Noninvasive Interrogation of DLL3 Expression in Metastatic Small Cell Lung Cancer

Small cell lung cancer (SCLC) is an extremely lethal form of lung cancer that accounts for 16% of all lung cancer cases diagnosed annually in the United States (1, 2). The majority of SCLC cases are metastatic at the time of diagnosis, necessitating effective systemic therapies (3). The standard of care for extensive stage SCLC consists of a combination of the cytotoxic agents, cisplatin or carboplatin with etoposide (4). While patients respond to these therapies, they often relapse shortly after the cessation of treatment, accounting in part for a dismal 5-year survival rate of 2% (5). These outcomes have not meaningfully improved in more than 3 decades (6). There are no FDA-approved targeted therapies for SCLC, and no approved therapies of any kind beyond topotecan in the second line. There is a desperate need for novel systemic therapies as well as diagnostic approaches to direct these therapies to the patients most likely to benefit from them.

We recently reported that delta-like protein 3 (DLL3) is a highly tumor-selective cell surface protein expressed in high-grade neuroendocrine lung tumors including SCLC (7). DLL3 is an inhibitory ligand of the Notch signaling pathway that is normally expressed exclusively on intracellular membranes, including those of the Golgi apparatus (8). However, marked induction of DLL3 expression in SCLC results in localization to the cell surface: this together with the absence of detectable cell surface DLL3 in nonmalignant cells opens a new window of opportunity for tumor cell–specific therapy. Of particular relevance to SCLC, DLL3 is implicated in the regulation of clonogenic and tumorigenic capacity (7). Exceptionally high clonogenic capacity, early metastatic spread, and rapid tumor repopulation after exposure to chemotherapy are hallmark features of SCLC (9).

Rovalpituzumab teserine (Rova-T; SC16LD6.5) is a DLL3-targeted antibody–drug conjugate (ADC) comprising a humanized anti-DLL3 mAb [rovalpituzumab (SC16)], a cleavable dipeptide linker, and a cell-cycle–independent pyrrolobenzodiazepine (PBD; D6.5) toxin (7, 10). SC16LD6.5 was reported to selectively target DLL3-expressing cells compared with an isotype-matched control antibody formulation. Treatment with the DLL3-targeted ADC completely and durably eradicated SCLC patient-derived xenografts (PDX) expressing high levels of DLL3 in a variety of preclinical models including those resistant to cisplatin and etoposide. Furthermore, a recently completed first-in-human phase I trial of Rova-T in patients with relapsed SCLC demonstrated encouraging clinical outcomes (11). Among the 67% of patients with >50% of cells expressing DLL3 (DLL3^Hi), a confirmed objective response rate of 39% and confirmed disease control rate of 89% were observed (11). Importantly, all patients with confirmed objective responses by investigator assessment were DLL3^Hi by IHC. This early clinical experience highlights the significance of DLL3 assessment as a predictive biomarker for DLL3-targeted agents.

Despite these encouraging early clinical results, IHC suffers from several limitations that may reduce its effectiveness as a clinical diagnostic for DLL3-targeted therapies. These limitations include (i) the lack of contemporaneous tissue biopsy, an especially acute problem in aggressive carcinomas like SCLC, where multiple biopsies are rarely performed; (ii) the sampling bias caused by intratumoral heterogeneity or heterogeneity between the primary tumor and metastases; and (iii) the inherently high false-negative rate of histopathologic assessment. Such limitations have led to the recent emergence and application of immuno-positron emission tomography (immunoPET) as a more reliable approach for the noninvasive evaluation of tumor-associated antigen expression in vivo (12). ImmunoPET may reflect physiologic drug binding more accurately than IHC of tumor sections ex vivo due to factors such as high intratumoral oncotic and hydrostatic pressures and variable perfusion that can limit delivery of antibody-based therapeutics. An increasing number of immunoPET strategies are now being translated into oncologic imaging protocols for patient evaluation prior to treatment with antibody-based targeted therapeutics (13–15). This trend can be attributed to the exquisite specificity of antibodies for tumor-associated molecular targets/antigens combined with the sensitivity and quantitative nature of PET (16).

We envisaged that a real-time, noninvasive, and quantitative approach to evaluate the in vivo status of DLL3 expression in patient tumors would have immediate clinical utility in the context of DLL3-targeted therapies. To this end, we have developed a ⁸⁹Zr-labeled, DLL3-targeted immunoconjugate leveraging the humanized antibody, SC16, to serve as a companion diagnostic immunoPET agent in neuroendocrine carcinoma patients. Here we report the PET imaging performance of this agent in preclinical mouse models of SCLC.

Gene expression data from 2,712 normal samples representing 55 different organs was downloaded from the Genotype-Tissue Expression (GTEx) project (release V4) as reads per kilobase of transcript per million mapped reads (RPKM). Raw RNAseq reads from primary SCLC and normal lung were aligned to the human reference genome GRCh37 with TopHat v1.1.4 assisted by GENCODE transcript model v18. RPKM values were calculated using RNA-SeQC v1.1.8 run in strict mode, consistent with published analysis methods for GTEx release V4. As both datasets include samples of normal lung, it was possible to apply batch effect correction prior to further analysis (17).

Robust Multi-array Average (RMA) and quartile normalized gene expression microarray data for 1,037 cancer cell lines was downloaded from the Cancer Cell Line Encyclopedia (CCLE). Two SCLC cell lines, H82 and H69, were identified as being representative of median and low DLL3 expression, respectively. A gene expression cutoff was set for this analysis using the approach of Zilliox and colleagues (18). A549, a non–small cell lung cancer line exhibiting DLL3 expression below the cutoff, was used as a negative control in the in vitro and in vivo experiments to validate the DLL3-specific tumor uptake of the SC16-based radioimmunoconjugates. All downstream analysis and figure generation was performed using the R statistical computing environment.

The cell lines used in this study were obtained from the ATCC in 2009 and grown under aseptic conditions in an incubator providing humidified atmosphere of 5% CO₂ in air. All cell lines were used between 3 and 9 passages after thawing to ensure complete revival. In addition to routine testing for mycoplasma, the identity and purity of the cell lines was validated by STR profiling. H82 and H69 cells were cultured in RPMI base media supplemented with 10% FBS, 2 mmol/L l-glutamine, 10 mmol/L HEPES, 1 mmol/L sodium pyruvate, 4,500 mg/L glucose, 1,500 mg/L sodium bicarbonate, and 100 U/mL penicillin and 100 μg/mL streptomycin. To facilitate the in vivo visualization and assessment of the growth and spread of tumor cells in orthotopic and metastatic tumor models for SCLC, H82 cells were transduced with firefly luciferase in the pLENTI6 lentiviral transfer vector (Invitrogen). After blasticidin selection, the resulting cell line, luc-H82, was used for in vivo bioluminescence imaging experiments. A549 cells were cultured in F-12K base media supplemented with 10% FBS and 100 U/mL penicillin and 100 μg/mL streptomycin.

All animals were treated according to the guidelines approved by the Research Animal Resource Center and Institutional Animal Care and Use Committee at Memorial Sloan Kettering Cancer Center (New York, NY). Four types of animal models were used in this study: (i) subcutaneous cell line xenografts; (ii) orthotopic cell line xenografts; (iii) subcutaneous PDX; and (iv) a distant organ metastasis model. See Supplementary Information for details.

A DLL3-targeted humanized mAb, SC16 (Stemcentrx Inc.), was chemically modified and radiolabeled with ⁸⁹Zr to generate a nonsite-selectively modified radioimmunoconjugate (⁸⁹Zr-DFO-SC16-NS). A single mutation in the heavy chain of SC16 was introduced to facilitate an unpaired cysteine that could be leveraged for site-selective, maleimide-based conjugation (⁸⁹Zr-DFO-SC16-SS). Both antibodies were used in the PET imaging of DLL3 expression in preclinical models of SCLC. See Supplementary Information for details.

PET imaging of the subcutaneous and orthotopic cell line xenograft mouse models of SCLC was conducted on a microPET Focus rodent scanner (Concorde Microsystems; ref. 19), whereas PET imaging of animals in the metastatic cohort was conducted on an Inveon PET/CT scanner (Siemens). The various xenografts were administered ⁸⁹Zr-DFO-SC16 radioimmunoconjugates [9.25–11.1 MBq; (250–300 μCi), 31–38 μg in PBS treated with Chelex resin] via intravenous tail vein injection (t = 0). See Supplementary Information for details.

In vivo biodistribution studies were performed to evaluate the uptake of the nonsite-selective versus site-selectively labeled ⁸⁹Zr-DFO-SC16 radioimmunoconjugates in mice bearing subcutaneous xenografts of H82, H69, A549, and PDX tumors. An isotype control radioimmunoconjugate – ⁸⁹Zr-DFO-hIgG-SS was injected in mice bearing subcutaneous H82 xenografts to estimate the nonspecific uptake of the radioimmunoconjugate in the DLL3-positive tumor. For animals in the orthotopic xenograft and metastatic model arms of this study, a single time point biodistribution analysis was performed upon euthanizing the animals 120 hours after the injection of ⁸⁹Zr-DFO-SC16-SS. See Supplementary Information for details.

Subcutaneously xenografted H82, H69, Lu64, Lu149, and Lu80 tumors (n = 4 per tumor type) were harvested from athymic nude mice that were not injected with any radioimmunoconjugates. Flash frozen tumor tissue was used for quantitative analyses of DLL3 expression (ng/mg of tissue) via Meso Scale Discovery (MSD) assay. Formalin-fixed, paraffin-embedded tumor tissue sections were used for immunohistochemical staining of DLL3 expression and attributed an H-Score upon analysis.

All animals (n = 16) in the metastatic cohort of this study injected with ⁸⁹Zr-DFO-SC16-SS [7.4 MBq; (200 μCi)] for imaging via PET and bioluminescence were euthanized via CO₂(g) asphyxiation and all vital organs including the head were harvested and analyzed for biodistribution of ⁸⁹Zr-DFO-SC16-SS and histopathology. The read outs from the histopathologic analyses were matched with the PET images to validate their anatomic correlation.

All data presented in this study are expressed as means ± SD. Where applicable, statistical differences were analyzed by unpaired, two-tailed, Student t test using GraphPad Prism 7 software. Multiple comparisons with adjusted P values <0.05 were considered statistically significant. To evaluate the various tumor-to-background tissue ratios between the two SC16-based radioimmunoconjugates, uncertainty propagation errors were used and a correction for multiple comparisons was done using the Holm–Sidak method to determine statistical significance.

To identify genes expressed in SCLC that are minimally expressed in normal tissues, we combined RNAseq data from 51 human SCLC samples and 2,712 normal samples, representing 55 different organs into a single gene expression dataset. Transcripts (27,195) with detectable expression in SCLC were ranked on the basis of the scaled difference between the median SCLC reads per kilobase of transcript per million mapped reads (RPKM) and the upper quartile RPKM of the highest expressing normal tissue. The ranked gene list of 3,915 genes with higher median expression in SCLC than any normal tissue was then filtered on the basis of gene ontology to include genes intrinsic to a membrane and exclude genes with likely intracellular localization. Of the 60 candidate genes meeting these criteria, and confirming previous reports (7), DLL3 had the most favorable gene expression profile, being highly expressed in SCLC with relatively lower levels in pituitary, brain, and testis (Fig. 1A). DLL3 expression in brain tissues was greatest in nucleus accumbens with less expression in hippocampus (Fig. 1B). No single normal sample exceeded the median expression in SCLC.

To identify appropriate cell lines for preclinical testing of a DLL3-targeted immunoPET agent, we turned to publicly available gene expression data from the CCLE. A mining expedition of the gene expression data from cell lines in the CCLE yielded an expression cutoff between cells that did not express DLL3, and the 104 of 1,015 (10.2%) that expressed various levels of DLL3 (Fig. 1C). Expression data for SCLC cell lines was extracted, among which, 32 of 53 (60.4%) expressed DLL3. Two DLL3-expressing cell lines, H82 and H69, were chosen for further analysis (Fig. 1C). H82 had median DLL3 expression and may represent the typical expression level found in human SCLC cell lines. H69 was selected to represent the lower quartile of DLL3-positive SCLC cell lines to challenge the sensitivity limits of DLL3 immunoPET imaging. In vitro cell binding assays using ⁸⁹Zr-DFO-SC16-NS with H82, H69, and A549 cells revealed the binding of the radioimmunoconjugate to bear excellent concordance to the predicted level of DLL3 expressed by the three cell lines (Fig. 1D). Furthermore, a MSD assay yielded 58.2 ng/mg, 14 ng/mg, and 1 ng/mg of DLL3 expressed by H82, H69, and A549 tumors, respectively. Similarly, IHC yielded H-scores of 195, 55, and 0, respectively, for the three tumors (Fig. 1F).

We synthesized two SC16-based immunoconjugates to evaluate as immunoPET tracers for the molecular imaging of DLL3 in preclinical models of SCLC. One of these immunoconjugates, DFO-SC16-NS, was synthesized through the random coupling of isothiocyanate-bearing DFO moieties to the lysines of the antibody, while the other, DFO-SC16-SS, was created via the site-selective conjugation of a maleimide-bearing variant of DFO with unpaired cysteine residues within the hinge region of the SC16 immunoglobulin (Fig. 2A). Mass spectrometry analysis revealed that DFO-SC16-NS had approximately 3.5 DFOs per antibody, with DFOs distributed between the light and heavy chains of the antibody (Supplementary Fig. S1A–S1D). A similar analysis of DFO-SC16-SS revealed that it had 2.8 DFOs conjugated per antibody (Supplementary Fig. S2A–S2D). Both the constructs showed ≥90% immunoreactive fractions as determined via surface plasmon resonance experiments (Supplementary Fig. S3A and S3B).

Upon radiolabeling with ⁸⁹Zr⁴⁺, both ⁸⁹Zr-DFO-SC16-NS and ⁸⁹Zr-DFO-SC16-SS were reproducibly obtained in high radiochemical yields (>95%) and high specific activities (7.5–9.0 mCi/mg). The radioimmunoconjugates were isolated in high radiochemical purity (>99%), and demonstrated ≥95% stability (⁸⁹Zr-DFO-SC16-NS) and 85% stability (⁸⁹Zr-DFO-SC16-SS) against demetallation when incubated in human serum at 37°C over a period of 5 days (Supplementary Figs. S4 and S5). Saturation ligand-binding assays performed on sections prepared from subcutaneously xenografted H82 and H69 tumors validated the nanomolar affinity (K_D) of the ⁸⁹Zr-DFO-SC16-NS construct for binding to DLL3 and yielded B_max concentrations of 14.7 ± 0.6 nmol/L (H82) and 6.0 ± 0.5 nmol/L (H69) to indicate the maximum number of DLL3 sites available per cell for binding of the SC16 antibody in the respective tumors. The nanomolar B_max concentration was converted to numeric values based on a mean SCLC cell diameter of 10 μm, which yielded approximately 14,000 and 6,000 DLL3 molecules available per cell in H82 and H69 tumor sections, respectively (Fig. 1E; ref. 20). A comparison of this range of expression levels of DLL3 with other tumor-associated antigens that have been targeted for the immunoPET imaging of cancer revealed that DLL3 expression in SCLC was lower by at least 2 orders of magnitude (Fig. 1G; refs. 21–30). Finally, although upregulated in SCLC relative to nonmalignant adult cells, DLL3 is a remarkably low-abundance protein on the surface of tumor cells, having on the order of only 10,000 molecules expressed per cell. We hypothesized that immunoPET imaging might still be feasible given the near-complete absence of DLL3 cell surface expression on normal tissues.

In vivo PET imaging studies with the two SC16-based radioimmunoconjugates clearly delineated H82 and H69 tumors in subcutaneous xenografts (Fig. 2B–E). A higher concentration of radioactivity was found in the H82 tumors than in H69 tumors. Conversely, even 120 hours after the injection of the ⁸⁹Zr-DFO-SC16-SS and ⁸⁹Zr-DFO-SC16-NS, a higher PET signal was found to persist in the systemic circulation of H69 xenografts (Fig. 2C and E). The in vivo biodistribution of the radioimmunoconjugates in subcutaneous H82 and H69 xenografts demonstrated a progressively increasing concentration of radioactivity in the tumors over time (Fig. 3A and D and Supplementary Figs. S6–S9). The tumoral uptake values of the radioimmunoconjugates in H82 and H69 tumors corresponded well with the level of DLL3 expression in these tumors as determined via in vitro and ex vivo analyses. Specifically, ⁸⁹Zr-DFO-SC16-SS yielded an uptake value of 27.3 ± 6.0 %ID/g in H82 tumors, whereas the H69 tumors afforded 16.2 ± 5.8 %ID/g. Furthermore, the concentration of radioactivity in both the tumors could be blocked by the coinjection of a 100-fold excess of unlabeled SC16 antibody (Fig. 3A and D; Supplementary Tables S1 and S2).

A comparison between the biodistribution of the site-selectively labeled versus nonsite-selectively labeled SC16-based radioimmunoconjugates in H82 xenografts revealed that ⁸⁹Zr-DFO-SC16-SS yielded a tumor uptake of 27.3 ± 6.0 %ID/g, whereas ⁸⁹Zr-DFO-SC16-NS afforded 19.5 ± 4.8 %ID/g (Fig. 3B; Supplementary Tables S1 and S3). Notably, a higher concentration of radioactivity was observed in the kidneys of H82 xenografts injected with the site-selectively labeled construct (6.2 ± 1.1 %ID/g) than those injected with ⁸⁹Zr-DFO-SC16-NS (3.5 ± 0.5 %ID/g; Fig. 3B; Supplementary Tables S1 and S3). A similar trend was observed for the biodistribution of the SC16-based radioimmunoconjugates in the tumors and background tissues of H69 xenografts (Fig. 3E and Supplementary Fig. S10; Supplementary Tables S2 and S4). Despite the slightly higher activity concentrations of ⁸⁹Zr-DFO-SC16-SS in the tumors and kidneys of H82 xenografts, comparable tumor-to-background ratios were obtained for ⁸⁹Zr-DFO-SC16-SS versus ⁸⁹Zr-DFO-SC16-NS injected in H82 xenografts in most tissues except the liver, which displayed a marginally significant difference (P_adj = 0.076; Fig. 3C; Supplementary Table S5). Furthermore, an evaluation of the ratio of the tumor-to-background ratios between ⁸⁹Zr-DFO-SC16-SS and ⁸⁹Zr-DFO-SC16-NS yielded tumor-to-kidney_(SS/NS) = 0.79 ± 0.31; tumor-to-liver_(SS/NS) = 2.14 ± 1.04; tumor-to-blood_(SS/NS) = 1.13 ± 0.48; and tumor-to-muscle_(SS/NS) = 1.36 ± 0.55. Finally, the biodistribution of ⁸⁹Zr-DFO-SC16-SS in subcutaneous xenografts of a DLL3-negative non–small cell lung cancer (A549) compared well with that of a site selectively–radiolabeled anti-hapten isotype-matched IgG (⁸⁹Zr-DFO-hIgG-SS) in DLL3-positive H82 xenografts. In both these cases, low and nonspecific uptake of the radioimmunoconjugates was observed on the basis of the in vivo enhanced permeability and retention effect (5–7 %ID/g at 120 hours postinjection; Fig. 3F; Supplementary Figs. S11 and S12; Supplementary Tables S6 and S7).

These data are particularly encouraging as the SC16 antibody is known to cross-react with murine DLL3 protein (7). The observed accretion of the SC16 radioimmunoconjugates in SCLC tumors is highly specific and attributed to the tumor-selective expression of human DLL3, and not due to the enhanced permeability and retention (EPR) effect in tumors or the lack of species cross-reactivity that can artificially lower the background signal in immunoPET imaging studies.

In an orthotopic model of SCLC, a high concordance was achieved between the bioluminescence signal from luc-H82 tumors engrafted in the left lung of athymic nude mice and PET imaging with ⁸⁹Zr-DFO-SC16-SS, which successfully delineated the lung tumors (Fig. 4A–C; Supplementary Fig. S13). Analysis of the biodistribution of ⁸⁹Zr-DFO-SC16-SS in this model showed a highly selective concentration of radioactivity within the orthotopically xenografted left lung versus the right lung (Fig. 4D).

In a mouse model developed to mimic distant organs metastases of SCLC, PET/CT imaging with ⁸⁹Zr-DFO-SC16-SS followed by necropsy for biodistribution studies at 136 hours after the injection of the radioimmiunoconjugate revealed a diverse set of findings. Although varying in intensity, the most prominent signal was obtained from the livers in all the mice that were examined in this cohort. Necropsy of the animals revealed the abundance of metastatic nodules to have a strong qualitative concordance with the intensity of the PET signal observed in the liver. Liver tissue with a semiquantitative histopathologic score of 2–3, yielded low intensity PET signal, and had an uptake of approximately 10.7 ± 1.2 %ID/g as measured from biodistribution analysis. On the other hand, livers with a histopathologic score of 5–6 yielded a relatively high intensity PET signal with an uptake of approximately 14 ± 1.5 %ID/g as measured in biodistribution analysis (Fig. 4K). The high intensity foci of PET signal from the liver correlated well with coalescing metastatic lesions found in the liver tissues examined via histopathology. Interestingly, 50% (8/16) of the animals in the metastatic experimental cohort showed unilateral and/or bilateral PET-positive ovaries (Fig. 4E, I, and K). There was a strong correlation between findings from PET imaging, necropsy, biodistribution analysis, and histopathology of the ovarian tissues. No false positives were obtained in our studies despite unilateral involvement of ovarian metastases in 37.5 % (3/8) of cases.

Among other sites of interest for metastases in our experimental cohort, 50% (8/16) of the mice showed distinct PET and bioluminescence signals within the head (Fig. 5A and B). Histopathologic analysis revealed a predominant infiltration of the hemimandibles (Fig. 5C and D) and sporadic infiltration of the pituitary gland (Supplementary Fig. S14). On the other hand, although the bone tissue was positive on PET images (Fig. 4E; Supplementary Fig. S15), the regional localization of the PET signal was somewhat discordant with that of neoplastic cells found in this tissue. Furthermore, a consistently high concentration of radioactivity was seen in the spleen, which appeared strongly PET-positive (Supplementary Fig. S16A and S16C) and was associated with uptake values >85 %ID/g as determined in the biodistribution analysis of all the mice within the metastatic model cohort. However, all the spleens examined were histopathologically negative for the presence of tumor cells (Supplementary Fig. S16B and S16D). Interestingly, tandem injections with the same batch of the radioimmunoconjugate in female athymic nude mice did not yield high intensity splenic PET signals (Supplementary Figs. S17 and S18).

When injected in mice bearing subcutaneously transplanted SCLC PDX, ⁸⁹Zr-DFO-SC16-SS clearly delineated the PDX tumors via PET imaging (Fig. 6A–C). In biodistribution studies, Lu64 tumors, which have the highest reported DLL3 expression (4.25 ng/mg) and sensitivity to treatment with SC16LD6.5 (7), demonstrated the highest tumor uptake of ⁸⁹Zr-DFO-SC16-SS (67.7 ± 13.8 %ID/g) at 120 hours after the injection of the radioimmunoconjugate in the current study (Fig. 6A, D, E, and F). On the other hand, Lu149 tumors, which represented a median concentration for DLL3 expression (2.71 ng/mg), demonstrated an intermediate tumor uptake for ⁸⁹Zr-DFO-SC16-SS (38.1 ± 1.8 %ID/g) at 120 hours after the injection of the radioimmunoconjugate (Fig. 6B, D, E, and F). Finally, Lu80 tumors were representative of a low DLL3 expression candidate (0.6 ng/mg) and afforded the least uptake (21.2 ± 2.4 %ID/g) in the tumors at 120 hours after the injection of the radioimmunoconjugate (Fig. 6C–F). Surprisingly, despite the low level of DLL3 expression in the Lu80 PDX tumors, a relatively high concentration of radioactivity was found associated with these tumors. To some extent, this may be due to perfusion from a sheath of blood that encased the subcutaneously xenografted Lu80 tumors unlike other PDXs examined in this study (Supplementary Fig. S19). Notably, many tumors in the SC16LD6.5-treated cohorts of Lu80 have previously been shown to escape disease control (7).

Early-phase clinical trials of Rova-T have assessed DLL3 expression from initial diagnostic biopsies by IHC. Although IHC appears to be an effective method to assess DLL3 expression status, it suffers from several practical limitations that may hinder its utility in the clinic, chief among them being the requirement for invasive research biopsies in previously treated SCLC, which may recur in inaccessible sites and for which clinical justification for rebiopsy is generally lacking.

Premised on the early success of Rova-T and the emerging role of precision medicine to improve clinical outcomes, we envisaged that a companion imaging diagnostic for this drug could potentially facilitate the identification, selection, and stratification of SCLC patients for treatment with Rova-T on the basis of their in vivo DLL3 expression status and serve as a tool with which to monitor for recurrence or determine whether Rova-T retreatment should be an option for patients upon recurrence. A DLL3-targeted molecular imaging agent could potentially provide clinicians with real-time information about the in vivo expression of DLL3 while concomitantly delineating SCLC tumors in a noninvasive whole-body PET scan.

The foremost challenge to image DLL3 expression via PET lay in its low abundance on the surface of tumor cells. Syndecan-1 and HER3 are among the few other tumor-associated antigens with similar expression densities that have been targeted for the PET imaging of cancer (24, 28, 29). In our hands, although the in vivo tumor uptake of the site-selectively labeled SC16 radioimmunoconjugate (⁸⁹Zr-DFO-SC16-SS) was not significantly different from its nonsite-selectively labeled cousin (⁸⁹Zr-DFO-SC16-NS), ⁸⁹Zr-DFO-SC16-SS represented a biochemically site-selective and functionally active radioimmunoconjugate that yielded favorable in vivo tumor-to-background ratios that might be further improved.

Among the chosen SCLC cell line tumors, the relatively low target sink provided by the H69 tumors resulted in a higher concentration of radioactivity to persist within the systemic circulation of H69 tumor-bearing mice even 120 hours after the injection of the radioimmunoconjugates. This observation may be critical to the dosing of patients with the Rova-T ADC and/or potential targeted-radiotherapy with ¹⁷⁷Lu/⁹⁰Y-labeled SC16 antibody, wherein patients with lesser tumor burden or lower in vivo tumoral expression of DLL3 are more likely to have an excess of the “targeted” drug lingering within the systemic circulation. In such patients, adjusted dosing of the DLL3-targeted ADC or radioimmunotherapeutic agent may be warranted to avoid treatment-related toxicities.

Finally, a critical benchmark that defines the clinical utility of immunoPET tracers lies in their ability to predict response to therapy based on in vivo imaging. In this context, ⁸⁹Zr-DFO-SC16-SS demonstrated rank order correlation in clinically representative PDX models of SCLC with previously documented responses to therapy with SC16LD6.5. Furthermore, the ability of ⁸⁹Zr-DFO-SC16-SS to noninvasively delineate micrometastatic foci with high sensitivity may potentially translate into the clinical detection of distant organ micrometastatic lesions in patients. Metastases to the brain are a common finding in SCLC patients and cannot be detected via ¹⁸F-FDG-PET due to the background caused by high uptake of glucose in the brain. The PET imaging performance of ⁸⁹Zr-labeled trastuzumab in the clinic encourages us to speculate that DLL3 immunoPET could potentially enable the identification of metastatic lesions in the brain of SCLC patients (31). Furthermore, our findings of the multinodular infiltration pattern in the ovaries of mice within the metastatic cohort were consistent with clinical reports of metastatic SCLC found in the ovaries of human patients (32). Sparing of the ovarian surface epithelium and bursa in all the cases that were examined is indicative of the hematogenous spread of the disease to the ovary. Although metastases to the mandible and alveolar bone is seldom observed in human patients, in our preclinical model the premolar–molar region may be providing an ideal site for metastatic seeding due to a predominance of red bone marrow and retardation of the circulation in this anatomic location.

Despite the excellent in vivo performance of ⁸⁹Zr-DFO-SC16-SS as a DLL3 PET imaging agent, our experiments highlighted some limitations of the immunoPET agent and the preclinical models that were used for its evaluation. Most importantly, the chelator-to-antibody ratio (CAR = 1–4) for the site-selectively labeled ⁸⁹Zr-DFO-SC16-SS construct used for PET imaging experiments differs from the drug-to-antibody ratio (DAR = 2) for the Rova-T construct used in clinical trials. This may be a result of a stronger reduction of the inter-chain disulfide bonds in the SC16 antibody with a 10-fold molar excess of TCEP used for the synthesis of ⁸⁹Zr-DFO-SC16-SS compared with the 1–2 fold molar excess of the mild-reducing agent used in the synthesis of SC16LD6.5 (7). In our case, a high CAR was desirable to yield a high specific activity immunoPET imaging agent that retains target binding capability and can delineate the expression of a low abundance tumor specific target such as DLL3.

Arguably, the elevated concentration of radioactivity in the kidneys of xenografts injected with ⁸⁹Zr-DFO-SC16-SS may be attributed to the labile nature of the thiol-maleimide linkage for in vivo hydrolysis. This is exacerbated by the tendency of the thioether bond to undergo retro-Michael addition reactions in the presence of reactive thiols of serum proteins such as albumin and/or free cysteines and glutathione present in the kidneys (33). Through this mechanism, the macrocylic chelator and the ⁸⁹Zr encased within it could be dissociated from the antibody and be systemically cleared through the kidneys. More stable thiol-reactive methylsufonyl-functionalized heteroaromatic linkers may improve the stability of the immunoconjugate (34). Another limitation revealed by immunoPET imaging with ⁸⁹Zr-DFO-SC16-SS in the metastatic SCLC model was the PET signal observed in the bone and the spleen, which had discordant histopathologic read outs for the presence of metastases in these tissues. These observations may most likely be related to the biology of the immunocompromised NSG mouse strain.

In conclusion, the work at hand describes the thorough preclinical evaluation of a DLL3-targeted companion diagnostic agent for potential use in the clinic. Such a tracer can noninvasively provide real-time information about the status of in vivo DLL3 expression and delineate the presence of SCLC tumors to facilitate the selection of patients for treatment with a therapeutic counterpart such as Rova-T and other DLL3-targeted therapeutics. In addition, several salient features related to the development of immunoPET tracers and preclinical animal models were highlighted through the experiments undertaken in this study. First, our study demonstrates that low abundance tumor-associated molecular targets/antigens such as DLL3 can be targeted for PET imaging, provided the target is unique to neoplastic cells and its in vivo expression in background tissues is minimal. Next, our study illustrates a comparative evaluation of using site-selectively modified immunoconjugates over their nonsite-selectively modified cousins for applications related to the targeted imaging and therapy of cancer. In this context, the critical role and limitations of the chemistry used in the preparation of such immunoconjugates was highlighted. Finally, the DLL3 PET imaging agent ⁸⁹Zr-DFO-SC16-SS consistently demonstrated highly sensitive and selective targeting of DLL3 to delineate SCLC tumors in vivo in a host of preclinical mouse models.

A.J. Bankovich is a senior director at Abbvie Stemcentrx LLC and has ownership interest (including patents) in Abbvie. S. Bheddah is a senior scientist III at Abbvie Stemcentrx. J. Sandoval has ownership interest (including patents) in AbbVie Stemcentrx LLC. C.M. Rudin is a consultant/advisory board member for AstraZeneca, Celgene, Abbvie, Harpoon, G1 Therapeutics, and Bristol-Myers Squibb. S.J. Dylla is the vice president of R&D and a chief scientific officer at AbbVie Stemcentrx, LLC and has ownership interest (including patents) in AbbVie Stemcentrx, LLC. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.K. Sharma, A.J. Bankovich, B.M. Zeglis, S.J. Dylla, J.T. Poirier, J.S. Lewis

Development of methodology: S.K. Sharma, S.D. Carlin, B.M. Zeglis, J.T. Poirier, J.S. Lewis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.K. Sharma, J. Pourat, S.D. Carlin, A.J. Bankovich, E.E. Gardner, K. Isse, J. Sandoval, D. Liu, J.S. Lewis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.K. Sharma, S.D. Carlin, A. Piersigilli, A.J. Bankovich, O. Hamdy, K. Isse, S. Bheddah, J. Sandoval, K.M. Cunanan, V. Sisodiya, S.J. Dylla, J.T. Poirier, J.S. Lewis

Writing, review, and/or revision of the manuscript: S.K. Sharma, A.J. Bankovich, O. Hamdy, K. Isse, K.M. Cunanan, V. Sisodiya, D. Liu, B.M. Zeglis, C.M. Rudin, S.J. Dylla, J.T. Poirier, J.S. Lewis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.K. Sharma, D. Abdel-Atti, E.E. Gardner, E.B. Johansen, V. Allaj, J.S. Lewis

Study supervision: J.T. Poirier, J.S. Lewis

This work was supported by a grant from the Druckenmiller Center for Lung Cancer Research to J.T. Poirier, J.S. Lewis and NIH grants U01 CA213359 to J.T. Poirier, J.S. Lewis and R01 CA213448 to J.T. Poirier, J.S. Lewis, C.M. Rudin. This work was made possible by using the MSKCC Small Animal Imaging Core Facility, the Radiochemistry and Molecular Imaging Probe core, the Anti-tumor Assessment Core, and the Tri-Institutional Laboratory of Comparative Pathology, which were supported in part by NIH grant P30 CA08748.

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.