Combinations of Single Chain Variable Fragments From HIV Broadly Neutralizing Antibodies Demonstrate High Potency and Breadth

Broadly neutralizing antibodies (bNAbs) are currently being assessed in clinical trials for their ability to prevent HIV infection. Single chain variable fragments (scFv) of bNAbs have advantages over full antibodies as their smaller size permits improved diffusion into mucosal tissues and facilitates vector-driven gene expression. We have previously shown that scFv of bNAbs individually retain significant breadth and potency. Here we tested combinations of five scFv derived from bNAbs CAP256-VRC26.25 (V2-apex), PGT121 (N332-supersite), 3BNC117 (CD4bs), 8ANC195 (gp120-gp41 interface) and 10E8v4 (MPER). Either two or three scFv were combined in equimolar amounts and tested in the TZM-bl neutralization assay against a multiclade panel of 17 viruses. Experimental IC50 and IC80 data were compared to predicted neutralization titers based on single scFv titers using the Loewe additive and the Bliss-Hill model. Like full-sized antibodies, combinations of scFv showed significantly improved potency and breadth compared to single scFv. Combinations of two or three scFv generally followed an independent action model for breadth and potency with no significant synergy or antagonism observed overall although some exceptions were noted. The Loewe model underestimated potency for some dual and triple combinations while the Bliss-Hill model was better at predicting IC80 titers of triple combinations. Given this, we used the Bliss-Hill model to predict the coverage of scFv against a 45-virus panel at concentrations that correlated with protection in the AMP trials. Using IC80 titers and concentrations of 1μg/mL, there was 93% coverage for one dual scFv combination (3BNC117+10E8v4), and 96% coverage for two of the triple combinations (CAP256.25+3BNC117+10E8v4 and PGT121+3BNC117+10E8v4). Combinations of scFv, therefore, show significantly improved breadth and potency over individual scFv and given their size advantage, have potential for use in passive immunization.


INTRODUCTION
Broadly neutralizing antibodies (bNAbs), isolated from a subset of HIV-1 positive individuals, are capable of neutralizing a wide range of HIV viruses. Crucially, bNAbs have been shown to provide protection in non-human primate studies and it is thought that such antibodies are needed for an effective HIV vaccine (1)(2)(3)(4). However, to date, no candidate HIV vaccines have been able to elicit bNAbs in humans (5)(6)(7)(8). This has led the field to actively explore the possibility of using bNAbs as biological drugs for passive immunization against HIV (9)(10)(11)(12)(13).
The results of the first efficacy trials of an antibody for HIV prevention tested in Africa and the Americas have recently been published (14)(15)(16)(17). These two Antibody-Mediated Prevention (AMP www.ampstudy.org.za) trials showed that VRC01 had 75% prevention efficacy in high-risk men and women if the infecting virus was sensitive to the antibody at <1mg/ml (IC 80 ). Therefore, to target the extensive envelope diversity, minimize escape, and provide sufficient potency a combination of multiple antibodies will be needed. Several studies have investigated the potential of antibody combinations and observed, as expected, an increase in breadth and potency (18,19). These studies demonstrate that the complementary neutralization profiles of individual bNAbs can improve the overall breadth and provide higher coverage of multiclade panels of viruses at much lower antibody concentrations (18,20). By using those antibodies that specifically target the HIV subtypes predominant in a specific area, a geographically relevant set of antibodies may be selected to provide optimal coverage and potency (20). For example, CAP256.25, which shows high potency against clade C viruses, is currently being assessed in combination with PGT121 and VRC07-523LS in dual and triple combination in the South African CAPRISA 012B trial (21,22). Similarly, there are several ongoing phase 1 trials, testing multispecific antibodies or dual and triple combinations. These trials test the aforementioned antibodies in addition to V3 (10-1074) and V2 (PGDM1400) antibodies and the broadly neutralizing MPER-targeting antibody 10E8v4 (23,24).
The ability to accurately predict the breadth and potency of antibody combinations without experimental validation enables the rapid identification of optimal combinations. Two models have been used to predict the IC 50 and IC 80 of antibody combinations based on single antibody titers, the Loewe Additive model, and the Bliss-Hill Independence model. Both models assume that there is no interaction between the different antibodies and that neutralization by combinations of antibodies will be additive, however, owing to different formulations of independence, their predicted results differ (18,25). These models can also be used to determine whether synergy or antagonism occurs by comparing the predicted data with experimental results. When using the Loewe Additive model most combinations of two antibodies were demonstrated to show additive potency where the experimental potency was close to the predicted IC 50 (19). This model in comparison to experimental results has in cases also indicated synergy or antagonism between antibodies targeting specific epitopes such as the CD4 binding site, MPER, and V1/V2 antibodies. However, comprehensive analyses have shown that the Bliss-Hill model tended to be better at predicting IgG combination titers (18,19) and this model did not predict synergy between these epitopes when predictions were compared to experimental results.
Experimental synergy between anti-HIV antibodies has only been rarely observed, and only in the context of bispecific antibodies. A bispecific antibody that simultaneously engaged the V2 and V3 epitopes showed moderate levels of synergy (26). Another bispecific employing a CAP256.25 scFv and the antibody binding fragment (Fab) of 10-1074, showed moderate levels of improved potency against a few viruses (26). In this case, neutralization was compared to single scFv-Fc or IgG rather than to the predicted combination titers or experimental combinations of the two arms, which may have overestimated the level of synergy (20,(26)(27)(28)(29)(30). More convincing evidence of synergy was observed when antibodies targeting a host cell protein and the viral Env protein were combined particularly as part of bispecific or trispecific antibody constructs. This effect is due to the localization of the anti-HIV antibody close to the host cell membrane through CCR5 or CD4 binding, for example, the 10E8-iMab, which targets the MPER region on the HIV virion and the CD4 receptor on the HIV target cell. This bispecific antibody has a geometric mean IC 50 potency of 0.002mg/mL compared to 0.4mg/mL for 10E8 and 0.05mg/mL for the iMab indicating a 25-fold improvement over the expected activity (27,31).
Single chain variable fragments (scFv) are small molecules, which contain the variable heavy and light chain of antibodies connected through a glycine linker. These molecules have some enhanced pharmacokinetic properties such as improved distribution and absorption into mucosal tissues despite a loss of half-life due to lacking an Fc region (32)(33)(34)(35). They may also display less steric hindrance when used in combination with other molecules and other bNAbs or scFv (36). This may be especially true for epitopes in close proximity such as the V2 and V3, or V3 and CD4bs. A recent study demonstrated that scFv targeting the V3 and CD4bs could display synergy when combined with fusion inhibitors (37).
We previously demonstrated that scFv of bNAbs retain significant breadth and potency against a multiclade panel of viruses despite potency differences linked to differential affinity for the epitope (38). In particular, 10E8v4 maintained the same breadth and most of its potency as an scFv. Single antibodies may be limited in their ability to prevent HIV infection, however, as evidenced by the recent AMP results. We, therefore, tested combinations of scFv that target major bNAb epitopes on the HIV trimer, namely CAP256.25 (V2 apex), PGT121 (N332supersite), 3BNC117 (CD4bs), 8ANC195 (gp120-gp41), and 10E8v4 (MPER) and show that they generally follow a model of additive potency and complementary breadth. No significant antagonism or synergy was observed compared to the models, although antibody combinations tested against individual viruses could show variation. Overall, combinations of three scFv antibodies reached considerable breadth and potency indicating that scFv in combinations should be further investigated for passive immunity purposes.

METHODS scFv Construction
scFv were designed and cloned previously as described (38). In short, single constructs containing the variable heavy and light chain interspaced with a 15 or 18 amino acid glycine-serine linker of five HIV-directed bNAbs (CAP256.25, PGT121, 3BNC117, 8ANC195, and 10E8v4) were generated through overlapping PCR or ordered from GenScript (New Jersey, USA) (38). These scFv genes were then cloned into a CMV/R expression plasmid (AIDS Reagent Program, Division of AIDS, NIAID, NIH. For the lambda chain of PGT121, the pBR322 based lambda expression vector was used (AIDS Reagent Program, Division of AIDS, NIAID, NIH).

scFv Protein Expression
The constructs were grown in JM109 bacterial cells and extracted using a plasmid Maxiprep kit (Qiagen, Hilden Germany). Sequences were confirmed using the Applied Biosystems 3500xL Genetic Analyzer. Constructs were expressed as previously described (39). In short, HEK293F suspension cells at 1.5x10 6 to 2x10 6 cells/ml were cotransfected with linear Polyethylenimine hydrochloride (molecular weight, 40,000) at a 3:1 ratio with 1mg of plasmid per 1 ml of culture. Supernatants were harvested after 6 days. scFv proteins were purified using Ni-Sepharose beads (GE Healthcare, Massachusetts USA), washed using a 30mM imidazole-phosphate-buffered saline (PBS) solution and eluted using 400mM imidazole in PBS. Glycerol was added to the elution at a final concentration of 5% to limit aggregation. Eluates were applied to Hiload Superdex 75 or Superdex 200 columns (GE Healthcare) equilibrated with PBS at pH 6.5 (5% glycerol with 0.02% sodium azide). The fractions corresponding to the size of the scFv were collected, pooled, and concentrated using Vivaspin concentrators or Vivapore static concentrators (GE Healthcare). The samples were dialyzed overnight at room temperature to remove sodium azide. Concentrations were measured on a NanoDrop device (Thermofisher, MA, USA), with extinction coefficients at 1% calculated using Expasy ProtParam (40) and characterized by SDS-PAGE. Molar weight was determined by using Expasy ProtParam (CAP256.25: 31.35kDa, PGT121: 28.89kDa, 3BNC117 28.47kDa, 10E8v4: 29.30kDa and 8ANC195 29.00kDa). scFv proteins were stored at -75°C.

IgG Production
IgG constructs were expressed in HEK293F cells as described previously (39). Supernatants were harvested after 6 days and purified using a protein A affinity column. Proteins were eluted using a 0.15M glycine buffer at pH 2.5 buffer into 1M Tris, pH 8, and were concentrated and dialyzed into PBS pH 6.5 containing 5% glycerol. Concentrations were measured on a Nanodrop using an Extinction Coefficient of 13.7 at a 1% solution. The molecular weight of the IgG was calculated using the Expasy ProtParam (40) of CAP256.25 (150.71kDa), 10E8v4 (147.29kDa), 3BNC117: (146.24kDa), PGT121: (146.2kDa), 8ANC195: (147.43kDa). IgG and proteins were stored at -75°C.

Pseudovirus production for TZM-bl Assay
Plasmids containing HIV-1 envelope (gp160) genes cloned in the pcDNA ™ 3.1D/V5-His-TOPO ® vector were co-transfected with pSG3 Denv into HEK293T cells and cultured for 48-72hrs at 37°C. Supernatants were harvested and filtered through a 45mm filter and frozen at -80°C. Virus stocks were titrated on TZM-bl cells using a luciferase assay to determine a dilution yielding RLU at least 10-fold above the "cell only" background (40,000-100,000 RLU).

Neutralization Assay
A panel of 43 viruses (41) plus BG505 N332 and CAP256_SU (CAP256.3mo.9C) (42,43) representing HIV-1 clades A, B, and C was used to compare neutralization titers of IgG and scFv. Neutralization assays were performed in TZM-bl cells as described previously (44)(45)(46). Proteins were tested at 200mg/ mL for the IgG (~146kDa) and 50mg/mL for the scFv (28-32 kDa). All assays were repeated at least twice. IC 50 and IC 80 of each antibody tested was calculated and geometric mean potency was calculated for both IgG and scFv using sensitive viruses only.

Experimental Testing of scFv Combinations
Combinations were tested by adding equimolar amounts of two or three scFv proteins in a neutralization assay as described above. A panel of 17 subtype A, B and C viruses were selected based on their sensitivity to at least 2 of the scFv in order to test neutralization of scFv combinations and confirm the Loewe Additivity and Bliss-Hill Independence models. Pre-dilutions containing scFv at 2mM and 10mM each were used to facilitate the assay set up. As a control, the single scFv were diluted to 10mM and 2mM as well and run alongside the combinations as a comparison. The highest concentration tested for the combinations was 30mg/mL or 1000nM. IC 50 and IC 80 of each experimental antibody combination was calculated, where each antibody is present at the concentration determined at IC 50 or IC 80.

Loewe Additivity and Bliss-Hill Independence Models
The IC 50 and IC 80 values from experimental combinations were compared to the predicted IC 50 and IC 80 based on the Loewe Additive model and the Bliss-Hill model as previously described (18,19).
The following formula is used to calculate Loewe Additivity.
Post analysis values were recalculated from nM into mg/mL based on the following formula.
Where (n) is the number of antibodies, Mw the molecular weight of the antibodies in the combination, and IC 50  The combination neutralization curve is then calculated using the Bliss Independence model, being the individual functions of the scFv antibodies. Combination molar IC 50 and IC 80 titers are calculated by setting f = 0.5 or 0.8 and assuming each scFv is present at the same molarity, and converted to µg/ml using the above formula. Dual/triple coverage was calculated by considering a virus resistant if less than 2/3 antibodies in the combination were able to neutralize that virus at set concentrations.
We used the following formula for both models to determine which is more accurate in predicting the combination potencies for IC 50 and IC 80 .

Synergy and/or Antagonism Predictions Based on Loewe Additivity and Bliss-Hill Independence Models
Synergy was predicted based on whether the experimental IC 50 and IC 80 were improved compared to the Loewe Additive model or the Bliss-Hill model. The formula below was used to characterize this effect, with the Bliss-Hill IC 50 replacing the Loewe IC 50 in the formula below.
Positive values indicate experimental titers lower (i.e. more potent) than predicted and imply synergy, while negative values indicate less potent experimental titers than predicted and imply antagonism. For a given combination of antibodies, the mean and the 95% Confidence interval were calculated using LogFold values for this combination against each virus in the panel.

Statistics
All statistics were done using the GraphPad Prism 8 software.
Fold differences of the experimental combinations were calculated with the most potent scFv in the mixture for both IC 50 and IC 80 . A Wilcoxon signed-rank test was performed to test which model was more accurate at predicting the experimental titers for both IC 50 and IC 80 . Pearson's Correlation between experimental neutralization titers and titers based on the two models was calculated using the GraphPad software. A nonlinear fit model (log-log) was used to predict the slope between the experimental and predicted data, using a robust regression. A log-fold with a 95% confidence interval was used to determine if the fold difference is significantly different from 0, indicating either synergy or antagonism. Breadth-potency curves were drawn using a survival model in GraphPad. Significance was tested using the log-rank Mantel-Cox significance test.

Dual and Triple Combinations of scFv of HIV bNAbs
To assess whether scFv of HIV bNAbs showed increased breadth and potency when used in combination, we tested five different scFv as part of dual and triple combinations. The scFv included those targeting the V2 (CAP256.25), the N332 supersite (PGT121), the CD4bs (3BNC117), and the MPER region (10E8v4), all of which were previously shown to retain significant activity compared to IgG (38). For this study, we added the antibody 8ANC195 that targets the gp120-gp41 interface so that all five major epitopes on the HIV trimer were covered (see single IC 50 and IC 80 data in nM Supplementary 1A and 1B and in µg/ml 1C and 1D respectively, (38)]. All scFv were expressed and purified by size exclusion columns and, with size and purity confirmed SDS-PAGE gels. scFv were stored in buffers containing 5% glycerol as determined previously to prevent aggregation (38).
Combinations of scFv were tested against a panel of 17 pseudoviruses from subtypes A, B, and C (n=3, 4 and 10, respectively). These were selected based on their sensitivity to at least three of the five scFv under investigation. Equimolar amounts (based on their molecular weights), of each scFv in combinations of two or three were tested, giving a total of 20 different combinations of antibodies (10 dual and 10 triple combinations), which were compared to single scFv neutralization titers. To standardize the output data, the concentration of the single scFv and the scFv combinations in mg/mL were calculated from the nM titers and the molecular weight of the scFvs (  most potent single scFv in that combination. There was little to no loss of potency noted compared to the most potent scFv for the entire panel for all 20 scFv combinations tested (a total of 340 single test titers). A few instances of potential gain of potency compared to the most potent scFv were noted, with most of these linked to specific scFv combinations (bolded values in Figure 1).(See Supplementary Figure 2 for IC 50 titers in nM (A) and mg/mL (B) and for IC 80 values in nM (C) and mg/mL (D) respectively). The combination of CAP256.25+10E8v4 scFv stood out, demonstrating an overall 4.9-fold improvement in geometric mean titer ( Figure 1B). This was driven by five viruses of which two were subtype A, one was subtype B and two were subtype C indicating this was not subtype-specific. Five other dual scFv combinations showed improved potency over single scFv for single viruses but this was not linked to any specific combination of antibodies. The 3BNC117+10E8v4 combination showed an overall significant improvement in the geometric mean titer although this was not seen for individual viruses indicating complementarity of neutralization potency of the scFv rather than synergy.
For the triple combinations, five of the ten combinations showed >3-fold improved geometric means compared to the most potent single scFv with a 12-fold improvement for the CAP256.25+3BNC117+10E8v4 combination ( Figure 1C). This was higher than for the dual combinations where only two out of ten showed an improvement in geometric mean potency and was driven by improvement in potency against single viruses (16 of the 170 virus-scFv combination pairings). This was due to the scFv (e.g. CAP256.25 and 3BNC117 scFv), in the combinations being potent against different viruses allowing for complementarity in neutralization potency and coverage. Triple combinations overall showed higher improvements in breadth compared to dual combinations or single scFv as a consequence of having more antibodies.
Some viruses appeared to be more sensitive to the effects of combined scFv. For example, a slight potency improvement was observed against the CAP61 and ZM249 viruses for the dual combinations, although for the latter loss of potency was also noted (fold difference <0.33 see Figure 1B). Some enhancement in potency was also seen with these viruses plus Du156 and CAP88 for the triple combinations ( Figure 1C).
Overall, we found either improved or similar titers for combinations compared to the most potent single scFv, particularly for triple combinations, indicating that combinations of scFv improve the coverage of a panel of viruses and the potency at which viruses are neutralized. We overlaid neutralization curves of the combinations with the single scFv used in the combination, to determine if there were distinct patterns associated with improved IC 50 s or where we observed similar IC 50 for the combinations compared to the single scFv. For most dual and triple combinations (314/340), an additive effect was seen where the combination curve overlapped with the best scFv in the combination (Figure 2A). Here IC 50 did not show a potency improvement compared to the single scFv. A few combinations shifted the curve to the left compared to the curve of the most potent single scFv corresponding with fold potency improvements as seen in Figure 1 and indicating potential synergy (10/170 virus-dual combination pairings and 16/170 in the triple combinations showed improved IC 50 s) ( Figure 2B). In one case we observed decreased potency for the combination compared to the most potent scFv possibly indicating some antagonism ( Figure 2C). With the addition of a third antibody this was partially negated (Figure 2C right panel). Overall, most combinations did not show significant potency improvements or loss of potency, suggesting that they largely followed an additive model of potency.

scFv Follow an Additive Model of Interaction in HIV Neutralization
To further explore whether combinations of scFv follow an additive model of interaction we compared the experimental results with predictions based on the single scFv titers using the Loewe Additive model and the Bliss-Hill models. As mentioned before, both these models assume no synergistic or antagonistic interactions between antibodies (18,19). We calculated the combination scFv IC 50 and IC 80 titers based on these 2 models (Methods) and based on the geometric mean IC 50 and IC 80 of the repeats of single scFv obtained in the experiment. All scFv predicted data titers were compared to the IC 50 and IC 80 titers of the experimentally tested combinations.
There was a strong significant (p<0.0001) correlation between the experimental IC 50 and IC 80 and the predicted IC 50 and IC 80 of the dual combinations for both the Loewe and Bliss Hill model (Loewe, Pearson's r=0.94 and r=0.79 respectively, Bliss-Hill, r=0.94 and 0.91 respectively) ( Figures 3A, B). Similarly, for the triple combinations, there was a strong correlation between the predicted values and the experimental values (Loewe r=0.78 and r=0.79, Bliss-Hill: r=0.70 and r=0.95 respectively) ( Figures 3C, D). For each combination, we also determined which model was better by comparing the mean absolute log difference between the experimental values and the predicted values for each individual combination. For the dual combinations, the Bliss Hill model was better at predicting two out of the 10 combinations (10E8v4 combined with CAP256.25 or PGT121), whereas no significant difference between the models was observed for any of the other combinations ( Figures 3E, F). Similarly, for the triple combinations, most combinations did not show differences between the two models in the IC 50 with the Bliss-Hill model being better at predicting two combinations (8ANC195 + 10E8v4 with either CAP256.25 or PGT121) ( Figure 3G). For the IC 80 of the triple combinations, 6/10 combinations were significantly better predicted by the Bliss-Hill model compared to the Loewe model ( Figure 3H). This indicated that for most combinations both Loewe and Bliss-Hill could predict combination IC 50 titers well with a slight advantage for the Bliss-Hill model. However, the Bliss-Hill model is significantly better at predicting triple IC 80 titers compared to the Loewe model, with the latter underestimating the potency of the triple combinations at IC 80 . We next sought to determine if any of the combinations showed significant potency improvement or potency loss in either IC 50 or IC 80 by comparing the experimental data to the expected titers based on the Bliss Hill model. These analyses allowed us to determine if there is potential synergy (higher potency than predicted), as indicated by a log-fold difference >0.4 (~2.5 fold) and/or antagonism (lower potency than predicted) as indicated by a log-fold difference <-0.4 ( Figure 4). The dual combinations had a mean log-fold difference close to 0 (log-fold -0.4<CI<0.4) for both IC 50 and IC 80 indicating that there was neither antagonism nor synergy, although a few antibody combinations showed some improved potency (red dots, IC 50 n=3, 1.7% and IC 80 : n=0, 0%) or decreased potency (blue dots, IC 50 and IC 80 : n=6, 3.5%) for individual viruses these were relatively few and most showed less than 0.4 log fold difference when compared to model predictions ( Figures 4A, B). Similarly, only a few individual viral titers showed a potential loss of potency for the IC 50 (n=8, 4.7%) and IC 80 (n=11, 6.5%) or a potential gain of potency (IC 50 and IC 80 : n=5, 3%) for triple combinations compared to the model (Figures 4C, D). None of the outliers in dual or triple combinations could be linked to  specific viral signatures (data not shown). Notwithstanding these few outliers, we, did not observe any significant synergy nor antagonism for dual or triple combinations of scFv.

Using the Bliss-Hill Model to Predict IC 50 and IC 80 Titers of scFv Combinations Against a Larger HIV Panel
In order to assess their breadth-potency on a larger virus panel, we used the IC 50 and IC 80 data of single scFv to predict neutralization titers for combinations of two and three scFv, now including 8ANC195, on a 45-virus panel described in our previous study (38) (see Supplementary Figure 1 for single IC 50 in nM (A) and IC 80 in nM (B) or mg/mL (C, D) titers) (38). These theoretical scFv combinations were ranked based on their geometric mean potency using both IC 50 and IC 80 data for dual and triple combinations ( Figures 5A, B, E, F). Combinations with the expected highest breadth and potency were plotted using breadth-potency curves (Figures 5C, D, G, H). The best dual combinations, namely PGT121+10E8v4 (red), 3BNC117+10E8v4 (pink), and CAP256.25+10E8v4 (orange) were able to neutralize most viruses barring one at IC 50 and IC 80 (Figures 5A, B). The breadth-potency curves of the three 10E8v4-containing combinations reached 100% neutralization at the IC 50 and 98% neutralization at the IC 80 (colored breadthpotency plots versus grey, Figures 5C, D) with most viruses neutralized between 0.1 and 1mg/mL at IC 80 . The fourth combination of 10E8v4 with 8ANC195 was less effective due to the low potency of both antibodies and the limited breadth of 8ANC195, [see Supplementary Figure 3 (38)]. The other dual combinations showed reduced breadth (up to 30% at 50mg/mL) or reduced potency especially at IC 80 (grey dots and lines). Some combinations such as PGT121+3BNC117 had a potent geometric mean below 0.1mg/mL at IC 50 , but there was a larger spread of neutralization titers and so breadth was lost at IC 80 ( Figures 5A, B). (See Supplementary Figure 4 for IC 50 in nM (A) and mg/mL (B) and IC 80 in nM (C) and mg/mL (D) titers of dual, and triple combinations predicted by the Bliss-Hill Independence model).
The triple combinations showed higher levels of potency and complete neutralization with 8/10 combinations showing complete neutralization of the viral panel at IC 50 , and 6/10 showing nearcomplete neutralization at IC 80 ( Figures 5E-H). Moreover, 7/10 triple combinations had a geometric mean potency below 0.1mg/mL at IC 50 ( Figure 5E, green, dark green, and blue). Of these, three combinations all had a geometric mean IC 80 below 1mg/mL and neutralized up to   Figures 5F, H). This indicated that triple combinations are able to reach nearly 100% neutralization at lower concentrations for both IC 50 and IC 80 (see Supplementary Figure 4 for IC 50 in nM (A) and mg/mL (B) and IC 80 in nM (C) and mg/mL (D) titers of dual, and triple combinations predicted by the Bliss-Hill Independence model).
The predicted breadth of each combination was next determined at four concentrations: 0.1, 1, 10, and 50mg/mL in addition to the geometric mean for the combinations tested (with a cutoff at 50mg/mL). Combining these scFv improved the coverage at all concentrations for both IC 50 and IC 80 ( Figure 6). Breadth for dual combinations ranged from 84% to 100% at IC 50 < 50µg/ml and 80% to 100% at IC 80 < 50µg/ml. For dual combinations, 100% breadth was reached for one combination at IC 50 < 1mg/mL, and for four combinations at IC 50 < 50mg/mL. For the IC 80 , 98% breadth was obtained at 10mg/mL for the same combinations. Most combinations of three scFv had 100% breadth at IC 50 < 50 µg/ml, except for PGT121+3BNC117+8ANC195 and CAP256.25+PGT121 +8ANC195, which had a breadth of 96% and 93% respectively. Breadth was especially improved for lower concentrations at IC 80 . For example, breadth for single scFv ranged from 0 to 44% for IC 50 < 0.1µg/ml and 0 to 24% for IC 80 < 0.1µg/ml, whereas dual combinations ranged from 22 to 62% for IC 50 and 2 to 29% for IC 80. At the same cutoff, the breadth for combinations of 3 antibodies ranged from 47% to 82% for IC 50 and 16 to 40% for IC 80 . Geometric mean potency for the panel also improved from 0.2-4.95 mg/mL for single scFv to 0.27-2.08mg/mL for dual combinations and 0.09-0.61mg/mL for triple combinations at IC 80 . This was expected as the titers were calculated as containing the same amount of each scFv at that concentration.
Since PGT121, 3BNC117, and CAP256.25 all had good potencies as scFv, using these together with 10E8v4 resulted in improved potency of the combination. Moreover, scFv such as CAP256.25 and PGT121 and/or 3BNC117 have complementary profiles, with 3BNC117, for example, being especially potent against subtype B viruses and CAP256.25 being potent against subtype C viruses (see Supplementary Data 1 for the IC 50 and IC 80 of single scFv) (38). 10E8v4 is the broadest scFv and that resulted in the combinations containing 10E8v4 having the highest breadth. As a result, four of the five triple combinations with the best geometric mean titers included 10E8v4. Conversely, introducing less potent or less broad antibodies such as 8ANC195 resulted in less favorable antibody combinations, where 5 out of 6 combinations with the lowest potency and breadth contained 8ANC195, consistent with the relatively lower potency and breadth of this antibody. Of the top five, three combinations, CAP256.25+PGT121+10E8v4  1, 1, 10, and 50mg/mL was calculated from single scFv titers and from predicted scFv dual and triple combination titers for the IC 50 and IC 80 . Colors indicate percentage neutralization with 1-19% in green, 20-49% yellow, 50-79% yellow-orange, 80-89% orange, 90-99% red and 100% dark red. Geometric mean at 50mg/mL is also given. Each scFv is present in the combinations at the titers indicated. and CAP256.25+3BNC117+10E8v4 and CAP256.25+8ANC195 +10E8v4 neutralized all viruses below 1mg/mL and 98% of viruses at IC 80 below 10mg/mL. The best ranked PGT121 +3BNC117+10E8v4 combination neutralized 96% of viruses below 1mg/mL and 98% of viruses below 10mg/mL at the IC 80 value. This analysis indicated a significant gain of potency and neutralization breadth at lower concentrations than single scFv at both IC 50 and IC 80 . This indicated that combinations of scFv were able to increase the coverage at lower concentrations, which will be vital in clinical settings.

Predicting the Efficacy of scFv Combinations
The AMP trial showed that viral isolates that were sensitive to VRC01 with an IC 80 <1mg/mL were prevented from establishing infection in 75% of cases. As indicated in Figures 5 and 6, scFv combinations can reach significant coverage by a single antibody at this potency with the best scFv combination potentially reaching a 96% protection level. However, other antibody classes may need higher levels of potency to provide similar protection as VRC01 and viral isolates which are neutralized close to this cut-off may allow for breakthrough. Dual coverage will be able to overcome these issues by neutralizing breakthrough viruses at higher potencies and providing a backup in case antibody levels drop. Therefore, using this cut-off, we calculated the predicted dual coverage (i.e. by at least two scFv) of the combinations using the thresholds of single scFv IC 80 < 1µg/ ml and < 10mg/mL. We found for dual combinations, the dual active coverage at 1mg/mL ranged from 20% with improvement at 10mg/mL to 51% for PGT121+10E8v4 ( Figures 7A, B), This ranged from 0-20% at 1mg/mL and 9-62% for other combinations of two antibodies (See Supplementary Figure 5). This was somewhat improved for triple combinations where we observed up to 58-80% dual active coverage at an IC 80 of  10mg/mL. The best 3-combinations,3BNC117+PGT121+10E8v4 showing the highest active coverage at 10mg/mL of 80%. Generally, triple combinations showed higher levels of active coverage by 2 antibodies and often included both 10E8v4 and 3BNC117. Similar to the dual combinations, low dual coverage was observed at 1mg/mL up to 38% (See Figure 7A, Supplementary Figure 5. This could not be significantly improved for quadruple combinations which showed 44% and 87% dual coverage respectively at 1 and 10mg/mL. This loss of active coverage by multiple antibodies is also demonstrated in the breadth potency curves (Figures 7C, D). These curves clearly show that although at 10mg/mL most of the dual coverage is maintained there is a significant drop off at 1mg/mL. This was mainly due to antibodies such as 3BNC117, PGT121, and 10E8v4 showing only 31-49% breadth at 1mg/mL and the CAP256.25 scFv losing significant potency compared to the IgG (38). IgG combinations including this antibody could show slightly better active coverage by 2 antibodies, up to 53% at an IC 80 of 1mg/mL and up to 91% at 10mg/mL (Supplementary Data 6A for predictions of IgG combinations titers in mg/mL and 6B for neutralization breadth at 4 concentrations and dual coverage by IgG) (38). The other IgG and scFv showed similar levels of potency with specifically the 10E8v4 scFv retaining all of its potency explaining why the IgG combinations did not show higher levels of dual active coverage at 1mg/mL (Figure 7, Supplementary Figure 6B). This indicated that potency of single IgG and scFv needs to be improved to obtain higher levels of active coverage by 2 antibodies at lower antibody levels.
The combination data demonstrate that the benefits of combinations of IgG targeting different HIV epitopes apply also to the smaller and more versatile scFv fragments.

DISCUSSION
Combinations of scFv antibodies were able to enhance the breadth and potency of HIV-1 neutralization compared to a single scFv. Specifically, several triple combinations reached 100% coverage at IC 50 and IC 80 due to complementary neutralization profiles. Dual and triple combinations of scFv demonstrated broad coverage in the potency range that correlated with prevention in the AMP trials (IC 80 <1mg/mL) and would therefore be expected to be highly effective in combating HIV transmission in high incidence areas. Generally, dual scFv combinations followed an independent action model rather than a synergistic model of potency. The potency of the combination was usually similar to the potency of the most potent scFv in that mix. Although scFv may show less steric interference between antibodies this did not result in scFv showing significant potency improvements when used in combination. The close positioning of certain epitopes may prevent dual binding such as between V3 and CD4bs targeting scFv indicating that steric hindrance may have still impacted the scFv. Given the distinct MPER epitope located on the gp41 subunit instead of the gp120 subunit, we also expected independent effects for antibodies combined with 10E8v4, which was somewhat indicated in comparing experimental results with the most potent scFv. However, besides a few outliers, this was not apparent using the Bliss-Hill model. It is likely that binding of the MPER antibody confers conformational changes upon binding to its epitope precluding neutralization by other antibodies (47)(48)(49)(50). scFv may be especially advantageous in bispecific constructs as their smaller size allows for easier expression. Bispecific products will also increase the local concentration of the second antibody and with the appropriate linker may in such cases allow for synergy. Similarly, minor losses of potency compared to the most potent scFv were overall a rare occurrence and were negated with the addition of a third antibody (in a triple combination with e.g. 10E8v4). The few cases of antagonism noted for scFv combinations may also be due to conformational changes rather than steric hindrance, which would occur after antibody binding. To explore these aspects, scFv-trimer complex crystallization studies would have to be done.
The Loewe model tended to underestimate the potency at IC 80 particularly of triple combinations. It also often predicted low-level synergy for combinations containing the MPER antibody 10E8v4 and the V2 antibody CAP256.25. The Bliss-Hill model did not indicate significant antagonism and/or synergy for any combination at either IC 50 or IC 80 . We did observe a few individual outliers where for example the combination showed some improvement in IC 50 and IC 80 (CAP256.25+3BNC117+10E8v4) or similarly (3BNC117 +PGT121) showed loss of potency against individual viruses but no specific viral signature could be identified. Some of these may be limitations of the model itself as these tended to occur more often with viruses where the neutralization curves were not sigmoidal or with a low slope (18). The Bliss-Hill model tended to be better at predicting the IC 80 titers and both models were matched for the IC 50 titers. This finding is similar to IgG combinations where the Bliss-Hill model was also more accurate at predicting combination titers compared to the Loewe model, therefore the Bliss-Hill model was the preferred model for analyses of larger scFv datasets.
From a clinical perspective, the use of antibody combinations may mitigate the impact of suboptimal antibody levels.
Overlapping neutralization can enhance coverage against single viruses, which ensures that complete viral neutralization is achieved at lower concentrations of antibody. The selection of antibody combinations with high potency (IC 80 ) can more effectively counter declining antibody levels associated with passive administration. Although combinations of two scFv neutralized 100% of viruses at the highest concentration tested, the coverage was much reduced at lower concentrations, whereas triple antibody combinations demonstrated much higher coverage at lower concentrations by combining broad antibodies with very potent antibodies. For example, at an IC 50 of 1mg/mL, 100% of viruses were neutralized and at an IC 80 of 1mg/mL, 96% of viruses were neutralized. This demonstrates that complementary neutralization of the same virus is extremely advantageous, which was previously demonstrated for combinations of IgG (20).
Besides improved coverage, we also observed improved active coverage by 2 antibodies for triple combinations compared to dual combinations (19). In experimental combinations, neutralization slopes were also improved by combinations of antibodies, especially IC 80 titers and neutralization plateaus (less than 100% neutralization of a virus) were also less common with antibody combinations (data not shown) (18). It may also be beneficial to use antibodies or scFv that bind trimers in different states; for example, CAP256.25 binds early, whereas 10E8v4 likely binds trimers that are in an intermediate state allowing for complementarity in virus neutralization (51)(52)(53). Clinically, active coverage by multiple antibodies will prevent antibody escape when the serum titers drop to close to the neutralization range. The AMP trial demonstrated that if the VRC01 antibody levels dropped, protection against less sensitive viruses (IC 80 > 1mg/mL) was lost (17). This may differ for other antibodies or for antibodies targeting different epitopes and combinations may also provide dual active coverage of the viral quasispecies further improving this level of protection (54,55). This highlights the need to use more potent antibodies but also to improve individual antibody potency. In the future, bNAbs and scFv may also need to be assessed in PBMC neutralization assays which may more accurately resemble in vivo neutralization, and could more accurately predict potency (17,55). We observed that although overall coverage was improved, active coverage by multiple antibodies was low for both IgG and scFv using this set of antibodies. There are multiple efforts ongoing to improve antibody potency and bioavailability. Titers are likely to improve if newer scFv are developed which retain better potency compared to the IgG. Moreover, the selection of antibodies with optimal potency to cover geographically relevant viruses will ensure efficacy in the region.
For long-term antibody-mediated protection, adherence, and maintaining sufficient plasma levels in a wider population is critically important. At lower concentrations, scFv and IgG lose coverage especially in the ability to fully neutralize the virus (concentrations below IC 80 titers). This is also observed for IgG, indicating that all the antibodies in the combination must be kept at optimal levels. For IgG, one way to achieve this has been to introduce mutations in the Fc portion of the antibody to increase the half-life by up to 6 months (56,57). A second option is to introduce them into the body via a stable expression system through a vectored immunoprophylaxis (VIP) approach. Some bNAbs were recently expressed using AAV, showing modest levels in vivo (58, 59). However, due to the space constraints on vectors, constructs with multi-specific modalities would prove difficult. Bispecific products, which would partially overcome this, are also limited by space constraints. scFv in comparison with their much smaller size may be more readily expressed on AAV, allowing for the expression of multiple scFv continuously. scFv have a genetic size approximately 2.6-times smaller than IgG allowing for the expression of minimally three scFv per AAV, and improved expression levels (9,29,(60)(61)(62)(63)(64)(65)(66). AAV based scFv expression can help overcome the short half-life of scFv due to their lack of an Fc region. Some studies have shown success in the use of AAV based scFv for Alzheimer's disease, and Amyotrophic Lateral Sclerosis amongst others (61,67,68). In vivo data will be required to determine the antigenicity and bioavailability of scFv. However, scFv conjugated to short amino acid peptides or linked to targeting molecules were shown to have significant increases in bioavailability and stability (69)(70)(71)(72). These strategies mitigate some of the challenges facing scFv.
The scFv studied here demonstrated high breadth and potency when used in combination. Although only a limited number of scFv were tested on a small panel of viruses, we demonstrated their potential against a larger panel of viruses through modeling. By selecting bNAbs that target all five epitopes we also showed the potential for complementary effects. These scFv showed little to no antagonism. As more potent scFv are engineered combinations with even better efficacy may be found. In addition, as scFv are more readily expressed on vectors, this could in future provide, an alternative avenue for the use of bNAbs in passive immunization. The encouraging results of the AMP trials are expected to fast-track the need to consider long-term delivery approaches for those antibodies with clinical benefit.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding author.

ETHICS STATEMENT
Ethics approval was obtained as per local regulations (M160341). Antibody sequences were obtained from publicly available databases.

AUTHOR CONTRIBUTIONS
RD performed all experiments, data interpretation, figures and manuscript generation, and literature review. KW performed data analysis and advised on statistical analysis. PM and LM supervised the project and assisted with data interpretation and manuscript writing. All authors contributed to the article and approved the submitted version.