AXL-specific single domain antibodies show diagnostic potential and anti-tumor activity in Acute Myeloid Leukemia

Rationale: AXL expression has been identified as a prognostic factor in acute myeloid leukemia (AML) and is detectable in approximately 50% of AML patients. In this study, we developed AXL-specific single domain antibodies (sdAbs), cross-reactive for both mouse and human AXL protein, to non-invasively image and treat AXL-expressing cancer cells. Methods: AXL-specific sdAbs were induced by immunizing an alpaca with mouse and human AXL proteins. SdAbs were characterized using ELISA, flow cytometry, surface plasmon resonance and the AlphaFold2 software. A lead compound was selected and labeled with 99mTc for evaluation as a diagnostic tool in mouse models of human (THP-1 cells) or mouse (C1498 cells) AML using SPECT/CT imaging. For therapeutic purposes, the lead compound was fused to a mouse IgG2a-Fc tail and in vitro functionality tests were performed including viability, apoptosis and proliferation assays in human AML cell lines and primary patient samples. Using these in vitro models, its anti-tumor effect was evaluated as a single agent, and in combination with standard of care agents venetoclax or cytarabine. Results: Based on its cell binding potential, cross-reactivity, nanomolar affinity and GAS6/AXL blocking capacity, we selected sdAb20 for further evaluation. Using SPECT/CT imaging, we observed tumor uptake of 99mTc-sdAb20 in mice with AXL-positive THP-1 or C1498 tumors. In THP-1 xenografts, an optimized protocol using pre-injection of cold sdAb20-Fc was required to maximize the tumor-to-background signal. Besides its diagnostic value, we observed a significant reduction in tumor cell proliferation and viability using sdAb20-Fc in vitro. Moreover, combining sdAb20-Fc and cytarabine synergistically induced apoptosis in human AML cell lines, while these effects were less clear when combined with venetoclax. Conclusions: Because of their diagnostic potential, sdAbs could be used to screen patients eligible for AXL-targeted therapy and to follow-up AXL expression during treatment and disease progression. When fused to an Fc-domain, sdAbs acquire additional therapeutic properties that can lead to a multidrug approach for the treatment of AXL-positive cancer patients.


Len�viral transduc�on of cells
The murine Axl gene was cloned into a transfer plasmid (pHR') using Gibson cloning.In short, gBlocks were designed to encode the sequence for mus musculus AXL receptor tyrosine kinase (Axl), transcript variant 1 (Gene ID: 26362; NM_009465.4)and purchased from Integrated DNA Technologies (Leuven, Belgium).Len�viral vectors were produced by transfec�on of HEK293T cells with the transfer plasmid, an envelope encoding plasmid (pMD.G) and a packaging plasmid (pCMVΔR8.9),as previously described [1].The pMD.G and pCMVΔR8.9plasmids were both provided by D. Trono, Geneva, Switzerland.Len�viral par�cles were harvested 48 and 72 h following transfec�on and concentrated using ultracentrifuga�on.Subsequently, murine C1498 AML cells were transduced with len�viral vectors harbouring mouse Axl at a ra�o of 10 transducing len�viral par�cles per cell (mul�plicity of infec�on 10).Transduced cells were then subjected to magne�c-ac�vated cell sor�ng, using an APC-conjugated mouse AXL an�body (R&D Systems, FAB8541A) and an�-APC microbeads (Miltenyi Biotec, 130-090-855) to increase the purity of AXL-expressing C1498 cells, further referred to as AXL high .Non-transduced cells are referred to as AXL low .

Thermal stability assay
The mel�ng temperature (Tm) values of an�-AXL sdAbs and their controls were determined using a ThermoFluor ® assay, using CFX Connect TM Real-Time PCR (Biorad, Pleasanton, CA, USA) and the fluoroprobe Sypro ® Orange dye (Molecular Probes, Oregon, OR, USA), as previously described [2].For thermal denatura�on, the samples were heated from 10 to 95 °C, with a gradual increase of 0.5 °C per 30 s, and a 10 s hold step at every checkpoint, followed by fluorescence reading.The Tm values were subsequently calculated using the Boltzmann equa�on.

Compe��on assay
The blocking capacity of an�-AXL sdAb20 and sdAb20-Fc was determined using the Biacore-T200 device (GE Healthcare).The measurements were performed at 25°C using HBS as running buffer.First, GAS6 (R&D Systems, #885-GSB) was immobilized on a CM5 sensor chip using linkage chemistry with EDC and NHS, and remaining free EDC-NHS linkers were neutralized with 1 M ethanolamine-HCl pH 8.5.
A�erwards, a serial dilu�on of sdAb20, sdAb20-Fc and the control sdAbs R3B23 and R3B23-Fc (10,000 to 0.9766 nM) was made and incubated with 150 nM recombinant AXL protein in a 1:1 ra�o for 1 h at room temperature.Recombinant AXL protein without sdAbs was used as a posi�ve control.The dilu�ons were analyzed at a flow rate of 10 µL/min and a�er each cycle, the chip was regenerated for response units of the dilu�on to the blank (HBS only) and the posi�ve control (AXL protein only).Based on these results, a nonlinear regression dose-response curve was generated using GraphPad Prism.
Coupled with Google Colaboratory, the open-source so�ware ColabFold [4] allows accelerated predic�on of protein structures and protein-protein interac�ons by combining the fast homology search of MMseqs2 (UniRef + Environmental) with AlphaFold2.The sequences used for structural predic�ons can be found in Table S1 and were retrieved from the UniProt database (#P30530, #Q00993, #Q14393 and #Q61592).Structural visualiza�on and alignments were generated using the PyMOL Molecular Graphics System, Version 2.5.2, license invoice #51791 (Schrödinger, Mannheim, Germany).

Flow cytometry
To evaluate the cell binding capacity of sdAb20, 200 nM sdAb was pre-incubated with 1 µL mouse an�-his�dine tag an�body (Bio-Rad Laboratories, #MCA1396) in PBS supplemented with 0.5% bovine serum albumin and 0.05% sodium-azide for 1 h on ice at 4°C.In the mean�me, cells were counted and washed with cold PBS, a�er which the cells were incubated with the sdAb20-an�-his�dine tag mix for 1 h on ice at 4°C.Next, cells were washed with cold PBS to remove unbound sdAb20 and incubated with 1 µL APC-conjugated rat an�-mouse IgG1 an�body (Biolegend) for 30 min on ice.
To inves�gate the cell binding poten�al of sdAb20-Fc, we directly fluorescently labeled sdAb20-Fc using the Alexa Fluor TM 647 Protein Labeling Kit (Invitrogen Carlsbad, CA).Cells were first washed and incubated for 30 min with FcR blocking reagent (Miltenyi Biotec, Leiden) (Fc-block) at 4°C on ice.
A�erwards, the cells were stained with the labeled sdAb20-Fc construct for 1 h at 4°C on ice.

RNA isola�on and quan�ta�ve Real-Time PCR
RNA was extracted and purified using the NucleoSpin RNA Plus Kit (Macherey-Nagel, Düren, Germany), according to the manufacturer's protocol and RNA concentra�on was measured with Nanodrop TM (Thermo Fisher Scien�fic).Next, 500 ng/µL RNA was converted into cDNA using the Verso cDNA Synthesis Kit (Thermo Fisher Scien�fic).Real-�me PCR was performed in a volume of 25 µL with 1 µL primer mix, 10.5 µL nuclease-free water, 1 µL cDNA, and 12.5 µL SYBR Green (Applied Biosystems, Thermo Fisher Scien�fic) using the Quantstudio 12 K Flex Real-Time PCR System (Thermo Fisher Scien�fic).Gene-specific primers were purchased from Integrated DNA Technologies (Leuven, Belgium) and were used at a final concentra�on of 200 nM.Primer sequences are listed in Table S2.Primers for human GAPDH were obtained from Qiagen (#QT00079247) (Westburg, Leusden, Netherlands) and was used as a housekeeping gene for data normaliza�on.Differen�al gene expression levels were determined using the compara�ve ΔΔCt method.

Supplementary Figure 1 .Supplementary Figure 3 .Supplementary Figure 4 .
(A-B) Shown are Biacore-T200 sensorgrams obtained for the binding of anti-AXL sdAbs to immobilized recombinant mouse (A) and human (B) AXL protein, after instantaneous background depletion.The sdAbs were injected at ten di fferent concentrations (shown in different colors), ranging from 1.9 to 500 nM.(C) Percentage unfolded sdAb20 and R3B23 at di fferent temperatures, determined by the ThermoFluor® assay.(D) Graphical representation of the ex vi vo biodistribution profile of R3B23 and sdAb20 in naive CB17-SCID mice.Organs were isolated 90 minutes post-injection of radiolabeled ( 99m Tc) sdAb20 or R3B23 control tracer (n=3).Results are presented as mean %IA/g ± SD. p ≤ 0.05 (*) w as considered statistically significant.Prediction aligned error (PAE) score for models ranked 1 to 5.This score quantifies the discrepancy between the predicted distances for pairs of amino acid residues.The x and y axes of the graph represent the positions of individual amino acids.The level of uncertainty in the predi cted distance bet ween t wo amino acids is depi cted by a color gradient ranging from blue (0 Å) to red (30 Å), as indicated in the accompanying legend.The color at the intersection of a verti cal line originating from one amino acid's position on the x-axi s and a horizontal line from another amino acid's position on the y-axis represents the error in the predicted di stance bet ween these t wo residues.PAE graphs consistentl y exhibit a diagonal blue line, reflecting the fact that amino acids closely positioned in the primary sequence are also proximate in the 3D structural arrangement.(B) Number of homologous sequences identi fied per position -30 to 100 sequences are requi red for best performance prediction.(C) Predicted local distance difference test (LDDT) score per position for the five models generated by Alphafold2.The graph illustrates the relationship between amino acid positions and their corresponding predi cted LDDT scores.When pLDDT values exceed 90, it suggests a very high level of accuracy, equivalent to experimentally determined structures.Conversely, values falling w ithin the range of 50 to 70 indicate lower accuracy, but it i s probable that the predictions for individual secondary structures remain correct.(D) Alternative AlphaFold2 predicitons of the interaction betw een sdAb20 and AXL.Rank_3 is show n in Figure 1.(A) Schematic representation of the lentiviral transduction of C1498 cells.HEK293T cells w ere lentivirally transfected with the third generation pCMV-ordered-mCO2-Axl1-PGK-PURO vector, expressing the first variant of the AXL1 coding sequence, together w ith an envelope-expressing plasmid (pMD.G) and a packaging plasmid (pCMV delta R8.9).Subsequently, harvested lentiviral particles were used to transduce C1498-AXL low cells.(B-C) C1498 cell transduction was confirmed by flow cytometry using sdAb20-APC (B) and a conventional APC-labeled mAb (C) (n=5).(D) Flow cytometric confirmation of the expression of end-stage tumorload marker CD45.2 on AXL-positive C1498 cells (n=3).(A) Size exclusion chromatography (SEC) confirms the addition of at least one DTPA molecule to sdAb20-Fc as indicated by the near-complete shi ft of the peak from 12 to 22 min, as measured at a wavelength of 280 nm.(B) Saturation binding curve of [ 111 In]In-DTPA-sdAb20-Fc on recombinant mouse Axl antigen using radio-ELISA, confirming the intact functionality of the construct.Speci fic binding (KD) was cal culated as the di fference bet ween total and non-speci fic binding.Data are expressed as mean ± SD. (C) Instant thin layer chromatography (iTLC) of [ 111 In]In-DTPA-sdAb20-Fc demonstrated high radiochemical purity.Radioactivity (green) chromatograms revealed greater than 90% of the radioactivity at the origin (Reg #1).Unbound, free 111 In, is visualized as the small red bump at region 2 (Reg #2) and moves w ith the solvent front.(D) SEC chromatograms confirmed the successful radiolabeling of 111 In to the DTPA-sdAb20-Fc construct, especially after NAP-5 column purifi cation.The construct remained stable up to at least 72 h.All SEC profiles are aligned at the main [ 111 In]In-DTPA-sdAb20-Fc peak.(E) The purity of the sdAbs was confi rmed by radio-SDS-PAGE under reducing conditions, follow ed by autoradiographic analysis and Coomassie blue staining.CPM = counts per minute, CPS = counts per second

M
111 In] af ter NAP5 [ 111 In] af ter f iltration 8c re a s A d re n a ls K id n e y L K id n e y R S to m a c h S .In t L .In t T e s ti c le /O v a ry M u s p h N o d e H e a r t T h y r o id T h y m u s L u n g s G a l B l a d d e r L i v e r S p l e e n P a n c r e a s A d r e n a ls K i d n e y L K i d n e y R S t o m a c h S .I n t L .I n t T e s t ic l e / O v a r y M u s l a d d e r L i v e r S p l e e n P a n c r e a s A d r e n a ls K i d n e y L K i d n e y R S t o m a c h S .I n t L .I n t T e s t ic l e / O v a r y p h N o d e H e a r t T h y r o id T h y m u s L u n g s G a l B l a d d e r L i v e r S p l e e n P a n c r e a s A d r e n a ls K i d n e y L K i d n e y R S t o m a c h S .I n t L .I n t T e s t ic l e / O v a r y M u s c le B

Table 1 . Sequences used for protein structure prediction using AlphaFold2.
Supplementary Table 3. AML patient's characteristics.