High-valency anti-CD99 antibodies toward the treatment of T cell acute lymphoblastic leukemia

Abstract

T-cell acute lymphoblastic leukemia (T-ALL) is an aggressive form of leukemia that currently requires intensive chemotherapy. While childhood T-ALL is associated with high cure rates, adult T-ALL is not, and both are associated with significant short- and long-term morbidities. Thus, less toxic and effective strategies to treat T-ALL are needed. CD99 is overexpressed on T-ALL blasts at diagnosis and at relapse. Although targeting CD99 with cytotoxic antibodies has been proposed, the molecular features required for their activity are undefined. We identified human antibodies that selectively bound to the extracellular domain human CD99 and the most potent clone, 10A1, shared an epitope with a previously described cytotoxic IgM antibody. We engineered clone 10A1 in bivalent, trivalent, tetravalent, and dodecavalent formats. Increasing the antibody valency beyond two had no effects on binding to T-ALL cells. In contrast, a valency of ≥3 was required for cytotoxicity, suggesting a mechanism of action in which an antibody clusters ≥3 CD99 molecules to induce cytotoxicity. We developed a human IgG-based tetravalent version of 10A1 that exhibited cytotoxic activity to T-ALL cells but not to healthy peripheral blood cells. The crystal structure of the 10A1 Fab in complex with a CD99 fragment revealed that the antibody primarily recognizes a proline-rich motif (PRM) of CD99 in a manner reminiscent of SH3-PRM interactions. This work further validates CD99 as a promising therapeutic target in T-ALL and defines a pathway toward the development of a selective therapy against T-ALL.

Graphical Abstract

Introduction

T-cell acute lymphoblastic leukemia (T-ALL) is a devastating disease in both children and adults, and current therapies come with a high risk of treatment related toxicity ^{1; 2}. T-ALL originates in T-lymphocyte progenitors and is characterized by the accumulation of leukemia cells in the bone marrow and reduced capacity to produce mature hematopoietic cells ³. Current therapies for T-ALL include chemotherapy, radiation, targeted therapy, and stem cell transplant ³. Current treatment options have dramatically improved patients’ prognoses, with pediatric T-ALL patients showing a 75% overall survival rate ⁴. However, about 20% of these patients relapse ⁵. Adult patients show poorer response to T-ALL therapies, with a complete remission rate of about 30% for primary relapsed or refractory T-ALL ^{6; 7; 8}. T-ALL relapse can be attributed to mutations that activate different oncogenic pathways, which in turn cause patients to have diminished responses to therapies ^{9; 10; 11}. New treatment options that are safer and more effective in all T-ALL patients, regardless of the underlying driving mutations, are necessary to overcome this disease.

CD99 has been proposed as a therapeutic target in T-ALL because it is overexpressed in leukemia cells relative to normal cells in nearly all T-ALLs ^{12; 13; 14; 15} and is overexpressed at both diagnosis and relapse ¹⁶. The CD99 antigen is highly expressed on the surface of T-ALL cells with cell-surface CD99 levels about seven times higher in T-ALL blasts compared with normal T cells, and thus this protein can be used as a marker to assess minimal residual disease ¹⁷. CD99 is also overexpressed in other leukemia types, including acute myeloid leukemia (AML) ¹⁸ and Ewing sarcoma ¹⁹. It is expressed at significantly lower levels on the surface of normal cells including endothelial cells in the blood vessels (19), suggesting a possible therapeutic window for anti-CD99 directed leukemia and cancer therapies.

Previous studies demonstrated that a mouse anti-CD99 IgM antibody has a cytotoxic effect both in vitro and in mouse models of AML ¹⁸, as well as in vitro models of T-ALL ¹⁵. Although these studies demonstrated that CD99 is a promising target in T-ALL, it was not clear whether specific molecular features of the mouse IgM antibody were essential for its cytotoxic effect. Given the challenges associated with producing IgM antibodies (NCT01123304) and the necessity of humanization, one might wish to generate therapeutic antibodies in a format that has been successfully tested in a clinical setting, such as IgG and its derivatives.

Here, we report the development of a human anti-CD99 antibody and the demonstration that the cytotoxic effect requires an antibody valency of three or more. Based on this mechanistic insight, we developed an IgG-based human antibody with a valency of four that shows cytotoxic efficacy selective to T-ALL cells. The crystal structure of an antibody-antigen complex rationalizes the antibody specificity and provides a guide for further development.

Results

Development of CD99-binding human antibodies

The CD99 extracellular domain (ECD) is rich in glycine, serine, and proline residues (Fig. S1A) and predicted to be disordered ^{20; 21; 22}. We produced the CD99 ECD as SUMO- and GFP- fusion proteins to facilitate production and purification. We screened a human naïve antibody library in the single-chain variable fragment (scFv) format using yeast surface display technology ^{23; 24}. Because antibodies to a disordered antigen tend to have low affinity, we performed library sorting using tetramerized CD99 ECD by coupling the biotinylated antigen to streptavidin in order to enhance the binding by the avidity effect (Fig. S1B) ^{24; 25}. We identified three clones, named 10, 22, and 30.

To facilitate rapid characterization of these antibody clones, we produced them in a scFv-Fc format that mimics the bivalent IgG architecture, but does not require the production of separate heavy and light chains (Fig. S2A) ^{26; 27; 28; 29}. These antibodies bound the immobilized CD99 ECD with apparent K_D values ≤ 24 nM Fv (Fig. 1A); K_D values are expressed in terms of nM Fv to minimize ambiguity across different antibody formats throughout this paper. To assess the specificity of the antibodies, we examined their cross-reactivity to CD99 homologs. CD99 is a member of the CD99 family and has two homologs, CD99L2 and Glycoprotein Xg, with their ECDs showing relatively low homology, with 35% and 34% sequence identity to that of CD99, respectively ^{29; 30}. The three antibodies bound to the CD99 ECD, but not to those of CD99L2 or Glycoprotein Xg (Fig. 1B). We next assessed cross-reactivity to CD99 orthologs. Mouse and monkey CD99 exhibit 44% and 90% sequence identity to human ECD of CD99, respectively. Clone 10 cross-reacted with mouse and monkey CD99, but clones 22 and 30 did not, suggesting the epitopes of the latter antibodies are in regions that substantially differ between human and monkey CD99 (Fig. 1C, S1F).

Figure 1.

Generation of human anti-CD99 antibodies. (A) Binding titrations of three antibody clones in the scFv-Fc format to purified SUMO-CD99 fusion protein and SUMO protein (a negative control) immobilized on beads. The apparent K_D values are from curve fitting of a 1:1 binding model. Values indicated are the means and standard deviations from curve fitting (n=3). K_D values are expressed in terms of nM Fv to minimize ambiguity across different antibody formats. (B and C) Binding of the antibodies to human CD99 and close homologs in human using 600 nM Fv of scFv-Fc (B) and to CD99 from different species using 1 μM Fv of scFv-Fc (C). n=3. (D-E) Flow cytometry analysis of antibody clones following staining of a T-ALL cell line that expresses CD99 (Jurkat, left), and ExpiCHO cells that do not express CD99 (right). (D) Flow cytometry etection of CD99 on Jurkat and ExpiCHO cells using a commercial mouse anti-CD99 antibody, HO36-1.1. The black dashed lines mark thresholds defined with the data with only the secondary antibody. ExpiCHO cells serve as a negative control. (E) Flow cytometry detection of CD99 on cells using clones 10, 22, and 30. The black dashed lines mark thresholds defined with detection using mouse Fc, equivalent to an isotype control. (F) Binding titration curves of clone 10 (top) and affinity matured clone 10A1 (bottom) in the Fab format to purified SUMO-CD99 fusion protein immobilized on beads (n=3). (G) Binding titration curves of clone 10 (top) and 10A1 (bottom) in the scFv-Fc format to endogenously expressed CD99 on Jurkat cells (n=3). The data for nonbinding mouse Fc are shown as open circles. The 10A1 antibodies are shown as filled circles. The apparent K_D values are from curve fitting of a 1:1 binding model. Values indicated are the means and standard deviations (n=3).

CD99 is a 32 kDa O-linked glycosylated transmembrane protein, with almost half the mass, 14 kDa, corresponding to the sugar molecules ³¹. To assess if the antibodies bind to CD99 with posttranslational modifications, we tested their binding to CD99 on cells. First, we confirmed CD99 expression on Jurkat cells using a commercially available anti-CD99 IgM antibody, clone HO36-1.1 (Fig. 1D). All three antibodies detectably bound to Jurkat cells, and clone 10 bound more efficiently than clones 22 and 30 (Fig. 1E). In contrast, none bound to CHO cells, hamster cells that do not express human CD99 (Fig. 1D, E).

Because the ECD of CD99 is predicted to be disordered, we predicted that the antibodies bind to a linear epitope. To test this hypothesis, we performed Western blots of recombinant CD99 with the antibodies. Each of the antibodies indeed produced strong bands in Western blotting (Fig. 2, S4). To further delineate the epitopes, we produced truncated forms of the CD99 ECD, and tested antibody binding by Western blotting. Clone 10 bound to four different fragments, allowing us to deduce that its epitope encompassed amino acid residues 63 to 76 (Fig. 2A-B). Met76 is one of three residues within the epitope (residues 63-76) that differs between human and rhesus CD99 (Table S1), suggesting that residue 76 is important in the interaction between clone 10 and CD99. Similar analysis determined that clone 22 bound to residues between 33 and 52 (Fig. 2C) and clone 30 bound to residues between 43 and 52 (Fig. 2C). The overlapping epitopes for clones 22 and 30 include Thr41, a known O-glycosylation site ³², whereas the epitope for clone 10 includes no known or predicted post-translational modification sites. These results and the fact that we used aglycosylated antigens for antibody discovery suggest that glycosylation at Thr41 weakened binding of clones 22 and 30 to Jurkat cells. We also determined that the epitope of HO36-1.1, a mouse IgM antibody with cytotoxic function in T-ALL and AML ^{15; 18}, overlaps with that for clone 10 (Fig. 2A-B). Ultimately, clone 10 was chosen for further characterization, because it exhibited the strongest binding to endogenously expressed CD99 and it binds to an epitope that overlaps with that of the cytotoxic IgM antibody.

Figure 2.

Epitope mapping of human anti-CD99 antibodies using truncated forms of CD99 and Western blot detection. (A) Scheme depicting truncated CD99 proteins used and deduced epitopes. Black bars: CD99 fragments that were not bound by any of the antibodies; Red bars: fragments bound by clones 22 and 30; Green bar: a fragment bound by clones 22 and 30 and at a lower degree to HO36-1.1 and clone 10; Purple bars: fragments bound by clones 10, 22, and 30 and HO36-1.1; Brown bar: a fragment bound by clone 10 and HO36-1.1 and to a lesser degree by clone 30; Blue bars: fragments bound by clone 10 and HO36-1.1. The locations of the deduced epitopes are highlighted. (B and C) Western blotting data. Anti-His antibody (bottom) was used to confirm sample loading. NegGFP denotes negative control GFP.

To identify possible off-targets of clone 10, we performed a BLAST search of its epitope, residues 63 to 76, and found 17 segments from ten proteins that are expressed extracellularly (Table S1). These segments are predominantly proline-rich sequences, but all have ≤50% sequence identity to the clone 10 epitope. Binding of clone 10 to monkey and mouse CD99, with 79% and 43% sequence identity, respectively, was significantly diminished (Fig. 1C, S1F) (Table 1). Together, these data suggest that off-target binding is unlikely, as clone 10 only weakly bound to monkey CD99 that exhibits the highest sequence identity to human CD99.

Table 1.

Summary of cell-based assays utilizing antibodies and complexes that bind to CD99 highlighting: valency, homotypic cell aggregation, cytotoxicity, IC₅₀ values, and apparent K_D values.

	Fab	IgG-LALA	IgG-HCscFv	lgG-E345R	IgG-RGY	A2D2	A3D1	A4
Valency in solution	1	2	4	2	12	2	3	4
Valency upon antigen binding	1	2	4	12	12	2	3	4
Cell aggregation?	No	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Cytotoxicity?	No	No	Yes	Yes	Yes	No	Yes	Yes
IC50 (nM Fv)	>2000	>2000	82 ± 2	299 ± 61	442 ± 20	ND	ND	ND
IC50 (nM molecule)	>1000	>1000	21	150	37	ND	ND	ND
KDapp (nM Fv)	>250	5.4 ± 0.6	5.3 ± 4.3	16 ± 5	6.2 ± 2.8	5.2 ± 1.8	3.9 ± 0.9	5.6 ± 1.2
KDapp (nM molecule)	>250	2.7	1.3	8.0	0.5	2.6	1.3	1.4

We then produced clone 10 in the Fab format and measured its binding to CD99 immobilized on streptavidin beads ^{33; 34} and determined its apparent K_D to be 53 ± 8 nM Fv. To enhance the affinity, we performed random mutagenesis of the heavy and light chain variable regions and identified a clone with enhanced binding, termed 10A1, with apparent K_D = 31 ± 1 nM Fv (Fig. 1F, S5). Interestingly, 10A1 in the scFv-Fc format exhibited greater enhancement of binding to endogenously expressed CD99 on Jurkat cells, with apparent K_D = 18 ± 2 nM Fv, a fourteen-fold improvement over that of Clone 10 (266 ± 98 nM Fv) (Fig. 1G).

Valency of two is insufficient for inducing cytotoxicity

We next examined whether clone 10 and 10A1 elicit cytotoxicity in T-ALL cells. We treated Jurkat cells with 5 μM of Clone 10 scFv-Fc antibody, a concentration 37 times greater than the apparent K_D for cell-based affinity, and found no effects in increasing apoptosis after 24 hours (Fig. S6). Similarly, we did not detect any significant changes in cell viability or apoptosis with 5 μM 10A1 scFv-Fc (Fig. S2B, S6). These results stand in stark contrast to the ability of HO36-1.1 to induce cytotoxicity in AML and T-ALL cells, particularly because all these antibodies bind to overlapping epitopes and HO36-1.1 has a lower apparent affinity to the recombinant ECD of CD99 (226 ± 147 nM, Fig. S7), compared with clones 10 and 10A1 (53 ± 8 and 31 ± 1 nM, respectively; Fig. 1F). These results suggest that affinity to CD99 is not a major determinant of the difference in cytotoxicity between our antibodies and HO36-1.1. Clone 10 and 10A1 in the scFv-Fc format are bivalent, whereas HO36-1.1 in the IgM format is decavalent, suggesting that a higher valency of a CD99-binding antibody may be crucial for inducing cytotoxicity.

Higher valency enhances leukemia cell cytotoxicity but not cell surface binding in vitro

To examine if a valency greater than two is required for inducing cytotoxicity we generated four antibodies containing the same VL and VH domains from 10A1: IgG with a valency of two, IgG-HCscFv with a valency of four, and IgG-E345R ‘inducible hexamer’ and IgG-RGY ‘constitutive hexamer’ with a valency of twelve (Fig. 3A, S2, S3) ^{35; 36}. When incubating KOPT-K1 cells, a human T-ALL cell line, with the antibodies we observed that those with a valency of two or more induced homotypic cell aggregation (Fig. S8, S9), a phenomenon previously seen when incubating Jurkat or CD4⁺CD8⁺ T cells with antibodies against CD99 ^{37; 38; 39}. Cell aggregation can lead to inaccurate measures of cell concentration and Annexin V positive cell percentage, because aggregated cells are counted as one in flow cytometry. Thus, we sought to find a method to disaggregate these cells. EDTA and DNaseI, reagents known to promote cell dissociation ^{40; 41}, did not dissociate the cells. In contrast, an addition of excess recombinant CD99 ECD caused the cells to dissociate (Fig. S8G), likely through inhibition of antibody-CD99 interactions. Consequently, we added excess CD99 ECD to antibody-treated cells prior to analyzing them with a flow cytometry.

Figure 3.

Cytotoxic effect and binding characteristics of the anti-CD99 antibody, 10A1, in different formats. (A) Dose-dependent cytotoxic effects on KOPT-K1, as measured by Annexin V staining after an 18-hour treatment. From left to right: Fab, IgG-LALA, IgG-HCscFv, IgG-E345R, and IgG-RGY. The antibody formats are schematically shown. Data for the nonbinding isotype controls are shown as open circles, and those for the 10A1 antibodies are shown as filled circles (n = 3). (B) Binding titrations curves of 10A1 antibodies to endogenously expressed CD99 on KOPT-K1 cells. From left to right: Fab, IgG-LALA, IgG-HCscFv, IgG-E345R, and IgG-RGY. Data for the nonbinding isotype controls are shown as open circles, and those for the 10A1 antibodies are shown as filled circles (n=3). (C) Cell surface levels of CD99 of healthy donor PBMC, T-ALL samples, and the KOPT-K1 cell line. (D) Cytotoxic effects of 18-hour treatment with 2.0 μM Fv (equivalent to 500 nM molecule) 10A1 HCscFv on healthy control PBMCs, and leukemia cells derived from T-ALL patients, and KOPT-K1 cells (n=3). (E) Cell surface levels of CD99 of healthy donor PBMC samples, and KOPT-K1 cell line. (F) Cytotoxic effects of 18-hour treatment with 2.0 μM Fv 10A1 HCscFv on healthy control PBMC samples and KOPT-K1 cells (n=3). (G) Cell surface levels of CD99 of OCI-AML3 (AML cell line), CCRF-SB (B-ALL cell line), and MOLT3 and KOPT-K1 (T-ALL cell lines). (H) Cytotoxic effects of 2.0 μM Fv 10A1 HCscFv on T-ALL and B-ALL cell lines after 18 hours. (I) Cytotoxic effects of 2.0 μM Fv 10A1 RGY on AML and T-ALL cell lines after 18 hours. P values were calculated using 2-way ANOVA with Tukey's multiple comparisons test.

We tested cytotoxicity of the 10A1 antibodies in different formats using KOPT-K1 cells. We detected a dose-dependent increase in the apoptotic cell percentage (Fig. 3A) and a concurrent reduction in the viable cell percentage (Fig. S9) for the antibodies with a valency greater than two. The IgG format shows only marginal levels of cytotoxicity even at the highest concentrations. The tetravalent antibody demonstrated the highest efficacy, with the rank order of IgG-HCscFv > IgG-E345R > IgG-RGY, as the IC₅₀ values in terms of the antigen-binding site concentration were 82, 299, and 442 nM Fv, respectively (Table 1). We also found that the 10A1 HCscFv was more effective at inducing cytotoxicity than HO36-1.1 (Fig. S10). These data indicate that a valency of four is sufficient for inducing T-ALL cytotoxicity.

To assess if enhanced cytotoxic effect is the consequence of enhanced binding to cells, we measured binding of these antibodies to KOPT-K1 cells (Fig. 3B). Contrary to our expectation, the apparent K_D value of the bivalent antibody was in the low nanomolar range, close to the higher valency antibody formats (Table 1). The apparent K_D values were similar among IgG-RGY, IgG-HCscFv and IgG (5~6 nM) except that IgG-E345R showed about 3-fold higher K_D value (i.e. lower affinity) than those antibodies. Thus, the lack of potent cytotoxic function in the IgG format is not because it had a lower affinity to the cells.

We observed that the maximum fluorescence intensities of bivalent and dodecavalent antibodies were similar but that of the tetravalent antibody was about a half as much (Fig. 3B). Because all these antibodies share the same Fab portion and they were detected using anti-Fab secondary antibodies, these results suggest that the anti-Fab secondary antibody does not effectively bind to the scFv portion of the tetravalent antibody and that approximately twice as many copies of the bivalent antibody bind to the cells as the tetravalent antibody. This interpretation in turn suggests that all the CD99-binding sites on the tetravalent antibody are engaged with CD99 on the cell surface.

Among the three formats with potent cytotoxicity, the IgG-RGY antibody has the highest affinity to KOPT-K1 cells, but the lowest cytotoxic effect. The size exclusion chromatography data showed that the RGY hexamer dissociated into IgG monomers below 4800 nM Fv, with 50% hexamer at 1895 nM Fv, and 5% at 500 nM Fv (Fig. S3), which is consistent with a previous report that the IgG-RGY antibody forms a hexamer in a concentration dependent manner ⁴². These results explain the loss of its cytotoxic activity at concentrations below 500 nM Fv, because the antibody is likely to be in the monomeric IgG format, i.e., bivalent, at these concentrations (Fig. 4A, S3). The concomitant loss of cytotoxic activity and the loss of a high valency structure further supports the view that an antibody needs to engage more than two CD99 molecules to elicit T-ALL cytotoxicity.

Figure 4.

A valency of three is required and sufficient for inducing KOPT-K1 cytotoxicity. (A) Schematic depiction of bi, tri, and tetravalent Streptavidin (SAV)-Fab complexes. A, wildtype SAV with ‘alive’ biotin-binding site; D, mutant SAV with ‘dead’ biotin-binding site. A dead biotin binding site is shown as a blue circle, and an active biotin binding site as a red circle. (B) Cytotoxicity, as measured by Annexin V staining, of bi-, tri-, and tetravalent SAV-Fab complexes (400 nM Fv) in KOPT-K1 cells treated for 18 hours (n=3). P values were calculated using ordinary 1-way ANOVA with Tukey's multiple comparisons test. (C) Binding titration curves of bivalent, trivalent, and tetravalent SAV-Fab complexes to endogenously expressed CD99 on KOPT-K1 cells (n=3).

10A1 HCscFv is effective against patient-derived T-ALL cells and nontoxic to healthy cells

To further investigate the efficacy of 10A1 HCscFv as a potential T-ALL therapy, we tested it with T-ALL patient samples. The patient samples we tested expressed CD99 at a lower level compared with the KOPT-K1 cell line (Fig. 3C, S11A-B). Strikingly, the 10A1 HCscFv antibody increased leukemia cell apoptosis, indicating its effectiveness to T-ALL patient samples (Fig. 3D, S11C). We treated healthy patient peripheral blood mononuclear cells (PBMCs), which expressed CD99 at a lower level than KOPT-K1 cells (Fig. 3C,E, S11A-B), with 10A1 HCscFv. After treatment, we observed no cytotoxicity, suggesting that the antibody does not lead to on-target, off-cancer toxicities in healthy cells (Fig. 3D,F, S11C).

Next, we tested 10A1 HCscFv with different cell lines that displayed lower levels of CD99 compared with KOPT-K1 cells, including DND-41 and MOLT-3 (T-ALL cell lines), OCI-AML3 and MOLM13 (AML cell lines), and CCRF-SB (B-ALL cell line) (Fig. 3G). We observed that the antibody induced cell death in the T-ALL and B-ALL cell lines, less efficiently than KOPT-K1 cells (Fig. 3H, S11F-K). In contrast, we saw no effects in AML cell lines, which expressed lower levels of CD99 than KOPT-K1 and DND-41 (Fig. S11D-K). Because the levels of CD99 are very low in OCI-AML3, we hypothesized that an antibody in a larger size and a higher valency is more effective at ligating more than two CD99 molecules that may be separated with a greater distance. Treatment of OCI-AML3 and KOPT-K1 cells with 10A1 RGY, at high concentration to ensure forming a hexamer, for 18 hours induced apoptosis in both AML and T-ALL cell lines (Fig. 3I). These results suggest that the level of 10A1 cytotoxicity can be finetuned toward different levels of cell-surface CD99 by utilizing different antibody formats.

CD99 is expressed on T-ALL blasts both at diagnosis and at relapse. Although CD99 is not required for T-ALL engraftment or self-renewal in an NSG xenograft model, CD99-negative T-ALL cells turn CD99 positive over time, and ultimately all engrafted cells become CD99 positive regardless of their initial surface display levels ¹⁶. These observations suggest that T-ALL cells could modulate CD99 expression and thereby escape from anti-CD99 therapy. To determine if treatment with the 10A1 tetravalent antibody affects CD99 surface display levels, we treated KOPT-K1 cells with a single dose and escalating dose and measured CD99 surface expression (Fig. S12). We found no changes in CD99 surface expression, suggesting that blasts do not readily escape the treatment by downregulating CD99 expression on the cell surface.

A valency of three is required and sufficient for inducing T-ALL cytotoxicity

To conclusively investigate the minimum valency required for inducing cytotoxicity, we generated streptavidin (SAV)-Fab complexes with a valency ranging from two to four. We utilized a streptavidin mutant with a ‘dead’ biotin-binding site ⁴³. We formed SAV tetramers containing two, three, or four functional subunits, i.e., tetramers comprising A2D2, A3D1 and A4 subunits, respectively, where A and D denote wild-type (alive) and mutant (dead) subunits (Fig. S13). Next, we conjugated biotinylated 10A1 Fab, to these SAV variants to yield bi-, tri-, and tetra-valent molecules (Fig. 4A). KOPT-K1 cells were treated for 18 hours, at concentrations more than seventeen times the apparent affinity values of the SAV-10A1 Fab complexes (Fig. 4B-C). Similar to the 10A1 IgG formats, we observed homotypic cell aggregation following incubation with the bi-, tri-, and tetravalent SAV-10A1 Fab, which was readily dissociated using excess CD99 antigen (Fig. S13A-F). We detected cytotoxic effects with the trivalent and tetravalent 10A1 molecules but not with the monovalent or bivalent molecules (Fig. 4B, Fig. S13G-J). The apparent K_D values of these molecules were all in the low nanomolar range (Fig. 4C, Table 1), similar to the 10A1 IgG formats, confirming our finding that cytotoxicity is not directly dependent on the effective affinity to the cell (Table 1, Fig. S13). The Fab portion is linked to the SAV via a flexible linker in these constructs, and the efficacy of the tri- and tetravalent constructs indicate that the close association of two Fab portions in the IgG format is not essential for cytotoxicity. Taken together, these data reveal that a valency of three is necessary and sufficient for inducing T-ALL cytotoxicity.

X-ray crystal structure of 10A1 Fab in complex with a CD99 peptide

To determine the structural basis for the recognition of CD99 by 10A1 Fab, we determined the crystal structure of 10A1 Fab in complex with a CD99 peptide containing the 10A1 epitope (residues 62-76; Fig. 2, S15A) at a resolution of 3.10 Å (Fig. 5A, Table S2). As expected, 10A1 takes on the canonical global fold of Fab. Of the 15 residues of the CD99 peptides, we observed electron density for residues 70-76, PNPPKPM (Fig. S15B), but not for the N-terminal region of the peptide. The observed peptide segment formed a left-handed helix, showing striking similarity to other proline-rich motifs (PRMs) that also form left-handed poly-proline type-II (PPII) helices (Fig. 5B) ⁴⁴. PRMs are known ligands for Src homology (SH) 3, WW, Enabled VASP Homology 1 (EVH1) domains as well as CD2 binding protein (CD2BP2) and profilin ⁴⁵. However, these well-documented interactions occur predominantly among intracellular proteins. There are precedents of proline rich motifs in extracellular domains of proteins. One of these proteins, mucolipin-1, contains proline rich regions (Table S1) that forms a PPII helix PDB ID: 5WJ5; ⁴⁶.

Figure 5.

The crystal structure of the 10A1 Fab in complex with a CD99 peptide, PDB ID: 7SFX. (A) Cartoon depiction of the complex. (B) Overlay of proline-rich peptides from CD99 residues labeled in magenta (this study, magenta & purple), SH3 domain binders: 2VWF (salmon, chain B: IQPPVN), 1AZE (red, chain B: PVPPRR), and 2W0Z (brown, chain B: APPPRPPKP); WW-domain binders: 2HO2 (light blue, chain B: PPPPPPPL) and 1EG4 (blue, chain A: TPYRSPPPYVP); and the proline-rich tail of CD2: 1L2Z (orange, chain B: SHRPPPPGHR). (C) Surface representation of 10A1 Fab showing two distinct binding pockets for the CD99 peptide enclosed circles. Heavy chain in marine, light chain in cyan. (D) The 10A1/CD99 interface highlighting key residues in the 10A1 Fab. Residues in yellow showed a substantial loss in CD99 binding upon mutation, as shown in panel E and fig. S15. Residues in gray tolerated mutations. ((E) 10A1 fab mutagenesis, in the yeast display format. Binding signals with 100 nM SAV-CD99 complex are shown (n=3). SAV650, Streptavidin DyLight650.

The peptide sits in a cleft at the junction between the heavy and light chains (Fig 5C). The interface buries 444 Ų of the solvent accessible surface of the peptide (Table S3). The interface is mainly composed of Van der Waals interactions, with the exception of one possible hydrogen bond between the CD99 peptide residue M11_P and Y31 _L ^{47; 48}. We identified residues of 10A1 that directly made contacts with CD99: K54 of the heavy chain, and Q30, Y31, and F95 of the light chain (Fig. 5D). Mutation to alanine of these residues substantially reduced CD99 binding (Fig. 5E). In addition, mutating D50_L and D91_L in CDR-L2 and -L3, respectively, to serine abolished binding (Fig. 5D, S15C), indicating these residues play important roles in the binding to CD99, although they do not directly contact the CD99 peptide in the crystal structure (Fig. 5D). D91_L is adjacent to residue V90_L, which was one of the residues mutated during affinity maturation (Fig. S1), suggesting that the A90V mutation contribute to the affinity increase by improving the position of D91_L. Mutation analysis also revealed six residues located immediately outside the paratope that show marginal effects upon Ala substitution (Fig. 5E). Altogether, these mutation results validate the crystal structure.

As we observed the similarity between the CD99 epitope and PRMs bound to modular domains, we compared their interface characteristics (Fig. 5B). Although the CD99 peptide structure resembles that of the other PRMs and buries surface areas comparable to these complexes (Table S3), the affinity of 10A1 to CD99 is much higher than those for the natural ligands of the modular interaction domains (micromolar K_D values) ⁴⁹. Our results demonstrate the feasibility of engineering high-affinity interfaces for RPMs.

Discussion

This study provides support for anti-CD99 antibodies as potential therapy in T-ALL. There are currently no FDA approved monoclonal antibodies approved for use in T-ALL ⁵⁰. This is unfortunate because the current chemotherapeutic treatment options for T-ALL are not selective and lead to patient toxicities ^{1; 2}. Even after intensive chemotherapy and radiation, some T-ALL patients relapse or become refractory. The best chemotherapeutic option for relapse/refractory T-ALL patients is nelarabine, but it is associated with a high risk of neurotoxicity ⁵¹. Alternative treatment options, such as stem cell transplants, are available for high-risk patients, but toxicity also plays a role in the lack of efficacy, with transplantation-related mortality in older patients ⁵². There is clearly an unmet need for more effective and less toxic therapies in T-ALL.

The tetravalent antibody that we have developed in this work, 10A1 HCscFv, has multiple favorable attributes that make it a promising candidate for further therapeutic development. It is a fully human antibody in a manufacturable format ^{53; 54}. It is selective to human CD99 (Fig. 1, Table S1), directly elicits cytotoxicity in T-ALL cells independent of effector-mediated function, such as antibody-dependent cellular cytotoxicity, and it is non-toxic to healthy cells (Fig. 3). It has weak, but substantial, cross-reactivity to monkey and mouse CD99, which can be enhanced via protein-engineering approaches for enabling preclinical studies. The structure of the 10A1-CD99 peptide complex will help improve the cross-reactivity and facilitate the development of variants that bind better to monkey CD99. The epitope includes PNPPKPM and a truncation of PM abrogates binding (Fig. 2). Monkey CD99 has a PNPPKPK motif, and we can alter the pocket for Met in such a way that it better accommodates Lys in monkey CD99. In contrast, Mouse CD99 has a distinct QPDPKPP motif and we envision that substantial engineering of 10A1 may be required to achieve high-affinity binding to mouse CD99.

The T-ALL cell lines utilized in this study contain an activating mutation in the NOTCH1 signaling pathway ⁵⁵. This pathway is a major oncogenic driver in T-ALL, with mutations in ≥50% of T-ALL patients ^{55; 56; 57}. γ-Secretase inhibitors (GSI), which inhibit the Notch signaling pathway, are being evaluated in combination with chemotherapy (NCT02518113 and NCT01363817). Interestingly, DND-41 and KOPT-K1 cell lines, but not MOLT-3, are GSI-sensitive ⁵⁸. All of these T-ALL cell lines responded to the 10A1 HCscFv treatment (Fig. 3A, S11F-G), suggesting that the antibody is potentially efficacious to T-ALL with diverse driver mutations.

We observed that 10A1-mediated cytotoxicity depends on CD99 cell-surface levels (Fig. 3C-G). In one example, we did not observe a cytotoxic effect in AML cell lines treated with the 10A1 tetravalent antibody (Fig. S11H-K), which is likely due to the lower CD99 surface levels in AML cells than T-ALL cells (Fig. 3E, Fig. S11D-E). When utilizing the 10A1 ‘constitutive hexamer’ antibody, we observed a cytotoxic effect in both AML and T-ALL cell lines, again with the higher CD99 expression associated with higher antibody mediated cytotoxic activity (Fig. 3G). The strong correlation between cytotoxicity mediated by the 10A1 antibodies and CD99 expression levels suggests that CD99-targeted therapeutics will exhibit minimal toxicity to healthy or low CD99-expressing cells. Furthermore, their efficacy to cells with a lower CD99 level than that of T-ALL cells may be enhanced by changing antibody valency and/or changing the affinity of the Fv unit to CD99, which potentially expand their utility beyond T-ALL.

Our studies reveal that a valency of three is necessary and sufficient for inducing cytotoxicity in T-ALL (Fig. 4B), which suggests a mechanism of action in which the engagement of a minimum of three CD99 molecules induces apoptosis. In contrast, a valency of two is sufficient for achieving strong binding of 10A1 antibodies, as we observed that the bivalent, trivalent, and tetravalent antibodies have similar apparent K_D values to T-ALL cells (Fig. 3B, 4C, Table 1), indicating that the higher valency is required for inducing cytotoxicity, but not for engaging the antibodies to the surface of T-ALL cells. Indeed, we found striking evidence that all antigen-binding sites of the tetravalent antibody are engaged with four CD99 molecules when it binds to T-ALL cells (Fig. 3B). Together these results conclusively establish the basis for the requirements of a valency ≥3 for anti-CD99 antibodies to induce apoptosis by bringing ≥3 CD99 molecules to close proximity on the cell surface.

The requirement of receptor trimerization is reminiscent of the key mechanism that drives signaling of the tumor necrosis factor receptor superfamily (TNFRSF) ^{59; 60}. In a striking parallel to our study, a bivalent antibody targeting Fas on a lymphoma cell line does not affect cell viability, whereas crosslinking of the same anti-Fas antibody, which creates a tetravalent molecule, causes cell death ⁶¹. In another example, an anti-OX40 ‘constitutive hexamer’ antibody, but not the bivalent version, exhibited agonistic activity, highlighting the importance of high valency in activating members of the TNFRSF ⁶². Interestingly, we found that the CD99 transmembrane domain contains a sequence similar to the oligomerization motifs of TNFRSF, ΦPxΦ and GxxxG, where Φ is a hydrophobic residue (e.g., Leu, Ile, Val and Phe) and x is any amino acid compatible with partitioning in the membrane (Table S4) ⁵⁹. Future research will determine whether CD99 also forms a trimer to drive signaling and whether there is a natural ligand that trimerizes CD99.

Antibodies against distinct regions on CD99 have previously been developed and some have been characterized in terms of cytotoxic activity ^{15; 63; 64}. In our study, we identified both 10A1 and HO36-1.1 studied here bound residues 63 to 76 and exhibited cytotoxicity (Fig. 2A-B). DN16, mouse IgG1, binds to residues 32 to 39 ⁶⁵, and exhibits cytotoxicity upon crosslinking ⁶⁴. 0662, mouse Igg3 that form non-covalent oligomers ^{66; 67}, binds to residues 88 to 97 ⁶⁸ and also exhibits cytotoxicity ⁶³. Taken together, these findings indicate that the 10A1 epitope is not unique in inducing cytotoxicity and that a high valency format of an anti-CD99 antibody is crucial for achieving this effect.

We found that 10A1 antibodies with a valency of ≥2 triggered cell aggregation at concentrations above 200 nM Fv (Fig. S8). As discussed above, these antibodies likely ligate a large fraction of CD99 molecules on the cell surface, which suggests a mechanism in which CD99 oligomers promote cell-cell interactions through homotropic interactions across cells. The antibody is unlikely to bridge CD99 molecules on separate cells, as our flow cytometry analysis revealed that all the CD99-binding sites on the tetravalent antibody were occupied with the antigen on single cells, leaving no free antigen-binding sites available for bridging cells (Fig. 3B). CD99 was previously shown to be involved in T cell adhesion and rosette formation between T cells and erythrocytes ^{38; 69}. We speculate that our antibodies may mimic a yet-to-be-identified ligand of CD99 and that this ligand brings together multiple CD99 molecules, inducing cell-cell interaction and ultimately cell apoptosis. A systematic search may discover such a ligand.

Although we now know that a valency of three is sufficient for cytotoxicity (Fig. 4B), we did not proceed with a trivalent antibody, as the production of such an antibody is cumbersome. We proceeded with the tetravalent antibody because unlike the hexamer IgG formats, its valency is independent of antibody concentration, which may contribute to it achieving the highest cytotoxic activity among the formats tested (Fig. 3A, Table 1). This attribute is beneficial in in vivo applications, as an antibody may be diluted in the blood and in the bone marrow. The 10A1 antibody has a lower IC₅₀ and greater cytotoxic effect on KOPT-K1 than the mouse anti-CD99 IgM, HO36-1.1 (Fig. S10). Although recent advances have demonstrated the feasibility of manufacturing IgMs for therapeutic uses ⁷⁰, it is still difficult to produce IgM antibodies. Our tetravalent antibody balances the requirement to achieve high valency and production feasibility. We envision anti-CD99 antibodies with high valency will contribute to the establishment of targeted therapies for T-ALL and potentially other hematological malignancies.

As we were completing this study, Shi et al. reported a CAR-T therapy targeting CD99 that showed selective efficacy toward T-ALL cells and in mouse xenograft models ⁷¹. Their study and our present study offer complementary strategies for eradicating T-ALL by targeting CD99. Moreover, the 10A1 antibody and its derivatives may also be used for constructing CAR-T cells and other types of biologics.

Experimental Procedures

Target cloning, expression, and purification from Escherichia coli.

GFP, SUMO, and GFP and SUMO fused with the extracellular domain of CD99 (ECD; residues 23-122 of UniProt ID P14209) containing a 6xHis tag, AviTag, and TEV cleavage site at the N-Terminal region were cloned into pHBT plasmids ⁷². BL21(DE3) cells were transformed, separately, with plasmids. The target proteins were expressed via isopropyl β-D-1-thiogalactopyranoside (IPTG) induction at 18°C for 20 h. The targets were purified via immobilized metal affinity chromatography using a Ni-Sepharose resin (Cytivia). The purified proteins were biotinylated in vitro using in-house prepared recombinant BirA enzyme and further purified by size-exclusion chromatography using a Superdex 75 10/300 GL column (Cytivia).

Single chain Fv selection using yeast display

Purified biotinylated targets were used for yeast display selection from a naïve scFv display library ²³ as previously described ^{24; 73}. Briefly, in the first round of selection Dynabeads M-280 Streptavidin (ca# 11206D, Invitrogen) were coated with biotinylated SUMO-CD99 and selection was performed by magnetic bead capture of target binding yeast clones. Yeast clones were then expanded, and two rounds of fluorescent activated cell sorting (FACS) was performed. In the first round, 100 nM NeutrAvidin 650 (ca# 84607, Invitrogen) complexed with SUMO-CD99 tetramer was incubated with the library then sorted. In the second round, 20 nM of Streptavidin 650 (ca# 84547, Invitrogen) complexed with GFP-CD99 tetramer was incubated with the library then sorted. Cell sorting was performed on a S3e Cell Sorter (Bio-Rad).

Detection of scFv expression on yeast was measured by binding signal of 5 μg mL⁻¹ anti-cMyc antibody (ca# A21281, Invitrogen) in PBSE (9.6 mM phosphate, 137 mM NaCl, 2.7 mM KCl pH 7.4 and 1 mM EDTA) supplemented with 0.5%BSA (PBSE/BSA) for 30 min at 4°C, with shaking. After incubation, cells were washed three times with 100 μL PBSE/BSA and staining with 4 μg mL⁻¹ goat anti-Chicken IgY Alexa Fluor 488 (ca# A11039, Invitrogen) for 30min at 4°C, with shaking. The cells were then washed with PBSE/BSA, and resuspended in 100 μL of PBSE/BSA and analyzed on a HyperCyt screener (Sartorius). Signals reported are median fluorescence intensities. Preferential binding towards the extracellular domain of CD99 was confirmed on yeast by comparing binding signal to 10 nM tetramerized SUMO compared to SUMO-CD99 recombinant proteins.

Bead-based binding assay of Fab clones

The binding titration of the Fab clones was performed on Dynabeads M-280 Streptavidin (ca# 11206D, Invitrogen) immobilized with biotinylated antigen, as described previously ^{33; 34}. Binding of Fab clones to the targets on the beads was analyzed on a HyperCyt screener (Sartorius). Signals reported are median fluorescence intensities.

Mammalian cell culture

KOPT-K1 cells were a gift of Dr. Leandro Cerchietti (Weill-Cornell, New York, NY), DND-41 cells were a gift from Dr. Ioannis Aifantis (NYU Grossman School of Medicine, New York, NY), MOLM-13 (ca# ACC554) and OCI-AML3 (ca# ACC582) were purchased from Leibniz Institute DSMZ, Jurkat (ca# TIB152), MOLT-3 (ca# CRL1552), and CCRF-SB (ca# CCL120) cells were purchased from ATCC. Human T-ALL specimens were obtained from patients at NYU Langone Health. Healthy human PBMC specimens were purchased from STEMCELL Technologies, ca# 70025.1. All cell lines were grown in 5% CO₂ at 37°C under media conditions described by the vendor or the source laboratory. Free of mycoplasma contamination was confirmed using the e-Myco plus Mycoplasma PCR detection kit (ca# 25234, Bulldog Bio).

Cell-based binding analysis using cells endogenously expressing CD99

Cells were confirmed to express CD99 by incubation at 4°C for 30 min with anti-CD99 mouse IgM, HO36-1.1 (ca# ab212605, Abcam) and matched mouse IgM isotype, MM-30 (ca# ab18400, Abcam), then staining with Goat anti-Mouse IgM Cross-Adsorbed Secondary Antibody, DyLight 650 (ca# SA5-10153, Invitrogen) at a 1:100 dilution. Cells were incubated with antibody clones and controls for 30 min, washed three times with PBS (2.7 mM KCl, 1.5 mM KH₂PO₄, 138 mM NaCl, 8.1 mM Na₂HPO₄-7H₂O) supplemented with 2% FBS (ca# 100106, GeminiBio) (PBS/2%FBS), followed by staining with a secondary antibody. Goat anti-mouse IgG (H+L) conjugated with Dylight 650 (ca# 84545, Invitrogen) and goat Anti-Human IgG, F(ab')₂ conjugated with Alexa Fluor 647 (ca# 109605097, Jackson ImmunoResearch Inc) were used for detecting the binding of scFv-Fc at a 1:100 dilution, and the human IgG and Fab, respectively. The cells were washed three times, and resuspended in 100 μL of PBS/2%FBS and analyzed on a HyperCyt screener (Satorius). For healthy donor PBMC and T-ALL patient samples cells were stained with anti-human CD45 conjugated with BB515 (ca# BDB564585, BioRad), anti-human CD34 conjugated with BV421 (ca# BDB562577, BioRad), anti-human CD2 conjugated with APC (ca# 300214, Biolegend), and anti-human CD7 conjugated with PE/Cy7 (ca# 343114, Biolegend) antibodies. Gating for flow cytometry consisted of FSC vs SSC cell gating, followed by single cell gating in FSC-H or FSC-A vs FSC-W to remove aggregates. For all samples cells were gated for Propidium Iodide negative. T-ALL patient samples cells were then gated for human CD45 positive followed by CD34 negative, and finally CD7 positive. For healthy donor PBMC samples cells were gated for human CD45 positive followed by CD34 positive and CD34 negative. Data collection performed on ZE5 Cell Analyzer (BioRad) or Guava easyCyte (Luminex) flow cytometers. Signals reported are median fluorescence intensities.

Affinity maturation by random mutagenesis

Random mutagenesis for affinity maturation was performed by error-prone PCR. To increase the mutation rate of Taq DNA polymerase, MnCl₂ was added to the PCR reaction ⁷⁴. Briefly, two error prone PCRs using GoTaq Flexi DNA polymerase (Promega) were performed with final concentrations of MnCl₂ of 0.1 and 0.3 mM. The variable heavy chain and light chain were amplified separately, then combined and amplified via overlap PCR. The yeast cells were transformed with the amplified PCR products and the digested vector, and they were assembled via homologous recombination in the yeast cells. Two rounds of FACS were preformed using 100 nM of tetramerized SUMO-CD99, as described above. Preferential binding of the scFv clones to CD99 was assessed as described above, except 5 nM tetramerized targets were used. High affinity clone was identified by comparing binding signal of single clones, using 5 nM tetramerized targets, to the parental clone. Titrations of clones showing enhanced binding to tetramerized CD99 were performed to assess apparent affinity of the scFv displayed on the yeast. Clone 10A1 had the highest affinity to tetramerized CD99 on yeast and it was selected for further antibody characterization.

Production of Fab antibody fragment in E. coli

The Fab fragments of clones were produced by attaching the V_H and V_L fragments of the scFv to the C_H1 and C_L fragments, respectively, as previously described ⁷⁵. The antibody Fab fragments also contained a C-terminal AviTag (Avidity), on the heavy chain, to facilitate biotinylation using the BirA biotin ligase. They were expressed in E. coli 55244 (ca# 55244, sATCC) and purified using a HiTrap Protein G HP column (Cytivia) using 20 mM sodium phosphate buffer (pH 7.0) and eluted with 0.1 M Glycine-HCl (pH 2.7). A portion of the eluted Fab clones were biotinylated in vitro using BirA. All biotinylated and non-biotinylated Fabs were further purified on a Resource S column (Cytivia) using a linear gradient of NaCl in 50 mM sodium acetate buffer (pH 5.0). Purity was assessed by SDS-PAGE.

Production of scFv-Fc and human IgG1-based antibodies

The scFv-Fc antibodies were produced by attaching the scFv fragment to a mouse IgG1 Fc fragment containing C-terminal AviTag and His tags ^{26; 27; 28; 29}. The human IgG1-based antibodies were produced by attaching the V_H and V_L fragments to the IgG1 CH₁CH₂CH₃ fragment and C_L fragments, respectively. The human antibodies contain LALA-PG mutations: L234A and L235A mutations to eliminate binding to Fc receptors, and P329G to eliminate complement binding ⁷⁶. In addition, the ‘inducible’ hexamer antibody contains an E345R mutation to promote IgG hexamerization upon antigen binding and the ‘constitutive’ hexamer antibody contains E345R, E440G, and S440Y mutations that promote the formation of IgG hexamers in solution, in a concentration dependent manner ^{35; 36}. The tetravalent IgG HCscFv antibody was produced by attaching a GS₆GS₂GS₄ linker connected to the scFv onto the C-terminal heavy chain. The variable heavy chains were cloned into pFUSEss-CHIg plasmid (ca# pfusess-hchg1, Invivogen) containing the mutations described above, the variable light chains were cloned into pFUSEss-CLIg-hk plasmids (ca# pfuse2ss-hclk, Invivogen).

The ExpiCHO-S cells (Gibco) were transiently transfected with the plasmids using the ExpiFectamine CHO Transfection kit, (ca# A29129, Gibco) according to manufacturer’s protocol. The cells were incubated at 37°C with 8% CO₂ and harvested after 3-7 days. The cell culture was clarified by centrifugation and supernatant was supplemented with Complete ultra tablets, mini EDTA Free, (ca# 5832731001, Roche) and 1 mM PMSF. The supernatant was filtrated and recombinant proteins were purified. The scFv-Fc antibodies were purified by affinity chromatography using a HisTrap excel column (Cytivia) followed by size exclusion chromatography using a Superdex 200 10/300 column (Cytivia). The human IgG1-based antibodies were purified by affinity chromatography using a Protein A column (ca# PROA102, GORE) followed by cation exchange chromatography using a Resource S column.

Target protein truncations for epitope mapping using western blotting

Kunkel mutagenesis was utilized to make truncated CD99 ECD fragments (23-122, 33-122, and 43-122 ) and a combination of Kunkel mutagenesis and restriction and insertion cloning were utilized to produce the other CD99 fragments (53-122, 63-122, 73-122, 83-122, 93-122, 103-122, and 113-122) ⁷⁷. The proteins were expressed and purified as described above, except they were only purified using a Ni-Sepharose resin (Cytivia). SDS-PAGE of 50 ng of each purified protein was performed followed by transfer to Immobilon-P^SQ PVDF membrane (ca# ISEQ00010, Millipore). For Western blot detection, the membranes were blocked with 5% skim milk overnight at 4°C, then washed three times with 1X TBST (50 mM Tris pH 7.5, 150 mM NaCl, and 0.05% Tween-20). The membranes were then incubated with 14 nM of clones 10, 22, and 30, 1 nM of HO36-1.1, or 1:1000 of mouse anti-His (ca# 37-2900, Invitrogen) for 1.5 hours at room temperature, then washed as described above. The membranes were then incubated with goat anti-mouse IgG Fc conjugated with horseradish peroxidase (HRP) (ca# 31437, Invitrogen) or goat anti-mouse IgM-HRP (ca# 626820, Invitrogen), for 1 hour at room temperature, then washed as described above. Pierce ECL 2 Western Blotting Substrate (ca# 80198, Thermo) was added according to manufacturer protocol. Signal detection was analyzed on ChemiDoc Touch imaging system (Biorad).

Production of SAV-Fab dimer, trimer, and tetramer

To form streptavidin complexes with two to four functional biotin-binding sites we utilized a streptavidin triple mutant (N23A, S27D, S45A) with a ‘dead’ biotin-binding site ⁴³ and wildtype ‘alive’ streptavidin containing N-terminal 6xHis tag. BL21(DE3) cells were transformed, separately, with the ‘dead’ and ‘alive’ SAV expression plasmids. The target proteins were expressed via isopropyl β-D-1-thiogalactopyranoside (IPTG) induction at 37 °C for 18 h, then purified from the insoluble fraction after resuspension with GuHCl buffer (20 mM TrisHCl, 6 M GuHCl, pH 8.0). The dead and alive SAV were mixed at a 1:1 ratio and refolding was performed by rapid dilution into TBS (50 mM Tris pH 7.5 and 150 mM NaCl). The mixture was then filtered and purified using immobilized metal affinity chromatography using a Ni-Sepharose resin (Cytivia), this allowed for the capture of only A1D3, A2D2, A3D1, and A4 SAV complexes since they contain His-tags, whereas the D4 does not. The SAV complexes with two, three and four functional biotin binding sites were further purified on a Resource S column (Cytivia). A1D3, A2D2, A3D1, and A4 have an increasing number of 6XHis tags, one, two, three, and four, respectively. Consequently, the increasing number of 6XHis tags make the complexes more positively charged, causing the A1D3, A2D2, A3D1, and A4 to elute from the cation-exchange chromatography, in that order. Once purified the SAV complexes were dialyzed into PBS buffer. To form the SAV-Fab complexes the A2D2, A3D1, and A4 SAV were separately mixed with biotinylated Fab at a ratio of 1:4 then incubated at 4°C for 30 min.

In vitro antibody cytotoxicity assay

For cell line cytotoxicity assays, 25,000 cells per well were treated with antibody or control. For primary sample cytotoxicity assays, 1 x 10⁶ cells per well were treated with antibody or control. A final volume of 100 μL per well in a 96-well, non-treated, flat-bottom microplates (ca# 0877253, Falcon) was incubated in 5% CO₂ at 37°C for 18 hours. After the 18-hour treatment cells were visualized under a light microscope to assess cell aggregation. Cells were dissociated by incubating with recombinant CD99 at ≥ 5 μM and Accumax (ca# 07921, StemCell) for 15 min, then stained with Annexin V 488 (ca# A1320, Life technologies) according to manufacturer’s instructions and propidium iodide at a final concentration of 5 μg/mL (ca# P4170-25MG, Sigma) at room temperature for 15 min. SureCount standard 3 μm microspheres (ca# CC03N, Bangs Laboratories, Inc) were diluted 1:10 in PBS buffer and added to stained cells to measure cell count. Fluorescence and cell count were analyzed by flow cytometry. For healthy donor PBMC and T-ALL patient samples cells were stained with anti-human CD45 conjugated with BB515 (ca# BDB564585, BioRad), anti-human CD34 conjugated with BV421 (ca# BDB562577, BioRad), and anti-human CD7 conjugated with PE/Cy7 (ca# 343114, Biolegend) antibodies. Gating for flow cytometry consisted of FSC vs SSC cell gating, followed by single cell gating in FSC-H or FSC-A vs FSC-W to remove aggregates. T-ALL patient samples were then gated for human CD45 positive followed by CD34 negative, and finally CD7 positive. For healthy donor PBMC samples cells were gated for human CD45 positive followed by CD34 positive and CD34 negative. For all samples gating of Annexin V positive and Propidium Iodide positive was performed to assess percentage of apoptotic cells. Data collection performed on ZE5 Cell Analyzer (BioRad) or Guava easyCyte (Luminex) flow cytometers.

X-ray crystallography and structural determination

The CD99 peptide containing the 10A1 epitope, Ac-GENDDPRPPNPPKPM-amide (new england peptide), was mixed with the 10A1 fab at a 2:1 molar ratio of peptide to fab to a final concentration of ~ 7 mg/mL. The condition at which crystals formed consisted of 0.2 M lithium sulfate monohydrate, 0.1 M HEPES, pH 7.3, 25% PEG3350, (ca# HR2-144, Hampton Research) and 0.1 M L-Proline (ca# HR2-428, Hampton Research). The crystal used for x-ray diffraction had a low three-dimensional shape and was a large two-dimensional plate (Fig. S15D). Diffraction data were collected at Advance Photon Source at the Argonne National Laboratory using the 19-ID beam line. Data processing was performed using HKL2000 ⁷⁸ and the starting model was built using SWISS-MODEL ⁷⁹ and molecular replacement was performed using Phaser ⁸⁰. Phenix refinement ⁸¹ and Coot ⁸² utilized for refinement and the h, -k, -l, twinning law applied. We used the 'Protein interfaces, surfaces and assemblies' service PISA at the European Bioinformatics Institute, PDBePISA (http://www.ebi.ac.uk/pdbe/prot_int/pistart.html), to determine the buried interface ^{47; 48}.

• Human antibodies to CD99, an upregulated surface marker of T-ALL

• Antibody valency of ≥3 is necessary for inducing cytotoxicity

• Development of an IgG-based high-valency antibody with cytotoxic efficacy

• Crystal structure reveals antibody recognition of a proline-rich segment

Data availability

X-ray coordinates for 10A1 Fab in complex with CD99 peptide have been deposited in the RCSB PDB under accession code 7SFX. All other data related to this work are contained within the manuscript.