ratio (R), which in this case is defined as

Since natural abundance isotope ratio measurements typically vary little (usually in the third to fifth decimal place) and precise isotope abundances are very difficult to quantify, natural abundance measurements are most often reported in terms relative to an agreed upon standard. In these cases, relative abundance is most often described in terms of parts per thousand or permil (alternatively “per mill,” “per mil,” or “per mille”) and designated symbolically as ‰. The relative abundance of a ¹³C abundance measurement can be expressed in permil as

where R_sam is the isotope ratio of the sample of interest, and R_std is the isotope ratio of the standard (e.g., Table I). Fundamentally, these standards are usually internationally agreed upon. An abbreviated table of common, internationally agreed upon isotope ratio standards is presented in Table I.

In order for the proper interpretation of SIMS data to be possible, an understanding of the quality of the data with respect to analytical uncertainty must also be achieved. As illustrated in Fig. 3, accuracy and precision are independent of one another. Fitzsimons et al.⁵⁹ presented an accessible discussion of analytical precision of SIMS stable isotope measurements. We will briefly revisit these concepts and extend them to include MSI. Although the terms are often confused in the literature, we subscribe to a philosophy that error, precision, and uncertainty have different meanings. In the literature, “error” is often used synonymously with accuracy as well as precision. Because error is often used to refer to different concepts, we will dispense with this term in this review and use “accuracy” instead. In this case, accuracy is a measure of agreement between an analytical estimate and the true, or agreed upon, value. Further, precision is a measure of the closeness of agreement between multiple independent measurements, which is irrespective of the true value. Uncertainty is a component of a reported value [expressed as a standard deviation (SD), see below], with an associated probability that theoretically encompasses the true value. Uncertainty includes all relevant random and systematic components of precision.

When we speak of internal precision, we are referring to the precision associated within a single analytical measurement. This measurement may involve a single estimate, or may be comprised of a number of repeated, dependent measurements that can be averaged to form a single estimate. External precision, on the other hand, refers to the reproducibility associated with multiple, independent analyses of identical samples. Recognizing that here we are simplifying the distinction between sample and population SD,⁵⁹ we define a SD where multiple measurements are averaged as

where x_i is the value of an individual measurement, and $x ¯$ is the average of all individual measurements. This formula is valid for calculating internal or external SD. The SD of the mean (also commonly known as standard error) is defined as

Assuming there is no systematic bias in the estimate of $x ¯$⁠, σ _{$x ¯$} is a measure of how well an increasing number of independent measurements approach the true value. σ _{$x ¯$} is often used as an estimate of internal precision when multiple, dependent measurements are averaged into a single analysis.

When an analytical estimate has more than one component of uncertainty or the result is expressed as a function comprised of more than one variable, it is important to know how the precision of individual variables can be combined to give an estimate of precision or uncertainty associated within the final result. Assuming negligible covariance (which is the case for isotope ratio measurements),⁵⁹ the SD associated with a function (F) that contains two components of variability (a and b) can be expressed as

where in our C-isotope example, ¹²C counts may be represented by a and ¹³C counts by b. Note, these variances do not have equal weight. The partial derivatives that govern Eq. (8) are weighting factors known as sensitivity coefficients. This treatment is a greatly simplified treatment of the law of propagation of uncertainty.

NanoSIMS measurements involve the counting of atomic or diatomic ions using either a FC (inferred from current measured), or more commonly by an EM (pulse counting detector). Since the circuits that amplify the FC signal have inherent noise associated with them, the limit of precision for FC measurements are governed by the signal-to-noise ratio, and thus FCs are used primarily at high count rates. The fundamental limitation on the precision of measurements using EMs is the total number of ions counted and is governed by Poisson statistics, often referred to as “counting statistics” or “shot noise.” The predicted SD of a SIMS measurement of a single ion species a is given by

where N_a is the total counts recorded for ion species a in a single analysis. Using this knowledge and substituting this value in Eq. (8), it can be shown that for an isotope ratio measurement, the theoretical precision, σ_R, is given by

where R¹³C is the mean ¹³C to ¹²C ratio for a given measurement. This average may refer to temporally or spatially resolved measurements or, simply, a single measurement with no temporal or spatial component. A temporally resolved measurement is more than one analysis cycle that is averaged to make up a single measurement, for example. An example of a spatially resolved measurement might be a region of interest (ROI) in which multiple pixels are averaged to give a mean value. In terms of permil, the relative precision of a measurement is given by

For isotope systems in which the minor isotope abundance is significantly less than that of the major isotope, as is the case with C, a useful approximation is

This expression gives rise to the well-known relationship that states in order achieve a theoretical precision of 1‰, 1 × 10⁶ ions of the minor isotope must be counted. Note that precision increases as the square of the number of counts accumulated such that to achieve a theoretical precision of 0.5‰, 4 × 10⁶ ions must be accumulated.

A useful exercise is to ask what the maximum theoretical precision that might be expected for a C isotope ratio measurement of a single bacterium. As an example, Vrede et al.⁶⁰ estimated the total C in marine phytoplankton harvested during exponential growth phase at about 150 fg. Thus, we can calculate the theoretical precision based on the ¹³C content of a single cell.

Applying this total number of ¹³C ions in a signal phytoplankton to Eq. (12) yields

Consequently, in theory at least, there exist enough ¹³C ions in a rapidly growing bacterium to achieve a degree of precision using NanoSIMS that rivals bulk analysis by conventional combustion isotope ratio mass spectrometry (IRMS). Of course, many factors such as ionization efficiency, instrument transmission, and detection efficiency reduce the number of ions actually detected dramatically, essentially reducing the discussion of precision of such measurements to the empirical.

Inherent to how they operate, EMs create potential artifacts that must be taken into account in order to optimize accuracy. Deadtime (DT) is the time between ion pulses in which an EM detector system is incapable of accepting another pulse. In the case of the NanoSIMS, an electronic filter known as a discriminator is used to set the minimum electronic pulse that is accepted (removing electronic noise) as well as the DT. Although possible DT values vary between 20 and 84 ns, most NanoSIMS laboratories opt for the default value of 44 ns. Although DT is nonlinear over orders of magnitude in count rate, the effect is minimal for count rates encountered in most NanoSIMS experiments and a simple linear correction model can be used, which is

where C_cor is ion counts per second (cps) corrected for DT, C_raw is raw cps, and $τ$ is the detector DT in ns. If higher accuracy is required, it is possible to measure DT for each detector using an element that has at least three stable isotopes (e.g., measuring ⁴⁶Ti, ⁴⁷Ti, and ⁴⁸Ti).⁶¹ 

A similar concept to DT is that of quasisimultaneous arrivals (QSA). Here, instead of underestimating counts waiting for the detector to be available for another ion pulse, counts are underestimated because more than one secondary ion strikes the detector at essentially the same time from a single primary ion impact and are registered in the counting electronics as a single count. Slodzian et al.⁶² published a theoretical correction scheme based on a Poisson probability of secondary ion ejection from a single primary ion impact. In a simplified form, the correction is of the form

where N_cor is the real number of ions reaching the detector, N_raw is the observed number of ions reaching the detector, K is the number of secondary ions ejected per primary ion impact, and β is a correction factor with a theoretical value of 0.5. However, Slodzian and others have shown that β can range from 0.6 to 1.0,^62,63 depending on the isotope system in question. Despite the obvious problems inherent in QSA corrections, we generally apply the correction to imaging isotope ratio data using a theoretical β value of 0.5. Since images can have pixel intensities that vary over orders of magnitude, it is important that the DT correction be done on a pixel-by-pixel basis. Likewise, ion yields change from location to location due to a number of subtle effects. These effects include topography, secondary ion energy, sample charging, and so-called other “matrix effects” (e.g., where the local chemical environment can cause nonconcentration-dependent changes in the signal intensity of an ion of interest). As such, pixel-by-pixel QSA correction is desirable. OpenMIMS, Look@NanoSIMS, and L'IMAGE have this functionality built in (see Sec. IV A).

It is a common practice for NanoSIMS laboratories to use a “working standard” that is externally calibrated to an international standard, to calibrate measurements for instrumental mass fractionation (IMF) on a frequent basis. For example, when performing isotope ratio measurements for ¹³C in our laboratory, we commonly use a yeast working standard that has been calibrated for δ¹³C using bulk isotope ratio analysis (−11.0‰ ± 0.1‰ relative to VPDB; Dr. James Moran, PNNL, personal communication). We prefer to take a daily imaging measurement of our yeast standard to calibrate for IMF, to estimate external uncertainty of ROI measurements, and, perhaps most importantly, to make sure the instrument is properly tuned. Essentially, this is valuable insurance that an analytical session will not be wasted producing data that are not of a useful quality.

There are a number of calibration schemes possible for isotope ratio data. An example of NanoSIMS images and associated ROIs used for calibration is presented in Fig. 4, where Table II illustrates one possible data reduction scheme from these images. Providing there has not been an interruption in measurement integrity between standard analyses (e.g., notable changes in room temperature, accidental changes in tuning voltages, etc.), these data provide for a calibration factor and uncertainty estimates for all subsequent ROI measurements in the analytical session. Here, the average raw R¹³C (pixel-by-pixel corrected for DT and QSA) is 0.02217 with a σ_ext = 0.00018. A δ_raw can be assigned to these using Eq. (5), where R_std is the ¹²C¹³C⁻/¹²C¹²C⁻ ratio corresponding to VPDB (doubled value for measuring C₂, see Table I). Although the δ-scale is nonlinear, it is usually adequate to calculate the ROI values relative to VPDB by simply adding the working standard value (−11) to δ_raw, such that

where $C raw 13 ‰$ is the raw ¹³C/¹²C ratio of a sample and $C corr 13 ‰$ is the ratio corrected for IMF.

The use of mass spectrometry and stable isotopes (e.g., ¹⁵N) for studying metabolism has been around for almost 80 years.^64,65 However, it was with SIMS that scientists were first able to visualize stable isotope incorporation, indicative of metabolic uptake into organisms and tissues, at the subcellular level.^66,67 Peteranderl and Lechene,⁶⁸ using the prototype NanoSIMS, first demonstrated the ability of NanoSIMS to track the incorporation of important biomolecules at the subcellular level. The benefit of this approach, over previous SIMS imaging endeavors within this scope, was the NanoSIMS is capable of imaging multiple isotopes at once, as opposed to one isotope per scan.^66,68 A couple of years later, the Lechene group built on this concept, so-called “multi-isotope imaging mass spectrometry” (MIMS), showing its many applications and making it easier to use with the addition of a novel quantitative image-analysis software⁶⁹ (now called OpenMIMS, see Sec. IV A). In this study, they reported the use of MIMS and the new software in many different bioapplications, ranging from analysis of human cells to microbial cells, as well as tracking whole cells in a larger system to tracking the movement of nucleic acids. These studies inspired an entire field of biological and bio-related mass spectrometry imaging using the NanoSIMS. As noted earlier, this review focuses on those studies and advancements since the publication of Boxer et al. review in early 2009.⁹ Accordingly, our definition of recent is within the past eight years since this publication. We grouped these studies into subsections based on the type of organisms. These subsections are ordered, in general, chronologically with respect to their evolutionary timeframe. Table S1 provides an overview of the experimental and analytical parameters for each of these studies.

In the past decade, NanoSIMS has increasingly become an essential tool in the study of ancient life since it allows the collection of isotopic information in and between different microparticles. Also, with its ability to simultaneously detect multiple atomic and diatomic species with high sensitivity and precision, less of this often delicate and/or limited material is required for data collection. Typically in these studies, conclusions are drawn from presence or absence of certain chemical features indicative of biological activity (e.g., nonrandom distributions of bio-related material) in the matter being studied.

There have been a couple of recent publications that explore this topic for the interested reader.^4,70 Specifically, a book chapter by Wacey details how isotopic information collected by NanoSIMS is used to understand Archaean life, presents results from several studies using this approach, and provided an example case study in order to further explain the use of different instruments in this type of study.⁷⁰ Around the same time, Oehler and Cady published a review that also covered the contribution of NanoSIMS to study ancient life forms, but they also provided a prospective about the history of its use in this field, as they were pioneers in employing NanoSIMS for these types of studies. Further, Oehler and Cady related how their studies later fit in with the studies performed in the Wacey group.⁴ The interested reader is encouraged to refer to these articles for more information. Since these publications cover a significant portion of the studies within this field using NanoSIMS up to 2014, for this subsection section alone, we primarily focus on studies since 2014 in order to provide an updated account of the application of NanoSIMS to study early life and how it relates to Earth's biological history.

In one study, Pacton et al. used NanoSIMS to understand the role of viruses in organomineralization.⁷¹ Here, they measured the abundance of ¹²C¹⁴N⁻ to identify organic material and ¹⁶O⁻ and ²⁸Si⁻ to identify mineralized material. Based on the locations of these species, the authors found primary nucleation sites for mineral precipitation occurred around particles similar to viruses. These particles formed magnesium silicates upon mineralization then altered to magnesium carbonate nanospheres during the process of diagenesis. This showed carbonate nanospheres found throughout the geological record could be due to viruses, though they state other explanations in a later paper, such as membrane vesicles trafficking these minerals.⁷² 

Wacey et al. recently described a novel way to detect, measure, and evaluate the potential biogenicity of framboidal pyrite.⁷³ NanoSIMS proved useful here because, as compared to other methodologies, it was able to show the abundance of C and N within and between fine pieces of pyrite grain, where the colocalization of these two ions and the absence of certain other ions suggest pyrite grains came from preserved organic material.

Another interesting study was that done by Kaźmierczak et al.⁷⁴ Here, well-preserved microfossils containing tubular structures were collected from the Omdraaivlei Formation in South Africa and analyzed using various technologies. NanoSIMS was used to measure the abundance of ¹²C⁻, ¹²C¹⁴N⁻, ²⁸Si⁻, ³²S⁻, ²⁷Al¹⁶O⁻, and ⁴⁰Ca¹⁶O⁻, which helped to show the presence of organic material, the relationship between C, N, and S in the sample, and investigate the mineral coatings associated with the tubular structures. Due to the characteristics of these structures, it was concluded they were of algal affinity, pointing to the first possible identification of a eukaryote in the Neoarchean era, 1.5×10⁹ years older than the previously discovered eukaryotic microfossils.

Other recent studies employing NanoSIMS to studying paleobiology material have acquired further evidence of Paleoarchean life in the Strelley Pool Formation. These results provided new conclusions about the environment during this era.⁷⁵ Further, they report the discovery, and later application, of foraminifera shells' ability to record the ratio of sulfate to calcium found in their environment during their lifetime.^76,77

NanoSIMS has proven to be an essential tool in microbiology, since it offers information about nutrient uptake and the metabolism of individual microbes. In combination with species-specific identification techniques, NanoSIMS has provided an overall better understanding of the role each species plays within its community.^{32,52–54,78–165} The majority of these studies focus on microbial metabolism,^{32,52–54,78–147,162–165} with many specifically using the NanoSIMS to study carbon and/or nitrogen fixation performed by autotrophic species (e.g., cyanobacteria)^{32,52,53,78–87,163} and nutrient exchange between species within consortia.^{52,54,84–112,162–165} Pett-Ridge and Weber extensively covered the application of NanoSIMS in microbiology in a 2012 publication.³¹ Here, we cover many of the same reports as Pett-Ridge and Weber, but primarily focus on some of the new results following their publication.

Nutrient exchange between species is an important aspect of understanding microbial community dynamics. NanoSIMS has proven to be a uniquely valuable tool for this application, since the transfer of stable isotopes between species can be monitored at high lateral resolution. Accordingly, many studies within the past seven years have taken advantage of this application.^{52,54,84–112,162–165} With direct imaging at high resolution, nutrient exchange in uncultivable species have been studied using NanoSIMS,^89,90 as well as in symbiotic relationships^{82,84,86–88,90–98,111–116,148–151,162–165} and complex samples, usually with the assistance of taxonomy identification techniques.^{54,87,89,99–110,115,116,137–144,152} One study that made a significant impact in understanding microbial interactions was reported by Foster et al. in 2011.⁸⁴ Here, they were the first group to demonstrate the transfer of nitrogen from a diazotroph to a unicellular partner using ¹⁵N stable isotope labeling methods. In a later study that used a similar method, they focused on carbon and nitrogen fixation in colonial unicellular cyanobacteria versus those growing on their own.⁵³ In another series of studies, the Orphan group investigated anaerobic methane-oxidizing archaea (ANME) with the use of the NanoSIMS.^100–102 In their initial study, they tracked the assimilation and distribution of ¹⁵N in cells and found ANME could not only fix nitrogen, but also passed these nutrients to sulfate-reducing bacteria in their community.¹⁰⁰ They later continued this study by focusing on nitrogen fixation in the community and the role of ANME.¹⁰¹ 

NanoSIMS has also been used to study nutrient exchange in relation to human–microbe interactions.^91,149–151 Berry et al. isotopically labeled human proteins and studied their uptake by microbial flora common to the human intestine in order to elucidate which species readily harvest compounds secreted by the intestine.⁹¹ Gouzy et al. used NanoSIMS, along with other complementary bulk assaying and high lateral resolution imaging methods, to study Mycobacterium tuberculosis. This provided a better understanding of its methods of nitrogen assimilation, and they proposed new methods to combat tuberculosis.^149,150 Last, Kopf et al. used a recently developed technique,¹¹⁷ where the NanoSIMS is used to track heavy water (D₂O) and ¹⁵N in single cells, to measure the anabolic activity of Staphylococcus aureus in the sputum of pediatric cystic fibrosis patients.¹⁵¹ 

NanoSIMS has been useful in cell toxicity studies by providing direct observation of toxin intake, effects, and localization within microbial cells. In a recent study,¹²⁸ this was used to investigate the antimicrobial properties^166,167 of medicinal clay. Morrison et al. were able to use NanoSIMS to visualize the cellular binding sites of reduced forms of aluminum and iron in Escherichia coli cells when exposed to the medicinal clay, where the harmful free cations of these metals (Fe²⁺ and Al³⁺) naturally seep out of the clay and allow for its antimicrobial properties.¹²⁸ This showed E. coli cells remained enriched with these metals over time, due to their constant seepage from the clay. In conjunction with other analytical methods, the authors concluded the Fe²⁺ enters the cell and creates free radicals that attack intercellular proteins and nucleic acids, whereas and the Al³⁺ binds and misfolds cell membrane proteins (Fig. 5). Another study looked in the effect of transmembrane versus monolayer pores in cell plasma membranes, using the NanoSIMS to image isotope enrichment of pores across different conditions.¹⁶¹ 

Last, worth noting in this subsection, is a report by Schreiber et al.¹⁴⁷ where they provide evidence that limited resources in a natural community increase phenotypic heterogeneity in cellular metabolism. Specifically, NanoSIMS was the only imaging modality used in this study, and it was the key in observing the changes in function of individual Klebsiella oxytoca cells as nitrogen, in the form of ammonia, was limited to varying degrees.

Employment of the NanoSIMS has contributed greatly toward the study of eukaryotic biology.^{22,24,25,29,36–39,41–43,90,91,114–116,126,131,148–152,168–239} Specifically, these studies have helped elucidate metabolic mechanisms between human–microbe interactions,^91,149–151 causes and effects of brain injury and disease,^168–176 processes related to cell division,^{41,151,177–180} modes of protein tracking,^29,181–187 effects of nanoparticles,^{131,174,188–192} and plant and coral metabolite and elemental uptake,^{24,25,115,204–220,240} among other things. Here, we discuss these applications in subsections focused on intra- and intercellular processes.

A variety of tissues, such as heart,^{179–181,196} brain,^{126,169–177,187,193–195} plant and coral tissues,^{24,25,114,115,148,182,204–224,240} have been studied using NanoSIMS in order to better understand the biochemical processes and transformations happening within these organisms. For example, a group at the Center for NanoImaging (CNI) at Brigham and Women's Hospital (BWH) studied the rate of cardiomyocytes renewal (Fig. 6).¹⁸⁰ Previously, the origin of new cardiomyocytes in the adult heart had been incredibly difficult to study due to their slow turnover rate. Here, they developed an isotopic-labeling technique, amenable to NanoSIMS analysis and overcame previous limitations of tissue imaging by fluorescence microscopy, which could be administered over long periods of time to identify new cells. Accordingly, this methodology provided evidence that most new cells in adult heart tissue are generated from pre-existing cardiomyocytes. This study was an excellent example of how NanoSIMS adds value to biological studies by enabling research that could not have been done previously. Another application of NanoSIMS in tissue analysis is its utilization to understand what happens to the brain tissue postinjury, as demonstrated by the Fitzgerald group.^171–174 In their initial studies, they used NanoSIMS to quantify the rapid concentration changes of calcium leading to secondary degeneration resulting from a partial transection of the optic nerve.^171,172 They were then able to apply their findings and, with the continued use of NanoSIMS, develop a treatment for secondary degeneration.^173,174

NanoSIMS has also proven useful in plant and coral studies by tracking the sites of accumulation of nutrients and toxins,^{24,25,115,204–220,240} studying symbioses,^{114,148,216–224} and performing studies pertaining to general human health.^{182,204–211} For example, one study used NanoSIMS to find the sites of uranium (U) accumulation within roots of the plant species Arabidopsis thaliana, showing U would always colocalize with phosphorous and was kept in the outer layers of plant roots (Fig. 7).²⁵ Another study used ¹⁵N labeling to track the origin of protein substrates in wheat, affording a first look into the pattern of protein accumulation.¹⁸² Overall, NanoSIMS is providing key insights into how plants survive and store nutrients, which is essential to advancing agriculture and human health sciences as well as, in the case of Krueger et al.,²⁴⁰ global warming studies. A short review focused on the application of NanoSIMS in this field that can provide further information was published previously.²²⁵ 

While cellular distributions of proteins and other biomacromolecules can be readily observed using affinity labels (e.g., labeled antibodies or nucleic acids) or fluorescent protein constructs, mapping the spatial locations of small molecules within cells is less straightforward, as conventional microscopy labels can influence the behavior of these molecules in biological systems.^241–244 Because SIMS allows direct compositional information about samples without the need for molecular labels, it has become ideal for spatially elucidating small molecules within cellular systems. Accordingly, one area where use of the NanoSIMS has had considerable impact is in imaging lipid distribution within the plasma membrane and in intracellular compartments. A comprehensive review covering the application of SIMS, which includes SIMS modalities aside from NanoSIMS, to image lipids within cell and model membranes was recently published for the interested reader.³⁹ 

The Boxer group^226–228 first demonstrated the use of synthetically created stable isotope labeled lipids to image and quantitatively analyze lipids within supported membranes using the NanoSIMS, and this method was later employed in the analyses of phase-separated model membranes more analogous to the eukaryotic plasma membrane.²²⁸ With the establishment of this technique, more complex biologically relevant questions could be addressed relating to how cholesterol affected distribution of lipids within cellular membranes, for example (Fig. 8). The Kraft group first explored how cholesterol modulated phase separation behavior of model lipid membranes [Fig. 8(a)].²²⁹ They then exploited known metabolic pathways to isotopically label lipids of interest in actual cells, where they mapped the lipid distribution of sphingolipids laterally across the cell membrane,³⁸ and later determined that cholesterol was not colocalized with the observed nonrandom distribution of sphingolipids [Figs. 8(c)–8(e)].³⁶ In separate studies,^230,232 Lozano et al. supported their findings by showing cholesterol associates with the ganglioside G_m1 and not sphingomyelin.²³⁰ Recently, the Kraft group used this analytical approach to map the 3D intracellular distribution of sphingolipids and cholesterol [Figs. 8(f) and 8(g)].³⁷ Using stable isotope lipids in conjunction with antibody labels that can be detected directly using SIMS, they were able to determine hemagglutinin protein clusters were neither enriched with cholesterol nor did they colocalize with sphingolipids in the plasma membrane.^42,43 Recently, using a similar methodology, He et al. were able to quantify the term “accessible cholesterol” (i.e., ¹⁸O-cholesterol-bound to ¹⁵N-proteins), showing that it was enriched within microvilli in the plasma membrane (areas of cell protrusions) as seen in Fig. 9.²⁴⁵ 

Supported lipid membranes continued to be probe using NanoSIMS, by the Boxer group and others, to understand physical mechanisms of lipid membrane dynamics, as well as physical processes related to SIMS ionization. Rakowska et al. used this approach to understand how antimicrobial peptides interacted with lipid membranes, enabling them to track the formation and characteristics of pore like structures within the membrane.²³³ In a recent publication from the Boxer group,²³¹ they developed a method that allows the localization of labeled small molecules below the lateral resolution of the NanoSIMS within these membranes. Using this method, they were able to address mechanisms related to secondary ion recombination,³⁰ which they suggested can be applied for measuring lipid proximities at lateral resolutions well below the primary ion beam size.

The data generated from NanoSIMS analysis are relatively simple in comparison to other MSI methods that collect entire mass spectra at every location. NanoSIMS data provide a two-dimensional map of an ion's abundance over the area analyzed, which can be viewed as an image and used to compare with the images of the other preselected ions. Typically, multiple scans are taken of an area within a single imaging run. This is done for reasons relating to external environmental conditions (e.g., slight changes in room temperature causing slight shifts in primary ion beam location during an imaging experiment) and often for experimental reasons (e.g., control of imaging depth). A benefit of generating multiple imaging planes in a single run is the ability to have sharply focused features by aligning images, which minimizes issues of beam drift, and to generate three-dimensional information about the composition of a sample of interest.³⁷ 

There are a variety of ways to process NanoSIMS data, which typically depends on the given hypothesis. Further, there are a number of software tools available to process raw NanoSIMS data. When possible, quantitative image analysis of NanoSIMS data and automated spatial statistic methods should be the principal source of information with which to test hypotheses. However, it has long been recognized that the human visual system is exceptionally capable of qualitatively comparing images, and can often be more successful than quantitative-based methods for identifying features and patterns.^246–249 Because data generated from NanoSIMS are spatially resolved counts of discrete ions, as opposed to spectra of human visible electromagnetic radiation, visual representations of the data for human interpretation is not straightforward. Thus, one of the most important steps in processing NanoSIMS data is to map ion counts to a human vision compliant form (e.g., colormap representations). Ideally, an image representation would allow the viewer to easily map the image back to the scalar values in their mind.²⁵⁰ This is not a trivial task. In this section, we discuss the different software tools available and methods of processing and visually representing NanoSIMS data. We also dedicated part of this section toward discussing trends in the literature for image representations of NanoSIMS data, consider some of the shortcomings of current methods, and provide suggestions for data representation based on the current understanding of the human visual system (also see Sec. V C).

As discussed, quantitative image analysis of NanoSIMS data and automated spatial statistic methods should be employed when possible. Toward that end, four primary NanoSIMS-specific software tools have been used extensively for processing NanoSIMS data: OpenMIMS,^251,252 L'Image (L. R. Nittler, Carnegie Institution of Washington), Look@NanoSIMS,²⁵³ and WinImage (CAMECA). Here, we briefly discuss these tools and their features.

OpenMIMS is a cross-platform, free, open source ImageJ plugin courtesy of the U.S. National Institutes of Health and the BWH CNI (which in 2016 emerged from the National Institute of Biomedical Imaging and Bioengineering National Resource for Imaging Mass Spectrometry). OpenMIMS is currently maintained and still actively developed by the BWH CNI (github.com/BWHCNI/OpenMIMS). Originally released November 2010, OpenMIMS was created for opening, processing, and analyzing images with up to seven isotopes.²⁵¹ Particularly useful is its ability to modify image stacks, implement QSA and dead time corrections, display data from regions of interest (ROI) including automated ROI creation using support vector machine segmentation, generate ratio images with custom thresholding, and, recently, generate stitched images for stacks collected adjacent to each other using nrrd Mosaics (see: github.com/BWHCNI/workflow).

L'Image, also known as Larry's SIMS Image Processing Program, is an interactive, Microsoft Windows-based, NanoSIMS data viewing and processing platform. It is available from its developer, Dr. Larry Nittler of the Carnegie Institution of Washington. L'Image was written in RogueWave Software's PV-WAVE language, which is optimized for visual data analytics. Similar to OpenMIMS, L'Image is able to open, align, stich, and perform corrections of NanoSIMS data. A strength of L'Image is its interactive display for profiling and plotting. The quantitative processing techniques include automated ROI creation, ROI tiling, and multiple plotting methods such as scatter plots and histograms.

Look@NanoSIMS is a matlab-based program for the analysis of NanoSIMS data. It is open source and the code is freely available; however, like L'Image, it is dependent on commercially available proprietary software. Look@NanoSIMS's capabilities are very similar to L'image, in terms of viewing, correcting, analyzing via ROIs, and plotting statistics. Where Look@NanoSIMS excels is its ability to create ROIs from external images [e.g., FISH, scanning electron microscopy (SEM), etc.] and provide detailed statistical analysis across multiple datasets simultaneously. The software is still being developed and recently advanced ROI generation features (e.g., logic expressions, size thresholding, and patterning), as well as more advanced plotting features (e.g., automated annotations, 3D plotting), have been added (see: nanosims.geo.uu.nl/nanosims-wiki/doku.php/nanosims:lans).

WinImage is a SIMS image processing software provided commercially by CAMECA for use on Microsoft Windows machines (cameca.com/support/winimage.aspx). This package is designed to work seamlessly with CAMECA SIMS data collection instruments. Its features are the most limited out of the NanoSIMS-specific tools, and most groups have opted for using one of the other packages, which nearly recreate the identical capabilities of WinImage and much more.

Out of the four commonly used software packages, OpenMIMS is currently the most popular, being present in over one third of manuscripts since 2009, with increasing popularity as the most common software over the past three years.

While the NanoSIMS specific software tools are the most common method for data processing and visualization, some groups have used generic image analysis packages or programming languages for custom approaches. For example, from the Research Systems, Inc., family of software, Interactive Data Language^120,160 and Environment for Visualizing Images²⁰⁰ provide generic capabilities for processing and visualizing data. Similarly, ImageJ and Fiji, a plugin replete distribution of ImageJ, have been used directly without reliance on OpenMIMS.^{25,138,175,181,208,236} Finally, other groups have generated custom analysis scripts using matlab,²⁸ python,^54,176 or similar programming languages. python has well-established and documented scientific libraries and tools, including those specifically for quantitative image analysis. These include winpython, a free, open-source, and portable full-featured python-based scientific environment,²⁵⁴ numpy and scipy,²⁵⁵ scikit-image,²⁵⁶ opensource computer vision for advanced image processing,²⁵⁷ matplotlib,^258,259 and others^260–263 for creating publication quality figures, and many other relevant libraries.^264–268 In our NanoSIMS data processing, we commonly preprocess images in OpenMIMS (using their dead-time correction, stack summing, and visual features) then use these python libraries alongside ipython,²⁶⁹ an enhanced python shell, running within the Scientific python Development Environment (spyder) for interactively creating figures. python is only one example of the different languages and development environments that are now available; there is no longer a need to rely on proprietary or commercial software for NanoSIMS data analysis.

Following in the footsteps of other research fields, we believe the NanoSIMS community should quickly begin to embrace open data and transparent processing methods. The benefits of open data, open-source software, and shared processing scripts have become increasingly clear over the past decade.^270,271 Adopting these approaches, where the community is able to access both the raw data and the processing scripts, can accelerate discovery and enable projects to have longer lasting impact.²⁷² Our perspective is similar to that of Sandve et al., that, at minimum, publications should be distributed with enough information so that other researchers can easily understand each step from sample creation to analysis, and from instrument raw data to publication figure.²⁷³ Ideally, researchers should be able to independently recreate the publication from the raw data (e.g., all raw data, as well as processing scripts, are supplied). Therefore, we see a benefit in using and supporting freely available open sources packages and languages, such as OpenMIMS and python, and “notebook documents,” such as Jupyter notebooks (jupyter.org)²⁷⁴ and R Markdown documents (rmarkdown.rstudio.com), which contain human-readable descriptions and discussion of processing, inline executable processing code, and on-the-fly results.

Universally, dynamic SIMS literature represents the data as 2-dimensional images of either single ions (e.g., ¹²C¹⁴N⁻, ³¹P⁻) or isotopic ratios (e.g., ¹²C¹⁴N⁻/¹²C¹⁵N⁻, ¹²C₂⁻/¹²C¹³C⁻), with normalized or thresholded intensities mapped against grayscale or pseudocolor maps. Table S1 shows a summary of the maps commonly chosen, with the most popular maps being grayscale and rainbow maps, such as hue-saturation-intensity (HSI) colormaps (Fig. 10). For example, of the popular NanoSIMS data processing programs, OpenMIMS defaults to either grayscale (“Display Ratio” under the Process tab) or the rainbow-type HSI color map (“Display HSI” under the Process tab). In general, intensity mapping and colormap8 choice have been arbitrarily chosen, following previous examples or journal requirements (e.g., grayscale for journals that do not print in color).

For many studies, where the features of interest have clear boundaries and high contrast to surrounding areas, the colormap chosen for image representation may not have any consequence to interpreting the data or on the study's conclusions. However, it is known that image data interpretation can be skewed by the chosen colormap, patterns can be obfuscated, and misinterpretation can occur.²⁷⁵ The failings of poorly designed colormaps has been discussed in detail.^250,276 Briefly, the main issues result from failing to consider: (1) human eye spectral sensitivity where different wavelengths of light produce different perceived luminosities (perceived brightness),²⁷⁷ which manifests as inconsistent color ordering between humans and as Mach bands (exaggerated contrast between colors),^250,278 (2) color blindness, which affects up to 8% of the male population and 0.5% of the female population and creates a distorted color vision sensory experience,^279,280 and (3) chromatic and viewing conditions adaptations such as the effect of simultaneous contrast and white light normalization.^281,282 Unfortunately, rainbow-type color maps (e.g., HSI, Jet, L'Image's default colormap, etc.), which are especially common throughout the sciences, fail to consider all of these issues and were created primarily out of computational ease and aesthetic appeal. Having wildly nonlinear brightness and hue, rainbow-type color maps lead to confusion by retarding the mind from mapping the image back to the underlying data, hiding variation in low contrast regions, and reducing perceived image sharpness (Fig. 10).²⁵⁰ Also, these can lead to confusion and misinterpretation when colors are ordered in new ways. For example, Jet and HSI place red at a higher value than yellow whereas L'Image's default colormap does the opposite. This can be incredibly confusing in a community where all three colormaps are used often, especially when two differently ordered colormaps are used in a single paper, as seen in Takado et al.¹²⁶ and Tourna et al.¹²⁷ While HSI is common in the literature, NanoSIMS data could be better assessed qualitatively using improved colormaps designed based on our current understanding of the human visual system (see Sec. V C for further discussion).

One of the critical steps in NanoSIMS data analysis is to identify features and then, typically, define ROIs to interrogate the isotope distribution and enrichment of that feature's area. In its simplest form, regions of interest can be defined without regard to feature locations by arbitrary or repeating patterns of tiled ROIs. For example, both software packages L'Image and Look@NanoSIMS provide a method for automatically generating grids of ROIs. This type of analysis can be helpful for quickly observing the overall statistics of a sample. Elemental segmentation can be used to select pixels with a specific signature or intensity, for example, by using automated thresholding technique's like Otsu's method, adaptive thresholding techniques, or by using logical expression (as available in Look@NanoSIMS) on specific NanoSIMS channels or mathematical transforms of channels. These segmentation techniques are also basic features of python and other computer language image processing libraries. In a few lines of code, custom and specific segmentation methods can be created, moving beyond what is possible in the currently available NanoSIMS-specific software packages. In general, languages like python do provide more complex data analysis capabilities, with the added benefit of complete scientific transparency and improved reproducibility.^269,283,284

Morphological segmentation attempts to use the form and structure of the underlying data to separate ROIs into biologically meaningful areas. While we refer to morphological segmentation as a quantitative approach here, it is similar to what researchers typically do when defining manual ROIs using the human visual system. Unfortunately, setting up the correct program or code to do morphological segmentation on a variety of disparate samples has been a significant enough of a barrier that, in most cases, researchers are still using manual methods to define ROIs. This leads to the potential for researcher bias, low sample analysis throughput, and often poorly defined edges (the case where portion of the real underlying feature are neglected either due to a forced shape ROI tool, such as a square or circle, or difficulty matching the edge with a freehand tool).

For quantitative morphological segmentation, a variety of mathematical operations are performed, typically post-thresholding, to isolate key features into ROIs. These operations can include erosion, dilation, opening, closing, topological skeletonization, edge detection, and many other techniques. As a very useful feature, Look@NanoSIMS allows morphological or other segmentation methods to be done on external image data, such as confocal FISH images, and then can apply that data as a template for ROI definitions in the NanoSIMS data. Furthermore, recent advances in machine learning, registration, and library matching have automated some aspects of segmentation.^54,285–287

In Sec. II C, we discuss the different units and nomenclature used to represent processed NanoSIMS data. It is also ideal to take into consideration the different levels of uncertainty, errors, and precision in any given measurement (as outlined in Sec. II D). Numerous examples have been provided in this review that illustrates the breadth of ways that NanoSIMS data can be presented and transduced. For example, NanoSIMS images can be presented to demonstrate APE or permil isotopic-enrichment across the entire imaging field of view. Lechene et al. also visualized the distribution of protein turnover in hair follicles,¹⁸³ by calculating the proportion of newly synthesized protein at each pixel using this following equation:

where ¹²C¹⁵N and ¹²C¹⁴N are the counts of the respective ions, A_n is the natural abundance of ¹⁵N (0.37%, Table I), and A_f is the relative abundance of ¹⁵N in the material used in the feed. Further, data from image ROI can be distilled to simple charts and tables that illustrate difference in elemental abundance or isotopic enrichment, for example.

NanoSIMS fills a unique niche as a MSI tool for biological analysis by providing unmatched lateral resolution of elemental and isotopic distributions of samples of interest, as has been established throughout this review. Its impact in biological sciences will only continue to grow as more NanoSIMS instruments are installed worldwide. New developments in instrumentation and data processing should expand the amount of information attainable from these measurements. In this section, we highlight some of the leading-edge instrumentation, what scientific questions future multimodal investigations may address, and how we can better visualize these results.

Scientists can expect to achieve even higher lateral resolution measurements in the future, as NanoSIMS instruments worldwide are fitted with cutting-edge ion sources. A new radio frequency (RF) plasma oxygen source, which is commercially available, is about an order of magnitude brighter than the conventional duoplasmatron source and can be focused below that of the Cs⁺ source.^288,289 Further, RF generated ion sources possess notably longer lifespans than that of the cold cathode duoplasmatron source, which has a notoriously unpredictable usage time before needing service (50–500 h, typically dictated by life of the cathode).²⁸⁹ Recently, Malherbe et al. demonstrated the application of this source compared to the duoplasmatron source, where in one example they were able to more clearly resolve the intracellular distribution of important metals within Chlamydomonas reinhardtii (algae) cells using the RF source (Fig. 11).²⁸⁸ Further, they demonstrated that the beam size of the RF source was even smaller than that of the Cs source, 37 vs 47 nm, respectively. This study was followed up with correlative TEM and TEM/x-ray energy-dispersive spectroscopy to confirm the spatial distribution of trace metals within these algae cells.¹³³ The newest commercial RF plasma source is now capable of dual polarity, where O⁻ allows for electropositive secondary ion analysis and O²⁺ provides high lateral resolution depth profiling.

As the new RF plasma oxygen source was pioneered within the focused ion beam community,²⁹⁰ perhaps one can expect new Cs beam technology from there as well. One potential example is the cold atom-based beam system.²⁹¹ Here, neutral Cs atoms are laser collimated then photo-ionized to produce a microbeam. Viteau et al. recently demonstrated the potential use of this Cs beam in ion microscopy efforts, where they were able to produce sub-40 nm lateral resolution images with low beam energies.²⁹² Currently, others are exploring redesigns of the conventional Cs source, which generates ions from vaporized Cs that is ionized as it flows over a tungsten plate, with the hopes of gaining higher lateral resolution images of electronegative secondary ions.

In many of the areas that NanoSIMS has made the most significant impact, it has been employed as part of a multitechnique process with correlative data streams. Numerous examples of NanoSIMS coupled with fluorescence, electron, x-ray, and force microscopies have demonstrated the power of these imaging techniques used in combination. FISH-NanoSIMS has provided unique insights into metabolic processes within complex communities and has a common microbiology methodology.⁸⁴ X-ray microscopy methods like scanning transmission x-ray microscopy (STXM) can provide high lateral resolution information about chemical speciation and redox states, which provides unique insights into to environmental processes when utilized in conjunction with NanoSIMS, as was described previously in a review by Berhens et al.³ The combination of force microscopy and NanoSIMS has provided insights into the sputtering process through biological material, giving more accurate three-dimensional information about cellular features.²⁹³ It should be noted that Wirtz and coworkers designed a modification to the NanoSIMS that integrated a scanning probe microscope into the analysis chamber.^294–296 Perhaps more modifications to NanoSIMS instrumentation that will facilitate multimodal imaging with greater ease can be expected.

Aside from high resolution imaging multimodal investigations, low- or no-lateral resolution mass spectrometry techniques used in tandem with NanoSIMS are also very valuable. IRMS can provide bulk isotopic enrichment within entire samples that can be correlated to submicron processes identified by NanoSIMS. Laser ablation-IRMS (LA-IRMS) can spatially locate hotspots within a sample (e.g., microbial mats)²⁹⁷ that can be further interrogated. If metal abundance is of interest, then inductively coupled plasma mass spectrometry (ICP-MS) and LA-ICP-MS can be used similarly.²⁹⁸ In general, it is often best to start with these methods before NanoSIMS analysis to ensure that samples contain isotopic enrichment or metal content of interest, respectively. Omics-based stable isotope probing (SIP) methods can be used to link high-lateral resolution isotopic distributions with molecular expression. For example, tools like MetaProSIP can elucidate isotopically enriched proteins within microbial consortia,²⁹⁹ providing automated reconstruction of what metabolic pathways were activated, which can then be correlated to spatial processes from NanoSIMS imaging.

NanoSIMS is increasingly being used in human health-related studies,^45,47,300 and this is perhaps the field where it will have the most impact in the near future. For example, the analytical toolbox for pathobiologists has been expanded with mass spectrometry based immunohistochemistry, which has the potential to improve disease diagnosis and theory.⁵ Next generation drug design can benefit from incorporation of element/isotope and fluorescence labels that permit correlated optical and isotopic nanoscopy,⁴⁸ where drugs can be tracked with greater detail and down-stream processes can be measured with bulk omics SIP measurements. The same NanoSIMS-based methodologies used in microbiology are translatable to exploring human-microbe interactions,^91,149,150 and will be of particular importance as the human microbiome is more thoroughly explored. Finally, isotopic activity-based probes could be utilized to simultaneously monitor protein expression and metabolic activity, for example.^5,301–303

The design and selection of the best colormap is a challenge across all science fields that rely on image representation and segmentation.³⁰⁴ While there is not a single “best” colormap, there are colormaps that perform better than those commonly being used commonly to represent NanoSIMS data, which can begin to address the issues discussed throughout Sec. IV. Grayscale maps avoid the issues associated with color perception and can provide linear luminance with relation to underlying values. However, grayscale suffers from some viewing condition adaptations and has a small dynamic range due to the lower discernibility of the shades of gray to the human visual system.

The human eye can discern orders-of-magnitude more of color shades, and color provides information in addition to intensity.³⁰⁴ This is part of the original justification for HSI used in OpenMIMS, as discussed by Lechene et al.⁶⁹ The HSI colormap cycles through the hues of the HSI color model and tries to match psychometric definition of saturation, and is more intuitive to human vision than other older color maps. HSI was created to be computationally efficient, while simultaneously being effective for segmentation. However, it is not appropriate for relaying quantitative spatial data or for displaying images that must be optimized for the human visual system. Furthermore, computers are no longer a bottleneck for creating colormaps, allowing us to move beyond HSI to better designed colormaps.

Isoluminant colormaps also avoid most of the issues above, as they have apparent equal brightness across all values. However, as discussed by Moreland,²⁵⁰ isoluminant color maps do not take advantage of the sensitivity of the human eye to changes in luminance, which results in difficulty discerning hue changes in some data. One benefit of isoluminant colormaps is they may be used simultaneously with additional data types (e.g., morphological), where the primary data type is displayed in relation to an isoluminant colormap and a second data type controls the brightness of each pixel. For example, this can be used to show morphological data (depth and shading) while preserving the colormap hue related to the NanoSIMS count or ratio data. Our python code, provided in the supplementary material, can turn an existing colormap into an isoluminant colormap and use it to create mixed images (Fig. 12), where the lightness values are adjusted based on the secondary electron image to show morphological changes. This method is known as referred to as “relief shading.”³⁰⁵ Also included in this code is the ability to take in the name of a colormap and analyze it by plotting it in CIECAM02-UCS colorspace (described below) and calculating the visual deltas, as was shown in Fig. 10. We have purposely designed this code to be easy to use both the high-level functions (mixing images and analyzing a colormap) and low-level functions (such as the conversion and math functions) in order to promote user experimentation and a better understanding of the colormap design. For python code used to create isoluminant colormaps, visit our GitHub repository at github.com/jamienunez/cmaputil.

Linear perceptually uniform colormaps are likely the ideal colormaps for most digitally reconstructed images (e.g., NanoSIMS data). They are “linear” in that the perceived brightness varies linearly (thus even when printed in grayscale, the perceived data are similar), and “perceptually uniform” in that a difference in color space is intended to closely correspond monotonically to human perception of color difference.³⁰⁵ These types of colormaps can be created to provide perceptual interpolation that is sequential and scalar, be accessible for color deficiencies (i.e., colorblindness), and still be aesthetically pleasing. Several teams have designed linear perceptually uniform colormaps, with the main differences being the color space chosen to construct the maps, and the slope and location of trajectories of the colormaps through the chosen color space. The International Commission on Illumination L*a*b* (CIELAB) color space has been frequently used. However, a more modern color appearance model (CAM), CIECAM02, corrects for some defects in CIELAB, and has recently been used to design linear, perceptually uniform colormaps (see Fig. 12). For example, a new colormap becoming more popular in the python community is viridis, which is based on the state-of-the-art CAM02-UCS model of perceptual distance, a CIECAM02-based color space.³⁰⁶ 

SIMS methods can readily provide subcellular information of biological processes, and the NanoSIMS 50/50L offers the highest lateral resolution of the tools. Here, we discussed the history of SIMS and the place of the NanoSIMS within that history. Further, we demonstrated that careful considerations must be taken in sample preparation and data acquisition, analyses, and visualization for successful NanoSIMS explorations. We have found that fruitful NanoSIMS utilization is typically based on relatively binary hypotheses, for example, is element/isotope accumulation or enrichment localized with a certain biological feature or another biological process? NanoSIMS should not necessarily be used as a stand-alone discovery tool. This is one reason that NanoSIMS analysis is especially powerful in multimodal investigations. NanoSIMS methodologies and instrumentation will continue to evolve as accessibility to these instruments continues to be ubiquitous. In the future, we expect NanoSIMS to play a significant role in further contributing to our understanding of complex environmental and human health-related biological processes.

The authors would like to thank Francois Horreard and Francois Hillion for the NanoSIMS-related documentation and useful discussions. They would also like to thank Matthew Steinhauser for useful discussions and assisting in access to numerous publications, and Lee Ann McCue for her suggestions and review edits. This research was supported by the Genomic Science Program (GSP), Office of Biological and Environmental Research (OBER), the U.S. Department of Energy (DOE), and is a contribution of the Pacific Northwest National Laboratory (PNNL) Foundational Scientific Focus Area (SFA). A portion of the research was performed using the Environmental Molecular Sciences Laboratory, a DOE Office of Science User Facility sponsored by the OBER and located at PNNL. PNNL is operated for DOE by Battelle Memorial Institute under Contract No. DE-AC05-76RL01830.