ImmunoPET, [⁶⁴Cu]Cu-DOTA-Anti-CD33 PET-CT, Imaging of an AML Xenograft Model Free

Acute myeloid leukemia (AML) is a highly aggressive hematopoietic malignancy with an extremely poor prognosis, as reflected by an overall 5-year survival rate of 40% to 45% in young adults and <10% in the elderly (>65 years of age; ref. 1). Research over the past decades has helped us understand the pathobiology, classification, and genomic landscape of the disease, which has resulted in improving current treatment options. Despite advances, the prognosis for elderly patients, who account for the majority of new AML cases, remains discouraging (2). More than 70% of elderly patients with AML (>65 years old) will die of their disease within 1 year of diagnosis and treatment (3). Therefore, new diagnostic and therapeutic approaches are necessary to improve outcomes.

Currently, the diagnostic criteria for AML is the presence of ≥20% blasts in the bone marrow or peripheral blood (4). AML diagnosis and prognosis are currently achieved by single-point bone marrow biopsies (iliac crest) followed by cytogenetics and mutation analysis. However, the iliac crest may not always be accurately representative of disease distribution within the entire body, particularly in the context of extramedullary disease, which is common in AML. Hence, there is need for new diagnostic tools that are noninvasive, specific, and sensitive to AML in the whole body, including extramedullary organs, and useful to longitudinally monitor disease and treatment response.

CD33, or SIGLEC3, is a cell surface marker found on myeloid stem cells, monoblasts, myeloblasts, monocytes/macrophages, and granulocytic precursors. However, CD33 is not expressed on erythrocytes, platelets, B cells, T cells, or NK cells, making it a suitable myeloid marker and therefore commonly used in the diagnosis of AML. CD33 has been shown to be expressed on more than 85% of AML cells (blasts; ref. 5), and an increased level of CD33 has been correlated with poor survival (6). An anti-CD33 antibody–drug conjugate (ADC; Mylotarg) is an FDA-approved immunotherapy for AML.

Although an anti-CD33 mAb is used for therapeutic purposes, no effort has been reported for developing it as a quantitative diagnostic and prognostic imaging biomarker. An anti-CD33 imaging modality may significantly improve these therapies, from patient stratification to posttherapy monitoring. Therefore, in this study, we aimed to develop an AML-specific diagnostic imaging modality; we hypothesized that an anti-CD33 mAb would be an ideal candidate for immuno-PET imaging of AML.

Previous clinical studies have reported whole-body-gamma imaging of patients with myeloid leukemia using anti-CD33 mAb (HuM195) conjugated with Iodine-131 and Bismuth-213, but these studies were not intended for diagnosing AML but rather studying pharmacokinetics and the targeting of the mAb to the bone marrow, the primary site of the disease (7–10). Additionally, the imaging modality used was planar or 2D and hence only qualitative.

On the other hand, whole body PET-CT is an important dual-imaging modality used in nuclear medicine. In PET-CT diagnostic 3D imaging, PET imaging provides physiological and biochemical quantitative information to identify normal versus malignant lesions, and inclusion/optimization of the whole body microCT imaging may provide spatial distribution of the disease in the whole body. Furthermore, PET-CT is currently used in several investigative noninvasive whole body diagnostic imaging studies of hematologic and nonhematologic malignancy (11). Particularly in AML, there are several clinical trials ongoing using PET-CT diagnostic imaging (e.g., NCT02392429 and our NCT03422731), suggesting an unmet clinical need for an AML-specific imaging modality.

The primary goal of this study was to evaluate CD33 as a diagnostic imaging marker of AML using an anti-CD33 mAb in a preclinical mouse model. For proof of concept, we first used the murine [⁶⁴Cu]Cu-DOTA-anti-CD33 (clone p67.6) mAb for immuno-PET-CT imaging of CD33⁺ AML cells (diagnostic) in a mouse xenograft model. However, besides detecting AML, this imaging modality interestingly also provided information about the spatial distribution of disease in the whole body. Then, with a translational objective, we produced a humanized anti-CD33 mAb (clone Hu-M195) and showed detection of CD33⁺ AML in preclinical models using the newly developed CD33 PET-CT imaging.

Murine antihuman CD33 clone p67.6 is an IgG1 kappa mAb that targets human CD33-positive cells of the myeloid lineage (12, 13). The hybridoma was produced in a hollow fiber bioreactor and purified by Protein G and cation exchange chromatography. Antihuman CD33 (Clone WM 53, #555450) was obtained from BD Biosciences. Antihuman CD45 (clone: 2D1, #368516) was obtained from BioLegend.

The murine anti-CD33 M195 mAb was humanized by CDR grafting to reduce human antimouse antibody responses (14, 15). The scFv was reformatted to a human IgG1 antibody by cDNA synthesis, transiently expressed in HEK293 cells and purified by Protein A chromatography.

The mouse and humanized antihuman CD33 mAb were conjugated with the metal chelator 1,4,7,10-tetraazacyclododecane-N,N′,N″,N′″-tetraacetic acid (NHS-DOTA; Macrocyclics) as described previously (16). The detailed methods are provided in the Supplementary Materials and Methods section.

Human AML cells (HL60, MV4-11, Kg1a), multiple myeloma cells (MM.1S), and Daudi cells (lymphoma) were cultured using standard tissue culture condition techniques. The use of human samples was approved by the City of Hope Institutional Review Board, and informed consent was obtained from patients. Human subjects research was in accordance with the Declaration of Helsinki. Detailed information regarding the cells and culture conditions are described in the Supplementary Materials and Methods section.

Flow cytometry was used to analyze CD33 and CD45 expression in human AML (MV4-11, HL 60, and Kg1a) and multiple myeloma cells (MM1S), using mouse antihuman CD33 and CD45 antibodies. Staining and flow cytometry analysis were performed as per standard protocols. Data were acquired from a BD Fortessa cytometer and analyzed with FlowJo V 10.0 software.

Mice bearing CD33⁺ AML cells (MV4-11, HL-60) or CD33⁻ MM cells (MM.1S) or nonleukemic control mice were injected intravenously with murine [⁶⁴Cu]Cu-DOTA-anti-CD33 (100 μCi/10 μg), or [⁶⁴Cu]Cu-DOTA-anti-CD33 (100 μCi/10 μg) + 500 μg of unlabeled anti-CD33-DOTA (1:50). For the humanized anti-CD33 mAb, human IVIg was injected (∼1 mg/mouse, i.p.) 2 to 3 hours prior to injecting humanized [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb (100 μCi/1 μg). However, murine mAb did not necessitate human IVIg for imaging and biodistribution studies. Each group consisted of at least 5 mice; representative data for each group are presented. Static PET scans were acquired at 1 day (40-minute scan) and 2 days (60-minute scan, with whole body CT at 100 μm resolution) postinjection using InVeon PET-CT (Siemens). MicroCT parameters were adjusted (voltage 80 kV, current 500 μA, 0.5 mm Al filter, 220 projection, scan time 264 seconds) to obtain anatomical details of skeleton. Image reconstruction was performed using Inveon Acquisition Workplace (Siemens). Post-processed image display and analysis was performed by Vivoquant (InviCro). For biodistribution studies, mice were euthanized at 24 hours and/or 48 hours, as both time points showed activity. Various organs were obtained from control and AML leukemia-bearing mice. Wet weighs of each organ were determined, and radioactive counts from each tissue/organs were measured using a WIZARD2 automatic gamma counter (PerkinElmer). Biodistribution and imaged PET activity of [⁶⁴Cu]Cu-DOTA-anti-CD33 was presented as the percent injected dose per gram of organ/tissue (% ID/g). MV4-11 with cold-CD33 block and CD33⁻ MM.1S served as negative controls. [⁶⁴Cu]Cu-DOTA-anti-CD33-PET-CT imaging and biodistribution studies were conducted at least twice independently and the data are representative of one experiment (n ≥ 5).

All animal experiments were carried out in accordance to the guidelines of Institutional Animal Care and Use Committee (IACUC). The NSG (NOD-scidIL2Rg^null) mice were purchased from the Jackson laboratory and were bred in a City of Hope animal breeding facility. The NSG mice were treated with 1 to 2 Gy radiation 24 hours before transplant as a preconditioning regime to ensure faster engraftment. Human AML and multiple myeloma cells (1–2 × 10⁶ cells) were injected via tail vein, and engraftment was determined using bioluminescent imaging (BLI) 7 to 10 days posttransplant. Biodistribution and [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb imaging studies and immunotherapy interventions were performed around 2 to 3 weeks posttransplant.

Humanized anti-CD33-DOTA mAb was radiolabeled as mentioned before with Indium-111 (Triad; Isotope). The blood activity clearance of murine and humanized [¹¹¹In]In-DOTA-anti-CD33 mAb was tested by collecting blood from tail veins at different points post injections (0, 2, 4, 24, 48, 72 hours) and using a gamma counter. The detailed method is described in the Supplementary Materials and Methods section.

The in vivo CD33 targeting potential of humanized anti-CD33 mAb as an ADC was evaluated indirectly using anti-human IgG (Fc) 2° Ab conjugated to monomethyl auristatin E (MMAE; Moradec LLC). The in vitro cytotoxicity assay was performed using a 1:1 molar ratio of anti-CD33 and 2°-ADC antibody. The primary and secondary antibody were first allowed to bind (∼15 minutes) in complete IMDM medium and were then added to cells and incubated at 37°C, 5% CO₂. The cytotoxicity was assayed after 5 days by measuring Annexin V and propidium iodide (PI) staining in these cells. The antibody amount used ranged from 1 ng to 2.5 μg. As control, CD33⁻ Daudi cells (lymphoma) and secondary Ab-ADC alone (1 μg) were used.

For in vivo study, 2.5 μg of humanized anti-CD33 antibody and 2.5 μg of secondary Ab-ADC were mixed, incubated for 15 minutes at 4°C, and then resuspended in 100 μL of PBS 0.1% BSA and retro-orbitally injected into mice bearing MV4-11 leukemia on day 0 and day 3. As controls, leukemic mice were untreated or given secondary Ab-ADC alone. Disease progression was monitored using BLI at indicated time points. The whole body of the mouse was contoured, and average BLI intensity (photons/s) for each group was plotted at different time points.

Humanized anti-CD33-DOTA mAb was radiolabeled as mentioned before with Actinium-225 (Oak Ridge National Laboratory) at a specific activity of 50 nCi/μg. For therapy, 3 different activities, 100, 200, and 300 nCi of [²²⁵Ac]Ac-DOTA-anti-CD33, were used, with one single dose given to AML-bearing NSG mice (n = 5). Untreated AML-bearing mice served as controls.

Statistical analysis was performed using ANOVA and the Student t test. The Pearson correlation coefficients were calculated assuming a Gaussian distribution. The difference was considered significant when the P value was <0.05. All graphs and statistical analysis were generated using GraphPad Prism software V 7.2.

Detailed methods for antihuman CD33 (clone P67.6 Hu-M195) mAb DOTA and Copper-64 conjugation, CD33 surface expression quantification, BLI, lentiviral production and transduction in MV4-11 cells, CD33 antibody immunoreactivity, in vitro and in vivo stability, and blood clearance and diagnostic accuracy calculations are provided in the Supplementary Materials and Methods section.

For use as an immunoPET tracer, the antihuman CD33 mAb, mouse hybridoma P67.7, was produced, purified, conjugated to the metal chelate DOTA, and radiolabeled with Copper-64. Coomassie stained SDS-PAGE gel electrophoresed under reducing conditions showed the purity, with 2 bands corresponding to the light and heavy chains (Supplementary Fig. S1A). Post-DOTA conjugation, iso-electrofocusing gel analysis revealed a shift to a more acidic pH, confirming the conjugation process (Supplementary Fig. S1B). The radiolabeled [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb was analyzed by size exclusion chromatography (SEC), and the radiochromatogram showed a single peak, with a retention time corresponding to a mAb (Supplementary Fig. S1C). The immunoreactivity of DOTA-anti-CD33 mAb was evaluated by incubating the [⁶⁴Cu]Cu-DOTA-anti-CD33 with soluble CD33-Fc antigen and analyzed by SEC. The radiochromatogram showed a faster retention time (∼30 minutes), indicating an increase in molecular size consistent with binding to CD33 Fc soluble antigen (67–85 kDa; Supplementary Fig. S1D and S1E). The [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb was shown to be stable in serum both in vitro and in vivo for 48 hours (Supplementary Fig. S1F and S1G).

The murine DOTA-anti-CD33 mAb and a validated commercial anti-CD33 antibody (clone WM53) were tested for binding and specificity by flow cytometry using CD33 positive MV4-11, HL-60 cells, Kg1a, THP-1, and the CD33-negative MM.1S-GFP cells. The DOTA-anti-CD33 antibody showed immunoreactivity towards CD33-positive AML cell lines but not in CD33-negative MM.1S cells (Supplementary Fig. S2A and S2B). The immunoreactivity of the antibody was further confirmed by immunofluorescence: HL60 cells showed brighter staining than MV4-11, whereas negative control MM.1S showed no staining (Supplementary Fig. S2C). The total number of CD33 cell surface receptors per cell was determined using the BD Quantibrite PE Kit (Supplementary Fig. S3A–S3C). The HL-60 AML cell line expressed ∼55,000 cell surface CD33/cell, versus ∼26,000/cell in MV4-11 (Supplementary Fig. S3C).

[⁶⁴Cu]Cu-DOTA-anti-CD33 mAb immuno PET-CT in vivo imaging was performed in NSG mice bearing CD33⁺ AML cells and CD33⁻ MM cells, and in nonleukemic control mice. Mice were injected with [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb (100 μCi/10 μg) 24 to 48 hours prior to PET imaging. As an additional control for specificity, [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb (100 μCi/10 μg) + 500 μg of unlabeled DOTA-anti-CD33-mAb (1:50) were injected into mice bearing CD33⁺ AML cells. Bioluminescent imaging of AML and MM.1S cells in NSG mice was carried out 24 hours prior to [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb injections. The engraftment of AML and multiple myeloma cells was also measured in the femur, L-spine, and spleen using flow cytometry upon harvesting tissues/organs after imaging and biodistribution studies. 3D intensity projection based PET-CT images clearly show CD33⁺ activity only in MV4-11 AML mice, but not in mice given cold-blocked anti-CD33, mice with MM.1S, and nonleukemic control mice (Fig. 1A and B), whereas the BLI images of the respective mice shows engraftment (Fig. 1C). CD33⁺ PET activity was detected in the epiphysis/metaphysis of the femur, tibia, and humerus; L-spine; and pelvic bone in CD33⁺ MV4-11 bearing mice (Fig. 1D). Although BLI and flow cytometry indicated high levels of myeloma cells in MM.1S bearing mice, there was no detectable CD33⁺ activity; similarly, the mice injected with cold-blocked unlabeled CD33 antibody showed no CD33⁺ PET signal, suggesting high specificity (Fig. 1A–C).

Biodistribution studies were carried out in mice from respective groups by harvesting different organs/tissues 24 to 48 hours postinjection of the [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb, and the activity was measured using a gamma counter. The %ID/g of tissue/organs was plotted to determine the activity. The biodistribution of [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb was similar in all groups for blood, liver, lung, heart, muscle, intestine, and kidney (Fig. 2A and B). However, the %ID/g was particularly high in bones (femur, humerus, tibia, and lumber spine) in CD33⁺ bearing MV4-11 AML mice, but not in CD33⁻ MM mice, cold-blocked AML mice, or control nonleukemic mice (Fig. 2C). The mean %ID/g in all skeletal sites showed no significant difference between control, MM.1S bearing mice, and cold blocked MV4-11 bearing mice. However, there was a significant difference in mean %ID/g between control and MV4-11 AML bearing mice in skeletal sites: femur (2.4%ID/g vs. 4.77%ID/g, P < 0.00001); humerus (2.8%ID/g vs. 5.83%ID/g, P < 0.00001); tibia (2.35%ID/g vs. 4.07%ID/g, P ≤ 0.00001) and L spine (2.1%ID/g vs. 4.41%ID/g, P < 0.00001). There was an average 2-fold increase in PET signal in CD33⁺ AML bearing mice over nonleukemic control mice. [⁶⁴Cu]Cu-DOTA-anti-CD33 targeting was validated independently using a second AML cell line, HL-60, and similar biodistribution results were obtained in the skeleton: femur (2.3%ID/g vs. 9.9%ID/g P < 0.00001), humerus (2.7%ID/g vs. 10.3%ID/g, P < 0.00001), L spine (2.1%ID/g vs. 8%ID/g, P = 0.0013), and tibia (2.2%ID/g vs. 8%ID/g, P = 0.0015; Fig. 2D and E). However, there was an average 4-fold increase in CD33⁺ activity in skeletal sites of HL-60 bearing mice.

The sensitivity and specificity were calculated from biodistribution of the [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb injected AML mice. The ROC curves showing sensitivity versus 100-specificity for biodistribution data from the femur indicate high sensitivity (∼95%) and specificity (100%; Fig. 3A).

Further, the femur and L-spine of AML-bearing mice were contoured in PET-CT images. The extent of engraftment from these bones was then determined by flow cytometry of human CD45⁺ cells/MV4-11-RFP. A high correlation was observed between PET-CT signals and percent engraftment (R² value for femur and L-spine is 0.9854 and 0.8027, respectively; Fig. 3B and C). Additionally, the femur of AML bearing mice was contoured both in BLI and PET-CT images, and a high correlation were observed between BLI and PET-CT signals (R² = 0.9262; Fig. 3D). However, spatial localization was evident in PET-CT in comparison to BLI images, which were diffuse.

Besides detecting a CD33⁺-specific signal, anti-CD33 PET-CT imaging also indicated the spatial heterogeneity of AML. The adjustment in whole body MicroCT allowed a better anatomical visualization of the skeletal system (Supplementary Fig. S4A). CD33⁺ PET activity was significantly heterogeneous within the femur; for example, the distal and proximal femur showed higher CD33 activity compared with that in the long bone area in mice with low leukemia burden; however, heightened activity in the long bone was observed when leukemic burden increased (Fig. 4A; Supplementary Fig. S4B and S4C). A similar localization pattern was seen in other bones including the tibia, humerus, and L-spine (Fig. 1D). CD33 activity was mostly concentrated in the proximal/distal end of the bone. Further supporting the concept of heterogeneous spatial distribution of AML, in the L-spine (Fig. 4B) at areas of low leukemic burden (5%–10%), the disease is strongly localized to one of the segments (L1–L5) of the L-spine. However, with increasing leukemic burden (>15%), the leukemia is distributed across the L-spine and can be seen even in the T-spine, which was initially undetectable in the low leukemia group (Fig. 4B), suggesting a preferential niche for AML during the early stages of disease progression.

After validating the use of CD33 mAb as an imaging agent, with the intent of translating this promising technology to humans, we produced a humanized anti-CD33 mAb, as described in the Materials and Methods section. The radiolabeling, immunoreactivity and stability (48 hours, in vivo) were analyzed as mentioned for the murine version (Supplementary Fig. S5A–S5D). We tested the humanized [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb immunoreactivity and found that it bound specifically to CD33⁺ AML cell lines (MV4-11, HL-60, Kg1a, THP-1) and human patient samples (Supplementary Fig. S6A and S6B). We then compared the murine and humanized CD33 mAb blood retention and the ability of the mAb to detect AML in mice bearing CD33⁺ AML cells. Significant differences between humanized and murine antibody clearance were observed at 4, 24, 48, and 72 hours postinjection (P = 0.0021, 0.00044, 0.00073, and 0.00015, respectively), and faster blood pharmacokinetics was observed with the humanized antibody. The concentration of humanized anti-CD33 mAb in blood dropped to <5% of initial concentration within 24 hours postinjection, whereas the murine antibody maintained >25% activity of initial concentration even at 72 hours postinjection (Supplementary Fig. S6C).

The humanized [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb PET-CT images (Fig. 5A and B) clearly show targeting of AML cells to the skeletal system similarly to that of the murine Ab, and the respective BLI image showed AML engraftment (Fig. 5C). As in the murine Ab PET imaging, humanized mAb also signaled CD33⁺ PET activity localized to the epiphysis/diaphysis of the bones, which was dependent on disease burden, indicating spatial heterogeneity (Fig. 5D). The humanized Ab also revealed CD33⁺ activity in the spleen of mouse, with ∼20% AML engraftment (Fig. 5A), suggesting its ability to detect extramedullary disease. Biodistribution studies were carried out in mice from respective groups as mentioned before. There was no significant difference in the biodistribution of [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb in both groups in the heart, lung, muscle, and kidney (Fig. 5E); however, the signal was significantly reduced in the blood while increased in the liver. Also, noticeably, there was reduction of the PET signal in the heart of the AML-bearing mice with the humanized antibody in comparison to that from the murine Ab, which may be due to lower blood activity. The %ID/g was particularly high in bones in CD33⁺ bearing AML mice, but not in control nonleukemic mice: femur (2.4%ID/g vs. 22.4%ID/g, P < 0.0001), humerus (2.2%ID/g vs. 14.5%ID/g, P = 0.004), L spine (1.7V vs. 15.2%ID/g, P < 0.0001), and tibia (1.9%ID/g vs. 13.4%ID/g, P < 0.0001; Fig. 5F). An average of 7- to 10-fold higher %ID/g PET activity was observed in bones of AML-bearing mice over non-AML–bearing control mice. Furthermore, the humanized mAb (21.2%ID/g) showed nearly 3 to 4 times higher CD33⁺ activity in bones than with murine mAb (4.7%ID/g; Supplementary Fig. S6D).

Additionally, to further asses in vivo CD33-specific targeting of the humanized anti-CD33 mAb, as a proof of concept we have used 2 methods, first an ADC, an antihuman IgG (Fc) 2° Ab conjugated to MMAE, and second, [²²⁵Ac]Ac-DOTA-anti-CD33 radioimmunotherapy (CD33-RIT). The in vitro cytotoxicity assay on MV4-11 cells using anti-CD33 mAb + 2°-ADC Ab (CD33-2°-ADC) was carried out as mentioned in the Materials and Methods section. As control, 2°-ADC Ab alone (1 μg) was used, which showed no significant cell killing over untreated cells, whereas the same concentration of CD33-2°-ADC yielded ∼55% cell killing (Fig. 6A), and maximum (>95%) cell killing was observed at 55 nmol/L (2.5 μg). Lymphoma cells (Daudi) were used to show that the CD33-2°-ADC had no significant cell killing on the CD33⁻ cell line, indicating specificity. For in vivo study, we used 2.5 μg CD33-2°-ADC, which showed maximum cell killing in the in vitro study. The MV4-11 AML-bearing mice treated with CD33-2°-ADC showed lower BLI intensity 24 days posttreatment than in mice treated with 2°-ADC alone (P = 0.02) or in the untreated AML-bearing mice (P = 0.0008; Fig. 6B; Supplementary Fig. S7A).

Next, CD33-RIT was carried out using 3 different Actinium-225 activities (100, 200, and 300 nCi). Similar to the ADC study, CD33-RIT treated mice showed a significant decrease (P < 0.00001) in leukemia burden 14 days post intervention (Fig. 6C; Supplementary Fig. S7B). These results suggest that the humanized anti-CD33 antibody targets CD33⁺ AML cells and can be used in anti-CD33 therapy.

Currently, a cGMP-grade humanized antibody is under preparation for a planned phase I/II clinical trial of anti-CD33 PET-CT imaging of AML. Efforts are also in progress for generating a primary anti-CD33 mAb-ADC, to explore treatment response monitoring post anti-CD33 immunotherapy using our newly developed anti-CD33 PET-CT imaging.

We have developed a noninvasive anti-CD33 immunoPET-CT imaging method for the in vivo detection of AML disease with high sensitivity and specificity. Diagnostic and prognostic markers facilitate stratifying patients for treatment management. However, the current clinically approved diagnostic method is invasive, relying on single point (iliac crest) biopsies, which may (i) not always be representative of the actual disease state, and (ii) limit the number of times it is performed. Therefore, in leukemia, noninvasive PET-imaging using PET tracers including ¹⁸F-flourodeoxyglucose (FDG; metabolic activity) and ¹⁸F-flourothymidine (FLT; cell proliferation) have been tested for diagnosis and monitoring treatment response (11, 17–19). Although ¹⁸F-FDG-PET has shown some success in diagnosing extramedullary disease in AML (20), it has also yielded highly inconsistent results because of changes in the metabolic activity of normal/tumor cells post treatment (21, 22). However, these PET tracers are nonspecific in that they detect metabolically active or highly proliferating cells, which could be affected by treatment, a contrast to CD33-PET imaging, which specifically detects cells expressing CD33, an accepted biomarker for AML. Studies have shown that even relapsed AML blasts post anti-CD33 therapy have CD33 expression (23, 24). Therefore, CD33-based imaging is expected to improve detection of AML with high specificity in the whole body. This approach may also provide a blueprint for selecting biopsy sites (image-guided) that would be greatly useful for early detection and also in longitudinal monitoring of treatment response.

In this study, [⁶⁴Cu]Cu-DOTA-anti-CD33 mAb PET-CT was used to detect AML in vivo, showing very high specificity and sensitivity to CD33⁺ AML. The spatial information was achieved with whole body CT-based 3D anatomical imaging. Notably, a spatial heterogeneity in the distribution of AML within the skeletal regions was observed, which would not be established with point biopsies. For example, CD33 activity was especially prominent in the L-spine and epiphysis/diaphysis of femurs, indicating a preferential skeletal niche for the disease at early stages. However, when the disease progresses, this spatial heterogeneity was reduced, and the disease appeared to be more systemic. This finding suggests, contrary to previous beliefs that leukemia was a systemic disease, that leukemia may be initiated in multiple preferential niches, characterized by multifocal disease before becoming more systemic. This imaging observation warrants future investigation to understand the role of the microenvironment in fostering preferential disease localization. Also, a previous study has shown that chemoresistance was due to localization of leukemic cells to the endosteal niche in the bone (25), further indicating the relevance and importance of spatial localization of the disease in the whole body. One small clinical study suggested that a heterogeneous distribution of leukemia may indicate resistance to chemotherapy, as shown through ¹⁸F-FLT-PET-based imaging of cell proliferation (26). Also, a recent study indicated that ¹⁸F-FLT-PET could identify the risk of early relapse in the spine prior to evidence of relapse by detection of minimal residual disease (MRD; ref. 27). Therefore, future clinical studies may be designed to assess the prognostic value of this anti-CD33 oncoPET imaging. As immuno-PET combines the sensitivity and resolution of a PET imaging system with the high specificity of a mAb, it may represent what is described as a “comprehensive immunohistochemical staining in vivo” (28), therefore guiding precise locations for biopsies. Also, precisely specifying sites of disease could yield valuable information about the tumor in terms of location, phenotype, treatment susceptibility, and response (29). The spatial heterogeneity of leukemia also suggests caution should be taken when interpreting data from single-point biopsies.

As the initial study was conducted using a murine anti-CD33 mAb, clone p67.6, for translational purposes, a humanized anti-CD33 mAb (Hu-M195) was generated. We carried out preclinical PET-CT imaging using humanized anti-CD33 mAb and showed that targeting was significantly improved over that of the murine version. Currently, a cGMP grade mAb has been made for future phase I/II clinical trials. The Hu-M195 mAb labeled with Iodine-131 or Bismuth-213 has been shown to be useful for whole-body-gamma imaging of patients with myeloid leukemia (7–10, 30). However, this modality is planar or 2D and is hence qualitative and lacks 3D quantitative information with spatial heterogeneity assessment of the disease. Also, none of the previous whole-body gamma imaging studies were conducted with an intent for AML diagnosis. They were used for studying targeting and pharmacokinetics of the radiolabeled anti-CD33 mAb. In contrast, our study was mainly conducted to evaluate CD33 as a noninvasive diagnostic imaging marker for AML. Additionally, the imaging modality used is PET-CT, which is more advanced than the gamma imaging conducted previously. Our study used PET-CT 3D imaging, which is quantitative and provides high-resolution images that reveal information about spatial distribution of the CD33⁺ AML in the whole body, hence resulting in a significantly improved diagnostic imaging of AML.

Additionally, we showed CD33 targeting of HuM195 anti-CD33 mAb using CD33-2°-ADC or CD33-RIT. Both treatment studies resulted in reduced disease burden in treated groups over untreated AML bearing mice, suggesting CD33-specific targeting. Such approaches are very effective against receptors like CD33, which is endocytosed after Ab binding, enabling the cytotoxic payload to internalize and result in cell killing (31, 32). Furthermore, after CD33 antibody therapy, blasts in relapsed/refractory disease continue to express CD33, suggesting that AML cells may not use downregulation of CD33 as an epigenetic “escape mechanism” (24), in contrast to the use of anti-CD38 mAb daratumumab which may result in relapse due to a CD38⁻ myeloma clone (33). Anti-CD33 mAb has been clinically validated as an ADC (34–36) and for RIT (37–39, 40). Therefore, CD33 PET imaging may also be useful in monitoring treatment response post intervention (such as ADC, RIT, or any other treatment modalities).

Several investigational CD33-targeted therapeutics have been developed for AML (41), including the recently FDA-approved anti-CD33 monoclonal ADC gemtuzumab ozogamicin (GO; Mylotarg; ref. 42), bispecific antibodies such as CD33/CD3 antibodies (43), and CD33/CD123-directed CAR T cells (44). Although positive results have been reported from clinical trials of CD33-targeting drugs, dose-limiting toxicities such as hepatotoxicities (45) during treatment are of concern. Therefore, an imaging tool as presented here would help determine optimal doses as well as monitoring treatment response noninvasively, providing an opportunity to improve treatment efficacy while decreasing toxicities.

There are some limitations of this study. Although CD33 is expressed in all patients on at least a subset of AML blasts, high interpatient variability in terms of level of expression (>2-fold) is known (6, 46, 47). This differential expression is not random but correlated with the presence of particular mutations; for example, very high CD33 expression has been shown in AML with NPM1 and FLT3/ITD mutations, whereas lower expression has been associated with core binding factor translocations (6, 47, 48). A properly controlled imaging study with subset (mutation) analysis will be essential to understand whether specific mutations impact imaging intensity and/or heterogeneity.

In conclusion, to the best of our knowledge, this is the first preclinical report of an anti-CD33-PET-CT imaging modality to successfully detect AML in vivo. The imaging tool may be used for diagnosis as well as monitoring treatment response in the whole body, including extramedullary organs. The molecular imaging method also detected heterogeneity in the spatial distribution of the AML, warranting caution in interpreting results from single-point biopsies. Therefore, the CD33 PET imaging modality also could provide guidance for preferential sites for biopsy when there is large disease heterogeneity, and it may also indicate location of chemo/radio-resistant leukemic cells posttreatment that have been suggested to be responsible for relapse (25). Future studies to understand the relationship between genomic mutation (like NPM1, FLT3-ITD, etc.) and spatial heterogeneity (as observed in the CD33 PET-CT imaging) would be beneficial in the diagnosis and prognosis of AML. In summary, anti-CD33 PET-CT is a viable AML diagnostic imaging method with significant translational potential.

A.S. Stein reports receiving speakers bureau honoraria from Amgen, Celgene, and Stemline, and is an advisory board member/unpaid consultant for Amgen. D.M. Colcher holds ownership interest (including patents) in NIH-NCI. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Sargur Madabushi, J. Brooks, D. Zuro, B. Kumar, P. Vishwasrao, A. Miller, J.Y. C. Wong, A.S. Stein, D. Colcher, P.J. Yazaki, S.K. Hui

Development of methodology: S. Sargur Madabushi, J. Brooks, D. Zuro, B. Kumar, P. Vishwasrao, K. Poku, N. Bowles, A. Miller, J. Molnar, D. Colcher, P.J. Yazaki, S.K. Hui

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Sargur Madabushi, J. Brooks, D. Zuro, M. Orellana, P. Vishwasrao, I. Nair, J. Chea, K. Poku, N. Bowles, D. Colcher, P.J. Yazaki

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Sargur Madabushi, J. Brooks, J. Sanchez, P. Vishwasrao, J. Chea, K. Poku, J.Y. C. Wong, J.E. Shively, P.J. Yazaki, S.K. Hui

Writing, review, and/or revision of the manuscript: S. Sargur Madabushi, J. Brooks, D. Zuro, J. Sanchez, P. Vishwasrao, K. Poku, J. Rosenthal, J.Y. C. Wong, A.S. Stein, D. Colcher, P.J. Yazaki, S.K. Hui

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Sargur Madabushi, J. Brooks, D. Zuro, L.E. Parra, P. Vishwasrao, I. Nair, J. Chea, K. Poku, T. Ebner, D.A. Vallera, J.Y. C. Wong,

Study supervision: S. Sargur Madabushi, L.E. Parra, P. Vishwasrao, P.J. Yazaki, S.K. Hui

Other (did everything related with mice work (treatment, supervision of health status, weight, blood collection, necropsies, tissue collection, and processing among others): L.E. Parra

Other (bench work): M. Orellana

Other (the Principle Investigator of the CD33 imaging and theranostics for AML): S.K. Hui

We appreciate administrative support at our institutions. Research reported in this publication included work performed by the Small Animal Imaging Core for PET-CT imaging and imaging precision radiation delivery system supported by the NCI of the NIH under award number P30CA033572 and partly supported by NIH grant 1R01CA154491-01. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.