Cryo-focused ion beam for in situ structural biology: State of the art, challenges, and perspectives

Cryogenic-focused ion beam (cryo-FIB) instruments became essential for high-resolution imaging in cryo-preserved cells and tissues. Cryo-FIBs use accelerated ions to thin samples that would otherwise be too thick for cryo-electron microscopy (cryo-EM). This allows visualizing cellular ultrastructures in near-native frozen hydrated states. This review describes the current state-of-the-art capabilities of cryo-FIB technology and its applications in structural cell and tissue biology. We discuss recent advances in instrumentation, imaging modalities, automation, sample preparation protocols, and targeting techniques. We outline remaining challenges and future directions to make cryo-FIB more precise, enable higher throughput, and be widely accessible. Further improvements in targeting, efficiency, robust sample preparation, emerging ion sources, automation, and downstream electron tomography have the potential to reveal intricate molecular architectures across length scales inside cells and tissues. Cryo-FIB is poised to become an indispensable tool for preparing native biological systems in situ for high-resolution 3D structural analysis.


Introduction
Cryogenic electron microscopy (cryo-EM) has become the most widely used method for determining the structures of biological macromolecules and complexes at 2e4 A ˚resolution [1].Cryo-EM allows for the study of complexes in their native environment, yet the thickness of cells and tissues impairs its applicability.Cryo-EM can provide a snapshot of samples whose thickness is limited to 500 nm due to the penetration depth of 300 keV electrons in biological material [2].Cryo-EM's tomographic acquisition counterpart, cryo-ET, requires tilting the sample up to 70 , further limiting the sample thickness to 200e300 nm for most samples.
Cryogenic-focused ion beam (cryo-FIB) instruments, introduced about 20 years ago [3,4], were developed to overcome the thickness limitation by preparing thin lamellae in vitrified samples.FIBs (regardless of the temperature at which they operate) use accelerated ions to sputter away atoms from the surface of the specimen, and in TEM sample preparation, this is done until the sample is milled to a slab thin enough so that the inelastic scattering is not anymore limiting, called a lamella.Considering the abovementioned reasonable limit for imaging biological samples, if cryo-ET is required, lamellae may be prepared with a maximum thickness of 200 to 300 nm.This effectively allows imaging of any area of the cell using cryo-ET/EM in its near-native state [5,6].Cryo-FIB preparation has led to multiple examples of high-resolution structural determination in situ [7e9].This method has also proven to help in sample preparation for microelectron diffraction in cases where crystal sizes are too small for X-ray and too thick for EM [10,11].Over the past 15þ years, cryo-FIB technology has rapidly advanced and has become accessible enough to be used by most laboratories with expertise in cryo-EM, yet many challenges remain [12].
Reliable vitrification of tissues is a key step in the cryo-FIB workflow, posing a significant barrier for some projects.For samples thicker than w10 mm, it is challenging to produce vitreous ice through plunge-freezing; here, high-pressure freezing (HPF) is the method of choice, which is capable of reliably freezing up to 300 mm [13].
Although it is in theory possible to vitrify biological spesimen that are hundreds of microns thick, consistently vitrifying samples thicker than 100 mm comes with significant challenges and, in some cases, with low reproducibility.Recently, methods such as the Waffle method [14] and customizing planchette-based carriers [15] have been shown to improve vitrification and throughput in cell culture samples [14,16].Typically, specimen thickness is restricted to w30e50 mm depending on the type of spacer used.In tissue, these dimensions may not be readily achievable due to the minimum dissectible sample where the requirement is to prepare specimens with a minimum thickness of 100e200 mm.Researchers have found that adding cryoprotectant to thicker samples increases the depth of vitrification [17,18].Once a tissue sample small enough for vitrification is obtained, methods for preparing thin lamellae may be used directly on the sample support [18e20] or by cryo-lift out [17,21e23], depending on the specific case.Automation of these methods is still being developed to ensure their complexity does not reduce their adoption.These automation softwares will be discussed later in this review This review describes the state-of-the-art capabilities and applications of cryo-FIBs, discusses the remaining challenges and limitations, and explores promising future directions to make cryo-FIBs more effective, increase throughput, and expand accessibility.This review focuses on critical developments in instrumentation, imaging modalities, automation, sample preparation, and correlative targeting techniques that underpin recent advances in this technology.

Optimized beam chemistry
Historically, FIB instruments were based on liquid metal ion sources such as Ga or Au/Si [24].These instruments have been optimized to meet the semiconductor industry's needs.With the expansion of applications and the challenges brought by milling softer and nonconductive materials, including ion implantation, charging, damage, and curtaining, there has been a push to explore alternative ion beam sources [18,25,26].
Various ion species, such as Xe, Ar, Ne, and N2, have been investigated as alternatives to Ga in FIB systems [25e27].These ion species have different atomic weights and interaction volumes within the sample, which can influence the extent of beam-induced damage and milling efficiency.In general, higher atomic weight ions have smaller interaction crosssections [28], helping reduce damage in sensitive biological samples compared to lighter ions.
When an ion beam impinges on a sample, the ions undergo a series of collisions that can result in deviations from the original trajectories (straggle), each collision may give rise to knock-on damage, the generation of secondary electrons, and the implantation of the beam ions in areas far from the original point of entry in the material.This can lead to a collision cascade and the formation of radicals, spreading the damage beyond the initial impact site [28].
Frozen-hydrated biological samples are among the most sensitive materials to mill; they are composed of light atoms, and the density of covalent bonds is lower than in typical semiconductors.Under cryogenic conditions, chemical reactions between the beam ions and the sample are unlikely due to the significantly reduced thermal energy.However, local chemical modification can still occur through ion implantation into the sample, altering the local composition and potentially causing structural damage.
The beam energy and the atomic weight of the ions are the two key variables that influence the interaction volume and, consequently, the extent of beam-induced damage.Lower beam energies and higher atomic weight ions result in smaller cross-sections of the interaction volume, which helps minimize the damage [28,29].Recently, it has been measured that the damage in biological samples is consistently found to be w50 nm from the polished face [30] in both Ga and Xe milled samples.This effect is reduced to w30 nm when using a lower energy beam (e.g. 8 keV), as well-described in the material science literature and quantified for cryo-EM by Yang et al. [31].
Tuijtel et al. [32] have shown that sub-nanometer resolution can be achieved in cryo-lamellae regardless of the ion-induced damage and that the thickness of the lamella does not necessarily scale with the final resolution.This conclusion is well supported by the fact that the damage is stochastic, and averaging can overcome it when enough sub-tomograms are available.The effect of FIB damage is comparable to the damage induced by electron beam pre-exposures in single-particle cryo-EM [31].
Finally, the reduced damage has not been the only reason for exploring different sources, plasma-based beams can deliver significantly larger currents compared to liquidmetal ion sources without sacrificing resolution or current density.The potential of using heavier atoms in the primary beam provides the added benefit of an increased sputtering yield (defined as the number of atoms from the substrate sputtered by each ion from the FIB), which, coupled with a higher current makes milling faster [26].Although the data available now are still limited, it has been shown that the sputtering rate and the milling efficiency increase with the energy and atomic weight as expected.The effect of milling efficiency on the effective milling time has not been established yet, as it appears that curtaining is more pronounced when using heavier atoms, therefore requiring longer polishing stages to produce highquality lamellae.

Volume electron microscopy
Cryo-FIBs are primarily used to produce thin lamellae for cryo-ET/EM imaging.Additionally, the presence of coincident high-resolution scanning electron microscope (SEM) on many cryo-FIB instruments provides the means for investigating cellular and tissue ultrastructure in three dimensions with resolutions of a few tens of nanometers [33e35].This method, called FIB/SEM tomography, consists of iteratively FIB milling off a few nanometers from the sample surface followed by SEM imaging of the newly exposed region.FIB/SEM tomography is a popular choice in metrology [36] where it can provide a three-dimensional map of the sample with a resolution on the order of a few nanometers [37].Since biological samples are inherently poor conductors and most of the comprising elements have comparable atomic weights, this technique results in reduced contrast, thus the resolution is limited to 10e20 nm [20].This resolution, although modest, is sufficient to identify cellular compartments and molecular aggregates [38].The samples are negligibly damaged due to the low energy used for imaging by the SEM beam (typically 1e2 keV).Thus, cryo-FIB/SEM tomography combined with cryo-ET can provide three-dimensional context around the region of interest.Recent software advances make it possible to correct the artifacts that afflict this method, such as charging and contrast changes [39].If resolutions between 4 and 10 nm are required while maintaining the three-dimensional micron-scale spatial context, alternative preparation and imaging modalities should be explored, including using FIB/SEM tomography in resin-embedded samples [40].Resinembedded samples will be limited in resolution by the heavy metal staining yet the ultrastructure will be preserved.
A recent development in cryo-liftout preparation involves serial extraction of slabs of specimen, which are then thinned individually into lamellae.This method, called Serial Lift-Out [41], delivers segmented volumetric context by allowing for w300 nm lamellae to be created every several microns from a single specimen..While serial cryo-sectioning allows for adjacent lamellae to be created, it suffers from compression artifacts and section warping.Comparatively, Serial Lift-Out is simpler to automate and some open-source packages such as OpenFIBSEM have pipelines available [42].

Targeting
Cryo-EM/ET imaging is cursed by the size of the imageable field of view, which recent methods are increasing [43e46], and by the difficulty of assigning identities to objects based only on individual images/ tomograms [47,48].Orthogonal approaches are required to help delineate objects of interest [49].Thanks to the specificity of fluorescence microscopy enabling targeting specific cellular events, cryo-FIB milling can be guided with a precision better than 500 nm in all three dimensions [50].Considering that the thickness of lamellae is in the order of 200e300 nm and that many target events are rare, precise and robust correlative methods are critical [51].
The use of cryo-correlative light and electron microscopy (cryo-CLEM) has become routine practice, where fluorescent tags are first located by light microscopy, and then those locations are registered to the coordinate system of the cryo-FIB/SEM images to target the milling of lamellae [52].While routine, cryo-CLEM is not trivial due to the lack of facilitating software and the risks of handling cryo samples throughout the multiple imaging steps.Cryo-FIB/SEM and cryo-light microscopy (cryo-LM) can be performed using either the same (integrated) or separate instruments, with cryo-CLEM accommodating both vacuum and non-vacuum environments and the potential to take advantage of super-resolution fluorescence microscopy.The benefits of integrated cryo-correlative systems [53] are contamination-free imaging, seamless coordinate transfer across modalities, and alternating imaging modalities between FIB and SEM that provide realtime feedback on the targeting.Although the resolution is, in principle, identical to that achievable with external microscopes, external microscopes have the benefit of allowing simultaneous use of the cryo-FIB and fluorescence microscope for separate experiments, therefore improving overall throughput.
Several solutions have been developed for combining cryo-FIB/SEM milling and cryo-LM in the same instrument, both where the LM is coincident with the FIB/ SEM beams [51,54,55] and where it is off-axis [53,56,57].Additionally, manufacturers are beginning to offer cryo-FIB/SEM instruments with a coincident cryo-LM already built in, such as the ThermoFisher Scientific (TFS) Arctis.Currently, these in-chamber cryo-LMs offer poor resolvability e they are typically capable of allowing the operator to identify cells and cell components in the plane of lamellae but are incapable of reliable targeting in the z-direction out of the plane.Higher-resolution external cryo-LM imaging solutions include the use of commercial super-resolution LMs [58], custom-built super-resolution LMs [59,60], and custom-built cryo-single molecule fluorescence microscopes [61].

Automation
Manual cryo-FIB milling requires an expert operator.User-defined automated routines that perform most steps hands-off were initially introduced by Buckley [62] and Zachs [63] significantly lower the expertise barrier.These automate all tedious processes, including trench milling, coarse thinning, and final polishing without supervision, improve reproducibility, increase throughput, and provide the means for batch and multiday sample preparation.
Currently, multiple options are building on this initial innovation, allowing the automated milling of cryosamples.The SerialFIB [64] package provides a flexible, Python-based platform supporting predefined protocols and customizable scripting for advanced applications.The OpenFIBSEM API [42] enables complete cross-platform control of cryo-FIB/SEM systems with a unified interface extending to controlling the manipulators for cryo-lift-out operations.On the commercial side, TFS's AutoTEM [22], Tescan's Shark-SEM, and Zeiss's SmartSEM are proprietary software for automated lamella preparation.Despite their limited flexibility, they provide stable environments and simple guided interfaces.

Remaining challenges
In scarcely over a decade, cryo-FIB has ascended from a speculative concept to an established sample preparation technique in the cryo-EM repertoire.Ongoing adoption is facilitated by expanding instruments and expertise in the field, automating significant workflow portions, and by continued demonstration of biologically relevant results [65e73].Further optimization of hardware, software tools, and imaging parameters should proportionally scale its impact.When combined with fluorescence-based targeting, expanded volume capabilities, emerging ion sources, and downstream cryo-ET, cryo-FIB enables the pursuit of structural biology investigations directly in situ.These intersecting advances not only promise to reveal intricate architectures sustaining life across length scales in cells and tissues but also offer insights into their disease states.However, several challenges that currently limit more widespread adoption of cryo-FIB remain to be addressed -an overview of the workflow bottlenecks is provided in Figure 1 with an attempt to quantify the severity of the bottlenecks.
All activities on the FIB are time-consuming, even when automated.Even the simplest preparation of lamellae on-grid can take anywhere from 5 to 40 minutes per lamella, and a lift-out session produces up to one lamella every 30e45 min depending on the method used and the skill of the operator.Accordingly, most of the biological targets studied and presented in the literature do not require precise CLEM, as most FIB targeting is done at the cellular rather than the molecular level.A significant improvement would be provided by novel CLEM alignment procedures and automated alignment software that does not require user intervention.The current throughput associated with CLEM limits the number of lamella sites that can be prepared during a session.New instrumentation capable of batch milling multiple grids, real-time CLEM, and full automation will be productive directions for the field.Applying the emergent field of machine learning and image interpretation to this problem could automate object identification and tracking in multi-modal imaging.
Ice contamination, another significant problem associated with all cryo-workflows, builds up over time, particularly on charged surfaces.Ice contamination is often visible on freshly milled lamellae as well as on areas that have been imaged by cryo-LMs.This charging-based contamination limits the use of superresolution methods that require long exposures.The most recent cryo-FIB anti-contaminator designs is a promising step in the right direction [74].Future steps will likely involve the development of highly automated cryo-LM systems kept under high vacuum that are capable of super-resolution imaging and have robust transfer systems.Such instruments may resolve the conundrum of balancing between high-resolution LM imaging and limiting ice contamination common in nonintegrated cryo-LM systems.Although this problem is minimized with integrated CLEM systems, integrated cryo-LMs reduce sample preparation throughput in the cryo-FIB/SEM.Faster detectors, optimized beam parameters, and machine learning-based imaging algorithms will improve data collection efficiency [33].Plasma-FIBs have the potential to allow for smoother, artifact-minimized milling [75].While quantifying the benefits of each alternative ion species requires in depth experiments and analyses, data is becoming increasingly available and the existence of multi-ion sources will make these comparisons easier.
While cryo-FIB once required customized instrumentation restricted to specialists, automation has recently enabled mainstream usage.With improved precision, speed, and usability, cryo-FIB is poised to become a routine technique for resolving native 3D biological structures in situ, unlocking new biological insights.Recent methods for tiled imaging in cryo-ET [43e45] coupled with serial lamella preparation such as Serial Lift-Out [41] will enable the investigation of tissue across large volumes.We envisage that these advancements will extend cryo-FIB adoption beyond specialized labs in the coming years.As methods become more reliable and the entry barrier for new users decreases, an increase in the utility of cryo-FIB and, more broadly, cryo-EM in a structural cell biology context may be expected.Improvements in cell culture on grids [76,77,77], sample handling [78], and smart preparation methods avoiding on-grid culturing [14] have led to significant improvements in throughput.Method developments are poised to bridge information across multiple length scales, from CLEM to volume imaging to cryo-ET.These methods may rapidly shift towards tissue biology, allowing for the possibility of investigating organoids and fresh tissue isolates from animals and patients [18,19,35,79].

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Another barrier to wide user adoption is the significant investment required to acquire cryo-FIB-related instrumentation and significant hands-on training requirements.Here, centralized user facilities play a significant role in reducing financial stress by making more efficient use of the instrumentation and personnel and by reducing the training time by establishing streamlined protocols and hands-on workshops.Examples of large, shared facilities are NCCAT and NCITU in New York, PNCC in Portland, SLAC in Stanford, eBIC in Oxford, EMBL/Imaging Centre in Heidelberg, and SUSTech in Shenzhen.Centralized service and training facilities are increasingly becoming a reality worldwide, providing more sustainable training and data acquisition frameworks compared to isolated facilities.