To test whether it is possible to modify an Nb such that its intracellular protein level is strongly dependent upon antigen co-expression, we used the GFP-binding Nb, GBP1, for proof-of-concept experiments (Kirchhofer et al., 2010; Rothbauer et al., 2006) (Figure 1B,C). We generated a Moloney murine leukemia virus (MMLV) library encoding randomly mutagenized variants of GBP1 fused to the blue fluorescent protein, TagBFP (Subach et al., 2008). t-HcRed (Gurskaya et al., 2001) was co-expressed via an IRES to report infection. TagBFP and t-HcRed bear little amino acid similarity to Aequorea-derived GFP and its derivatives. We infected 293T cells with this library, and combined FACS with super-infection by a GFP-encoding recombinant adeno-associated virus (rAAV) to isolate GBP1-TagBFP variants whose blue fluorescence depended upon GFP expression (Figure 1B; Materials and methods). One hundred GBP1 variants were then individually screened for enhanced TagBFP expression in the presence of yellow fluorescent protein (YFP), a GFP derivative known to also interact with GBP1 (Rothbauer et al., 2008; Tang et al., 2013). Some variants showed fusion TagBFP aggregates within well-transfected cells when YFP was absent, but became soluble in the cytoplasm when YFP was present (Figure 1—figure supplement 1A). Notably, a GBP1 variant carrying 6 amino acid changes (A25V, E63V, S73R, S98Y, Q109H, S117F) gave little to no TagBFP fluorescence, with no signs of aggregation in the absence of YFP. We focused our efforts on this variant, which will hereafter be referred to as destabilized GBP1 (dGBP1). dGBP1-TagBFP showed strong fluorescence and protein level when co-expressed with GFP or YFP, but became weakly detectable or undetectable when antigen was absent (Figure 1D,E and Figure 1—figure supplement 1). In contrast, unmodified GBP1-TagBFP showed strong fluorescence and protein level regardless of antigen co-expression (Figure 1D,E). Interestingly, we detected an increase in the level of wildtype GBP1-TagBFP protein in the presence of YFP (Figure 1—figure supplement 1C). In an electroporation experiment using the mouse retina, dGBP1-TagBFP fluorescence and protein level were detected only upon GFP co-expression in vivo (Figure 1F–H, Figure 1—figure supplement 2). Strikingly, the efficiency of TagBFP stabilization by GFP expression was nearly 100%, i.e. almost every GFP+ cell was TagBFP+ (Figure 1—figure supplement 2). The efficiency of the TDDOG and CRE-DOG systems was, at the highest, ~60% in similarly designed electroporation experiments (Tang et al., 2013, 2015). This difference likely reflects the requirement for delivery of a greater number of components for T-DDOG and CRE-DOG experiments. Taken together, these data show that one can create a highly destabilized Nb whose protein level is dependent upon co-expression with its cognate antigen in vitro and in vivo.

We previously created Nb-based, antigen-dependent systems that use the antigen as a scaffold for the assembly of split protein domains or fragments (Tang et al., 2013, 2015). Complex assembly can be inhibited when excessive antigen levels saturate antigen-binding sites in Nb-fusion proteins (Tang et al., 2013, 2015). In contrast, a single polypeptide, dNb-fusion protein should not suffer the same limitation. Indeed, YFP promoted dGBP1-TagBFP stability in a dose-dependent manner, with no adverse effects even when YFP plasmid was transfected at ten-fold excess of dGBP1-TagBFP plasmid (Figure 1I). To investigate the mechanism of dNb destabilization, dGBP1-TagBFP-transfected 293T cells were treated with the ubiquitin proteasome inhibitors, MG132 or Bortezomib (BTZ) (Kisselev et al., 2012). dGBP1-TagBFP protein was evident following addition of either inhibitor and was absent without inhibitors, indicating that it was degraded by the ubiquitin proteasome system (UPS) (Figure 1J).

Discosoma-derived mCherry fused to dGBP1 (dGBP1-mCherry) also showed antigen-dependent stabilization. Unlike dGBP1-TagBFP (Figure 1E), some aggregation of dGBP1-mCherry occurred inside cells when antigen was absent (Figure 2—figure supplement 1). We use dGBP1-mCherry as a sensitized reporter to map the key residues involved in GBP1 stability, by comparing the level of fluorescence and aggregation of the fusion proteins in cells. (Figure 2—figure supplements 1,2). C/S98Y and S117F showed strong destabilizing effects, as seen in both sufficiency and necessity experiments. S73R and Q109H also had destabilizing effects in single mutant analyses. GFP rescued the destabilization phenotype of all mutants. Thus, specific single dGBP1 mutations had clear destabilizing effects, which could be enhanced by combination with other destabilizing mutations.

The dGBP1 mutations mapped onto the structurally conserved framework regions of Nbs (Muyldermans, 2013), and 99–100% of Nbs (n=76) shared the same residue as GBP1 at each of the 3 most destabilizing positions (3maj: S73R, C/S98Y, S117F) (Figure 2A; Figure 2—figure supplement 3A). Further, a survey across 76 unique Nb-antigen interfaces, gathered from a total of 102 crystal structures, indicated that Nb positions corresponding to those of dGBP1 A25V, S73R, S98Y and S117F were universally located outside of all Nb-antigen interfaces (Figure 2—figure supplement 3B). Nb positions corresponding to dGBP1 Q109H fell outside of 99%, or 75 of 76 unique Nb-antigen interfaces. Positions equivalent to dGBP1 E63V were found in 22%, or 17 of 76 unique Nb-antigen interfaces, and in close proximity to the interface in 9%, or 7 of 76 of the cases. Given these results, we hypothesized that the destabilizing framework mutations could be transferred across Nbs to rapidly create antigen-dependent stability. We transferred all dGBP1 mutations (6mut) and the 3maj mutations to Nbs targeting the HIV-1 capsid protein (αCA) and Escherichia coli dihydrofolate reductase (αDHFR), respectively. dNbs created by mutation transfer behaved similarly as dGBP1 in that TagBFP fusion fluorescence and protein level both depended upon expression of the cognate antigen (Figure 2B–E). Destabilization also depended on degradation by the UPS (Figure 2F). We then explored whether dNb-TagBFP expression had an adverse effect on antigen expression. Using western blots to quantify protein levels, we found that dGBP1-TagBFP did not have an obvious effect on YFP protein level, when compared to the negative control condition whereby αCA^6mut-TagBFP replaced dGBP1-TagBFP (Figure 2—figure supplement 3C).

Figure 2 with 3 supplements see all

Download asset Open asset

Figure 2—source data 1

Source data for fluorescence quantifications of Nb-TagBFP tests.

This excel file contains numerical values for the % Nb-TagBFP fluorescence parameter shown as a colored heat map in Figure 2G.

https://doi.org/10.7554/eLife.15312.007

Download elife-15312-fig2-data1-v2.xlsx

To further investigate the generality of the mutation transfer approach, we transferred the 3maj mutations to 9 Nbs that recognize epitopes of intracellular origin (Figure 2G; Materials and methods). All dNb-TagBFPs showed strongly reduced fluorescence relative to their unmodified Nb counterparts and occasionally formed faint fluorescent punctae in cells over-expressing the fusion proteins (Figure 2B,D and 2G). Importantly, whereas no unmodified Nb showed >2 fold increase in TagBFP fluorescence in response to antigen co-expression, 8 of 9, or 89% of dNbs, passed this threshold (Figure 2G). Notably, mutations that destabilized a Camelus dromedarius Nb (GBP1) had very similar effects on Nbs derived from Vicugna pacos and Llama glama, indicating that destabilizing mutations can be transferred across Nbs from different camelid species to create antigen-dependence (Figure 2A and G; Table 1).

To explore if additional fusion proteins could be rendered antigen-dependent, we engineered dNbs onto two popular site-specific recombinases, Cre and codon-optimized Flp (Flpo) (Luo et al., 2008; Raymond and Soriano, 2007). Both enzymes were rendered GFP-dependent after fusion to dGBP1, but not GBP1 (Figure 3A–B). By increasing the number of dGBP1 domains fused to either enzyme, residual GFP-independent recombination of the initial fusions was decreased without affecting GFP-dependent recombination (Figure 3A–B and Figure 3—figure supplement 1A–B). Notably, Flpo fused to tandemly repeated dGBP1 (dGBP1x2-Flpo, or Flp dependent on GFP [Flp-DOG]) had insignificant background signal and over 600 fold induction by GFP (Figure 3B). We also rapidly constructed an Flpo fusion protein dependent upon the C-terminal portion of HIV-1 CA (C-CA) using mutation transfer. This construct was functional in vitro and in vivo(Figure 3C and Figure 3—figure supplement 1D). Further, both GFP- and C-CA-dependent Flpo responded to antigen in a dose-dependent manner (Figure 3—figure supplement 1C–D). Thus, dNbs can confer antigen-dependent control over different types of effector proteins, and can adequately reduce the activity of a highly sensitive enzyme when antigen is absent.

Figure 3 with 1 supplement see all

Download asset Open asset

The ability to control Flpo activity with tandem dNb-fusions raised the possibility of imposing dual regulation on effector protein activity using two dNbs, each targeting a different antigen (Figure 3D–E and Figure 3—figure supplement 1E). We tested this by generating Flpo fused to dGBP1 and αCA dNb^6mut (dGC-Flpo) or dGBP1 and αDHFR dNb^3maj (dGD-Flpo). Strong Flpo recombination was triggered only when both antigens were present (Figure 3D–E). This shows the feasibility of using dNb-fusion proteins to create synthetic circuits whereby dual inputs are integrated entirely at the protein level.

We further tested whether it was possible to perform genome targeting and editing under the control of specific antigen(s) (Figure 4A). We created a fusion between two αCA dNb^6mut and Cas9 (dCC-Cas9) and delivered the construct to an engineered human cell line that expresses β-galactosidase upon removal of a loxP-stop-loxP cassette. We also delivered a guide RNA that can specifically target the loxP sites, leading to Cas9-mediated deletion of the stop cassette (dCC-Cas9-loxPgRNA) (Figure 4B–C). Co-expression of C-CA with dCC-Cas9 and loxPgRNA triggered genome-editing events, while little to no β-galactosidase expression was detected when C-CA was absent (Figure 4D). The efficiency of C-CA-dependent genome editing approached that of control Cas9 (Figure 4D). This result demonstrated the feasibility of using intracellular epitopes to initiate genome editing or targeting.

Figure 4

Download asset Open asset

To evaluate the usefulness of dNbs, we applied dNb-fusion proteins in situations where we could not control antigen level. GFP and its derivatives (Tsien, 1998) are widely used to label cell types, with specificity dependent upon cellular features such as gene transcription (Chalfie et al., 1994) or neuronal connectivity (Beier et al., 2011; DeFalco et al., 2001; Ekstrand et al., 2014; Lo and Anderson, 2011; Wickersham et al., 2007). Genetic manipulation of GFP-labeled cells can reveal their functions, but current approaches require delivery of 2 or more αGFP Nb-fusion proteins (Tang et al., 2015, 2013). Here, we used electroporation and rAAV to deliver the one-component Flp-DOG along with Flp-dependent constructs to the retinas of Tg(CRX-GFP) (Samson et al., 2009) and cerebella of Tg(GAD67-GFP) (Tamamaki et al., 2003) lines, respectively. In both instances, robust Flpo recombination was detected in GFP+ tissues, but not in GFP-negative tissues labeled with electroporation, infection or injection markers (Figure 5A–C; Figure 5—figure supplement 1, and Figure 5—figure supplement 2A–E). We used rAAV-delivered Flp-DOG to induce ChR2-mCherry expression in GABAergic Purkinje cells (PCs) of Tg(GAD67-GFP) cerebella (Figure 5C). Under conditions in which infection did not alter spontaneous firing frequency or input resistance, we evoked excitatory photocurrents and synaptic inhibitory currents in ChR2-mCherry+ PCs, with inhibitory inputs from neighboring ChR2-mCherry+ neurons that contacted the recorded PCs (Figure 5; Figure 5—figure supplement 2F–G). GFP+ neurons that did not express ChR2-mCherry, as well as control ZsGreen+ neurons in GFP-negative animals, never showed light-evoked photocurrents, indicating antigen-specificity of the system (Figure 5D). Thus, Flp-DOG provides a much simpler approach to manipulate GFP-defined cell types over pre-existing methods. Overall, these results demonstrate that dNb-fusion proteins enable functional manipulation of antigen-expressing cells in vivo.

Figure 5 with 2 supplements see all

Download asset Open asset

Next, we tested whether dNbs could be used to detect and isolate live, antigen-expressing cells. ACH-2 (Folks et al., 1989), a human T-cell line chronically infected with HIV-1, is widely used to study HIV-1 persistence (Clouse et al., 1989). Destabilized αCA fused to either TagBFP or TagRFP (Matz et al., 1999) were expressed in ACH-2 or the uninfected parental cell line (CEM), under conditions in which HIV-1 was reactivated with phorbol 12-myristate 13-acetate (PMA) (Poli et al., 1990) (Figure 6A and Figure 6—figure supplement 1). Using flow cytometry, we detected fluorescence of the destabilized fusions selectively in ACH-2, but not CEM cells (Figure 6B–C and Figure 6—figure supplement 1B). ACH-2-specific fluorescence was dependent upon CA recognition, as the effect was not observed with dGBP1-TagBFP. Importantly, unmodified αCA Nb-TagBFP fluoresced strongly in both cell lines and could not be used to distinguish between the two (Figure 6B–C). As positive controls, we confirmed that αCA dNb^6mut-TagBFP could be stabilized by C-CA co-expression in CEM cells (data not shown) and that the HIV-1 CA antigen was specifically detected by immunofluorescence in ACH-2, but not CEM cells (Figure 6—figure supplement 1C). Thus, dNbs make possible detection of intracellular viral epitopes without the need for cell fixation or membrane permeabilization, enabling live monitoring of intracellular viral protein expression and specific isolation of infected live cells with a choice of spectrally distinct fluorescent proteins.

Figure 6 with 1 supplement see all

Download asset Open asset

A key aspect of this study is the demonstration that these Nb-based, conditionally stable reagents were effective even when the experimenter did not choose the antigen levels. This was true for experiments conducted in human cell culture and in mice. Importantly, the detection of HIV-1 CA+ cells was achieved using a dNb that was rapidly generated by mutation transfer rather than by isolation from a screen. In future work, one may create fusion proteins that specifically manipulate or kill infected cells, e.g. by fusing a conditional Nb to a cellular toxin.

Studies of model organisms often take advantage of transgenic lines that express an exogenous protein in specific cell types (Luo et al., 2008). Driver molecules, such as transcription factors and site-specific recombinases, can respond to the introduction of DNA cassettes to enable the manipulation of gene expression in a cell type-specific manner. Here, we used GFP, which has no naturally known regulatory abilities, as a novel driver molecule. The GFP/dGBP1 binary system is thus analogous to the popular GAL4/UAS, Cre/loxP and Flp/FRT systems. Flp-DOG is immediately useful for studies in model organisms such as the mouse, by making use of existing transgenic GFP reporter lines (>1000 lines in the mouse) (Chalfie, 2009; Gong et al., 2003; Heintz, 2004; Siegert et al., 2009; Tang et al., 2015, 2013) or virally labeled neural circuits (Beier et al., 2011; DeFalco et al., 2001; Ekstrand et al., 2014; Lo and Anderson, 2011; Schwarz et al., 2015; Wickersham et al., 2007) for cell-specific manipulation studies. In addition, one can combine GFP and the popular Cre recombinase for intersectional Cre + Flp cell targeting studies (Dymecki et al., 2010; Fenno et al., 2014).

Destabilized nanobodies were originally developed with the desire to simplify the delivery of GFP-dependent reagents as well as to improve their performance. Indeed, we observed that the most direct output of the GFP/dGBP1 system – a dGBP1 fusion protein (dGBP1-TagBFP), could be stabilized by YFP with close to 100% efficiency amongst electroporated cells of the retina (Figure 1—figure supplement 2B and C). In comparison, the efficiency of direct reporter output using the T-DDOG and CRE-DOG systems was ~60% in similarly designed experiments, likely due to the need to co-deliver multiple reagents to the same cell (Tang et al., 2015, 2013). Also, whereas the previous dimerizer-based systems were inhibited by excessive amounts of GFP, the activity of Flp-DOG continued to rise along with increasing GFP levels (Figure 3—figure supplement 1) (Tang et al., 2015, 2013). These results highlight the improvements in the GFP/dGBP1 approach compared to previous approaches. However, there are several potential caveats of the GFP/dGBP1 system. First, although Flp-DOG is relatively easier to deliver to tissues as a single-component coding sequence, enhanced construct delivery may lead to enhanced background activity. Indeed, we found that it was necessary to remove the woodchuck hepatitis virus post-transcriptional regulatory element (WPRE) sequence from the rAAV expression cassette in order to avoid background Flp-DOG activity with rAAV infection. Second, tandemly repeated dGBP1s were fused to Flpo in order obtain tight GFP-dependent recombination. The requirement for two GFP molecules to stabilize a single Flp-DOG construct is predicted to reduce the sensitivity of Flp-DOG for lower GFP expression levels. However, in practice, Flp-DOG had adequate activity for the targeting of GFP cell types of the two different transgenic GFP lines tested. Nevertheless, these considerations suggest that one should establish an appropriate level of construct delivery, such as the amount of DNA plasmid or virus to deliver for optimal Flp-DOG performance.

Beyond fluorescent proteins, endogenous proteins should be usable as driver molecules to trigger sensor or effector activity. Endogenous proteins would enable one to selectively target specific cell types in wildtype animals for experimentation, without requiring any knowledge of cell type-specific promoters or creation of knock-in alleles. Such an approach would especially benefit studies of non-model organisms, with the only demand being a method to introduce genetic constructs, e.g. via viral vectors. As a possible caveat, the range of biological activities that could be driven by an endogenous protein may be limited by the protein’s natural functions and/or sub-cellular localization. For example, it may be problematic to use a membrane-localized protein to trigger DNA recombination events in the nucleus. One may overcome this by choosing a dNb-fusion protein that, when stabilized, is biologically active in the same sub-cellular localization as that of the antigen. Another possibility is that an antigen-stabilized complex may travel from the site where stability is conferred to another sub-cellular locations to exert its function. These possibilities will likely be idiosyncratic to the fusion protein and antigen. An additional caveat is the effect that a dNb fusion protein might have on the targeted endogenous protein. There could be cases where antigen-Nb interactions lead to less antigen activity. Reduction in activity will depend on several variables. The ratio of Nb fusion to antigen, the binding site of the Nb fusion, and perhaps the particular Nb fusion structure, all have the potential to create changes in antigen activity. Whether or not a reduction in cellular function will be effected will depend upon the sensitivity of the cell to the level of antigen activity. In most heterozygous loss of function mutations in mice and humans, a phenotype is not noted. Although we cannot predict the frequency with which Nb fusions will result in a phenotype, we believe that the method is strong enough to encourage its continued development and application.

Although dGBP1-TagBFP showed virtually no background signal, fusions of dGBP1 to some fusion partners gave significant background signals. Background signals could be addressed by simply increasing the number of dNbs fused to the protein partner. Additional engineering efforts could further reduce background activity of particular fusion constructs. For example, background activity of dGBP1-Cre could be further controlled by fusion to an ERT2 domain to create small-molecule dependency (Feil et al., 1997). Lastly, one could perform additional screens to isolate novel destabilizing mutant combinations that enhance the antigen-specificity of a wider variety of sensor/effector fusion partners. Such an approach could help eliminate the background fluorescent aggregates seen with some dNb-fluorescent protein fusions.

Additional improvements may be made to enhance the response of dNb fusion constructs to antigen co-expression. First, the activity of a protein might be affected by fusion to the Nb. For example, Cre activity was reduced upon binding of GFP to GBP1-Cre, possibly as a result of steric hindrance. Second, the sensitivity of the dNb fusion construct might be sub-optimal. Although the proportion of CA+ cells detected by the dNb sensor was approximately 1/3 to 1/2 of that detected by a mouse monoclonal antibody, the efficiency of antigen detection may be improved by optimization of the dNb fusion construct or of the gene delivery protocol. Possible optimization steps may include exploring different fusion orientations, linker lengths or linker compositions.

The fusion of protein binders to fluorescent proteins enables visualization of antigen localization in living cells. Optimal signal-to-noise detection requires that the fluorescent fusion proteins be strictly co-localized with the antigen. This may not occur if the number of fluorescent fusion proteins exceeds the number of target antigens. One could address this by designing a transcriptional feedback mechanism to control the level of a fluorescent fusion protein (Gross et al., 2013). This method requires that the antigen be localized outside of the nucleus. The use of dNb fluorescent protein fusions is not limited by this requirement, as the mechanism for background reduction involves protein degradation rather than transcriptional feedback. Indeed, we found that a dNb-TagBFP construct became strictly localized to the nucleus upon co-expression with its NLS-tagged antigen (data not shown).

The finding that one can fuse an effector protein to two dNbs led to the development of a strategy wherein two different antigens bound to distinct dNbs could stabilize the effector protein, Flpo. With further development, dual antigen dependence may enable one to precisely target specific cell populations in ways similar to established intersectional strategies, but using proteins that do not necessarily have any defined regulatory abilities (Dymecki et al., 2010; Luo et al., 2008).

Here, we demonstrated the ability to integrate dNb with CRISPR/Cas technology to perform genome editing selectively in antigen-expressing cells. A concern with the expression of Cas9 and gRNA in cells is that there is non-specific genome editing, and several methods are being developed to address this problem (Hsu et al., 2014; Sander and Joung, 2014). The strategy developed here, wherein dNb-Cas9 activity is suppressed until antigen can stabilize the fusion protein, offers a novel strategy to restrict the cell populations that might suffer from off-targeting events.

The screening strategy described here should enable the creation of additional protein-responsive reagents useful for control of fusion protein activity in specific cell populations. A key feature of our screen is the use of rAAVs to deliver the antigen to cells. Virtually all MMLV-infected cells in culture can be super-infected by rAAV, and the cells remain viable for subsequent culture expansion and FACS. In principle, this screening strategy can be extended to generate a diversity of protein-responsive, destabilized binders based on the Nb scaffold or other protein scaffolds (Wurch et al., 2012; Helma et al., 2015).

Over the past 20 years, ~100 crystal structures featuring Nb-antigen complexes have been solved. We leveraged this resource to establish a phylogenetic and structural basis for transferring destabilizing mutations across Nbs. All successfully modified Nbs were derived from camelid species different from that of GBP1, demonstrating the broad transferability of the mutations discovered here. As one would expect, dNb-TagBFPs generated by mutation transfer showed a spectrum of fluorescence fold-change in response to antigen co-expression. This is likely due to multiple factors, including variable Nb affinity for antigen, variable antigen stability, and variable Nb stability even before destabilization. In addition, Nbs might have variable tolerance to the destabilization mutations tested. Thus, although the high percentage of successful mutation transfers indicate that the strategy is generally applicable, it would be beneficial to derive novel combinations of destabilizing mutations that are even better tolerated across Nbs. Lastly, although dNb generation may be limited by the availability of Nbs isolated from immunized animals, additional Nbs and dNbs may be isolated from in vitro screening technologies that are constantly being improved upon.

Beyond Nbs, multiple classes of artificially derived binding proteins that are amenable to expression in living cells are being developed for antigen-recognition (Helma et al., 2015; Wurch et al., 2012). As epitope-specific binders are typically generated by varying loops or surfaces on a common structural scaffold, it should be possible to generate epitope-responsive properties by incorporating a common set of mutations onto conserved and non-epitope binding regions of the scaffold. Future developments building upon this work should expand our ability to rapidly generate sensors and effectors against a diversity of intracellular epitopes, for cell- or antigen-specific applications in biology and medicine.

The Institutional Animal Care and Use Committee at Harvard University approved all animal experiments. Timed pregnant CD1 (Charles River Breeding Laboratories, Boston, MA) were used for electroporation experiments. Tg(CRX-GFP) (Samson et al., 2009) and Tg(GAD67-GFP) (Tamamaki et al., 2003) were kept on a C57/BL6J background.

pCAG-GFP (Addgene plasmid 11150) (Matsuda and Cepko, 2004). pCAG-YFP (Addgene plasmid 11180) (Matsuda and Cepko, 2004). pCAG-DsRed (Addgene plasmid 11151) (Matsuda and Cepko, 2004). pRho-GFP-IRES-AP (referred to as Rho-GFP) (Emerson and Cepko, 2011). pCAG-nlacZ (Cepko lab, Harvard Medical School) pCAGEN (Addgene plasmid 11160) (Matsuda and Cepko, 2004). pCALNL-DsRed (Addgene plasmid 13769) (Matsuda and Cepko, 2004). pCAFNF-DsRed (Addgene plasmid 13771). (Matsuda and Cepko, 2004). pCALNL-luc2 (Tang et al., 2015). pRL-TK (#E2241; Promega, Madison, WI).

Antibodies used were rabbit anti-TagRFP (also targets TagBFP; 1:5,000 dilution for immunoblot, 1:1,000 for immunohistochemistry) (AB233; Evrogen, Moscow, Russia), mouse anti-βgal (1:50 for immunoblot) (40-1a supernatant; Developmental Studies Hybridoma Bank, University of Iowa), chicken anti-βgal (1:1,000 for immunohistochemistry) (ab9361;Abcam, Cambridge, MA), rabbit-anti-GFP (1:500 for immunohistochemistry) (A-6455; Invitrogen, Carlsbad, CA), rabbit anti-GAPDH (1:10,000 for immunoblot) (A300-641A; Bethyl Laboratories, Inc., Montgomery, TX), mouse anti-FLAG M2 (1:1,000 for immunoblot) (F1804; Sigma-Aldrich, St. Louis, MO), mouse anti-KC57-RD1 (5ul per 1 million cells) (6604667; Beckman Coulter, Danvers, MA). Secondary antibodies used were goat anti-chicken Alexa Fluor 647 (1:500 of 50% glycerol stock) (102371; Jackson ImmunoResearch Laboratories Inc., West Grove, PA), goat anti-rabbit DyLight 649 (1:500 of 50% glycerol stock) (111-495-144; Jackson ImmunoResearch Laboratories Inc.), anti-rabbit or anti-mouse IgG-Horseradish Peroxidase (GE Healthcare, Little Chalfront, UK).

293T cells (Cepko lab stock). ACH-2 (NIH AIDS Reagent Program, Germantown, MD). CEM (NIH AIDS Reagent Program). HIV-1 integration sites in ACH-2 cell line were authenticated by integration site analysis of HIV-1 genome, confirming the major integration site is on chromosome 7. Using immunostaining followed by flow cytometry analysis, HIV-1 CA protein was confirmed to be present in ACH-2 cells, and to be absent in CEM cells.

All antigen and Nb protein sequences, except YFP, were acquired from Protein Data Bank (PDB). Protein sequences were backtranslated into DNA sequences, using codons optimized for Mus musculus. The list of tested Nbs and their antigens are listed in Table 1. In general, an antigen sequence was synthesized as gBlock fragments, which were inserted into an EcoRI/NotI digested pCAG vector via Gibson Assembly, giving pCAG-antigen plasmids used for co-expression of antigen in cells. In general, Nb sequences were synthesized as gBlock fragments, and individually inserted into an EcoRI/SphI digested pCAG-TagBFP vector via Gibson Assembly, giving pCAG-Nb-TagBFP plasmids. To destabilize Nbs, mutations were introduced into residue positions that aligned with the dGBP1 mutation positions. Equivalent residues were easy to identify since surrounding amino acid sequences were highly conserved. For 6mut combo, the dGBP1 mutations were A25V, E63V, S73R, C/S98Y, Q109H and S117F. For 3maj combo, the dGBP1 mutations were S73R, S98Y and S117F. Note that C/S98Y in GBP1 was originally a cysteine, but was mutated to serine in earlier studies to avoid complications with disulfide bond formation.

pBMN-GBP1-TagBFP - A GBP1-TagBFP construct was inserted into a BamHI/NotI digested pBMN DHFR(DD)-YFP (a gift from Thomas Wandless; Addgene plasmid # 29325) (Iwamoto et al., 2010), replacing the DHFR(DD)-YFP insert and generating pBMN-GBP1-TagBFP. This became the host vector for mutagenized GBP1 inserts.

pBMN-dGBP1-Cre and pBMN-dGBP1-Flpo –pBMN-dGBP1-TagBFP were digested with SphI/SalI, liberating TagBFP as well as the IRES-t-HcRed element. PCR-amplified Cre and Flpo fragments were then inserted into the digested vector via Gibson Assembly.

pBMN-GBP1-Cre and pBMN-GBP1-Flpo – PCR fragments of GBP1 were inserted into BspEI/SphI digested pBMN-dGBP1-Cre and pBMN-dGBP1-Flpo by Gibson Assembly, resulting in pBMN-GBP1-Cre and pBMN-GBP1-Flpo, respectively. dGBP1 sequence was removed by BspEI/SphI digest.

pBMN-dGBP1x2-Cre, pBMN-dGBP1-GBP1-Cre, pBMN-dGBP1x2-Flpo, pBMN-dGBP1-GBP1-Flpo – pBMN-dGBP1-Cre or –Flpo plasmids were digested with SphI. A gBlock fragment encoding a codon modified dGBP1 was inserted into this site via Gibson Assembly, generating pBMN-dGBP1x2-Cre or –Flpo. Using a GBP1 gBlock fragment instead of dGBP1 gave pBMN-dGBP1-GBP1-Cre or –Flpo.

pCAFNF-luc2 – An EcoRI-Kozak-luc2-NotI DNA fragment liberated from pCALNL-luc2 (Tang et al., 2015) was sub-cloned into EcoRI/NotI digested pCAFNF-DsRed vector, giving pCAFNF-luc2.

pCAG-dGBP1-TagBFP – Using PCR, an AgeI-Kozak-dGBP1-TagBFP-NotI was generated from pBMN-dGBP1-TagBFP. This fragment was sub-cloned into AgeI/NotI digested pCAG-GFP, giving pCAG-dGBP1-TagBFP and removing GFP from the construct.

pCAG-dGBP1-TagBFP-FLAG – A gBlock fragment encoding Kozak-TagBFP-FLAG was inserted into SphI/NotI digested pCAG-dGBP1-TagBFP via Gibson Assembly, giving pCAG-dGBP1-TagBFP-FLAG and removing untagged TagBFP from the construct.

pCAG-YFP-FLAG – A gBlock fragment encoding Kozak-YFP-FLAG was inserted into EcoRI/NotI digested pCAG-αCA-dNb^6mut-TagBFP, giving pCAG-YFP-FLAG and removing αCA dNb^6mut-TagBFP from the construct.

pCAG-dGBP1-mCherry – PCR amplified mCherry was inserted into a SphI/NotI digested pCAG-dGBP1-TagBFP vector, resulting in replacement of TagBFP with mCherry. The vector became pCAG-dGBP1-mCherry.

pCAG-GBP1-mCherry – A gBlock fragment encoding GBP1 was inserted into a EcoRI/SphI digested pCAG-dGBP1-mCherry vector, resulting in replacement of dGBP1 with GBP1. The vector became pCAG-GBP1-mCherry.

pCAG-αCA-Nb-TagBFP, pCAG-αDHFR-Nb-TagBFP, pCAG-αCA-dNb^6mut -TagBFP and pCAG-αCA-dNb^3maj-TagBFP –A gBlock fragment carrying either the αCA Nb or αDHFR Nb coding sequence was inserted into an EcoRI/SphI digested pCAG-TagBFP vector via Gibson Assembly, resulting in pCAG-αCA-Nb-TagBFP or

pCAG-αDHFR-Nb-TagBFP. gBlocks carrying these mutations in the respective Nbs were introduced into the EcoRI/SphI digested pCAG-TagBFP vector via Gibson Assembly, giving either pCAG-αCA-dNb^6mut-TagBFP or pCAG-αDHFR-dNb^3maj-TagBFP.

pCAG-dGC-Flpo and pCAG-dGD-Flpo – A gBlock fragment carrying either αCA-dNb^6mut or αDHFR-dNb^3maj coding sequence were inserted into SphI digested pCAG-dGBP1-Flpo vector (Tang, J.C.Y., Cepko lab) via Gibson Assembly, giving either pCAG-dGC-Flpo or pCAG-dGD-Flpo, respectively.

pCAG-dGBP1x2-Flpo – An AgeI-Kozak-dGBP1x2-Flpo-NotI fragment was generated by PCR using pBMN-dGBP1x2-Flpo as a template. This fragment was sub-cloned into AgeI/NotI-digested pCAG vector, giving pCAG-dGBP1x2-Flpo.

pCAG-αCA-dNb^6mutx2-Flpo – Two gBlock fragments, together encoding αCA-dNb^6mutx2, was inserted into EcoRI/SphI-digested pCAG-dGBP1x2-Flpo, giving pCAG-αCA-dNb^6mutx2-Flpo and replacing dGBP1x2 from the construct.

pCAG-αCA-dNb^6mut-TagRFP – A gBlock fragment carrying the TagRFP coding sequence was inserted into SphI/NotI digested pCAG-αCA-dNb^6mut-TagBFP via Gibson Assembly, giving pCAG-αCA-dNb^6mut-TagRFP and removing TagBFP from the construct.

pAAV-EF1α-dGBP1x2-Flpo-NW- A BamHI-Kozak-dGBP1x2-Flpo-EcoRI PCR fragment was inserted into BamHI/EcoRI digested pAAV-EF1α-N-CretrcintG (Tang et al., 2015), giving pAAV-EF1α-dGBP1x2-Flpo. The WPRE element was subsequently removed from this plasmid via EcoRV/AfeI digest and re-ligation, giving pAAV-EF1α-dGBP1x2-Flpo-NW.

pAAV-CAG-FLEX^FRT-ChR2(H134R)-mCherry- A Chr2(H134R)-mCherry DNA fragment was digested with NheI and inserted into NheI digested pAAV-CAG-FRTed-SynGFPreverse-WPRE (Pivetta et al., 2014) (a gift from Sylvia Arber) (Pivetta et al., 2014). A clone with ChR2(H134R)-mCherry inserted in the reverse direction relative to CAG promoter were chosen, giving pAAV-FLEX^FRT-ChR2(H134R)-mCherry.

pCAG-C-CA and pCAG-DHFR– A gBlock fragment carrying either the HIV-1 C-CA coding sequence (encoding residue 278–352 of HIV-1 gag polyprotein) or E.coli DHFR coding sequence was inserted into EcoRI/NotI digested pCAG-GFP vector via Gibson Assembly; C-CA or DHFR replaced GFP in the cassette.

For sample size, we chose to reproduce all our results in at least 3 independent experiments (equal to at least 3 biological replicates, or transfected wells). We consider this to be a sufficient sample size for demonstrating reproducibility of our findings. For statistical analysis, we chose to increase our independent experiments (equal to biological replicates) to 8.

293T cells were seeded onto 24 or 96 well plates and used for transfection when the cells reached between 60–95% confluency, usually 1–2 days later. Transfections were achieved with polyethyleneimine (PEI) at a 1:4 DNA mass:PEI volume ratio. PEI stock was 1 mg/ml. A total of between 100 and 400 ng of DNA were transfected into single wells of 96 well plates for fluorescence analysis of destabilized mutants. Approximately 70 ng total DNA were transfected into single wells of 96 well plates for luciferase analysis. Approximately 400 to 520 ng of total DNA were transfected into single wells of 24 well plates for fluorescence imaging and western blot analysis.

We exhaustively searched the Protein Data Bank (PDB) and, at the time of analysis, identified 77 unique camelid single-chain antibody fragments (VHH or VH, here collectively referred to as nanobodies (Nbs)) that have been co-crystallized with their respective antigens. We removed one structure (PDB ID 3J6A) from the analysis because it was a low-resolution structure produced by cryo-electron microscopy. We used the PDBePISA online server tool (http://www.ebi.ac.uk/pdbe/pisa/) to evaluate whether residue positions equivalent to those of dGBP1 mutations are located outside of antigen-Nb interfaces across different crystallized complexes. PDBePISA produces an analysis of buried surface area (BSA), defined as the solvent-accessible surface area of the corresponding residue that is buried upon interface formation, in Ų. We considered an Nb residue to be in the interface with the antigen if its BSA value is above 0 Ų. We confirmed that this metric is a reliable indicator of an Nb residue’s proximity to the antigen by examining all structures by eye using tools such as PyMol. In a few cases the Nb bound to more than one antigen. We took this into consideration by analyzing the interfaces formed between an Nb and each of the two antigens. We used protein alignment to determine the residue positions corresponding to the mutations located in dGBP1. The same 76 Nbs were used to determine the extent of GBP1 residue conservation across Nbs.

Analysis across 76 unique Nb-antigen interfaces, and a total of 102 uniquely crystallized interfaces, indicated that Nb positions corresponding to those of dGBP1 A25V, S73R, S98Y and S117F were universally located outside of all Nb-antigen interfaces (Figure 2—figure supplement 3B). Nb positions corresponding to dGBP1 Q109H fell outside of 99%, or 75 of 76 unique Nb-antigen interfaces. Positions equivalent to dGBP1 E63V were directly found in 22%, or 17 of 76 unique Nb-antigen interfaces, and in close proximity to the interface in 9%, or 7 of 76 of the cases.

16 identical or highly similar Nb-antigen complexes had been crystallized under similar or differing conditions, allowing us to validate the mutation mapping results by comparing across identical or related crystal structures. There was agreement in mutation mapping results between redundant or similar crystal structures for 94%, or 16 of 17 unique Nb-antigen interfaces. The lone exception concerned a unique Nb (PDB ID 4KRM and 4KRL). The Q109H equivalent position in the Nb of structure 4KRL was identified to be at the interface. This discrepancy may be explained by the fact that the two structures were crystallized under very different pH conditions.

For the mutation transfer experiment, we tested 18 Nbs that, when fused to TagBFP, showed strong blue fluorescence and soluble, diffuse localization in 293T cells. Some available nanobodies were too problematic to be included for analysis because they recognized problematic antigens such as the Ricin toxin. Nevertheless, Nbs that bound to proteins originating from both intracellular and extracellular locations were selected. We used a pCAG expression vector to express antigen deposited in PDB for each crystal structure. During analysis, we noticed a strong correlation between dNbs that failed to be stabilized by antigen (<2-fold TagBFP fluorescence induction by antigen) and the use of extracellular epitopes. dNbs targeting extracellular epitopes were thus excluded from evaluation of mutation transfer generality.

To determine whether the antigen used for TagBFP stabilization assays derived from intracellular or extracellular proteins, we studied the annotation and literature reports of each antigen’s cellular localization.

In all in vivo experiments, biological replicates are defined in terms of cells, retinas or animals. Technical replicates are defined in terms of whole brain sections. We consider 3 biological replicates to be a sufficient sample size for demonstrating reproducibility of our findings. As an exception, we had 2 biological replicate for injection of green fluorescent beads/rAAV mix into GFP and wildtype mouse brains (Figure 5—figure supplement 2). However, we deemed this sufficient as the results basically replicated our findings in an equivalent experiment using a slightly different injection mix (Figure 6). For statistical analysis, we used data that consisted of 7–21 cells.

rAAV (2/1) virus preparations were made from pAAV-EF1α-dGBP1x2-Flpo-NW and pAAV-CAG-FLEX^FRT-ChR2-mCherry. All rAAVs were injected in the range of 10¹³–10¹⁴ genome copies/ml, assayed by PCR of rAAV vectors at Boston Children’s Hospital (Zhigang He lab). Primers targeted the ITR region of AAV vectors, and were: Forward - 5’-GACCTTTGGTCGCCCGGCCT-3’, Reverse - 5’-GAGTTGGCCACTCCCTCTCTGC-3’. Note that we found this titering method gave about 10-100 fold higher numerical value in titer than other titering methods.

The human LoxP-LacZ cell line was obtained from Allele Biotech (San Diego, CA) (SKU: ABP-RP-CLACLOXE), and cultured as instructed in the product manual. Cas9 activity was assessed by detecting βgal-expressing cells in wells transfected with pX330-dCC-Cas9 and either pCAG-C-CA or pCAG-GAPDH-AU1 control construct (simply called AU1 in the main text). In addition, pCAG-mCherry is included as a transfection marker. For X-gal staining, cells were fixed on ice with 0.5% Glutaraldehyde for 5min. X-gal staining was performed as previously described. Cells were left at room temperature overnight for color development. Images were acquired by Keyence BZ9000 microscope. The number of mCherry⁺ and X-gal⁺ cells was quantified by Fiji software. The normalized Cas9 activity is calculated by dividing individual replicate values of specific conditions by the average number of X-gal+ cells induced by pX330-loxPgRNA alone.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank S Zhao and C Wang of the Z He laboratory (Boston Children’s Hospital) for rAAV production (core service supported by grant NEI 5P30EY012196-17). We thank M Springer, B Huang, M Lichterfeld, E Scully, A Tsibris, D Kuritzkes and members of the Cepko/Tabin/Dymecki lab for input on the manuscript, and the Neurobiology Imaging Facility (supported by NINDS P30 Core Center grant NS072030) for consultation and instrument availability. We thank WG Regehr for research support. We thank the Dana Farber Flow Cytometry facility for assistance with FACS. We thank S Arber and Addgene for plasmids. In addition to direct funding, SR was funded by the US National Institutes of Health (R01 NS32405 and R01 NS092707 to WG Regehr).

Animal experimentation: The Institutional Animal Care and Use Committee at Harvard Medical School approved all animal experiments conducted under protocols 428-R98, 04537 and 1493.

1. Received: February 17, 2016
2. Accepted: May 19, 2016
3. Accepted Manuscript published: May 20, 2016 (version 1)
4. Version of Record published: June 27, 2016 (version 2)

© 2016, Tang et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.