Intrabodies

Goal of the project: To develop genetically encoded probes (intrabodies) that can label endogenous proteins allowing their localization and trafficking to be visualized in arbitrary cell types in vivo in real time.

Why it is difficult to study protein localization in neurons using traditional antibodies

Fig. 1 A Antibody staining of endogenous PSD-95 in the cortex. It is difficult to determine the subcellular localization of PSD-95 from this image because of the lack of context. B Cortical slice containing a pyramidal neuron cotransfected with alkaline phosphatase (purple) and tagged PSD-95 (white) using biolistics. It is relatively easy to determine the subcellular localization of a tagged protein expressed in an isolated cell, even if the cell is in intact tissue. Scale bar is 10 mm. C Golgi stain of pyramidal neuron.

C

Fig. 1 A Antibody staining of endogenous PSD-95 in the cortex. It is difficult to determine the subcellular localization of PSD-95 from this image because of the lack of context. B Cortical slice containing a pyramidal neuron cotransfected with alkaline phosphatase (purple) and tagged PSD-95 (white) using biolistics. It is relatively easy to determine the subcellular localization of a tagged protein expressed in an isolated cell, even if the cell is in intact tissue. Scale bar is 10 mm. C Golgi stain of pyramidal neuron.

Golgi stain: How to study features of individual neurons in densely packed tissue

Fig. 2A Antibody staining of endogenous Kv4.2 in the hippocampus.(1) marks a discrete, punctate pattern of expression. Note that the arrow points to a puncta at the cell body. B Exogenous, tagged Kv4.2 expresses in a nonspecific manner in a hippocampal neuron in culture. It is clear that the tagged overexpressed construct is localized in a manner that does not accurately reflect the endogenous protein (4).

The problem of how to highlight the properties of individual cells that are part of a solid mass of cells was solved by 19th century neuroanatomists, who used the Golgi technique to stain neural tissues. When tissue is prepared using the Golgi technique, a very small number of random cells is stained with silver chromate crystals, which allows even the finest anatomical features of the stained neurons to be marked without being obscured by neighboring cells (Fig. 1C). A similar effect can be achieved molecularly by expressing an exogenous gene in a small percentage of random cells using either a transient transfection system such as biolistics (Fig. 1B;(3)) or by using a transgenic system in combination with a promoter such as Thy-1 that drives expression in a small percentage of cells (5).

Exogenous, tagged proteins as probes for protein localization and trafficking
Exogenous, tagged proteins have proven to be extremely popular for studying subcellular localization because of the following desirable properties:

1. When exogenous, tagged proteins are expressed in a small number of cells they can be easily visualized, in the same manner as neurons marked with a Golgi stain (Fig. 1B, C).

2. Exogenous, proteins can be tagged with GFP, allowing them to be visualized without the need for invasive procedures and thus allowing for proteins to be labeled in real time in vivo.

Many studies of protein targeting use such an approach, despite the fact that it suffers from two serious drawbacks:

1. Exogenous, tagged proteins must compete with endogenous proteins for any interacting proteins necessary for transport and targeting such as clathrin adaptor proteins, kinesins, as well as auxiliary subunits. The result is that, at best, they only partially replicate the expression patterns of the endogenous proteins (Fig. 2).

Fig. 3 A High magnification of a dendrite on a hippocampal neuron in culture expressing tagged, exogenous PSD-95. B. Wild-type cell from the same culture as A with smaller and fewer spines than cell expressing PSD-95. Scale bar 5 mm (2).

2. Overexpressed proteins introduce gain of function mutations (or, if an inactive form of the protein is expressed, dominant negative mutations) that can disrupt the physiology of the cell and can modify the morphology of structures to which they are targeted (Fig. 3). A particularly dramatic example of this occurs when synaptic proteins are overexpressed. PSD-95, SPAR, and many other postsynaptic proteins cause a dramatic increase in the number of dendritic spines and in spine head size (2, 6).

Using intrabodies as genetic probes for labeling endogenous proteins

The ideal probe for measuring subcellular localization of proteins would be genetically encoded, but would directly bind to endogenous proteins and thus avoid the problems associated with tagged, exogenous proteins (Figs. 3, 4). In vitro selection techniques such as phage display and mRNA display currently represent the best way to design such molecules (reviewed in (7, 8)).  In these methods, a large protein library (109 – 1013 members) is generated and used to isolate unique proteins that bind a target, using repeated rounds of selection and amplification.  Recent work by Perez et al., demonstrated the power of this approach, by designing single chain antibodies (scFvs) targeting activated forms of the G protein Rab6.  When linked to GFP and expressed in live cells, these antibody-reporter fusions enabled tracking activated Rab6 in real time, without noticeable effects on protein function (9, 10).  Unfortunately, antibodies are not ideal tools for intracellular recognition as they are subject to unfolding via reduction of their disulfide bonds.  The Rab6 binder seems to be resistant to this process, for reasons that are not entirely clear. In general though, the ability to design disulfide-free antibody-like proteins that bind with high affinity would be a powerful solution for creating intrabodies and tracking proteins in vivo.

mRNA display can be used to develop high affinity peptides and proteins
mRNA display, as designed and implemented by one of us (11, 12), represents a powerful approach to design peptides and proteins via in vitro selection experiments (reviewed in (8)).  This approach enables peptide and protein design by directedmolecular evolution, using libraries with more than 10 trillion independent sequences.

Fig. 4 Synthesis of a mRNA library. A mRNA template (black line) covalently attached to puromycin is used to program an in vitro translation reaction. After protein synthesis, the puromycin enters the ribosome in cis to form a covalent mRNA-protein fusion.

In mRNA display, mRNA molecules bearing a pendant 3’ puromycin are translated in vitro to generate covalent mRNA- protein fusions (Fig. 4). In order to perform directed protein evolution experiments, synthesis of the mRNA-protein fusions is incorporated into an in vitro genetic cycle (Fig. 5).  An initial library is created in the form of double stranded linear DNA using PCR.  Diversity in the library is generated using randomized DNA cassettes (e.g. bearing NNG/C codons; N = any of the 4 bases), mutagenic PCR (13), or both.  In vitro translation results in protein sequences covalently attached to their own mRNA. 

Fig. 5 In vitro selection cycle with mRNA display (adapted from (7,12). A double stranded DNA library is generated from synthetic oligonucleotides. Transcription, ligation of a puromycin linker (P) and translation, results in formation of a mRNA-peptide fusion (left). Extension of this product with reverse transcriptase results in a cDNA/mRNA-protein fusion (right). This product is then enriched for a peptide or protein target using affinity chromatography.

Typically, cDNA synthesis with reverse transcriptase is performed prior to selection to avoid formation of RNA secondary structures and subsequent isolation of RNA aptamers.

The ability to perform directed protein evolution entirely in vitro provides several critical advantages relative to the popular phage display and yeast two-hybrid systems (14, 15).  The obligate in vivo step in both methods commonly limits the number of molecules that can be examined to 10⁶ – 10⁹ sequences.  The in vivo step can also produce substantial library biases due to poor expression, folding, processing, subcellular localization, and proteolytic stability.  For example, murine antibodies are very poorly expressed and processed in bacteria (16).  Poor folding likely represents a common pitfall in bacteria where partially unfolded proteins are readily degraded (17).  Finally, the yeast system can be plagued by false positive signals while the phage system can be prone to avidity effects because neither gene VIII- nor gene III-based screening results in strictly monovalent libraries.

Our published work demonstrates that mRNA display is now a powerful and versatile tool for peptide and protein design (reviewed in (8)).  Using mRNA display, it is now possible to generate both peptide and protein libraries, generating novel functional polypeptides by an organized design algorithm. Examples from the Roberts lab include libraries targeting RNA (18, 19), bacterial antibiotic resistance genes (20), signaling proteins (21), and cell-surface receptors(22).  mRNA display may also be applied to random (23), patterned (24), and scaffold-based (25) protein design experiments. 

Design of Disulfide-Free Intrabodies:  A Trillion member protein Library Based on 10FnIII.

Fig. 6 Design of our antibody mimetic fibronectin 10FnIII-based scaffold library. The 10FnIII domain has a b-sandwich but no disulfide bonds. The 10FnIII FG and BC loops were randomized in our library construction and are 10 and 7 residues long, respectively.

 The 10FnIII domain represents an ideal, disulfide-free scaffold that resembles theantibody VH fold (Fig. 6). Scaffold-based libraries such as antibodies are clearly useful for recognizing protein surfaces and can achieve nanomolar to picomolar affinities (26). However, antibodies can be problematic from a protein design standpoint due to their size, complexity, disulfide bonds, and poor expression in bacterial hosts (16).  The fibronectin 10FnIII domain developed by Koide and coworkers represents a potentially far superior scaffold, particularly forwork inside cells (27, 28). Fibronectin-based scaffold libraries have also been reported in mRNA display experiments (25, 29). In particular, the 10FnIII fibronectin domain appears useful for protein design due to its 1) small size, 2) lack of disulfide bonds, 3) folding stability, 4) known 3-dimensional structure (30, 31), and 5) excellent bacterial expression properties.  As proof of their utility, fibronectin-based libraries have recently been used to isolate novel ligands that target the surface of tumor necrosis factor with picomolar affinity (25) and the surface of VEGF R2 with nanomolar affinity as well (29).

Our designed library has 17 random positions over the BC (7 residues) and FG (10 residues) loops (see Fig. 6) and lacks the unstructured N terminus {Olson, 2007 #2959}. We have also extensively tested the 10FnIII library and found its members to be stable in both intra- and extracellular environments. Our 10FnIII library will thus serve as the predominant scaffold in the directed protein evolution experiments proposed in this grant. We have characterized both the expression and folding of our 10FnIII library.  First, we validated that 10FnIII library members can be expressed readily as GFP-fusions in bacteria by cloning our library into the GFP folding/expression reporter assay developed by Terwilliger (32). We have also tested the stability of library clones using chemical denaturation experiments.  This work demonstrates that many of the clones in the library have chemical stabilities within 2-3 kcal of the wild type 10FnIII (DG(wt)unfolding = 7.7 kcal/mol). Finally, sequencing of the clones reveals that the codon usage and amino acid representation is very good and similar to well-expressed proteins. 

Generating a fibronectin-based protein that binds specifically to PSD-95
We chose a 20 amino acid hydrophilic region (AAs 271-290) that is unique to PSD-95 as a target and then generated a protein that binds to that target with high affinity using a selection procedure known as mRNA display (12). In this selection, convergence occurred after 6 rounds, at which point a single protein was found to account for 89% of the library, as determined by sequencing DNA generated from the mRNA-protein library members.

Fig. 7 Radiolabeled protein-mRNA hybrids from different rounds of the selection were exposed to the target peptide. By the 6th round 60% bound to the target + bead, while <1% bound to bead alone.

To test how well the pools of proteins from different rounds bound to the target, we performed a “hot binding assay”. This assay involves radioactively labeling mRNA-protein molecules from the pools selected in each round, and determining what percentage of them bind to the target. To control for nonspecific binding, the percentage of proteins from each round that bind to the target peptide on the bead is compared with binding to the bead without the target. By the sixth round almost 60% of the library bound to the target, but not to the beads alone, indicating that the selection had converged on a high affinity binder (Fig. 7). Note that because of inherent inefficiencies in transcription and translation, binding efficiency is never 100%. We retrieved the DNA encoding the high affinity binder (fibro2E) and placed it in both bacterial and mammalian expression vectors for further testing.

Fibro2E binds to and colocalizes with PSD-95

To test whether fibro2E could bind to PSD-95 in cells, we expressed a GFP-tagged version of it in COS cells along with exogenous, tagged PSD-95 and Kv1.4. Kv1.4 clusters with PSD-95 to produce small rafts on the cell surface. Fibro2E showed dramatic colocalization with PSD-95 indicating that it could bind to PSD-95 even though PSD-95 was part of a raft of proteins (Fig. 8A, B). To determine whether fibro2E could target proteins to postsynaptic densities in neurons, we expressed it in dissociated cortical neurons. We found that fibro2E-GFP targeted very specifically to the tips of dendritic spines, indicating that fibro2E is capable of targeting proteins to postsynaptic sites (Fig. 8C).

Fig. 8 (A) The PSD-95 binding protein fibro2E colocalizes with (B) PSD-95 when the two proteins are coexpressed in glial cells. (C) fibro2E-GFP localizes to post-synaptic densities in a dissociated neuron cotransfected with Kv1.4.

References

1. S. Jinno, A. Jeromin, T. Kosaka, Neuroscience 134, 483 (2005).
2. A. E. El-Husseini, E. Schnell, D. M. Chetkovich, R. A. Nicoll, D. S. Bredt, Science 290, 1364 (Nov 17, 2000).
3. D. B. Arnold, D. E. Clapham, Neuron 23, 149 (1999).
4. J. Kim, D. S. Wei, D. A. Hoffman, J Physiol 569, 41 (Nov 15, 2005).
5. G. Feng et al., Neuron 28, 41 (Oct, 2000).
6. D. T. Pak, S. Yang, S. Rudolph-Correia, E. Kim, M. Sheng, Neuron 31, 289 (Aug 2, 2001).
7. W. J. Dower, L. C. Mattheakis, Current Opinion in Chemical Biology 6, 390 (2002).
8. T. T. Takahashi, R. J. Austin, R. W. Roberts, Trends in Biochemical Sciences 28, 159 (2003).
9. C. Nizak et al., Science 300, 984 (May 9, 2003).
10. C. Nizak et al., Traffic 4, 739 (Nov, 2003).
11. R. Liu, J. Barrick, J. W. Szostak, R. W. Roberts, Methods in Enzymology 317, 268 (2000).
12. R. W. Roberts, J. W. Szostak, Proc. Natl. Acad. Sci. USA 94, 12297 (1997).
13. J. Tsang, G. F. Joyce, Methods in Enzymology 267, 410 (1996).
14. G. P. Smith, V. A. Petrenko, Chemical Reviews 97, 391 (1997).
15. S. Fields, Song, O., Nature 340, 245 (1989).
16. A. Pluckthun et al., in Antibody Engineering:  a Practical Approach H. Hoogenboom, J. McCafferty, D. Chiswell, Eds. (IRL Press, Oxford, 1996) pp. 203-252.
17. A. A. Pakula, R. T. Sauer, Annu. Rev. Genet. 23, 289 (1989).
18. J. E. Barrick, T. T. Takahashi, J. Ren, T. Xia, R. W. Roberts, Proc. Natl. Acad. Sci. USA 98, 12374 (2001).
19. T. Xia, A. Frankel, T. T. Takahashi, J. Ren, R. W. Roberts, Nature Structural Biology 10, 812 (2003).
20. S. Li, R. W. Roberts, Chemistry & Biology 10, 233 (2003).
21. W. W. Ja, R. W. Roberts, Biochemistry 43, 9265 (2004).
22. Z. Zhang et al., Science 303, 371 (Jan 16, 2004).
23. A. D. Keefe, J. W. Szostak, Nature 410, 715 (2001).
24. D. S. Wilson, A. D. Keefe, J. W. Szostak, PNAS 98, 3750 (2001).
25. L. Xu et al., Chemistry & Biology 9, 933 (2002).
26. E. T. Boder, K. S. Midelfort, K. D. Wittrup, Proc. Natl. Acad. Sci. 97, 10701 (2000).
27. A. Koide, C. W. Bailey, X. Huang, S. Koide, Journal of Molecular Biology 284, 1141 (Dec 11, 1998).
28. A. Koide, S. Abbatiello, L. Rothgery, S. Koide, Proc. Natl. Acad. Sci. USA 99, 1253 (Feb 5, 2002).
29. E. V. Getmanova et al., Chem Biol 13, 549 (May, 2006).
30. A. L. Main, T. S. Harvey, M. Barron, J. Boyd, I. D. Campbell, Cell 71, 671 (1992).
31. C. D. Dickinson et al., J. Mol. Biol. 236, 1079 (1994).
32. G. S. Waldo, B. M. Standish, J. Berendzen, T. C. Terwilliger, Nature Biotechnology 17, 691 (1999).