Advances in domain and subunit localization technology for electron microscopy

Abstract

The award of the 2017 Nobel Prize in chemistry, ‘for developing cryo-electron microscopy for the high-resolution structure determination of biomolecules in solution’, was recognition that this method, and electron microscopy more generally, represent powerful techniques in the scientific armamentarium for atomic level structural assessment. Technical advances in equipment, software, and sample preparation, have allowed for high-resolution structural determination of a range of complex biological machinery such that the position of individual atoms within these mega-structures can be determined. However, not all targets are amenable to attaining such high-resolution structures and some may only be resolved at so-called intermediate resolutions. In these cases, other tools are needed to correctly characterize the domain or subunit orientation and architecture. In this review, we will outline various methods that can provide additional information to help understand the macro-level organization of proteins/biomolecular complexes when high-resolution structural description is not available. In particular, we will discuss the recent development and use of a novel protein purification approach, known as the the PA tag/NZ-1 antibody system, which provides numberous beneficial properties, when used in electron microscopy experimentation.

Introduction

The 2017 Nobel Prize in chemistry for the development of cryo-electron microscopy (cryo-EM) was recognition that electron microscopy (EM) has come of age as a powerful technique for determining the structure of biological macromolecules and protein complexes (Frank 2018). Recent advances in microscope components (McMullan et al. 2016; Danev and Baumeister 2016) have been matched by improvments in analysis software (Zivanov et al. 2018; Grant et al. 2018), and novel sample preparation techniques (Razinkov et al. 2016; Kaledhonkar et al. 2018), which all together have allowed for the high-resolution structural determination of a range of complex biological machinery (Liang et al. 2017; Sugita et al. 2018; Qiu et al. 2018). At high resolutions, the identity of individual residues and atoms can be determined to some degree of certainty and the local and global structure of the target biomolecule can be understood within limits related to responsible interpretation of the experimental data. However, not all targets can be resolved at high-resolution due to various intrinsic or extrinsic properties (that may be intractable for a given experiment), and so in these cases, the identity of side chains, secondary structures, or even whole domains can remain ambiguous, requiring the use of other tools to correctly characterize the domain or subunit orientation and architecture. In this review, we will outline various methods that can provide localization information, sufficient to help understand the macro-level organization of protein and biomolecular complexes, where high-resolution features are not present. In particular, the recent development of a novel protein purification system, the PA tag, has found a number of applications in a range of EM experimental contexts due to the unique structural features it proffers (Fujii et al. 2014, 2016).

Identifying domains at intermediate resolutions

As with any high-resolution structural technique, cryo-EM can facilitate understanding of the relationship between structure and function in the biomolecular complex under investigation (e.g., Fislage et al. 2018). The higher the resolution, the easier it is to determine the correspondence between local and global subunit or domain topology and the function of a target biomolecule. However, if the target can only be determined at an intermediate resolution (considered to be between 10 and 20 Å) due to some combination of extrinsic factors, (technical limitations of the microscope, difficulties with analysis), or intrinsic factors (excessive conformational variation, small size or highly variable complex composition), then other methods will be required to fully determine the target structure. As an example, the order of domains for the small pseudosymmetrical protein LRP6 was not clear from intermediate resolution maps until a comparison was made with two different fragment antibody (Fab) bound structures to the unbound apo form (Matoba et al. 2017). The introduction of ligand (in this case an antibody) provided additional structural features that corresponded to the known binding site, allowing the subunit organization to be understood even at intermediate resolution. Collectively, strategies that leverage some other form of intermediate resolution information (compared with high-resolution features such as the side chains of amino acids) are referred to as EM labeling methods, as often they use an additional label to help identity a domain or subunit by comparing the labeled and unlabeled form and inferring localization information from the additional label density. EM labeling strategies can be classified coarsely as either modification-free methods, that use a non-covalently attached ligand, or alternatively, as modification-based methods, which require the genetic or chemical modification of the target (Table 1). Three important properties of EM labels are their specificity, affinity, and label occupancy. A high specificity ensures that only the target region is labeled, although this can lower the broad applicability of the label system to only targets with a specific epitope. A high affinity allows for extended reactions and preparation of the sample, whereas a low affinity label may dissociate before the sample has been prepared. The occupancy of a label is important as it will mean that lower levels of data are required to capture a particular labeled state. For example, if only 1% of the imaged proteins are labeled then data collection requirements would be 100 times greater than a comparable case were 100% of all particles have a label present. However, in complex systems these requirements are not the sole determinant of success and so any tuning of the specificity, affinity, and occupancy of a label system often needs to be balanced against other experimental requirements.

Table 1.

EM labeling tools showing targetable region, modification size, and type of tag

Name	Modification	Target site	Modified residues	Tag	Citation
BAmSA	No	Transmembrane	0	Streptavidin	Perry et al. 2019
Fab	No	Internal	0	Various Fab	Wu et al. 2012
Ligands	No	Various	0	Various	e.g., Zingsheim et al. 1982
MBP-UAA	Yes	Internal	1	MBP	Dambacher and Lander 2015
Deletions	Yes	Internal	Various	–	e.g., Bui et al. 2008
MBP	Yes	Terminal	396	MBP	Ciferri et al. 2012
Dynein interaction domain	Yes	Terminal	84–111	DID protein	Flemming et al. 2010
Efficient mapping by internal labeling	Yes	Internal	238	GFP	Ciferri et al. 2015
eGFP	Yes	Internal	238	eGFP	Zang et al. 2016, 2018
Domain Localization by RCT Sampling	Yes	Internal	15	Streptavidin	Lau et al. 2012
PA tag	Yes	Internal	13	NZ-1 Ab	Brown et al. 2018

Modification-free methods

Various ligands can be used as EM labels provided they have a known binding region, sufficient affinity to guarantee their presence in the imaged sample, and importantly, are large enough that they will be visible at the desired resolution of investigation. Compounds ranging from antibodies (Boisset et al. 1993), detergents (Perry et al. 2019), to even venom (Zingsheim et al. 1982) have been used to identify ambiguous domains and subunits. Modification-free EM labeling depends on the availability of an appropriate ligand, but have the advantage that, as there is no genetic modification required, there is a lower risk of destabilizing sensitive complexes that may not handle the introduction of a new domain. A recently developed strategy, known as BAmSA [standing for biotinylated amphipol (BA) and monovalent streptavidin (mSA)], exploits the hydrophobic properties of transmembrane regions to cluster biotinylated amphipol, that subsequently allows for labeling with streptavidin. These regions can then be identified in negative stain images, as well as cryo-EM 2D classes and 3D density maps (Perry et al. 2019). Other useful tools include monoclonal antibodies as they can have a high specificity and affinity (Boisset et al. 1993, 1995; Kelly et al. 2010; Matoba et al. 2017; Cormier et al. 2018). Antibody labeling can be broadly generalized with phage display technologies to develop antibodies for various target regions (Wu et al. 2012); however, there is a potentially higher experimental cost due to the extensive search of epitope-paratope space that is required to identify high-specificity and high-affinity antibodies.

EM labeling through modification of the target complex

If a deletion variant is available which lacks a particular subunit or domain, then this can be used to identify the deletion location by comparison with the wild type complex. This has been used extensively to understand the cross-sectional architecture of the cilia and flagella as there are many variants available that lack single or multiple elements relative to the wild-type (Bui et al. 2008; Heuser et al. 2009, 2012; Pigino et al. 2011). The accuracy of such experiments depends on the structural integrity of the variant being maintained when a subunit or component is missing so that the only difference between the wild-type and the deletion variant is the missing region. In the published cases involving the cilia and flagella, which are large multi-subunit components of the cellular machinery on the order of 1–5 μm, mutations that knock-out a single element are demonstrably stable (e.g., Pigino et al. 2011). However for smaller structures the conformation can be altered by removing a component, meaning that such a strategy may not be effective.

Targeted labeling is also possible by inserting new residues or domains into a complex. Insertions can range in size from single mutations, to small epitopes, up to entire domains. With smaller insertions, additional binding partners are necessary to allow for visualization of the insertion site. Site-directed mutagenesis of a single amino acid to certain specific unnatural amino acids (UAA) can can allow for chemical conjugation to maltose binding protein (MBP), a two domain 40 kDa protein that is visible under EM, providing localization of the targeted mutation (Dambacher and Lander 2015). While this process has the advantage of only requiring a single mutation to introduce an UAA, it does require additional reactions to cause the conjugation of MBP, which could introduce some experimental limitations (e.g., very sensitive complexes that cannot withstand additional reactions). Under these conditions, it may be more advantageous to directly insert a new domain so that it is expressed together with the target, without requiring additional tagging reactions. Again, MBP can be used here by directly inserting the entire protein into the terminal regions of the target complex allowing the localization of this domain (Ciferri et al. 2012; Lander et al. 2012). Provided the insertion site does not interfere with expression of the target molecule, the constitutive expression of a tag such as MBP is useful, as there is essentially 100% effective tagging. One drawback of using MBP is that the insertion site is limited to the terminal regions, and so other tags have been developed that can be targeted to central domains. Green fluorescent protein (GFP) has closely situated terminal regions and so can be inserted directly into surface exposed loops of a target protein, allowing tagging of central domains (Ciferri et al. 2012, 2015; Zang et al. 2016, 2018). While constitutive expression does not require any additional tagging reactions, there is a potential reduction in protein yield as the insertion can interfere with expression or protein folding. This can be compensated for by including additional linker residues between the target and the insertion to allow for flexibility and correct folding; however, this may reduce the resolvability of the inserted tag due to the increased flexibility of the domain (Zang et al. 2018). Insertion of a smaller epitope that can be labeled after expression is a way to overcome this potential limitation. Self-assembling tags that can be built after protein expression could solve this issue, such as the dynein interaction domain (DID), a terminal extension of 84 amino acids which self-assembles after protein expression thereby reducing the potential interference that may come from constitutively expressing a large tag (Flemming et al. 2010). Domain localization by RCT sampling (DOLORS) involves the insertion of the 15 amino acid AviTag, which can be inserted into central loops and then post-translationally biotinylatedand tagged with streptavidin (Lau et al. 2012). For a given experiment, the appropriate label depends on the stability of the target complex, if it is able to accept insertions of whole domains such as GFP, or MBP, or whether smaller insertions are more ideal but with the requirement of additional reactions to effect tagging. As we will discuss in the next section, protein purification systems offer a range of tools potentially suitable for this purpose.

Epitope-based tagging

Epitope-based purification systems and their conjugate antibodies can be used as labels in EM due to their high affinity and specificity. One area that requires some consideration is the placement of the epitope, due to the fact that inserting an epitope directly into a central domain can deform the epitope and render it unrecognizable by its conjugate antibody (Fujii et al. 2016). This limitation has been previously recognized when designing constructs for protein purification, as often only the C-, or N-terminal positions are considered for epitope insertion (Malhotra 2009), or alternatively, the insertion site must be carefully selected inside a long loop region (Dinculescu et al. 2002; Morlacchi et al. 2012) or with additional linker residues (Facey and Kuhn 2003; Kendall and Senogles 2006). In all of these cases, the insertion site is chosen to preserve conformational freedom of the epitope so that it can be recognized by the antibody. While some epitopes which have been used as an EM label (e.g., FLAG tag) have been placed in the terminal region (Kelly et al. 2010), other recently developed systems have a number of properties that allow central insertion. In particular, the PA tag and its conjugate antibody, NZ-1, have a particular degree of freedom for internal insertion (Brown et al. 2018) due to unique structural features of the peptide (Fujii et al. 2016).

PA tag/NZ-1 antibody system

The development and experimental characterization of the PA tag/NZ-1 antibody system is reviewed in detail elsewhere (Brown and Takagi 2018), but in brief, the monoclonal antibody NZ-1 was identified during a search for inhibitors of cancer induced platelet aggregation (Kato et al. 2006), and was shown to block aggregation by binding to a 38–51 amino acid long sequence in the PLAG domain of the type I transmembrane protein podoplanin (Ogasawara et al. 2008). Due to the high affinity of the interaction and the small size of the epitope, it was reasoned that the peptide could make a useful protein purification tool and so the minimum recognition site was identified as the dodecapeptide (GVAMPGAEDDVV), dubbed the PA tag (Fujii et al. 2014). The PA tag and NZ-1 antibody have a number of properties that are attractive as a purification handle, including a very high binding affinity relative to other purification tags (Table 2), mild elution conditions, and a very low dissociation rate that allows for extensive washing for removal of contaminants (Fujii et al. 2014). As a purification tool, the PA tag/NZ-1 system has been used extensively to purify a wide range of proteins (Kitago et al. 2015; Mihara et al. 2016; Suzuki et al. 2016; Umitsu et al. 2016; Meng et al. 2017; Nagae et al. 2018; Tabata et al. 2018).

Table 2.

Size and binding affinities of various purification tools

Tag name	Residues	KD	kD reference
FlAsH	6	1.00E-11	Wombacher and Cornish 2011
PA tag	13	4.00E-10	Fujii et al. 2014
G196	9	1.25E-09	Tatsumi et al. 2017
HA	9	1.60E-09	Fujii et al. 2014
MYC	10	2.20E-09	Fujii et al. 2014
SBP-tag	38	2.50E-09	Keefe et al. 2001
CBP	26	3.70E-09	Montigiani et al. 1996
Nanotag15	15	4.00E-09	Lamla and Erdmann 2004
AGIA	9	4.90E-09	Yano et al. 2016
CP5	5	7.49E-09	Takeda et al. 2017
TARGET	21	1.00E-08	Tabata et al. 2010
ABD	46	1.35E-08	König and Skerra 1998
FLAG	8	2.80E-08	Fujii et al. 2014
Strep-tag II	7	1.02E-06	Voss and Skerra 1997

To better understand the structural causes of the high affinity interaction between PA tag and NZ-1 antibody, the crystal structure of the apo and PA peptide-bound NZ-1 Fab (Fig. 1) were collected at 1.65 and 1.70 Å respectively (Fujii et al. 2016). Analysis of these data showed several features that suggest both the antibody and peptide are present in a binding competent conformation prior to interaction that could provide a favorable entropic contribution to binding. First, the structural difference between the apo and peptide-bound NZ-1 was low (~ 0.5 Å RMSD of C_α atoms) indicating that only a minimal conformational change is required upon binding with the PA peptide. Second, several water molecules that form part of the hydrogen bonding network that stabilizes the PA peptide in the binding pocket were also present in the apo structure. Finally, the structure of the PA peptide in the binding pocket of the NZ-1 Fab also offered an explanation for the high affinity interaction and implied some useful applications of the PA tag. The central proline-glycine residues in the peptide cause the formation of a type II β-turn in the NZ-1-binding pocket and the C-, and N-terminals are oriented in the same direction (Fig. 1). The type II β-turn in the PA peptide, its terminal proximity (~ 10 Å) and its orientation, implied that NZ-1 recognizes the PA peptide, not as a linear epitope, but as a compact form that could be inserted into surface exposed loops of central domains without epitope deformation. Such a 'mobile epitope’ would be a useful tool as it has a ready conjugate antibody with high affinity.

Fig. 1.

Crystal structure of the PA peptide (GVAMPGAEDDVV) in the binding pocket of NZ-1 Fab (PDB: 4yo0). The proline-glycine positions 5–6 of the peptide cause a type II β-turn that orients the terminal regions in the same direction

To demonstrate the mobile epitope functionality of the PA tag, the cell surface heterodimeric receptor αIIbβ3-integrin was selected as a test case (Fujii et al. 2016) for the reasons that (1) the heterodimer can be easily reconstituted (Takagi et al. 2002), and (2) the αIIb-subunit has an extensive extracellular region with four β-rich domains that contain multiple surface exposed loops where the PA tag could be inserted (Zhu et al. 2008). Ten different constructs were generated by inserting a single PA peptide directly into various surface exposed loops in the αIIb-subunit and 80% of these were successfully expressed. A non-exhaustive search of the literature found that the failure rate for inserting proteins or peptides into internal domains is 40% (Jarvik and Telmer 1998), and so the lower failure rate of the PA peptide was an initial indication that the small peptide generally does not disrupt protein expression and folding. Importantly, all of the PA tag containing constructs that could be expressed maintained reactivity with NZ-1 showing that the direct insertion of the PA peptide did not distort its epitope. In contrast, two common purification tags, FLAG tag (DYKDDDDK) and MYC tag (EQKLISEEDL), that were inserted into the same loop regions lost reactivity with their respective antibodies (M2 and 9E10) providing evidence that internal insertion distorts linear peptides but not the PA peptide. Interestingly, while the crystal structure of the PA peptide and NZ-1 Fab showed a 10.2 Å spacing between the C-, and N-terminals of the peptide some of the chosen insertion sites in the αIIbβ3-integrin had a spacing that was as small as 5.2 Å, suggesting a high degree of flexibility when choosing an insertion site. This interpretation is also supported by the crystal structure, which shows that the terminal regions of the PA peptide are not involved in antibody recognition. Given the mobile nature of the PA peptide and the high affinity of NZ-1, it was reasoned to be suitable for use as a tool for domain labeling in EM experimentation.

Labeling of internal domains using the PA peptide and NZ-1 Fab has been used to identify domains and subunits for a range of proteins (Wang et al. 2018a, b; Brown et al. 2018). The type II β-turn conformation of the peptide allows the insertion of the peptide into central domains to give a mobile EM labeling system. Such a system has other uses, for example if the PA peptide is carefully inserted into a rotary protein such as the V-ATPase, then binding of NZ-1 can be used to lock the complex into a particular conformation (Tsunoda et al. 2018). By locking the conformational state of a protein, previously rare structural states can be increased, allowing their structural determination using cryo-EM. This system has the advantage of (1) being a highly mobile epitope that can be inserted into central domains, (2) having a ready-made high affinity antibody in NZ-1, and (3) requiring a single labeling step that does not need any other reactions.

Conclusion

Despite the technological and analytical advances that have occurred with cryo-EM, there are still many experimental situations in which attainment of atomic resolution is a not realizable proposition. Tools that can be used to identify subunits or domains from intermediate resolution EM structures, or that can alter the conformation of a target in such a way as to reduce conformational variability are still needed. With respect to these situations, a range of methods exists that can be used to provide information about the location and organization of domains and subunits. These include those that can target broad chemical properties of a target system, such as hydrophobic transmembrane regions (Perry et al. 2019), to more specific approaches involving monoclonal antibodies that recognize a single epitope (Wu et al. 2012). Such a range of tools mean that, for a particular experimental situation, the appropriate properties of the labeling approach including, the type of modification, its size, its affinity and the nature of the chemisty used for carrying out the labeling reaction can all be optimized. One such tool, the PA tag/NZ-1 antibody system has been employed for a wide range of uses such as protein purification (Fujii et al. 2014), Western blotting, cell flow cytometry (Fujii et al. 2014), conformational reporting (Fujii et al. 2016), domain level tagging (Wang et al. 2018a, b; Brown et al. 2018), and to influence conformational states of target biomolecules (Tsunoda et al. 2018). We hope that this review has alerted the biophysics community to this toolbox of intermediate resolution level approaches, which can provide extra structural information in EM experimentation when full atomic detail is not attainable for experimental reasons.

Compliance with ethical standards

Conflict of interest

Zuben P. Brown declares that he has no conflict of interest. Junichi Takagi declares that he has no conflict of interest.

Ethical approval

This article does not contain any studies with human participants or animals performed by any of the authors.