Cytosolic delivery of inhibitory antibodieswith cationic lipidsHejia Henry Wanga and Andrew Tsourkasb,1

aGraduate Group in Biochemistry and Molecular Biophysics, University of Pennsylvania, Philadelphia, PA 19104; and bDepartment of Bioengineering,University of Pennsylvania, Philadelphia, PA 19104

Edited by Richard A. Lerner, Scripps Research Institute, La Jolla, CA, and approved September 26, 2019 (received for review August 12, 2019)

Antibodies can be developed to directly inhibit almost any protein,but their inability to enter the cytosol limits inhibitory antibodiesto membrane-associated or extracellular targets. Developing acytosolic antibody delivery systemwould offer unique opportunitiesto directly inhibit and study intracellular protein function. Here wedemonstrate that IgG antibodies that are conjugated with anionicpolypeptides (ApPs) can be complexed with cationic lipids originallydesigned for nucleic acid delivery through electrostatic interactions,enabling close to 90% cytosolic delivery efficiencywith only 500 nM IgG.The ApP is fused to a small photoreactive antibody-binding domain(pAbBD) that can be site-specifically photocrosslinked to nearly alloff-the-shelf IgGs, enabling easy exchange of cargo IgGs. We showthat cytosolically delivered IgGs can inhibit the drug efflux pumpmultidrug resistance-associated protein 1 (MRP1) and the tran-scription factor NFκB. This work establishes an approach forusing existing antibody collections to modulate intracellular proteinfunction.

antibody | protein delivery | cytosolic | intracellular | penetrating

Antibodies have become important research tools becausethey can be developed to bind nearly any exposed proteinepitope with high affinity and specificity through either tradi-tional immunization or in vitro display approaches (1). By bindingto an appropriate epitope, antibodies can also directly inhibit theirantigen’s biological activity by either sterically blocking the antigenfrom binding to interaction partners or locking the antigen inan inactive conformation (1, 2). Indeed, microinjected antibodieshave not only shown that antibody-dependent inhibition of intra-cellular proteins is possible, but have also been used to uncoverthe biological roles of oncogenes (3), stress response proteins (4),and regulatory proteins (5, 6). Although physical delivery tech-niques such as microinjection (3–7) or electroporation (7, 8) arevery effective at cytosolic antibody delivery, they are low through-put and can result in significant toxicity, significantly limiting theutility of cytosolic antibodies as research tools.Unsurprisingly, many alternative approaches have been ex-

plored for cytosolic antibody delivery (9, 10). For example, anti-body fragments or antibody-like binding proteins can be engineeredfor cytosolic stability and expressed intracellularly as intrabodies(9, 10). Antibodies or antibody fragments can also be fused to orincubated with cell-penetrating peptides (CPPs) to induce theirendocytic uptake into cells followed by endosomal escape into thecytosol (11, 12). Finally, antibodies with an innate ability to enterthe cytosol have recently been developed in which the deliverymoiety lies within the light chain variable region (13).Carrier-mediated approaches for cytosolic protein delivery, in

which cargo proteins are encapsulated by delivery lipids orpolymers, have also been explored and capitalize on advances innonviral nucleic acid delivery formulations that have alreadybeen proven effective (14, 15). Although carrier-mediated cyto-solic antibody delivery has been reported multiple times (9, 10),those claims must be evaluated cautiously. A stringent assess-ment of commercially available carrier-mediated antibody de-livery platforms revealed that none were capable of cytosolicdelivery to >6% of cells (8). However, recent progress in delivering

the Cas9 protein, which is similar in size to IgG antibodies, withlipid nanoparticles for genome editing (16) suggests that carrier-mediated approaches are still viable strategies for cytosolicantibody delivery.Inspired by strategies for complexing proteins with cationic

lipids (16), polymers (17–19), and nanoparticles (20), we hypoth-esized that IgGs functionalized with anionic polypeptides (ApPs)could mimic the polyanionic nature of nucleic acids and becomplexed with cationic lipids designed for nucleic acid de-livery. Rather than engineering IgGs directly, we fused ApPs toa photoreactive antibody-binding domain (pAbBD) that couldbe photocrosslinked to each heavy chain of an IgG to createhighly negatively charged IgG-(pAbBD-ApP)2 conjugates withoutperturbing binding affinity (Fig. 1) (21). Because functionality isbuilt into the pAbBD rather than the IgG, cargo IgGs can beeasily exchanged without genetic reengineering, allowing most off-the-shelf IgGs to be easily functionalized.Here we report that our cytosolic antibody delivery approach

enables close to 90% delivery efficiency at a concentration of only500 nM IgG. Our modular antibody functionalization strategy iscompatible with IgGs from many different species and isotypesand our complexation approach is compatible with a diverse set ofcationic lipids. Finally, we demonstrate that cytosolically deliveredIgGs are functional and can inhibit not only the drug efflux pumpMRP1 to sensitize cancer cells to chemotherapeutic drugs, butalso the transcription factor NFκB. These findings establish thatour method enables easy and efficient cytosolic delivery of almost

Significance

Antibodies are important research tools because they canbe developed to bind to as well as directly inhibit almost anyprotein. Unfortunately, antibodies cannot cross the plasmamembrane and are therefore limited to perturbing membraneor secreted protein activity. We found that by appending anionicpolypeptides (ApPs) to immunoglobulin G (IgG) antibodies, theycould be complexed with cationic lipids, originally designed fornucleic acid delivery, through electrostatic interactions to enableefficient cytosolic antibody delivery. By fusing ApPs to photo-reactive antibody-binding domains, we can rapidly functionalizealmost any off-the-shelf IgGwithout genetic reengineering of thecargo IgG. Our cytosolically delivered antibodies are capable ofinhibiting intracellular proteins, establishing a new approach forstudying intracellular protein function with inhibitory antibodies.

Author contributions: H.H.W. and A.T. designed research; H.H.W. performed research;H.H.W. and A.T. analyzed data; and H.H.W. and A.T. wrote the paper.

Competing interest statement: H.H.W. and A.T. have a pending patent on this technol-ogy. A.T. is a founder and owns equity in AlphaThera, a biotechnology company that sellspAbBD-based products.

This article is a PNAS Direct Submission.

Published under the PNAS license.1To whom correspondence may be addressed. Email: atsourk@seas.upenn.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplemental.

First published October 14, 2019.

22132–22139 | PNAS | October 29, 2019 | vol. 116 | no. 44 www.pnas.org/cgi/doi/10.1073/pnas.1913973116

Dow

nloa

ded

by g

uest

on

June

3, 2

021

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1913973116&domain=pdfhttps://www.pnas.org/site/aboutpnas/licenses.xhtmlmailto:atsourk@seas.upenn.eduhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/cgi/doi/10.1073/pnas.1913973116

any off-the-shelf IgG antibody and can extend the potential ap-plications of existing IgG antibody collections.

ResultsStringent Detection of Cytosolic Protein Delivery. When developingintracellular protein delivery technologies, it is imperative to usean assay that only detects cargo proteins that have been success-fully delivered to the cytosol. Endosome escape is the majorbottleneck for cytosolic delivery, resulting in large false-positiverates for assays that measure total cellular uptake of cargo pro-teins (22, 23). Accordingly, we used a stringent self-assemblingsplitGFP (24) reporter system in which one half of the splitGFP,the S11 peptide, is fused to pAbBD-ApP, whereas the other half,splitGFP(1–10), is expressed in reporter cells (25–28). Only onceIgG-(pAbBD-ApP-S11)2 is successfully delivered into the cytosoldoes splitGFP complementation and turn-on fluorescence occur(Fig. 1). Furthermore, fluorescence intensity is directly correlatedwith the amount of protein that is cytosolically delivered.We prepared pAbBD-ApP-S11 with 10 to 30 residues-long

polyaspartate or polyglutamate ApPs and confirmed that theycould photocrosslink to rituximab (Ritux) (SI Appendix, Fig. S1 Aand B). We chose rituximab because its antigen, CD20, is notexpressed in our reporter cells, removing a potential confound-ing factor for delivery. To validate that our protein cargos arecompatible with the splitGFP reporter system, we confirmed thatsplitGFP complementation occurred when either pAbBD-S11 orRitux-(pAbBD-S11)2 were incubated with purified splitGFP(1–10)or physically delivered into HEK293T splitGFP(1–10) reportercells by electroporation (SI Appendix, Fig. S1 C–F). Due to thetime required for chromophore maturation (SI Appendix, Fig. S1C andD), reporter cells were assessed for splitGFP fluorescence 6 hfollowing electroporation with either flow cytometry or live-cellfluorescence microscopy. As expected, fluorescence microscopyrevealed diffuse splitGFP fluorescence with both pAbBD-S11 andRitux-(pAbBD-S11)2, but fluorescence was depleted from thenucleus only with Ritux-(pAbBD-S11)2 (SI Appendix, Fig. S1F).Because pAbBD-S11 is only ∼11 kDa, it is capable of passivelytranslocating across the nuclear pore complex, whereas Ritux-(pAbBD-S11)2 is ∼170 kDa and is very inefficient at passivenuclear translocation (29).

Cytosolic pAbBD Delivery. Initially, we wondered whether pAbBD-ApP-S11 alone, without IgG, could be cytosolically delivered intoHEK293T splitGFP(1–10) cells once complexed with a commercially

available cationic transfection lipid, Lipofectamine (Lipo) 2000.We tested simple polyaspartate or polyglutamate ApPs thatwere 10 to 30 residues long. Live-cell fluorescence micros-copy showed robust splitGFP fluorescence, indicating substantialcytosolic delivery, once polyaspartate ApPs were at least 15residues long (D15), with a peak at 20 aspartate residues (D20)(Fig. 2A). With polyglutamate ApPs, splitGFP fluorescenceincreased with length up to 30 glutamate residues (E30) (Fig.2B). Even though the base pAbBD-S11 has a net charge of −7,no cytosolic delivery could be detected without ApPs (Fig. 2 Aand B).Using flow cytometry, we quantified splitGFP fluorescence as

either the percentage of splitGFP-positive cells, which reflectsdelivery efficiency, or the fold increase in splitGFP fluorescence,which reflects the amount of protein delivered (Fig. 2 C and Dand SI Appendix, Fig. S2). Flow cytometry confirmed the trendsseen between ApP length and delivery efficiency identified viamicroscopy. Increasing the ratio of cationic lipid to proteingenerally improved delivery efficiency, but at the cost of in-creased toxicity (SI Appendix, Fig. S2). We were able to identifyregimes, though, where viability remained greater than 90% withexcellent delivery efficiency (Fig. 2 C and D and SI Appendix, Fig.S2). The best polyglutamate and polyaspartate ApPs were D20and E30 with delivery efficiencies of 53.9 ± 2.6% and 50.5 ±5.4%, respectively, when 500 nM pAbBD-ApP-S11 was complexedwith 2 μL Lipo 2000 (Fig. 2 C and D).Together, these results demonstrate that a small protein,

pAbBD-S11 (∼11 kDa), can be efficiently delivered into thecytosol simply by fusing it to polyaspartate or polyglutamateApPs via complexation with cationic lipids. We believe that thisstrategy can be easily adopted for cytosolic delivery of smallantibody-like binding proteins such as affibodies (∼6 kDa), mono-bodies (∼10 kDa), or nanobodies (∼15 kDa), but ApP lengthmay need to be reoptimized for maximal delivery.

Cytosolic IgG Delivery. Next, we tested whether Ritux-(pAbBD-ApP-S11)2 could also be cytosolically delivered when complexedwith Lipo 2000. As expected, no splitGFP fluorescence could bedetected by microscopy with Ritux-(pAbBD-S11)2, which has abase net charge of +4 (Fig. 3 A and B). Once both polyaspartateand polyglutamate ApPs reached at least 20 residues long,however, microscopy revealed diffuse splitGFP fluorescence withnuclear depletion (Fig. 3 A and B). Because nuclear depletionindicates that the S11 reporter peptide remained linked to Ritux-(pAbBD-ApP-S11)2 following endosomal escape, we are confi-dent that the splitGFP fluorescence is reflective of significantcytosolic delivery of Ritux-(pAbBD-ApP-S11)2.Flow cytometry of splitGFP fluorescence corresponded well to

microscopy, but additionally revealed that delivery peaked with25 aspartates (D25), but plateaued with 20 glutamates (E20)(Fig. 3 C and D and SI Appendix, Fig. S3). Lipo RNAiMax, acationic lipid designed for siRNA delivery, was more effectivethan Lipo 2000 at delivering Ritux-(pAbBD-ApP-S11)2 withshort ApPs, but worse with longer ApPs (SI Appendix, Fig. S4).Similarly to pAbBD-ApP-S11 delivery, increasing the ratio of ei-ther lipids to Ritux-(pAbBD-ApP-S11)2 generally improved de-livery, but also increased toxicity (SI Appendix, Figs. S3 and S4).After significant optimization, we achieved a maximal deliveryefficiency of 65.7 ± 3.6% in HEK293T splitGFP(1–10) cellswith >90% viability when 500 nM Ritux-(pAbBD-D25-S11)2 wascomplexed with 2 μL Lipo 2000 (Fig. 3C). Interestingly, deliveryefficiencies for Ritux-(pAbBD-ApP-S11)2 were higher than that ofpAbBD-ApP-S11 alone, likely due to having 2 ApPs linked toeach rituximab molecule.To demonstrate the generalizability of our IgG delivery approach,

we tested several other commercially available cationic lipids andreporter cell lines. All conditions resulted in efficient cytosolic de-livery of Ritux-(pAbBD-D25-S11)2 and Ritux-(pAbBD-E25-S11)2,

Fig. 1. Schematic of antibody delivery approach. (1) pAbBD is purified as afusion to an ApP as well as the splitGFP S11 reporter peptide and thenphotocrosslinked to a cargo IgG. (2) IgG-(pAbBD-ApP-S11)2 conjugates arecomplexed with cationic delivery lipids. (3) Lipid-IgG complexes are takenup by cells via endocytosis. (4) Delivery lipids promote endosome escapeand cytosolic IgG-(pAbBD-ApP-S11)2 delivery. (5) Cytosolic IgG-(pAbBD-ApP-S11)2 delivery can be detected by splitGFP complementation and turn-onfluorescence.

Wang and Tsourkas PNAS | October 29, 2019 | vol. 116 | no. 44 | 22133

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

June

3, 2

021

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplemental

albeit to varying degrees (Fig. 4 A–C and SI Appendix, Figs. S5 A–C,S6, and S7). Delivery efficiency was significantly higher in HT1080and A549 splitGFP(1–10) cells, reaching up to 72.7 ± 3.9%and 86.4 ± 1.1%, respectively, with 500 nM Ritux-(pAbBD-D25-S11)2 (Fig. 4B and SI Appendix, Fig. S7). Other than rituximab(human IgG, hIgG1), pAbBD-D25-S11 could also photocrosslinkand cytosolically deliver hIgG2, mouse IgG2a (mIgG2a), mIgG2b,mIgG3, rat IgG2c (rIgG2c), and rabbit IgG (rabIgG) into A549splitGFP(1–10) cells (Fig. 4 D and E). Delivery efficiencyremained high with all tested IgG species and isotypes,except for rIgG2c, which had a moderate delivery efficiency(Fig. 4D).Finally, we measured how delivery efficiency varied with Ritux-

(pAbBD-D25-S11)2 and Ritux-(pAbBD-E25-S11)2 dose (Fig. 4F–H and SI Appendix, Fig. S5 D–F). At 100 nM Ritux-(pAbBD-D25-S11)2, a 5-fold decrease in concentration, delivery efficiencywas only slightly reduced to 73.5 ± 3.0% in A549 splitGFP(1–10)cells with Lipo 2000 (Fig. 4G). Even at 1 nM Ritux-(pAbBD-D25-S11)2 or Ritux-(pAbBD-E25-S11)2, cytosolic delivery was stilldetectable, albeit low (Fig. 4 F–H and SI Appendix, Fig. S5 D–F).Collectively, these results demonstrate the versatility of our

IgG delivery approach with regards to the following factors: 1)We can deliver off-the-shelf IgGs from a variety of species andisotypes. 2) It is compatible with all tested cationic lipids and celllines thus far. 3) Delivery efficiency is maintained at low IgG-(pAbBD-ApP-S11)2 concentrations. We observe that although

increasing the net negative charge of cargo proteins is critical forcomplexation and delivery, it eventually becomes unproductiveor even deleterious for delivery efficiency.

Cytosolically Delivered QCRL3 Can Inhibit MRP1, a Drug Efflux Pump.Having shown robust delivery, we next sought to demonstrate theutility of cytosolic IgGs by inhibiting MRP1, a drug-export pumpassociated with chemotherapy resistance (30). Although it isa transmembrane protein, we chose MRP1 because it can beinhibited by a well-characterized monoclonal IgG, QCRL3, whichinhibits via binding to one of MRP1’s cytosolic nucleotide bindingdomains (30–33).Initially, we used the calcein-efflux assay to assess MRP1 ac-

tivity in which MRP1 inhibition results in calcein fluorescenceretention (Fig. 5A). A total of 500 nMQCRL3-(pAbBD-D25-S11)2delivery resulted in calcein retention in HEK293T, HT1080, andA549 cells (Fig. 5 B and C), indicating inhibition of endogenousMRP1 activity. No inhibition was observed following delivery ofthe mIgG2a-(pAbBD-D25-S11)2 isotype control or simply incu-bating cells with QCRL3-(pAbBD-D25-S11)2. QCRL3-(pAbBD-D25-S11)2 delivery performed as well as MK571, a nonselectivesmall-molecule MRP1 inhibitor (30), in all except for HEK293Tcells (Fig. 5C). We attribute this to the high expression of otherefflux pumps that are also inhibited by MK571 in HEK293Tcells. Finally, photocrosslinking is necessary for IgG delivery, asdelivery of QCRL3 mixed with pAbBD-D25-S11 did not result in

Fig. 2. Optimizing ApPs for cytosolic pAbBD delivery. A total of 500 nM pAbBD-S11 (negative control) or pAbBD-ApP-S11 with either polyaspartate orpolyglutamate ApPs 10 15, 20, 25, or 30 residues long were complexed with 2 μL Lipo 2000 and added to HEK293T splitGFP(1–10) cells for 6 h. (A and B)Representative live-cell fluorescence microscopy images following delivery with polyaspartate (A) and polyglutamate (B) ApPs shows diffuse splitGFP fluo-rescence indicating significant cytosolic delivery. (C and D) Flow cytometry of splitGFP fluorescence following delivery with polyaspartate (C) and poly-glutamate (D) ApPs. (Left) Representative flow cytometry histograms. (Center) Percent of cells splitGFP-positive. (Right) Fold increase in median splitGFPfluorescence over negative control. The dotted line indicates either 90% of the cell population (Center) or no increase in fluorescence (Right). Viability wasdetermined with the LDH assay. Data are mean ± SEM, n = 4; *P < 0.05, **P < 0.01 (1-sided 1-sample t test of log ratios).

22134 | www.pnas.org/cgi/doi/10.1073/pnas.1913973116 Wang and Tsourkas

Dow

nloa

ded

by g

uest

on

June

3, 2

021

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/cgi/doi/10.1073/pnas.1913973116

calcein retention (Fig. 5D). This suggests that during complexa-tion, noncovalent interactions between pAbBD and IgGs aredisrupted.Next, we attempted to sensitize A549 cells to doxorubicin and

vincristine, which are chemotherapeutic drugs known to be MRP1substrates (30, 34). Cytosolic delivery of 500 nMQCRL3-(pAbBD-D25-S11)2 was able to sensitize A549 cells to doxorubicin by 3.7 ±0.45 fold and vincristine by 9.0 ± 2.0 fold (Fig. 5 E and F). Incomparison, MK571 treatment resulted in only moderate sensiti-zation, and delivery of the mIgG2a isotype control resulted in nosensitization (Fig. 5 E and F). Thus, cytosolically delivered IgGsremain functional following delivery and can inhibit biologicallyinteresting proteins.

Cytosolically Delivered IgGs Can Inhibit NFκB. Finally, we investi-gated whether cytosolically delivered IgGs could inhibit protein–protein interactions, which are particularly difficult to perturbwith small molecules. We targeted the transcription factor NFκB,which is a heterodimer between p50 and RelA (p65) whose nu-clear localization signals (NLSs) are normally masked by IκBα,sequestering NFκB in the cytosol. With TNFα stimulation, IκBα isdegraded, allowing NFκB to enter the nucleus to stimulate tran-scription (35). We hypothesized that an anti-RelA NLS IgG couldsterically block RelA from engaging with its cognate nuclear im-port factor (NIF) to prevent RelA nuclear translocation and

NFκB-mediated transcription (Fig. 6A). We also tested an anti-RelA C terminus IgG that bound to an epitope distinct fromthe NLS.Initially, we delivered 150 nM anti-RelA NLS-(pAbBD-D25-

S11)2, anti-RelA C-term-(pAbBD-D25-S11)2, or their isotypecontrols (mIgG3 for anti-RelA NLS, rabIgG for anti-RelAC-term) into A549 cells and assessed for RelA nuclear translocationfollowing TNFα stimulation. Immunofluorescence revealed thatboth anti-RelA NLS and anti-RelA C-term delivery reduced RelAnuclear translocation to 48.0 ± 0.8% and 60.1 ± 5.9% of that ofnormal cells, respectively (Fig. 6 B and C and SI Appendix, Fig. S8).Next, by using a NFκB-driven luciferase reporter plasmid, weshowed that delivery of both anti-RelA NLS and anti-RelA C-termreduced NFκB transcriptional activity to 52.4 ± 1.1% and 68.3 ±2.6% of that of normal cells, respectively (100 ng/mL TNFα) (Fig.6D). This degree of inhibition is excellent considering the deliveryefficiencies with 150 nM anti-RelA NLS and anti-RelA C-termwere 70.9 ± 0.8% and 29.7 ± 4.5%, respectively (SI Appendix,Fig. S9). Because anti-RelA C-term IgG delivery was capable ofinhibiting NFκB, antibody binding to non-NLS epitopes may besufficient to sterically block nuclear translocation of many targetproteins. We anticipate that cytosolic IgG-dependent cytoplas-mic sequestration can be easily adopted to inhibit other tran-scription factors or nuclear proteins.

Fig. 3. Optimizing ApPs for cytosolic IgG delivery. A total of 500 nM Ritux-(pAbBD-S11)2 (negative control) or Ritux-(pAbBD-ApP-S11)2 with eitherpolyaspartate or polyglutamate ApPs 10, 15, 20, 25, or 30 residues long were complexed with 2 μL Lipo 2000 and added to HEK293T splitGFP(1–10) cellsfor 6 h. (A and B) Representative live-cell fluorescence microscopy images following delivery with polyaspartate (A) and polyglutamate (B) ApPs showsdiffuse splitGFP fluorescence with nuclear depletion indicating significant cytosolic delivery. (C and D) Flow cytometry of splitGFP fluorescence followingdelivery with polyaspartate (C ) and polyglutamate (D) ApPs. (Left) Representative flow cytometry histograms. (Center) Percent of cells splitGFP-positive.(Right) Fold increase in median splitGFP fluorescence over negative control. The dotted line indicates either 90% of the cell population (Center) or no increasein fluorescence (Right). Viability was determined with the LDH assay. Data are mean ± SEM, n = 4; *P < 0.05, **P < 0.01, ***P < 0.001 (1-sided 1-sample t test oflog ratios).

Wang and Tsourkas PNAS | October 29, 2019 | vol. 116 | no. 44 | 22135

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

June

3, 2

021

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplemental

DiscussionEfficient cytosolic delivery of proteins, particularly antibodies,has long been sought for expanding the toolbox that one can usefor perturbing biological systems. By leveraging a modular anti-body functionalization technology (21), we were able to easilyappend ApPs onto off-the-shelf IgGs to enable their complexa-tion with a variety of commercially available cationic lipids andefficiently deliver them into the cytosol of cells. Cytosolicallydelivered IgGs remain functional and are capable of inhibitingdiverse proteins such as MRP1 and NFκB.When compared to CPP-mediated delivery of small peptides

or proteins, our approach enables cytosolic delivery of a muchlarger IgG cargo with similar or greater efficiencies at an ∼100-fold lower concentration (25–28). It is difficult to directly com-pare our technology to previous carrier-mediated approachesdue to the use of different reporter systems. We note, however,

that previous studies have relied on either a reporter that onlydetects total cellular uptake (17) or by delivering proteins ca-pable of greatly amplifying their signal, such as Cre-recombinaseor enzymes (16, 18–20). In contrast, our functional assays—MRP1 and NFκB inhibition—are more stringent and requiredelivery of close to stoichiometric amounts of IgG relativeto their target, which is more representative of most proteininhibition assays.Intrabodies have long been used to modulate cell biology,

ranging from simply inhibiting target proteins (8–10, 36, 37) tomarking target proteins for degradation (38). However, thoseapproaches require either expertise in in vitro display technolo-gies or antibody engineering to create fragments that can foldproperly in the cytoplasmic environment, which can be majorbarriers for easy adoption by researchers. Because IgG anti-bodies have been invaluable research tools for decades, there

Fig. 4. IgG delivery scope. (A–C) A total of 500 nM Ritux-(pAbBD-S11)2 (negative control) or Ritux-(pAbBD-D25-S11)2 was complexed with 2 μL of the in-dicated cationic lipid and added to the indicated splitGFP(1–10) reporter cells for 6 h. Representative flow cytometry histograms of splitGFP fluorescence areshown in A. Flow cytometry data were quantified as the percent of cells splitGFP-positive (B) and the fold increase in median splitGFP fluorescence overnegative control (C). (D and E) Same as for B and C but 500 nM IgG-(pAbBD-D25-S11)2 of the indicated species and isotype was complexed with 2 μL Lipo 2000and added to A549 splitGFP(1–10) cells. (F–H) Same as for A–C, but indicated concentrations of Ritux-(pAbBD-D25-S11)2 were complexed with 2 μL Lipo 2000and added to A549 splitGFP(1–10) cells. The dotted line indicates either no increase in fluorescence (C, E, and H) or 90% of the cell population (G). Viabilitywas determined with the LDH assay. Data are mean ± SEM, n = 4; **P < 0.01, ***P < 0.001 (1-sided 1-sample t test of log ratios).

22136 | www.pnas.org/cgi/doi/10.1073/pnas.1913973116 Wang and Tsourkas

Dow

nloa

ded

by g

uest

on

June

3, 2

021

https://www.pnas.org/cgi/doi/10.1073/pnas.1913973116

exists large and well-validated antibody collections (39, 40) that,using our modular IgG functionalization and cytosolic deliveryapproach, can now be repurposed for perturbing the activity ofintracellular proteins. Although IgG binding is not guaranteed toinhibit all target proteins, we believe that the large size of IgGs issufficient to sterically block many biological interactions.Cytosolic delivery of inhibitory antibodies also offers unique

advantages over traditional genetic or small-molecule approachesfor modulating protein function. Because genetic approaches forprotein knockdown or knockout do not directly act on targetproteins, they perform poorly against proteins with long half-lives(41) and can induce significant cellular compensatory responses(42). Small-molecule inhibitors and modulators can avoid theselimitations, but many proteins are not druggable by small mole-cules and for those that are, identifying potent compounds andthen validating their selectivity can be challenging (43). In con-trast, inhibitory antibodies directly bind to target proteins, act onfast time scales, and can be generated far more easily than smallmolecules. Finally, cytosolic inhibitory antibodies also offer theopportunity to discriminate between and modulate the activity of

proteins with specific posttranslational modifications or certainisoforms of a protein, which are generally not possible withtraditional approaches.Cytosolic IgGs have previously been reported to be capable of

engaging the TRIM21 E3 ubiquitin ligase to degrade their targetproteins (7), but pAbBD photocrosslinking sterically blocks theTRIM21 binding site (21, 44, 45) and prevents our cytosolicallydelivered IgGs from harnessing any endogenous protein degra-dation machinery. Future studies could address this by engineeringalternative antibody-binding domains (46) to photocrosslink toIgGs outside of the Fc region. In this study we complexed IgGswith commercially available cationic lipids. Although the resultingcomplexes perform efficiently in cultured cells, their poor phar-macokinetics render them unsuitable for in vivo studies. Futurestudies should test alternative delivery formulations for in vivodelivery, such as those containing ionizable and poly(ethyleneglycol)-conjugated lipids (14), which have shown promise for thein vivo delivery of siRNA. However, it is not clear whether theseformulations, when prepared with antibodies, will have adequatepharmacokinetics, be capable of extravasating from the vasculature

Fig. 5. Cytosolic QCRL3 delivery inhibits MRP1. (A) Schematic of assays that assess MRP1 inhibition. In the calcein efflux assay, cells are first loaded withcalcein, a fluorescent membrane-impermeable MRP1 substrate. Cells with high MRP1 activity will rapidly export calcein, whereas MRP1 inhibition results incalcein fluorescence retention. In the chemotherapeutic sensitization assay, MRP1 inhibition results in greater intracellular accumulation of doxorubicin orvincristine, resulting in sensitization to both compounds. (B) Representative flow cytometry histograms of calcein fluorescence after 16 h of export in calcein-loaded HT1080 cells treated with 20 μM MK571, 500 nM QCRL3-(pAbBD-D25-S11)2, 500 nM cytosolically delivered mIgG2a-(pAbBD-D25-S11)2, or 500 nMcytosolically delivered QCRL3-(pAbBD-D25-S11)2. (C) Calcein-efflux assay quantification across HEK293T, HT1080, and A549 cell lines. Only QCRL3 delivery andMK571 treatment resulted in calcein fluorescence retention. Data are mean ± SEM, n = 4, ***P < 0.001 (1-sided 1-sample t test of log ratios). (D) Same as for B,but in calcein-loaded A549 cells treated with cytosolic delivery of 500 nM QCRL3 with or without photocrosslinking to pAbBD-D25-S11. Calcein fluorescenceretention is only seen with photocrosslinked QCRL3, indicating that photocrosslinking is necessary for delivery. (E and F) A549 cell sensitivity to doxorubicin (E)or vincristine (F) following treatment with 20 μM MK571, 500 nM cytosolically delivered mIgG2a-(pAbBD-D25-S11)2, or 500 nM cytosolically delivered QCRL3-(pAbBD-D25-S11)2. Data are mean ± SEM, n = 4.

Wang and Tsourkas PNAS | October 29, 2019 | vol. 116 | no. 44 | 22137

APP

LIED

BIOLO

GICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

June

3, 2

021

to induce uptake into desired cell populations, or be capable ofpromoting escape from the endosome–lysosome system with ade-quate efficiency to impart a therapeutic effect in vivo.In summary, we have designed and rigorously validated a

modular cytosolic IgG delivery platform that allows previouslydeveloped IgG collections to now be used as intracellular tools.We believe that this capability will significantly expand the waysin which intracellular pathways can be perturbed and will shedunique insight into cell function in both health and disease.

MethodspAbBD Expression and Purification. All pAbBD variants were purified viaproximity-based sortase-mediated ligation (PBSL) (47). Using PBSL to purifyproteins consists of 2 steps: 1) PBSL resin preparation and 2) target proteinpurification. PBSL resin was prepared as previously described with 2 minormodifications due to the use of SpyCatcher-SrtA-His12 (47). Pellets werelysed by resuspending in lysis buffer (PBS + 1% wt/vol N-octyl-β-D-1-thio-glucopyranoside [OTG] + 200 μg/mL lysozyme + 4 μg/mL DNaseI + EDTA-freeprotease inhibitor mixture [Roche]) and rotating for 30 min at room tem-perature (RT). Following binding, the resin was washed with PBS + 20 mMimidazole 3 times followed by PBS once.

Plasmids encoding for various pAbBD variants were transformed in con-junction with pEVOL-pBpF into T7 express competent Escherichia coli cells(New England Biolabs). Starter cultures were grown in LB + 100 μg/mL am-picillin (amp) + 25 μg/mL chloramphenicol (cam) at 37 °C with shaking untilOD600 ∼ 0.6. The starter culture was added at a 1:1,000 dilution to auto-induction media (Formedium AIMLB0210 autoinduction media LB brothbase including trace elements supplemented with 0.6% vol/vol glycerol and

100 μg/mL amp) further supplemented with 25 μg/mL cam + 0.1% wt/volarabinose + 3.33 μM 4-benzoyl-L-phenylalanine (BPA, Bachem). All pAbBDvariants were grown at 37 °C with shaking for 24 h, except for pAbBD-D30-S11and pAbBD-E30-S11, which were grown at 25 °C with shaking for 48 h. Ex-pression cultures were then pelleted and stored at −20 °C.

Frozen pellets were lysed by resuspending in lysis buffer for 30 min at RT.Afterward, lysates were frozen at −80 °C and then thawed in a 37 °C waterbath. The lysates were clarified by centrifuging for 15 min at ≥14,000 × gand discarding the pellet. Clarified lysates were incubated with theSpyCatcher-SrtA-His12 resin prepared above while rotating for 25 min at RT.Following binding, the resin was transferred to a Poly-Prep chromatographycolumn (Bio-Rad) and washed with 1 column volume (CV) of PBS, 1 CV of PBS +20 mM imidazole, and 1 CV of PBS + 1 M NaCl + 20 mM imidazole. pAbBDvariants were then eluted from the resin by adding PBS + 250 μMCaCl2 + 2mMGly-Gly-Gly (triglycine) and incubating at 25 °C for 3 h. Following elution,pAbBD variants were buffer exchanged into PBS and concentrated to≥0.5mg/mLvia a 10k MWCO Amicon Ultra centrifugal filter (MilliporeSigma). The finalprotein was analyzed by SDS/PAGE for purity, tested for splitGFP comple-mentation, stored at −80 °C, and tolerated freeze–thaw cycles well.

See SI Appendix, Supplementary Methods, for details on plasmid gener-ation, splitGFP(1–10) purification, and splitGFP complementation assays.

Photocrosslinking pAbBD Variants to IgGs. For photocrosslinking, pAbBDvariants were added to IgGs at a 2:1 molar ratio in PBS. IgG concentration waskept at ≤5 μM and the pAbBD-IgG uncrosslinkedmixture was aliquoted in 2 mLclear polypropylene microcentrifuge tubes. The mixture was then placedin an ice bath and irradiated for 3 h with 365 nm UV light using a UVP CL-1000L UV crosslinker placed in a 4 °C cold room. After photocrosslinking,IgG-pAbBD2 conjugates were washed with PBS 3 times and then concentrated

Fig. 6. Cytosolic anti-RelA IgG delivery inhibits NFκB. (A) Schematic of NFκB inhibition. Anti-RelA IgGs inhibit NFκB transcriptional activity by preventing itsnuclear translocation following TNFα stimulation. (B and C) Representative immunofluorescence images (B) and quantification (C) of RelA nuclear trans-location following delivery of the indicated 150 nM IgG-(pAbBD-D25-S11)2 antibody and TNFα treatment. Only delivery of anti-RelA IgGs reduced RelAnuclear translocation. Data are mean ± SEM, n = 3, ***P < 0.001 (1-way ANOVA). (D) A549 cells were transiently transfected with a NFκB-driven fireflyluciferase reporter plasmid. NFκB transcriptional activity was detected by luminescence following delivery of the indicated 150 nM IgG-(pAbBD-D25-S11)2antibody and TNFα treatment. Only delivery of anti-RelA IgGs inhibited NFκB transcriptional activity. Data are mean ± SEM, n = 3; *P < 0.05, **P < 0.01,***P < 0.001 (1-way ANOVA).

22138 | www.pnas.org/cgi/doi/10.1073/pnas.1913973116 Wang and Tsourkas

Dow

nloa

ded

by g

uest

on

June

3, 2

021

https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1913973116/-/DCSupplementalhttps://www.pnas.org/cgi/doi/10.1073/pnas.1913973116

to ≥10 μM via a 100k MWCO Amicon Ultra centrifugal filter to remove anyuncrosslinked pAbBD. SDS/PAGE was used to confirm that >95% of IgGheavy chains were photocrosslinked and that any excess pAbBD was re-moved. The final protein was then tested for splitGFP complementation andstored at 4 °C for short durations (